• All across the world, there is a negative correlation between IQ and fertility.

• This means that populations are getting less intelligent as time goes on. Through some basic assumptions, across 62 nations for which we have data, one can estimate a decline in IQ per decade that ranges from as low as 0.01 points in Estonia and Switzerland to as high as 0.65 in Panama, Romania, and North Macedonia.

• In general, more developed countries see a weaker decline in intelligence than less developed nations.

• Selection for lower IQ does not appear to be a major long-term problem in the richest countries today. If at some point even 20% of the population uses embryo selection, it would be more than enough to reverse the current trends that we are seeing.

• There may be more reasons to be concerned about developing countries. The fact that the rate of decline is so fast implies that even if IQ differences between nations are completely environmentally determined today, over the coming decades there may still be a significant divergence between them. That said, the data here may exaggerate the scale of the problem if large family size depresses phenotypic IQ in the developing world.

Introduction

In the 2006 film Idiocracy, a young American soldier of average intelligence wakes up 500 years after a botched hibernation experiment and finds himself to be the smartest man in the world. A narrator explains that over the years, humanity got progressively dumber as the less intelligent outbred smarter individuals in one generation after another.

While a work of fiction, it is reasonable to ask whether the dystopia described in the film gets at an issue with real world significance. There is certainly a stereotype that more intelligent people have fewer children, and as we know, stereotypes generally have a basis in fact. Whether this one is true is certainly worth looking into, given how important intelligence is for human flourishing. Moreover, if we are concerned with humanity as a whole, it is important to analyze differential fertility by intelligence not only in United States and other Western countries, but across the world.

Along with some co-authors, I have gathered relevant data on this question from 62 countries. In general, our research shows that humans are likely getting dumber, with the problem being much more pronounced in the developing world. If the data is correct and reflects a decline in genotypic intelligence, it may cause problems for poor states trying to catch up to first world nations, and ultimately threaten their economic health and political stability.

Past Work on IQ and Fertility

Scholars have for generations argued that the fears expressed in works like Idiocracy are supported by empirical data. Early research on the relationship between fertility and intelligence found that the variables were negatively correlated, albeit weakly.1 Subsequent work vindicates these early studies, with a recent meta-analysis finding much the same thing.2 Despite this previous research, there is very little agreement on the magnitude of the decline in IQ per decade, with estimates ranging from 0.14 to 0.64 points.3 In addition, most of the data on fertility and intelligence comes from the USA and Britain, and few studies have been done on other regions.

Other critics of the differential fertility literature argue that the Flynn effect, the global increase in IQ scores of about 2-3 points per decade,4 is evidence against differential fertility resulting in decreases in intelligence in future generations. However, IQ tests do not measure the same construct across time,5 and a portion of Flynn effect increases can be attributed to increases in guessing.6 Besides these issues, genetic predisposition to educational attainment fell in Iceland by about 0.1 standard units, the equivalent of 1.5 IQ points, in the 20th century, confirming the hypothesis that differential fertility leads to changes in a population’s genetic makeup.7

Other researchers have estimated that average levels of intelligence in the West have fallen by the equivalent of 13 IQ points since the Victorian era based on declines in reaction time.8 This is much larger than what could be reasonably inferred based on differential fertility. Instead, the primary cause of this decline is hypothesized to be mutational load. However, this study suffers from certain methodological issues, notably that some of the early samples were not representative of the general population and that only 14 samples were available.9 Woodley et al. (2013) responded to some of these critiques by arguing that the difference in reaction time between 19th century and modern students is similar to the overall decline in reaction time observed from their samples.

Fortunately, there is a better study on this topic that compares secular trends in the number of digits people can remember (the ‘digit span’ test), depending on whether they must be repeated forwards or backwards.10 The authors involved gathered over 1,700 effect sizes in total and controlled for the effect of moderators, so this study does not share the same concerns as the previous paper mentioned. It found that performance increased for forward digit span tasks, but decreased for backwards digit span tasks in recent years. The latter correlates more with IQ because it is a more cognitively complex task, so this is suggestive evidence that general intelligence has fallen with time. This also holds true for corsi tasks, where the test taker is asked to tap a given sequence of blocks forwards or backwards. Using the summary statistics from the previous study, I calculated that, based on the change in backward digit span, the magnitude of the decline in general intelligence is about 2 points per century.

Moreover, a decline in intelligence should result in falls in quantitative and verbal abilities of students if the quality of education has stayed the same over time. Fortunately, the National Assessment of Educational Progress has tracked the achievement of American students in math and reading. Over time, Americans have hardly changed in their reading or mathematical abilities. In 1970, 13-year-old students scored 255 out of 500 in reading and 266 out of 500 in mathematics, and in 2023 they scored 256 in reading and 271 in mathematics. This is confounded a bit by immigration, but one might suspect that if anything that makes the arguments against a drop in IQ among Americans who have lived in the country for generations even stronger, given that most new residents over the past several decades have been low-skilled.

If mutational load had such a large effect on IQ, then we might suspect that there would be stronger evidence for its existence by now. However, trends in mathematical performance, reading comprehension, and backwards digit span suggest that the average IQ in the United States has not fallen by a large amount. Sure, these tests do not measure intelligence directly, but if intelligence really was dropping, then performance on at least some of them would be expected to fall as well.

There is a lot of uncertainty in the literature regarding the exact effect of mutational load on the decline in IQ per decade (95% CI: [0.14, 1.5]), based on statistics taken from Woodley (2015). Also, recent research on mutational load finds an effect that is severely attenuated and not statistically significant after controlling for birth order.11 Based on theory, mutational load should result in decreases in IQ over time, but the magnitude of the effect is difficult to measure.

New Cross-National Data

To add to the literature on this topic, I collaborated with Bryan Pesta, George Francis, and Emil Kirkegaard on a study that assessed the relationship between intelligence and fertility in as many countries as possible, and projected the change in the average IQ of each nation and the entire world based on the available data. We were able to gather 156 effect sizes (n = 419,444) from a combination of the academic literature and public data, in particular the datasets of the National Longitudinal Study of Youth (NLSY), General Social Survey (GSS), Programme for International Student Assessment (PISA), Progress in International Reading Literacy Study (PIRLS), and Program for International Assessment of Adult Competencies (PIAAC). Not all of these sources directly test IQ  — proxies such as performance in mathematics or reading had to be used in most cases. In some sources, such as the PISA and PIRLS datasets, the IQ of the children was tested instead of the IQ of the adults.

Unfortunately, only 62 countries had data on the relationship between intelligence and fertility, so imputations had to be made for the others. These imputations were estimated using the correlation between educational attainment and fertility, with, depending on the country, national IQ, socioeconomic development, or geographic estimates. For the sake of transparency, all of the figures in this article use data from countries that tested intelligence or a reasonable proxy for it (e.g., mathematical ability, reading comprehension). These projections of the average IQ of each country were then combined with the United Nations’ population projections to estimate the change in the world’s IQ from 2023 to 2100. Note that this analysis does not take into account the possibility that IQ can improve due to socioeconomic development or cultural changes. This may matter in developing nations, but is much less likely to in wealthier countries.

To measure the relationship between intelligence and fertility, we used a statistic called the selection differential: the difference between a reference population and the parents of that population weighted by the number of children each parent has. Then, this number is multiplied by the heritability of IQ to generate a different statistic called the response to selection, which is the expected genetic decline in a trait in one generation. So if the selection differential for IQ is 2 points, and the additive heritability of IQ is 80%, then the response to selection is 1.6, which is the expected genetic decline in that trait in the next generation. As the correlation between the average IQ of two parents and their child’s IQ is about 0.6,12 selection differentials found were shrunk by a factor of 0.6 to estimate the expected decline in genetic intelligence within the next generations. To calculate the projected decline in genotypic IQ per decade in each country, these selection differentials were divided by the average maternal age within each country to calculate the decline per year, and then that decline was multiplied by 10.

This may be an underestimate or overestimate of the true future decline depending on several factors, notably mutational load, environmental shifts, measurement error, and class based differences in generational turnover. Unfortunately, these variable cannot be modeled given the information that we have, so they have not been taken into consideration when projecting the future IQs of different countries.

While I was collecting the data, I noticed that the selection differentials for IQ varied considerably depending on the source. For example, the selection differential for IQ in the USA is about -2.4 based in the NLSY datasets, which were considered the best sources for that country due to the fact IQ was measured very accurately, they are large samples, most individuals were close to completed fertility (ages of 51-64 in NLSY79, 35-38 in NLSY97), and the samples are roughly representative of the nation as a whole. In comparison, the selection differential in the international datasets is -1.2.

It is unclear why this is the case. Perhaps the tests used are of lower quality, or the United States was not sampled in a representative fashion. The NLSY was relied on for the “regional” study of the US, meaning that it did not apply to other countries. It uses a very exhaustive measure of cognitive ability, the ASVAB, while the international datasets only have rough proxies for cognitive ability, which are comprised of 3 subtests in the case of the PISA dataset, one subtest in the case of the PIRLS, and 2-3 subtests in the case of the PIAAC dataset. In addition, the PIAAC dataset had uniquely high estimates of the selection differentials, with the average country having a selection differential for IQ 1.07 points higher than what would be normally expected. The differences in the magnitude of the selection differentials between datasets hold up well between countries, so this effect is not an artefact of different countries being present in each dataset.

To correct for the bias introduced by the international (PISA, PIRLS, PIAAC) datasets, two different methods were considered. One assumed that the bias is multiplicative, that is, the international datasets shrink effect sizes by a factor of 0.5; or that the effect is additive, and the international datasets reduce the selection differentials for IQ by 1.2 points. Because it’s not clear which method is better, the results of both were averaged into one final estimation. Besides this adjustment, the effect of specific datasets was also considered, so the effects of individual datasets wouldn’t bias the estimation of the decline between countries.

If the reason why the international datasets have attenuated effect sizes within the United States is sampling error, correcting for this bias will potentially lead to deflated selection differentials within smaller countries, which are easier to sample. The reasoning is as follows: if sampling error is leading to attenuated effect sizes within the international datasets, then correcting for the effect observed within a large country like the United States will overcorrect for the effect of sampling error within other countries that are not as large.

Based on our data, we found that the average country is expected to decline in IQ by 0.35 points per decade, while the whole world is expected to decline by 1.1 points per decade during this century. The reason that the global decline is larger than the decline within the individual countries is that lower IQ nations are expected have higher growth rates. Fortunately, the rate at which global intelligence is falling is slowing down; by the end of the century the decline will only be 0.8 points per decade.

Countries also vary considerably in differential fertility for intelligence. In Iceland, Finland, Denmark, Estonia, and Switzerland, fertility and intelligence are negligibly related, while Latin American and certain Middle Eastern countries have a stronger negative relationship. Part of this is because the latter countries have younger mothers, so selection for intelligence occurs faster independent of the relationship between fertility and intelligence within a generation. But this doesn’t explain the whole relationship, as countries only vary so much in their average maternal ages; the number ranges from 25.7 to 33.4, while the decline in IQ per decade varies much more, from 0.01 in Switzerland to 0.65 in Panama.

Selection differentials for intelligence correlated with both national IQ (r = 0.51) and socioeconomic development (s-factor) (r = 0.48). As s-factor and national IQ are highly correlated (r = 0.8 in this dataset), it is difficult to tell which of these variables is more likely to be driving the relationship.

Prior to conducting the study, I had hypothesized that socioeconomic development past industrialization had a positive effect on selection differentials for IQ, because developed countries have more resources, so a slower life history. In addition, wealthier nations have more variance in years spent in the education system,13 so if education has a negative effect on fertility, the relationship between the two variables should be stronger in developing countries because of lower restriction of range. According to our data, socioeconomic development and the correlation between fertility and education are correlated between countries (r = .27, p < .01), though the relationship is weak and potentially spurious.

Another reason that less developed countries have more negative relationships between intelligence and fertility is that, at the individual level, early births tend to be disproportionately from less intelligent individuals, while later births follow the opposite pattern.14 As developed countries tend to have higher average ages of births (r = .37, p < .001), differential fertility for IQ and socioeconomic development should positively correlate.

The cause of the relationship between national IQ and the selection differential for IQ is opaque — it could be because the average IQ of a nation also measures its social development, or because there is some causal relationship between the two variables.

Alternatively, the relationship between socioeconomic development and the selection differential could be an artefact that arises out of using the IQs of children to estimate the selection differential instead of the parents. If family size has a more negative causal effect on the IQs of children in the developing world, then a positive relationship between the estimated selection differentials and social development will be found, but only because the causality occurs in the opposite direction — that is, larger families cause lower IQs, rather than the other way around. Because the heritability of fertility is very low,15 this environmental effect won’t be transmitted in any large degree to future generations, so the decline in IQ of these nations may be overestimated.

Implications and Conclusion

My main takeaway from the study is probably going to surprise most readers: future reductions in intelligence due to differential fertility for IQ are not going to be a major issue in the developed world. This is because the process is happening too slowly for it to have a significant effect; by 2123, the most concerning drop in the developing world is the US falling from an IQ of 97.5 to 93.7. Most European countries will have hardly budged by 2123, besides a few in the Balkans: Romania, Macedonia, and Albania are projected to drop from 87 to 80.5, 80 to 73.4, and 82.1 to 77.5 respectively. All of this of course assumes that whatever rates of differential fertility by IQ exist now will continue into future generations, which will be true to a greater or lesser extent depending on the country.

Besides the fact that the effect is weak within most developed nations, embryo selection for intelligence, and potentially other technologies, can attenuate secular decreases in intelligence, though I doubt the practice will be adopted by most of the general population. The average member of the public estimates their likelihood of using embryo selection to boost intelligence at 35%, with 30% morally opposed to the practice.16 Data from Pew Research (2018) suggests that favorability among the public may be even lower, with only 19% of Americans saying that selecting embryos for intelligence is an appropriate use of medical technology, while 80% think that it takes science too far. Embryo selection is also very expensive at the moment, with clinics requesting between $4,000 to $20,000 for the service, so some of the population will be priced out.17 Nonetheless, it is historically common for a practice to be rejected by the public at first and then have opinion towards it shift as its availability increases and its benefits become clear. Therefore, polls should be seen as reflecting a lower bound estimate for how many people might use a technology.

Assuming that 20% of Westerners in the future select their embryos for intelligence, half of those 20% only select for intelligence, and 20% of the causal variance in cognitive ability can be estimated with genes, then embryo selection will boost the average IQ of the population by approximately 0.73 points per generation (math done with gwern’s help). This is roughly equivalent to an increase of 0.24 IQ points per decade, which is enough for most developed countries to maintain their current levels of intelligence, assuming that the relationship between intelligence and fertility remains the same. Theoretically, genetic predictors should be able to explain up to 50% of the variance in intelligence, so this may be an underestimate of the impact of embryo selection.

Some commentators like Anatoly Karlin argue that concerns about national IQ will not be as relevant once strong artificial intelligence has been developed, as lots of white collar work will then be automatable. This is true, but there are other reasons why national IQ would continue to be useful in the future; barring rapid technological breakthroughs, manual labor will still need to be done by humans, and job performance and IQ correlate at about 0.4, even within categories like rifleman (r = .59) and military policeman (r = .46).18

Across individuals, low IQ is a predictor of marital instability,19 obesity,20 and criminality,21 and the relationships are likely to be causal as they cannot be ascribed to confounders. This means that declines in national intelligence will lead to lower levels of wealth and social peace, and potentially civilizational collapse if the fall is too severe.

What is most concerning is the strong negative relationship between fertility and intelligence that is observed in some developing countries, like Turkey and Brazil. By the end of this century, these countries will have dropped by 5.5 and 6.4 points respectively. The effectiveness of embryo selection for intelligence drops with genetic distance from the reference population.22 This technique may therefore be effective for populations that are genetically close to Europeans like white Brazilians or Turks, but less so for people from Central America or Southeast Asia. Even worse, countries that are less developed are less likely to adopt embryo selection and similar practices in the first place, and will need to use the technology more to sustain their current levels of intelligence.

Figures for changes in IQ within nations and across the world should not be taken as exact estimates. They depend on many assumptions: for example, that the rate of differential selection by IQ doesn’t shift over time, the heritability of IQ stays constant, and nations do not see a rise or fall in IQ anywhere due to cultural or technological changes unrelated to patterns of fertility. It is certain that not all of these assumptions will hold for all countries indefinitely into the future. Nonetheless, given the importance of intelligence for accomplishing practically all human goals, it is worth undertaking back-of-the-envelope calculations to give us some idea of how worried we should be. The exercise is enough to determine that selection for low IQ is a minor problem in wealthier states, and potentially much more concerning in developing countries.

On the morning of February 24, 2022, as Russian troops marched into Ukraine, the world woke up to find out that it had entered a new era. By ordering the invasion of Ukraine, Vladimir Putin did not just destroy the lives of countless people in Ukraine and Russia, but also shattered what little was left of the hopes born at the end of the Cold War. Between 1989 and 1991, the Berlin Wall had fallen and the Soviet empire in Central and Eastern Europe had disappeared almost overnight, while the Soviet Union itself had followed suit shortly afterward. Almost everywhere in the Eastern bloc, communist leaders were replaced by democrats, who organized competitive elections and launched reforms to transition to a market economy. Even before the Soviet Union disintegrated, the communist leadership had started to democratize their country and had effectively put an end to the Cold War through negotiations with the US and its allies, resulting in vast reductions of armament both nuclear and conventional. For the first time in decades, people did not have to fear a war between NATO and the Warsaw Pact, which unless it was stopped in time could have led to a nuclear exchange and the destruction of modern civilization in both Europe and North America. The future seemed bright and people looked ahead to a new era of peace, democracy and cooperation. Unfortunately, thirty years later, we know that it wasn’t meant to be. This vision of a bright future did not unravel all of a sudden but slowly over many years and the invasion of Ukraine by Russia in 2022, which kicked off the largest war in Europe since WWII, was merely the conclusion of a long process.

Why did this happen? According to the prevailing narrative, the answer to that question is very simple. Unlike Germany or Japan, Russia never truly rejected imperialism and, though after 1991 it was temporarily weakened, it continued to aspire to dominate its neighbors. Out of naivety, and despite being told this repeatedly by Central and Eastern Europeans, the West refused to act forcefully against Moscow’s imperialist tendencies. In particular, after Russia invaded Georgia in 2008, the West did not give up the hope that it could reach some kind of accommodation with Moscow and refused to break off relations with it. Neither the US nor the EU took sanctions against Russia in response to the invasion. Instead, the US launched the so-called “reset” in 2009 to improve relations with Russia, which led to various political and military agreements. Meanwhile, Germany and several other Western European countries continued to build the Nord Stream pipeline under the Baltic Sea in collaboration with Russia, which allowed them to secure a supply of Russian natural gas that bypassed Ukraine and therefore weakened it by reducing its leverage over Russia. Even after Russia annexed Crimea in 2014, in violation of several international treaties it had ratified, the West reacted weakly and only adopted limited sanctions. This emboldened Putin and, when Ukraine refused to obey his diktat by adopting a federal constitution and granting autonomy to Donetsk and Luhansk (where Russia was covertly supporting the separatists against the Ukrainian armed forces), convinced him that he could get away with a full-blown invasion.

In a series of essays, I will argue that although this narrative has now become largely uncontroversial, it’s extremely simplistic and in some respects even gets things backwards. That is not to say, of course, that Russia is not imperialist and that this didn’t play a role in the sequence of events that resulted in the invasion of Ukraine. But there is a lot more to this story and, as I will argue, the deterioration of relations between the West and Russia that made the invasion possible in the first place would not have occurred if the West, not just Russia, had not made several mistakes after the end of the Cold War that needlessly aggravated Moscow. Nor does it mean that, as some people think, the US and its allies deliberately provoked Russia into invading Ukraine. Rather, bad luck and policy mistakes not only in Russia but also in the West and Ukraine conspired to create a perfect storm, which led to the events of 2014 and ultimately to the invasion of Ukraine in 2022. Nobody forced Russia to invade Ukraine, but historical events always have several causes and there is no such thing as a law of conservation of blame, so the fact that Russia bears the bulk of the responsibility for the war doesn’t mean that nobody else is to blame for it. As we shall see, there is an element of Greek tragedy in what happened, as nobody intended their actions to result in that outcome, but that is not to say that it was inevitable. I don’t think it was, but to understand how it could have been avoided, one has to understand how it came about.

The problem is that, in order to understand what happened and how it could perhaps have been avoided, one has to go back all the way to the end of the Cold War and carefully pore over more than thirty years of history, for without this context it’s simply impossible to understand more recent events and one is bound to misinterpret them. This makes it difficult for people who think it’s more complicated than generally understood to make their case effectively, because they have to challenge assumptions that are now deeply entrenched, and this requires launching into long historical explanations that vastly exceed most people’s attention spans. By contrast, proponents of the prevailing view have a narrative that is not only superficially compelling and consistent with people’s prejudices (since it fits the handful of facts they know or think they know about very well), but can be presented very succinctly. As historian Mariana Budjeryn wrote on a related issue, “complicated histories are unpopular, as they do not readily translate into political slogans”.1 Anyone who aspires to challenge this narrative has therefore his work cut out for him, but I shall nevertheless endeavor to do so and ask that you keep an open mind as I present my interpretation of the events I discuss in this work.

In order to challenge the prevailing view, it’s not enough to look at the facts;, the historical analysis has to be informed by clear thinking about morality and foreign policy. As Paul D’Anieri wrote after identifying some normative questions one has to answer to say how the events of 2014 could have been averted:

These are normative questions whose answers depend on further assumptions about the rights of great powers, the inviolability of sovereignty and international law, the boundaries of realpolitik, and so on. How one answers those questions will determine whose claim one believes has greater weight, who should therefore have backed off, and who, in the final analysis, is guilty of not backing off and therefore to blame for the conflict. Even in February 2014, violence could have been avoided as long as each side refrained from shifting to violence. Whether that move to violence should be blamed on protesters in Kyiv, on Yanukovych, or on Russia also falls back on normative assumptions. Thus, rather than history or analysis resolving who is to blame, how one assigns blame tends to shape how one writes or reads the analysis.2

Indeed, it’s not just the historical analysis embedded in the dominant narrative that is simplistic, but also the normative assumptions on which it implicitly rests. As we shall see, once those assumptions are made explicit and critically examined, many common arguments on what happened and who is to blame for it fall apart. I hope that, at the very least, I will convince even the skeptical reader that things are more complicated than they seem. Below you can find the first essay of the series in which I explain why the hopes of 1989 have not been fulfilled and instead Europe is now engulfed in the largest war on the continent since WWII.

Summary

Did the US and its allies promise the Soviet Union that there would be no NATO expansion at the end of the Cold War? This question has caused considerable controversy, as successive Russian leaders have made this claim over the past 30 years and accused the West of breaking their pledge by expanding NATO to Central and Eastern Europe.

As revolutions swept through Central and Eastern Europe and the Cold War drew to a close in 1989, German reunification was the most pressing issue to settle the conflict. But it required Moscow’s approval due to legal rights inherited from the end of WWII and the presence of a massive contingent of Soviet troops in the German Democratic Republic (GDR).

The central question was whether united Germany would be allowed to stay in NATO after reunification or whether it would have to leave the Alliance and adopt some kind of neutral status. Indeed, with the end of the Cold War, it wasn’t even clear whether NATO should continue to exist.

Most scholars have sided against Russia in arguing that no such pledge had been made during the negotiations on German reunification and more generally that the West didn’t have any obligation not to expand NATO as a result of the commitments made at the end of the Cold War.

The debate has focused on statements made by US and West German officials during preliminary talks held on the issue in February 1990. While everybody agrees that on this occasion they pledged not to expand NATO to the east if Germany was allowed to stay in the Alliance, people disagree about what they meant and what implications those exchanges had.

Critics of the Russian position argue that they were only talking about the territory of the GDR, that Gorbachev didn’t take even this limited no-expansion deal and that it was subsequently retracted anyway. Defenders of the Russian position argue that US and West German officials were talking about Central and Eastern Europe as a whole and that some kind of implicit deal ruling out NATO expansion to that area was in fact struck.

I argue that although they often overstate it, the Russians nevertheless have a strong case for a weak version of their claim, but that it’s not for the case defenders of their position typically make. Conversely, critics of the Russian position are right that US and West German officials were only talking about the GDR (with one important exception about which they misrepresent the evidence), but it’s not for the reasons they claim.

Moreover, they mistakenly conclude that Western officials didn’t make assurances that ruled out NATO expansion because, like defenders of the Russian position, they misconstrue the dynamic of the negotiations on German reunification and focus on the preliminary talks held in February 1990 at the expense of the rest of the negotiations.

I argue that Western officials never proposed a quid pro quo to their Soviet counterparts, at least not the quid pro quo that both critics and defenders of the Russian position have suggested. Gorbachev was not the naive negotiator usually portrayed, but was severely constrained by the fact that achieving his main policy goals required maintaining a cooperative stance with the West.

This led him to consent to German reunification in NATO, but he also accepted because he was assured repeatedly by Western officials that it would be followed by the creation of an inclusive post-Cold War European security order. I argue that NATO expansion, which instead created a NATO-centric security architecture that excluded Moscow, was a violation of those assurances.

The most significant aspect of this controversy, however, is not so much whether the US and its allies violated a pledge not to expand NATO made at the time, but the decision by the Bush administration to preserve NATO’s primacy in the post-Cold War era instead of pursuing a pan-European security agenda.

I conclude by reflecting on this road not taken at the end of the Cold War and argue that the Bush administration’s decision not to follow it, which I explain was made for both good and bad reasons, made the subsequent deterioration of relations between Russia and the West and a conflict between Ukraine and Russia, while by no means inevitable, much more likely.

In 1985, after the death of Konstantin Chernenko, Mikhail Gorbachev was elected General Secretary of the Communist Party of the Soviet Union (CPSU). In the years that followed, he launched a series of wide-ranging reforms that undermined the CPSU's monopoly on power, ended the Cold War and eventually led to the dissolution of the Soviet Union. Despite widespread misconceptions, Gorbachev decided to end the Cold War not because he had no choice (while the Soviet Union was already in crisis at the time it still had the means to continue to wage the Cold War), but rather because he came to reject the ideological underpinning of the East-West confrontation. Paradoxically, he was able to succeed in ending the Cold War because he found a willing and enthusiastic partner in Ronald Reagan, who then as now was considered the arch-Cold Warrior. Reagan came to power with the conviction that something had to be done to reduce the threat posed by nuclear weapons, but he was critical of the détente because he thought that it had disproportionately benefited the Soviet Union, so he adopted a policy that combined firmness to force the Soviets to the negotiating table with a genuine willingness to make substantial reductions in nuclear weapons once they did. Although initially skeptical, Reagan came to trust Gorbachev after meeting him and became convinced that he was serious about reforming the Soviet Union and democratizing it, leading to a fruitful cooperation and a transformation of US-Soviet relations that ended the Cold War.3

In 1987, the two leaders signed the Intermediate-Range Nuclear Force Treaty (INF Treaty), which closed the Euromissile Crisis started by the Soviet Union's decision to deploy SS-20 missiles targeting Western Europe at the end of the 1970s and was the first arms reduction agreement to eliminate a whole category of weapons. They also resumed the negotiations on the Strategic Arms Reduction Treaty (START Treaty), which had begun in 1982 but had stalled after Reagan announced the Strategic Defense Initiative in 1983, although it would not be signed until 1991 under George H. Bush. Moreover, Gorbachev started to reduce the number of troops that were stationed in Warsaw Pact countries, although hundreds of thousands remained. In December 1988, he made a speech at the UN in which he rejected the use of force in foreign policy and declared that Central and Eastern Europeans should be free to decide what kind of political and economic system they wished to live under, which amounted to a rejection of the Brezhnev Doctrine and made people in the West realize that he was serious about reforms.4 A few months later, his intentions were put to the test when Central and Eastern European countries moved toward liberalization and started the process that would result in the collapse of socialism in the Eastern bloc. But Gorbachev stood firm and, in a speech he gave at the Council of Europe in July, repeated his pledge that the Soviet Union would not use force to stop that process.5

In August 1989, a peace demonstration was held on the Austro-Hungarian border and the border was briefly opened, which set in motion a chain of reaction that resulted in the disintegration of the Soviet bloc. By the end of the year, after a series of revolutions, communist regimes had been swept away everywhere in Central and Eastern Europe and replaced by democratic political systems with competitive elections. In September, the Hungarian authorities decided to open the border with Austria, which led to the exodus of thousands of East Germans and further increased the pressure on the German Democratic Republic to open up. Mass protests erupted in East Germany and, by October, Erich Honecker, the leader of the German Democratic Republic (GDR), was forced to resign. On November 9, the spokesman of the regime gave a press conference to announce new travel regulations, but in talking to journalists he gave the impression that East Germans were now free to cross the border in Berlin. The news spread quickly and, a few hours later, a very large crowd had gathered at the Wall and demanded to cross to the West. Since no one was willing to take the responsibility to use force to disperse the crowd, the commander at one of the border crossings yielded and ordered the guards to let people through.6 Almost 40 years after it was erected, the Berlin Wall had fallen and the German reunification, which up until then had seemed but a distant hope, suddenly appeared inevitable.

However, it would require negotiations with the Soviet Union, which still had more than 300,000 troops on the territory of the GDR and legal rights over Germany due to the post-World War 2 arrangements under which Germany was to be administered by the victorious powers. These negotiations took place throughout 1990 and, in the years and decades that followed, have become the topic of a controversy between the West and Russia. Moscow claims that, during the negotiations, Western officials had pledged not to expand NATO to Central and Eastern Europe to convince Gorbachev to allow Germany to remain in NATO after reunification. As we shall see, although today this argument is commonly associated with Vladimir Putin (who frequently brings up this old grievance to justify his actions), it has a long history that started well before he rose to power and did not originate with him.7 It's important to examine this controversy because, although ultimately I will argue that it’s not as important as many people who are sympathetic to the Russian argument believe, I think it illustrates how Russia's complaints are summarily, and sometimes even dishonestly, dismissed in the West even when they have merits. This kind of attitude goes a long way toward explaining why Moscow came to distrust the West in the decades since the end of the Cold War, even if taken in isolation this particular episode didn't matter as much as the importance it has taken in the public debate suggests. This controversy also highlights how key decisions that were made by the US and its allies at the end of the Cold War and, beyond the particular issue of NATO expansion, shaped the relations between Russia and the West in the post-Cold War era.

1 The controversy about the pledge not to expand NATO

As I just noted, the controversy about the pledge that Western officials allegedly made not to expand NATO to Central and Eastern Europe at the end of the Cold War did not originate with Putin. Russian officials have offered several different versions of that argument over the years and it’s not always easy to pin down exactly what claims they are making. A version of that argument was made in 1993 by Boris Yeltsin in a letter that he sent to Bill Clinton:

I also want to call attention to the fact that the spirit of the Treaty on the Final Settlement with Respect to Germany, signed in September 1990, especially its provisions that prohibit the deployment of foreign troops within the eastern lands of the Federal Republic of Germany precludes the option of expanding the NATO zone into the east.8

This claim was not only vague but also pretty weak, since in particular he was just talking about the “spirit” of the treaty on Germany’s reunification, but later in the 1990s Russian officials made stronger arguments that referred more specifically to claims made by Western officials at the end of the Cold War.9 More recently, in the speech he made just before the invasion of Ukraine, Putin also referred to the promise that Moscow had allegedly received in 1990 and was more specific:

In 1990, when German unification was discussed, the United States promised the Soviet leadership that NATO jurisdiction or military presence will not expand one inch to the east and that the unification of Germany will not lead to the spread of NATO's military organization to the east. This is a quote.10

As we shall see, the quote he seemed to have had in mind was a statement made by James Baker, who in 1990 was Secretary of State. In 2009, Dmitry Medvedev, then President of Russia, had also accused the West of having broken promises that were made at the time:

After the disappearance of the Warsaw Pact, we were hoping for a higher degree of integration. But what have we received? None of the things that we were assured, namely that NATO would not expand endlessly eastwards and our interests would be continuously taken into consideration. NATO remains a military bloc whose missiles are pointed towards Russian territory. By contrast, we would like to see a new European security order.11

It’s interesting to note that, in Medvedev’s version of the argument, NATO expansion was part of a more general complaint about the West’s attitude toward Russia after the end of the Cold War. In order to understand what they were talking about, it’s necessary to go back to what happened after November 8, 1989.

Three weeks after the fall of the Wall, Helmut Kohl, then Chancellor of the Federal Republic of Germany (FRG), made a speech in the Bundestag where he presented a ten-point plan to reunify Germany. In this speech, he proposed in particular that, as a first step toward unification, a German confederation be created that would bring together the FRG and the GDR. But he made that proposal without consulting either Gorbachev or his Western allies, which raised concerns both in the Soviet Union and the West that he might try to act unilaterally and present them with a fait accompli. Gorbachev was particularly upset because, after the Wall fell, he’d sent a personal envoy to Kohl enjoining him to proceed with caution and promising that if he did then “anything might become possible”, so he thought that by making that speech without talking to him first Kohl had violated their agreement.12 As a result, he refused to discuss reunification for 2 months after that, which gave Kohl time to reassure his Western allies. They were also concerned that he’d made that announcement without consulting them and, ironically, suspected that he might have made a deal behind their back with Gorbachev. François Mitterrand, the French president, accepted reunification without enthusiasm but wanted to make sure that it would not imperil European integration and that post-WWII borders would not be questioned. Margaret Thatcher, the British Prime Minister, was opposed to reunification because she feared Germany would go bad again. While eventually she accepted that she couldn’t prevent it and aligned herself with the US position, she secretly hoped that Gorbachev would stop it. As for George Bush, the US president, he supported German reunification as long as it did not threaten NATO, which then as now was the main vehicle of US influence in Europe. In January, the Soviets reached the conclusion that reunification was inevitable, so they agreed to host US and West German officials in Moscow on February 9-10 for preliminary talks on the issue.

The controversy has focused on what US and West German officials told their Soviet counterparts during those preliminary talks in Moscow. As we shall see, they made some assurances that NATO would not expand to the east, but people disagree about what they meant by that. The Russians and people who defend their position claim that, when US and West German officials told the Soviets that if Germany could stay in NATO after reunification the Alliance would not move eastward, they were not just talking about the territory of the GDR but about Central and Eastern Europe in general. Most Western officials, on the other hand, insist that the assurances made in Moscow only pertained to the GDR. For instance, then Secretary of State Baker, who as we shall see played a crucial role in the negotiations over German reunification, repeatedly made that claim and denied that the Russian argument had any merit:

There was a discussion about whether the unified Germany would be a member of NATO, and that was the only discussion we ever had. And the Soviets signed a treaty acknowledging that the unified Germany would be a member of NATO. So I don't understand how they can have these ideas that somehow, now, we promised them there would be no extension of NATO. There was never any discussion of anything but the GDR.13

This is also the view expressed by Philip Zelikow, who in 1990 was on the staff of the National Security Council (NSC) dealing with those issues. In a 1995 op-ed, he claimed that “there is no evidence that in late January or early February of 1990 anyone — Mr. Genscher, James Baker or Mikhail Gorbachev — was even thinking, much less talking, about the possibility of NATO expansion even further into East-Central Europe”.14 Rodric Braithwaite and Hans-Dietrich Genscher, respectively UK ambassador to the Soviet Union and foreign minister of the Federal Republic of Germany (FRG) in 1990, more recently expressed a similar opinion.15

Until a few years ago, this position had also become a near-consensus among scholars. Its most influential defense is probably a paper by Mark Kramer published in 2009, where he argued that the idea that Western officials had pledged not to expand NATO if the Soviet Union agreed to allow reunified Germany to stay in NATO was a "myth".16 Angela Stent, citing that paper among other things, recently summarized that position:

What did the West promise Gorbachev at the time of German unification? This question has riled relations for more than twenty years, reinvigorated as more archival material has become available. Some Western participants in the German unification process insist that the United States promised Gorbachev that NATO would not enlarge after Germany was unified and accuse the West of reneging on commitments it made to the USSR. Many Russian officials and experts, including Mikhail Gorbachev himself, subscribe to this view. “According to the Two-Plus-Four Agreements under which Germany was unified,” he has said, “the United States, Germany, Britain and France promised us that they would not expand NATO east of Germany.” Yet a careful reading of the agreements that were signed in 1990 when a united Germany joined NATO reveals that NATO enlargement was not explicitly addressed. Gorbachev and his advisors may have with hindsight believed that promises had been made by the Americans. But the historical record shows that no explicit commitments about NATO not enlarging were made—simply because this issue was not on the table. Secretary of State James Baker had told Gorbachev in February 1990 (before Germany was unified) that NATO’s jurisdiction would not shift eastward from its present position, but he was referring to NATO’s jurisdiction over the territory of the GDR, not to NATO’s possible enlargement. In other words, Baker was talking about not stationing NATO troops on the territory of the former East Germany after 1990, not about anything else.17

For a long time, this view was virtually unchallenged in academic circles, but this changed a few years ago when new archival documents emerged and some scholars started to argue that, in light of that evidence, Kramer's position had to be revised.

This actually started in 2009, the same year Kramer published his paper, when Der Spiegel published a piece saying that "after speaking with many of those involved and examining previously classified British and German documents in detail, SPIEGEL has concluded that there was no doubt that the West did everything it could to give the Soviets the impression that NATO membership was out of the question for countries like Poland, Hungary or Czechoslovakia".18 But it took a few years before academics, using the documents cited by Der Spiegel and other recently declassified materials, started to push back against Kramer's view. In 2016, Joshua Shifrinson published a much noted paper in which he reopened the debate and, using recently declassified documents, argued that despite the consensus "this evidence suggests that Russian leaders are essentially correct in claiming that U.S. efforts to expand NATO since the 1990s violate the ‘spirit’ of the 1990 negotiations".19 Svetlana Savranskaya and Thomas Blanton made a similar argument in 2017, concluding that "multiple national leaders were considering and rejecting Central and Eastern European membership in NATO as of early 1990 and through 1991, that discussions of NATO in the context of German unification negotiations in 1990 were not at all narrowly limited to the status of East German territory, and that subsequent Soviet and Russian complaints about being misled about NATO expansion were founded in written contemporaneous memcons and telcons at the highest levels".20 More recently, Marc Trachtenberg published a paper in which he defends a similar view, concluding that "the Russian allegations are by no means baseless".21 Previously, Mary Sarotte had somewhat nuanced Kramer’s position, arguing that contrary to what he claims the expansion of NATO not only to East Germany but also to Central and Eastern Europe was briefly discussed in 1990, though in the end Gorbachev never secured a promise that it would not happen and even made concessions on East Germany in the final agreement that undermined the principle that NATO could not expand eastward.22 Kristina Spohr is generally close to Kramer's position, but she nevertheless acknowledges that several Western officials "did make comments to Soviet officials that might have been interpreted as more far-reaching and thus perhaps more consequential in terms of Soviet perceptions than has so far been acknowledged".23

Before Russia invaded Ukraine, those arguments had weakened the consensus in favor of Kramer's position, but since then it has become unpopular to acknowledge that some of Putin's complaints may have merits, so now people tend to talk as if the debate of the last few years had never happened and the view that Russia's position has no merits whatsoever once again goes largely unchallenged. In what follows, I argue against that view and conclude that the conclusion reached by Der Spiegel is essentially correct, while acknowledging that the case for the Russian position has nevertheless been exaggerated. In particular, I show that critics of that position are right about a crucial aspect of that debate (though for the wrong reasons), namely the scope of the assurances made in Moscow during the preliminary talks on Germany’s reunification in February 1990. With one exception, US and West German officials who participated in those preliminary talks were only talking about the territory of the GDR, but I argue that it doesn’t follow that NATO expansion to Central and Eastern Europe didn’t violate assurances made at the end of the Cold War because the negotiations on Germany’s reunification continued for several months after that and Western officials subsequently made broader assurances that it was very reasonable for the Russians to retrospectively consider incompatible with NATO expansion. But perhaps more importantly, this discussion will highlight a key but largely ignored fact about this controversy, which is that even if the debate has so far narrowly focused on the issue of NATO expansion, what violated the assurances made by Western officials at the end of the Cold War wasn’t so much NATO expansion per se but the gradual exclusion of Russia from the European security architecture it came to imply. In order to understand why, however, it’s necessary to go back to the controversy as it has played out in the literature and the arguments made by critics of the Russian position. In his paper on the controversy, Trachtenberg distinguishes three main arguments that critics of the Russian position have made:

First, they claim that the assurances applied only to eastern Germany, and not to Eastern Europe as a whole, and that even those assurances were superseded by arrangements worked out with the USSR later in the year. Second, they claim that the assurances in any event were not legally binding, and were thus not binding at all, because they were not embodied in any formal, signed agreement. And, third, they insist that whatever impression the Russians took away from what they had been told, western leaders had not deliberately sought to mislead them.24

He goes on to argue, often that each of those arguments is flawed. In what follows, I will also examine those arguments and explain why, though on most points his position is correct and I’m in broad agreement with him, I nevertheless think Trachtenberg sometimes overstates his case or doesn’t frame his position in the right way.

1.1 The assurances made at the beginning of the Two Plus Four process

Everybody agrees that, when he saw Gorbachev on February 9, 1990, Baker told Gorbachev that "if we maintain a presence in a Germany that is a part of NATO, there would be no extension of NATO’s jurisdiction for forces of NATO one inch to the east".25 But as we have seen, most former Western officials and scholars claim that this and other similar assurances given by Western officials in the course of the negotiations about Germany's reunification only applied to the territory of the GDR. Moreover, they argue that even those assurances about East Germany were superseded by concessions made by Gorbachev later in the negotiations, as reflected by the content of the Treaty on the Final Settlement with Germany or Two Plus Four Agreement by which France, the Soviet Union, the United Kingdom and the United States renounced the rights they held on Germany and allowed reunification between the FRG and the GDR. As Trachtenberg notes, when people claim that assurances made during the negotiations of this treaty only concerned the territory of the GDR and not the rest of Central and Eastern Europe, they actually make two distinct arguments.26 First, they argue that the expansion of the NATO beyond the Oder-Neisse line (which constituted the border between the GDR and Poland) was simply not an issue at the time, because at least when Baker said that nobody envisioned the dissolution of the Warsaw Pact. If that were true, then it would settle the question, because if nobody was even thinking about the possibility that NATO might expand to Central and Eastern Europe at the time then Baker couldn't have been talking about this when he told Gorbachev that NATO would not move "one inch to the east". As Trachtenberg points out, the second argument is narrower in scope and says that whatever the participants had in mind at the time, the issue of NATO expansion to Central and Eastern Europe never came up during the negotiations and the assurances to which people who find merits in the Russian argument refer applied only to the GDR.

1.1.1 The context of the preliminary talks in Moscow between US/FRG officials and their Soviet counterparts

Starting with the first argument, as Trachtenberg and others have pointed out, it's patently false that nobody envisioned that the Warsaw Pact would disappear and that NATO might expand to former members of that organization in February 1990. On the Soviet side, Kramer claims that in February 1990, "Gorbachev was still fully confident that the USSR would continue to ‘work with its allies’ in the Warsaw Pact", but Soviet archives of a Politburo meeting on January 2, the British record of a meeting between Eduard Shevardnadze and Douglas Hurd, respectively Foreign Minister of the Soviet Union and Foreign Minister of the UK, on February 12 and a diary entry by one of Gorbachev's aides for January 21 show that, on the contrary, many people in the Soviet leadership were already predicting the end of the Warsaw Pact by then.27 For instance, in his meeting with Hurd, Shevardnadze is reported as saying:

If [the GDR] ceases to exist, Soviet troops will be pulled out of Hungary and Czechoslovakia. Poland also will not want them. What purpose then would the Warsaw Pact have?

He said that at a time when, by everyone's account, the Soviet government had reached the conclusion that Germany's reunification was inevitable and that it would happen soon, so it's obviously not the case that nobody on the Soviet side was thinking about the end of the Warsaw Pact in February 1990. While Gorbachev himself was still talking as if he believed that at least part of it could be saved in a meeting with some of his advisors on January 26, as Trachtenberg notes, he was hardly "fully confident" that it would remain intact.28

In the West, not only had intelligence services been predicting the disintegration of the Warsaw Pact since the end of 1989, but the press was openly speculating that its days were numbered by the beginning of 1990.29 On February 4, just a few days before Baker met with Gorbachev and gave his infamous assurance that NATO would not move "one inch eastward", The Washington Post published a piece whose title was "Warsaw Pact—Endgame: In Eastern Europe, the Military Alliance Is Dead" and in which the author reported that "many top U.S. and European officials confidently predict" that "Soviet troops will be forced to withdraw entirely from Eastern Europe within a few years" and that "Hungary and perhaps Czechoslovakia may formally withdraw from the Warsaw Pact". A bit later in the article, he even added that "the idea of an entire nation defecting from the Warsaw Pact is hardly surprising, given that the alliance has already ceased to function substantively".30 On January 19, The Washington Post had already reported that "Hungarian and Polish leaders said today they want all Soviet troops out of their countries in a year or two, underscoring the increasingly rapid dissolution of the Warsaw Pact as a military alliance".31 During a public seminar organized by a German defense magazine on February 3-4, Gerhard Stoltenberg, the Minister of Defense of the FRG, declared that "we would commit a fatal strategic mistake if, faced with the increasing disintegration of the Warsaw Pact, we negotiated in exchange that of the Atlantic Alliance".32 Recently declassified British and US documents, as well as interviews with former Bush administration officials, also show that by the end of 1989 and the beginning of 1990, the White House was already thinking about how to increase US influence in Central and Eastern Europe after the Soviet Union retreated from the region, so obviously they expected this to happen.33 Thus, while nobody at the time expected that it would happen so fast, few people doubted that in the long run the Warsaw Pact would disappear or, at the very least, that it would not survive without undergoing a significant transformation that would open the door to more US and Western influence in Central and Eastern Europe.

But the fact that people expected the Warsaw Pact to disappear and more generally Soviet influence to wane in the region doesn't mean they were already planning to expand NATO. Indeed, while Bush and his advisors were already concerned with making sure to keep their options open when the negotiations over German reunification started, they did not begin to seriously consider offering Central and Eastern European countries the prospect of NATO membership until 1991 and it wasn’t until Bush’s final year in office that a consensus developed that it would have to happen eventually because it was the best way to strengthen the US position in the region and prevent a dangerous standoff between Russia and Germany did not emerge until Bush's final year in office.34 However, in a speech he gave in Tutzing on January 31, Genscher not only endorsed Gorbachev's idea of a pan-European security architecture, but he publicly called for NATO to "state unequivocally that whatever happens in the Warsaw Pact, there will be no expansion of NATO territory eastward, that is to say, closer to the borders of the Soviet Union".35 Immediately before that, he had noted that in “Poland, Czechoslovakia and Hungary there is a growing demand for the withdrawal of Soviet forces” and that “we cannot say for sure at present what impact this will have on the structure, and on the future, of the Warsaw Pact”, clearly implying that it might unravel. If he had just said that NATO would not expand "closer to the borders of the Soviet Union", this sentence would be ambiguous on whether he was merely referring to the territory of the GDR or more generally to Central and Eastern Europe, but the fact that he prefaced this claim by "whatever happens to the Warsaw Pact" immediately after, strongly implying that it might soon cease to exist, in addition to being more evidence that Western officials already suspected that the Soviet-led alliance would not survive at the beginning of 1990, makes it clear that he meant the latter.

Yet when he mentions that speech, Kramer paraphrases Genscher as saying that “a united Germany would be a member of NATO, but that NATO’s jurisdiction would not extend to the eastern part”, which falsely makes it sound as if he was only talking about the territory of the GDR.36 Nor was Genscher’s wording a mistake or something he’d let slip in the heat of the moment, for he repeated the same thing almost word for word in a speech he made in Potsdam on February 9, just one day before he met with Shevardnadze in Moscow.37 It’s true that just after the passage where he made that non-expansion assurance, Genscher also said explicitly that “any proposals for incorporating the part of Germany at present forming the GDR in NATO’s military structures would block intra-German rapprochement”, but it’s clear that at this point of the speech he’s moved on to a different, though related point.38 To be sure, it’s also possible to interpret Genscher as saying that NATO should commit not to expand to the territory of the GDR even if the Warsaw Pact collapsed (in which case one might think that Moscow would care less about what happened in the territory of the GDR), but this interpretation seems far less natural and, as we shall see shortly, there is more evidence that it’s not what Genscher meant and that he was talking about Warsaw Pact countries. Moreover, even if that interpretation were correct, it wouldn’t really help critics of the Russian position. Indeed, if the Soviet Union was so concerned about NATO that Genscher thought it was important to promise that it would not expand to the territory of the GDR even if the Warsaw Pact collapsed, then a fortiori this assurance would have applied to former Warsaw Pact countries in Central and Eastern Europe. Not only did the assurances made by Genscher extend far beyond the GDR, but officials in Washington seem to have noticed and were alarmed by it. Robert Hutchings, who in 1990 served as a special advisor to the Secretary of State, later wrote about Genscher’s speech in Tutzing that he was speaking “not about the GDR but about Poland, Czechoslovakia, and Hungary”.39 It would be very surprising if Soviet officials had not also noticed. Indeed, Gorbachev discussed in some detail the range of views that had been expressed at the symposium of the Protestant Academy in Tutzing on January 31 when he met Baker in Moscow on February 9, so it’s hard to believe that he didn’t read Genscher’s speech, which was delivered at the same event.40 Shevardnadze also claimed to have read the speech he made in Potsdam and even to have forwarded it to Gorbachev.41

Moreover, just one week before his fateful meeting with Gorbachev in Moscow, Baker went along with this idea or at least did nothing to suggest that he didn't even when the opportunity to do so came along. Indeed, after he gave that speech, Genscher decided to meet with Baker in Washington on February 2 to make sure they were on the same page and, according to a cable about the meeting sent by the State Department I discuss below, he told him there was a "need to assure the Soviets that NATO would not extend its territorial coverage to the area of the GDR nor anywhere else in Eastern Europe for that matter" and Baker apparently did not express a disagreement with that idea.42 Genscher then went on to repeat that point in the press conference he and Baker gave after their meeting, during which he made the following remarks as Baker was standing next to him:

We agreed that the intention does not exist to extend the NATO defense area toward the East. That applies, moreover, not just to the territory of the GDR, which we do not want to incorporate, but rather applies in general.43

He could hardly have been clearer that, contrary to what critics of the Russian position claim, he was not just talking about the territory of the GDR but about Central and Eastern Europe in general. As Trachtenberg notes, the fact that he used "we" is also significant, because it means that he was speaking for both he and Baker. If Baker had wanted to express his disagreement with that proposal, he could have done so during the press conference, but he did not.

One could argue that Baker may have missed the significance of that point since at the time everyone was still focused on the issue of whether Germany would be allowed to remain in NATO after reunification in the first place, which pushed the question of the conditions under which the Soviet Union might agree to that in the background.44 However, a summary of their discussion sent by the State Department to the US Ambassador to the FRG at Baker's request shows that he, or at least the author of that cable, Assistant Secretary of State for Europe and Canada Raymond Seitz, did not miss this part of Genscher's proposal at all:

Genscher confirmed that neutrality for a united Germany is out of the question. The new Germany would remain in NATO because NATO is an essential building block to a new Europe. In stating this, Genscher reiterated the need to assure the Soviets that NATO would not extend its territorial coverage to the area of the GDR nor anywhere else in Eastern Europe for that matter. (He made this point with the press after the meeting.)45

Moreover, in describing Baker's response to Genscher's proposals, the cable does not say that Baker expressed disagreement with that idea at any point during that conversation. As Sarotte explains, the reason why this cable was sent to Bonn is that Baker wanted the US Ambassador to the FRG to inform Kohl, the German Chancellor, of what had been discussed in Washington because he wasn't sure that Genscher, who didn't have a very good relationship with Kohl, would do so and he wanted to make sure that everyone both in Washington and Bonn was on the same page before the negotiations with the Soviet Union started. Vernon Walters, the US Ambassador to the FRG, did as instructed and briefed Horst Teltschik, Kohl's security advisor, the next day.46 Thus, not only was the State Department in Washington aware that Genscher wanted to give Moscow assurances that applied not just to the territory of the GDR but to Central and Eastern Europe in general, but so was the Chancellor's office in Bonn and as far as we can tell neither raised any objections.

Finally, to remove any lingering doubt that US officials may have missed that Genscher's proposal applied not only to the territory of the GDR but to Central and Eastern Europe in general, the State Department official transcript of his press conference with Baker makes Genscher's remarks even more explicit than they already were:

Perhaps I might add, we were in full agreement that there is no intention to extend the NATO area of defense and the security toward the East. This holds true not only for GDR, which we have no intention of simply incorporating, but that holds true for all the other Eastern countries. We are at present witnessing dramatic developments in the whole of the Eastern area, in COCOM, and the Warsaw Pact. I think that it is part (of) that partnership in stability which we can offer to the East that we can make it quite clear that whatever happens within the Warsaw Pact, on our side there is no intention to extend our area—NATO’s area—of defense towards the East.47

So people in the State Department, but probably also in the NSC, had duly noted the scope of Genscher's proposal and, if US officials did not express disagreement with it either publicly or privately, it's not because they had missed it.48

At the time, US officials were not only worried that the Soviet Union might adamantly reject that a reunified Germany remain in NATO, but they also feared that Gorbachev might offer Helmut Kohl a deal he "couldn't refuse" by proposing to allow reunification in exchange for neutrality.49 This would effectively spell the end of NATO and, as a result, of the US military presence in Europe, something US officials were determined to prevent. Thus, it's possible that Baker didn't actually intend to give the Soviet Union assurances that went beyond the territory of the GDR, but thought it best not to raise that issue with Genscher at this time to avoid a split between the US and Germany.50 In fact, this is more or less what James Dobbins, then Assistant Secretary of State for Europe, told the British Ambassador in Washington after Baker’s meeting with Genscher:

Immediately after the meeting Genscher announced to a hastily assembled press conference that he and Baker were in ‘full agreement’ that reunification would not involve the extension of NATO to the east. Dobbins said that Baker had not in fact blessed Genscher’s particular formula, even though it was the best available at present. The Administration has made its position clear in December with its four principles, and was now cautious about being too specific, lest it be interpreted by the German opposition as an imperial dictat [sic] or upset the Russians before Baker’s talk with Shevardnadze.51

Thus, it seems that at the time Baker wasn’t ruling out Genscher’s formula, but he may also have not been committed to it and simply remained silent for tactical reasons when Genscher publicly said he had endorsed it. However, it’s also possible that Baker was initially more aligned with Genscher’s position, that he revised his view later after other people in the Bush administration pushed for a less accommodating stance and that Dobbins was engaging in damage control after the fact with allies.52 This might also explain why, as we shall see, Baker remained somewhat ambiguous when he met Shevardnadze and Gorbachev in Moscow. The most important thing for US officials at the time was that Genscher had said that Germany had to stay in NATO after reunification and they wanted to make sure he would not move from this position.

In any case, it's remarkable that even though several critics of the Russian position actually quoted this passage of the press conference during which Genscher talks about the necessity to make clear that NATO would not expand to the east, they omitted the part where he makes clear that he isn't merely talking about the territory of the GDR but about Central and Eastern Europe in general. For instance, Zelikow and Rice claim that "Baker understood Genscher to say that Germany would remain in NATO, but the Soviets had to be assured that NATO's territorial coverage would not extend to the former GDR" and quote Genscher's comment that he and Baker "were in full agreement that there is no intention to extend the NATO area of defense and security towards the East", which comes from the State Department’s transcript of the press conference, but omit the next sentence in the transcript, where Genscher clarifies that it "holds true not only for GDR, which we have no intention of simply incorporating, but that holds true for all the other Eastern countries".53 Spohr wrote that Genscher’s discussion with Baker “left the territories east of the GDR untouched”, but as we have seen that is simply not true.54

Kramer even went further and, after quoting the exact same part of the transcript as Zelikow and Rice (without the next sentence that clarified the scope of his assurance that NATO would not expand "towards the East"), he added a gloss on that comment that completely changed the meaning of what Genscher had actually said:

At a joint press conference after their meeting, Genscher said that he and Baker "were in full agreement that there is no intention to extend the NATO area of defense and security toward the East," meaning eastern Germany. [emphasis is mine] When asked by journalists what exactly this meant, Genscher insisted that he was not talking about "a halfway membership [for a united Germany] this way or that. What I said is there is no intention of extending the NATO area to the East."55

Kramer cites a Frankfurter Allgemeine Zeitung article as the source of those quotes, but it doesn't actually contain them and they are identical to the State Department transcript, so either he checked the transcript directly or he found those quotes in another source that was quoting Genscher from the transcript.56 Whether he knew that Genscher had made clear that he was talking not just about the territory of the GDR but about Central and Eastern Europe in general and knowingly altered the meaning of his remarks by adding that gloss or was misled by another source that quoted Genscher partially and added that gloss in good faith, the fact is that he completely changed the meaning of Genscher's remarks in a way that makes them consistent with his view that Western officials never made any assurances that went beyond the territory of the GDR, when in fact they show that view to be false. As we shall see, adding interpolations to ambiguous statements made by Western officials pledging not to expand NATO eastward to make them sound like they support his view that assurances given about NATO expansion only ever applied to the territory of the GDR is part of Kramer's usual modus operandi, but in this case he actually did the same thing with a statement that was not ambiguous at all.57 As we have seen, he had already distorted Genscher’s Tutzing speech in a similar way.

On February 6, Genscher met with Douglas Hurd, the British Foreign Minister, in Bonn and told him the same thing. Here is how the latter summarized what Genscher had told him in a cable he sent to Christopher Mallaby, the British Ambassador to the FRG, after their meeting:

Genscher added that when he talked about not wanting to extend NATO that applied to other states beside the GDR. The Russians must have some assurance that if, for example, the Polish Government left the Warsaw Pact one day they would not join NATO the next.58

A German document summarizing the same meeting says that Genscher told Hurd something very similar, but using the example of Hungary instead of Poland:

The West could do a lot to facilitate the current developments for the SU. Of particular importance was the declaration that NATO had no intention of expanding its territory to the east. Such a declaration should not only relate to the GDR, but should be of a general nature. For example, the SU also needs security that Hungary will not become part of the western alliance in the event of a change of government.59

According to Hurd's cable, Genscher also told him that "the CSCE summit, devoted to the future of Europe, would be an important vehicle for helping the Soviet Union to come to terms with the erosion of the Warsaw Pact", which again show that both the dissolution of the Warsaw Pact and the possibility that Moscow might worry about the expansion of NATO in that area were very much on the mind of Western officials just before the negotiations about German reunification were set to begin.60

The possibility that NATO might expand to Central and Eastern European countries or at least that Western countries might expand their influence in that region as the Soviet Union retreated from it was certainly on the mind of leaders in those countries. Indeed, on February 4, The Washington Post reported that it was a "hot topic of discussion" among military leaders in the Warsaw Pact:

But the shopworn, bipolar concept of Europe no longer interests military leaders of the Eastern bloc, many of whom have been swept into power within the past two months. In their circle, the prospect of a militarily neutral Eastern Europe, or even one with a web of economic and military ties to the West, is suddenly a hot topic of discussion.61

Gyula Horn, Hungary's Foreign Minister, actually talked about the possibility that his country might join "NATO's political councils" publicly on February 24.62 He had even raised the idea directly to Lawrence Eagleburger, the Deputy Secretary of State, in a private meeting around the same time.63 In the months that followed, as they realized that Gorbachev would not use force to prevent the dislocation of the Warsaw Pact, leaders of Central and Eastern European countries started to be increasingly explicit about their interest in eventually joining NATO. Kramer is right that, in his statement, Horn was not talking about joining NATO's military structures but merely about some kind of political cooperation and that he said that more than 2 weeks after Baker met Gorbachev, but The Washington Postarticle makes clear that Eastern European leaders had already talking about it privately before that. Now, if leaders of Warsaw Pact countries were talking about this even before Baker went to Moscow, it's very unlikely that Soviet officials didn't know about it. Indeed, in a January 26 meeting with his advisors I already mentioned above, Gorbachev explained that unless the Soviet Union worked with "other socialist countries" they'd be "picked up by others", by which he presumably meant NATO.64

Thus, the argument that Western officials couldn't possibly have intended to give assurances about NATO expansion that applied to Central and Eastern Europe in general and not just the territory of the GDR in February 1990, because nobody at the time imagined that the Warsaw Pact would collapse and that NATO might expand in the region, is undoubtedly incorrect. To be sure, in both private and public conversations in the days and weeks leading up to the beginning of the negotiations over German reunification, the focus was on the status of the territory of the GDR, but officials in both the Eastern and Western blocs definitely had in mind the unraveling of the Warsaw Pact and the possibility that Western countries might take advantage of it, including by expanding NATO even though nobody was actively planning to do so at the time. As we have seen, Genscher even gave public assurances to that effect that it wouldn't happen on several occasions, including during a press conference where he claimed to be speaking for both he and Baker, yet as Trachtenberg aptly noted neither Baker during that press conference nor the State Department later "issue a clarification pointing out that Genscher had been speaking only for himself when he had made that remark and that the U.S. government did not necessarily share his views in that regard".65 Which brings us to the second, narrower argument that, whatever people were thinking when the negotiations about Germany's reunification started, nothing was said during the negotiations to the effect that NATO would not expand to Central and Eastern European countries and Western officials only gave assurances about the territory of the GDR.

1.1.2 What US and West German officials told their Soviet counterparts in Moscow

Baker went to Moscow on February 9 and first met with Shevardnadze before talking to Gorbachev. He tried to convince both that it would be in the Soviet Union's interest to allow Germany to stay in NATO after reunification, on the grounds that a neutral Germany would probably seek to develop an independent nuclear capability to ensure its security and might even return to the militarism of the past, which would be much less likely if it remained anchored in NATO. He told Shevardnadze that, in that case, "there would, of course, have to be iron clad guarantees that NATO's jurisdiction or forces would not move eastward" and that it "would have to be done in a manner that would satisfy Germany's neighbors to the east".66 This is ambiguous as to whether he just meant the territory of the GDR or Central and Eastern Europe in general, but the reference to the concerns of “Germany’s neighbors to the east” suggests he was just talking about the GDR. Indeed, if he had been referring to a possible expansion to Central and Eastern European countries, it’s hard to see why those countries would have been concerned about it, whereas it makes perfect sense if he was talking about NATO forces moving into the territory of the GDR, since in that case Czechoslovakia would have to deal with NATO’s presence not just on its western border but also on its northern one and NATO would suddenly appear on Poland’s western border. Although the Warsaw Pact was falling apart, nobody thought it would completely disappear so rapidly, so it made sense to worry about the reactions of those countries.

Baker also proposed a mechanism for the negotiations that would associate the FRG and the GDR to the four powers that still had legal rights over Germany, the so-called "Two Plus Four” mechanism, which eventually became the format under which the negotiations were conducted. Later in the conversation, he suggested that such a mechanism could be used to produce "the right kind of outcome" and went on to say that "it might be an outcome that would guarantee that there would be no NATO forces in the Eastern part of Germany" and even added that "in fact there could be an absolute ban on that". Trachtenberg argues that, since Baker had no problem mentioning the territory of the GDR explicitly in that part of the conversation, it's evidence that, when he used the more ambiguous phrase earlier by assuring Shevardnadze that NATO "would not move eastward" if Germany was allowed to remain part of the Alliance, he was talking about Central and Eastern Europe in general. However, one could argue that, on the contrary, this was a clarification of what he'd meant earlier. The truth is that, just based on what was said during that conversation according to the State Department memorandum that I have been quoting, it's impossible to tell for sure what Baker meant, but I think it’s more likely than not that he was only talking about the GDR. What Shevardnadze understood, on the other hand, is harder to determine.

After his meeting with Shevardnadze, Baker met with Gorbachev and, according to both the State Department memorandum of their conversation and the Soviet record of their meeting, made very similar points and was similarly ambiguous when he talked about not expanding NATO eastward. After explaining why he thought that it would be preferable for everyone, including the Soviet Union, that Germany remain in NATO after reunification, he gave the assurance I already quoted above that in that case the Alliance's jurisdiction would not move "one inch to the east":

We don't favorably view a neutral Germany. The FRG says that this is not a satisfactory approach. A neutral Germany in our view is not necessarily going to be a non-militaristic Germany. It could well decide that it needed its own independent nuclear capability as opposed to depending on the deterrent of the United States. All our allies and East Europeans we have spoken to have told us that they want us to maintain a presence in Europe. I am not sure whether you favor that or not. But let me say that if our allies want us to go we will be gone in a minute. Indeed, if they want us to leave we'll go and I can assure you that the sentiment of the American people is such that they will want us to leave immediately. The mechanism by which we have a US military presence in Europe is NATO. If you abolish NATO there will be no more US presence. We understand the need for assurances to the countries in the East. If we maintain a presence in a Germany that is a part of NATO, there would be no extension of NATO's jurisdiction for forces of NATO one inch to the east.67

In this part of the conversation, Baker is trying to convince Gorbachev that it would actually be in the Soviet Union’s interest to allow Germany to stay in NATO, because otherwise US troops will have to leave the country and Germany will likely become a military power again, which the Soviet Union doesn’t want. He is arguing that German’s continued NATO membership and, relatedly, the US military presence would be a factor of stability in Europe and would therefore benefit the Soviet Union. Again, the phrase Baker used when he said that NATO would not expand is quite general, which if not for the context would be more naturally interpreted as referring to Central and Eastern Europe in general and not just the GDR.

However, the fact that Baker said that immediately after saying that he understood that “countries in the east” would need assurances once again suggests that he was talking about the GDR.68 Indeed, if he was talking about the expansion of NATO to Central and Eastern European countries, then how would ruling out that possibility assuage the security concerns of those same countries? Trachtenberg interprets this passage as evidence that Baker’s assurance was about Central and Eastern Europe in general and not just the territory of the GDR, but that doesn’t make sense since he talked about “countries” in the plural and therefore couldn’t have been referring only to the Soviet Union.69 It’s possible that Baker meant that, if some but not all of them joined NATO, expansion would be seen as a threat to those who stayed outside, but this interpretation seems contrived. It’s much more likely that he was referring to the fact that Central and Eastern European countries, especially Poland which at the time was still worried that after reunification Germany might seek to revise the post-WWII border, would be concerned if NATO suddenly appeared on their borders.70 It’s surprising that, as far I know, no critic of the Russian position has ever made that point, even though it’s a very strong argument in favor of their position that despite the ambiguous language he used Baker was only talking about the GDR.

Still trying to convince Gorbachev that it would be in the Soviet Union's interest to allow Germany to remain in NATO after reunification, he returned to that point later in the meeting:

Baker: I want to ask you a question, and you need not answer it right now. Supposing unification takes place, what would you prefer: a united Germany outside of NATO, absolutely independent and without American troops; or a united Germany keeping its connections with NATO, but with the guarantee that NATO’s jurisprudence or troops will not spread east of the present boundary?

Gorbachev: We will think everything over. We intend to discuss all these questions in depth at the leadership level. It goes without saying that a broadening of the NATO zone is not acceptable.

Baker: We agree with that.71

This comes from the Soviet account of their conversation, but although the State Department memorandum of that meeting is redacted after Baker asks Gorbachev this question, they are virtually identical before that. Moreover, as we shall see shortly, a letter that Baker sent Kohl immediately after the meeting that was declassified by Germany confirms Gorbachev’s reply. Again, they used phrases that were quite general, but still ambiguous in the sense that in this part of the meeting they never explicitly refer to either the territory of the GDR or Central and Eastern Europe in general, so I don't think that we can reach a determination about what they meant or understood the other to be saying just based on the content of this part of the conversation. As before, one can interpret the fact that in other parts of the conversation he specifically mentioned the GDR as evidence that ambiguous phrases about not expanding NATO eastward referred to that, but one can also argue like Trachtenberg that it shows that Baker had no problem mentioning the GDR explicitly when he wanted to make a point specifically about it. Nevertheless, as I argued above, Baker’s reference to Germany’s eastern neighbors earlier in the conversation and during his conversation with Shevardnadze suggests that he was only talking about the GDR.

Robert Gates, Deputy National Security Advisor, met with Vladimir Kryuchkov, head of the KGB, the same day and also tried to convince him that it would be in the Soviet Union's interest to allow a reunified Germany to stay in NATO. In order to make that prospect more attractive to his counterpart, he also said that in that case NATO would not move eastward. Here is how a NSC memorandum of their meeting summarized this part of the conversation:

Events are moving faster than anticipated. We might see some GDR initiative after the 18 March elections. Under these circumstances, we support the Kohl-Genscher idea of a united Germany belonging to NATO but with no expansion of military presence to the GDR. This would be in the context of continuing force reductions in Europe. What did Kryuchkov think of the Kohl-Genscher proposal under which a united Germany would be associated with NATO but in which NATO troops would move no further east than they now were? It seems to us to be a sound proposal.72

As Shifrinson noted, this indicates more support for this idea within the Bush administration than previously recognized.73 On the other hand, since in the first part of his argument Gates apparently talked about not moving NATO forces to the GDR specifically (in contrast to Baker's more ambiguous phrasing), it also suggests that US officials only had in mind a more limited assurance restricted to the territory of the GDR. However, he also refers to Genscher's proposal in the next part of his argument, which as we have seen was not limited to the GDR but applied to Central and Eastern Europe more generally. What Gates meant really depends on whether he saw that proposal as the same as the idea that NATO forces would not move into the territory of the GDR after reunification or as a more general proposal that would just entail the latter. We saw that people in the State Department had noted that Genscher's proposal applied to Central and Eastern Europe more broadly, but it's possible that people in the NSC missed that aspect of the proposal and construed it as more limited than it really was. It's important to keep in mind that governments are not unitary actors, so Gates could have meant one thing and Baker another. In fact, as I’m about to explain, there is evidence that regardless of whether they had noticed that Genscher’s proposal wasn’t limited to the GDR but applied to Central and Eastern Europe in general, people in the NSC were more narrowly focused on the former, so it’s likely that Gates was also talking about the GDR.

Before he went to Moscow, Baker had agreed that he would tell Kohl, who was supposed to meet Gorbachev the next day, what had been said during his conversation with the Soviet leader, so he sent him a letter summarizing their meeting before going back to Washington. In that letter, he used the same language that we have discussed above:

And then I put the following question to him. Would you prefer to see a unified Germany outside of NATO, independent and with no US forces or would you prefer a unified Germany to be tied to NATO, with assurances that NATO's jurisdiction would not shift one inch eastward from its present position.

He answered that the Soviet leadership was giving real thought to all such options, and would be discussing them soon “in a kind of seminar”. He then added: “Certainly any extension of the zone of NATO would be unacceptable.” (By implication, NATO in its current zone might be acceptable.)74

Again, this letter cannot adjudicate the issue of whether this assurance referred only to the territory of the GDR or more generally to Central and Eastern Europe based on this letter, because the language used by Baker remained ambiguous on this point.

In any case, when people in the NSC heard about the question Baker had put to Gorbachev during his meeting with him, they worried that, in his desire to convince the Soviet leader that it was in Moscow's interest to allow Germany to stay in NATO after reunification, Baker might have preemptively made concessions on NATO's future that were not necessary to achieve the US goal of making sure that a reunified Germany would remain in NATO. So the White House drafted and Bush signed another letter to Kohl in which they warned him against making any concessions in advance of express requests by Soviet officials. In that letter, Bush declared himself in favor of a "special military status for what is now the territory of the GDR", but repudiated Baker’s formulation that NATO’s “jurisdiction” would not expand to the east after reunification.75 However, according to Zelikow and Rice, the problem wasn't that Baker's language was ruling out the expansion of NATO to Central and Eastern European countries but that he'd called for no extension of NATO's "jurisdiction" and people in the NSC worried that it would be difficult to reconcile that kind of language with Article 5 of the Washington Treaty, which commits NATO members to mutual defense in case of attack.76 As we shall see, Bush’s formulation is often presented as a hardening of Baker’s proposal, which is true in the sense that it removed any potential ambiguity as to the fact that NATO’s Article 5 would apply to Germany as a whole, including the territory of the GDR. However, the main issue people in the NSC had with his proposal seems to have been, not that it was overly generous, but that taken literally it made little sense, since it seemed to imply that Germany as a whole would be part of NATO yet Article 5 would only apply to some of its territory. Not only would this have been absurd from a political and strategic point of view, but it wasn’t even clear that it could work legally. In any case, after that intervention by the White House, Baker quickly stopped using that kind of language in negotiations with Soviet officials.77 The fact that, in his letter to Kohl, Bush talked specifically about the territory of the GDR and didn’t mention the rest of Central and Eastern Europe is yet more evidence that, despite the ambiguous language they sometimes used in Moscow, US officials were only talking about the GDR.

Thus, right before he met with Gorbachev to begin the negotiations about Germany's reunification, Kohl had received two discordant messages from the US. While Baker talked of the need to give the Soviets assurances that "NATO's jurisdiction would not shift one inch eastward", Bush and the NSC wanted to be clear that NATO would shift to the east of its current position (in the sense that unified Germany as a whole and not just the territory of the FRG would be covered by the Washington Treaty), allowing only for the much smaller concession that after reunification the territory of the GDR would have a "special military status" of some kind. In other words, NATO’s treaty and in particular Article 5 would apply to the entire German territory, even if some restrictions on the Alliance’s freedom of action in the territory of the GDR could be worked out as part of the deal that would allow Germany’s reunification. As Sarotte explains, faced with the choice of which formulation to use in his meeting with Gorbachev, Kohl decided to use "the language most conducive to achieving his goal of German unity" and therefore opted for Baker's formulation.78 As we have seen, this formulation was itself inspired by the language used in the speech Genscher had made in Tutzing and that he'd repeated on several occasions both publicly and privately, but unlike Genscher it seems that Baker never clarified that it applied not only to the territory of the GDR but to Central and Eastern Europe in general and, as I have argued above, it’s very likely that he was only talking about the former.

The Soviet and West German accounts of what Kohl told Gorbachev differ in how specific he was, but according to both, he said that NATO could not expand to the east, instead of using Bush's formulation that NATO would cover even the territory of the GDR, which would simply have a special military status. However, according to the Soviet account of their meeting, Kohl just said that NATO should not "expand its scope of action", whereas according to the West German account he was more specific and said that it "must not extend its sphere to the territory of today’s GDR".79 Kramer claims that "the discrepancy is of negligible importance" since "both transcripts show that Gorbachev would have understood the comment to refer to eastern Germany". Commenting on this discrepancy, Spohr also dismisses it as unimportant, on the ground that "both texts reveal that in delineating NATO’s future boundary the chancellor was referring solely to (eastern) Germany".80 But that is clearly not true, since according to the Soviet but not the German account, what Kohl said was just as ambiguous on that issue as what Baker had said the day before and one cannot conclude that he was only referring to the GDR without begging the question about how to resolve that ambiguity, which is precisely what is at issue here. That being said, since Kohl had read Bush’s letter just before he met with Gorbachev and in that letter the President talked unambiguously about “what is now the territory of the GDR”, I think Kohl likely was only talking about the GDR.

Indeed, while Kohl and Gorbachev were talking, Genscher was having a meeting with Shevardnadze, which Kramer completely ignores, during which he once again gave assurances on NATO expansion that, according to the German memorandum summarizing their conversation, were anything but ambiguous:

Neutralism for Germany as a whole was wrong. We were also thinking of the feelings of our neighbors. It would be better for our neighbors if a united Germany were firmly integrated into European structures. We are aware that NATO membership for a unified Germany raises complicated questions. For us, however, one thing is certain: NATO will not expand to the east. Of course, the newly elected government of the GDR would also have a say in this. It would then have to come to an understanding with the SU. Perhaps it would then turn out that a solution was not so complicated. If Soviet troops remained in the GDR, this was not our problem. The important thing is that we talk to each other in a spirit of trust. As far as the non-expansion of NATO is concerned, this applies in general.81

As you can see, after using the same ambiguous language to assure Shevardnadze that NATO would not expand "to the east", he briefly mentions the GDR to suggest that Soviet troops might be allowed to stay there depending on what the East German government decided, but then goes back to his assurance that NATO would not expand to the east to clarify that it applies "in general".

As Trachtenberg noted, this was the same language that he'd used before to make clear that his assurance that NATO would expand to the east didn't merely apply to the territory of the GDR but more generally to Central and Eastern Europe, both in public such as during his press conference with Baker on February 3 and in private such as during his meeting with Hurd on February 6.82 Spohr argued that, since Shevardnadze was not present at Genscher's meeting with Hurd, he couldn't possibly have known that and Genscher's language must have seemed to apply only to the GDR.83 However, even putting aside that it's hard to see what Genscher could have meant when he said that the non-expansion of NATO "applies in general" if not that it applied to Central and Eastern Europe in general and not just to the GDR, she only says that because she doesn't know that Genscher had also used that language publicly during his press conference with Baker on February 3. Indeed, when discussing that press conference, she cited Zelikow and Rice's book, which as we have seen omitted the crucial passage where Genscher used the same language to clarify that in ruling out the expansion of NATO eastward he didn't just mean the GDR but more generally Central and Eastern Europe.84 Thus, while other US and West German officials used language that, though quite general, was nevertheless ambiguous as to the precise scope of the assurance they were giving Soviet officials in Moscow, Genscher was just as clear that it didn't only apply to the territory of the GDR but to Central and Eastern Europe in general on this occasion as he'd been earlier both in public and private. In fact, on February 21, he publicly repeated this assurance yet again during a press conference.85 In his case at least, there can be little doubt as to what he meant, but as we have seen the same cannot be said about Baker, Gates and Kohl.

Some people on both sides of the debate argue that context definitely shows either that Western officials were only talking about the territory of the GDR or that on the contrary they were talking about Central and Eastern Europe in general, but I think that if we are honest we have to acknowledge that, except for Genscher, it's simply impossible to tell for sure what they meant. For instance, Kramer claims that the context of the negotiations and the expressions used by the participants “leave no doubt” that Gorbachev, Shevardnadze and Baker were only talking about the expansion of NATO into the territory of the GDR.86 However, that is clearly false since not only was the language used objectively ambiguous, but as we have seen the context of the negotiations makes it clear that Kramer is wrong when he argues that NATO expansion to any territory other than the GDR was simply not an issue at the time. He also misrepresents the assurance made repeatedly by Genscher, both in Moscow and before that in public and private, as applying only to the GDR. Nevertheless, as I argued above, a close examination of the record of the conversations that Western officials had with their Soviet counterparts in Moscow strongly suggests that, with the exception of Genscher, they were only talking about the territory of the GDR.

It’s possible that, although other Western officials were only talking about the GDR, Soviet officials were misled by Genscher’s more far-reaching position and interpreted them as making a broader assurance since again the language they used was ambiguous, but there is evidence that it was not the case. Indeed, Valentin Falin, then Chief of the International Department within the Central Committee of the CPSU and a key advisor on German affairs, sent a note to Gorbachev in April warning him against what he saw as a hardening of the position adopted by Western officials in the negotiations. He explains that although until March they had expressed a willingness to commit to “the non-expansion of NATO's sphere of activity to the GDR” they were now trying to reduce the scope of that assurance by saying it should not apply in “crisis situations”.87 Moreover, in that same note, Falin also pointed out that although Genscher still talked about something closer to the Soviet idea of a pan-European structure that would replace Cold War alliances from time to time, he was isolated and few people listened to him “except for the West German Social Democrats and left-wing parties in some countries of the [European] common market”.88 This document shows that, by April at least, not only did the Soviets interpret Western officials as talking only about the GDR, but they also understood that Genscher could not be assumed to talk for his colleagues and that his views didn’t reflect the Western position. It’s possible that Soviet officials interpreted their Western counterparts differently in February, but there is nothing to suggest that. Thus, the evidence strongly supports the view not only that Baker, Gates and Kohl were only talking about the territory of the GDR, but that Soviet officials interpreted them as doing so at the time. While they misrepresented Genscher’s position and the context in which the preliminary talks in Moscow took place, Kramer and other scholars who argued that US and West German officials were only talking about the territory of the GDR in Moscow are right, while on this point at least Shifrinson and Trachtenberg are wrong.

1.2 From Moscow to Moscow

As I will argue shortly, the preliminary talks in Moscow between US and FRG officials and their Soviet counterparts did not result in a deal by which the Soviet Union agreed to allow Germany to stay in NATO after reunification in exchange for the promise that NATO would not expand to Central and Eastern Europe, but they nevertheless had important consequences. During his conversation with Gorbachev, Kohl got him to agree that it was for the German people alone to decide about unification, without securing anything in exchange. It’s likely that Gorbachev didn’t realize the significance of that claim at the time, but Kohl did and he immediately moved to take advantage of that concession by calling a press conference during which he told everyone that Gorbachev had agreed that whether the Germans wanted to live in a unified state was a question for them to decide, making it difficult for the Soviets to walk back Gorbachev’s hastily made remark later. Before Kohl met Gorbachev, it was not clear that Moscow would allow Germany’s reunification to proceed after the elections in the GDR planned for March, but after Gorbachev’s concession and Kohl’s swift effort to publicize it that became a fait accompli. Had Gorbachev not made that concession and Kohl not ensured that Moscow could not walk it back, the Soviet Union could have obtained something in return for giving the green light to reunification, but now it would only be able to secure advantages in exchange for concessions it made, not about reunification itself, but about the conditions in which it could take place and in particular the external aspects of it.

Thus, as other Soviet officials lamented, Gorbachev had weakened his hand for the rest of the negotiations by throwing away a bargaining chip he could have used to secure more concessions from the West later.89 The other consequence is that, a few days later in Ottawa (where NATO and Warsaw Pact states convened for the Open Skies Conference), Shevardnadze accepted the Two Plus Four format to negotiate Germany’s reunification. This mechanism had merely been floated by Baker in Moscow, but Gorbachev had not accepted it, though he also had not rejected it. In Ottawa, Baker took advantage of the fact that the foreign ministers of all six nations concerned were present to mount a concerted effort to pressure Shevardnadze into accepting that mechanism, which the US expected to simplify the negotiations on Germany’s reunification by restricting them to the FRG and the GDR plus the four powers that held legal rights over Germany. Since Gorbachev had inadvertently given the green light for reunification without asking for anything in return, Shevardnadze was not in a good position to resist this effort, so he agreed and the Two Plus Four mechanism was adopted.90 The final agreement was still a long way ahead, but the outline of a deal had already started to emerge in Moscow.

1.2.1 What kind of deal was outlined during the preliminary talks in Moscow?

Everyone in the literature on this controversy agrees that Baker outlined a deal to Gorbachev in February by which the Soviet Union would agree to Germany’s continued NATO membership in return for the promise that afterward the Alliance would not expand to the east, but people disagree about whether he was only talking about the territory of the GDR or about Central and Eastern Europe in general and they also disagree on whether such a deal can be said to have been informally struck in Moscow. I will come back to this debate later, but for the moment the important point is that, on both sides of the debate, people interpret Baker as floating a quid pro quo by which the Soviet Union would have allowed Germany to stay in NATO after reunification in exchange for a restriction on further expansion. The problem is that, regardless of whether one takes Baker to have meant for that restriction to apply only to the territory of the GDR or to Central and Eastern Europe in general (I have just argued that he probably meant the former), this interpretation is flawed and I think has misled everyone in this debate about the logic of the negotiations on Germany’s reunification. Before I go over what happened during the Two Plus Four process, it’s therefore important to understand the nature of the deal outlined by Baker in Moscow, since otherwise one is bound to misinterpret what happened after that and how the final agreement was reached.

In order to understand why, it’s necessary to recall that, at the time, everyone understood that if Germany left NATO it would spell the end of a significant US military presence in Europe and, as a result, would effectively consign the Alliance to history. As we have seen, Baker even said as much during his conversation with Gorbachev in February. Moreover, unlike Baker’s talk of NATO’s “jurisdiction” not expanding to the territory of the GDR, Western officials did not drop that argument after February but on the contrary continued to make it throughout the Two Plus Four process. For instance, Margaret Thatcher, then British Prime Minister, made that point very clearly to Gorbachev during a meeting they had in June:

The American military presence in Europe is a stabilizing factor. But there is only one place where this presence is needed most: in Germany. France is not integrated into NATO's military chain of command, the Netherlands and Belgium are too small to station American troops there, Spain is too far away, and there are already enough American forces on British territory. That leaves Germany.91

So how could Soviet officials have believed that, unless the Soviet Union agreed to Germany’s continued NATO membership after reunification, the Alliance might expand to Warsaw Pact states? The notion that Baker was offering Gorbachev a quid pro quo by which the West would promise not to expand NATO in exchange for allowing Germany to stay in the Alliance only makes sense if, in the event that he refused, NATO could still have expanded. But as Baker himself told Gorbachev in the same conversation, the US would be forced to leave Europe and NATO would effectively be dead if Germany couldn’t remain part of it, so it obviously couldn’t have expanded.

It’s also clear that Gorbachev understood that since, during his meeting with Kohl the day after his conversation with Baker, he complained that the argument made by US and West German officials that without Germany NATO would disappear applied equally to the Warsaw Pact:

They say: what is NATO without the FRG. But we could also ask: what is the WTO [Warsaw Treaty Organization] without the GDR? This is a serious question. There should be no divergence in military matters. They say NATO will fall apart without the FRG. But it would be the end of the WTO without the GDR, too. If we agree about the main things, it is important that we are in accord here as well.92

Thus, in order to interpret Baker in the way both critics and defenders of the Russian argument do (setting aside their disagreement on whether Baker was only talking about the GDR or about Central and Eastern Europe in general, which is irrelevant to the point I’m making here), we’d have to assume that he was offering Gorbachev a completely nonsensical deal, asking him to make a concession in exchange for something that, if he refused to make the concession in question, had essentially no chance of happening. You don’t make a quid pro quo if, by refusing to give the quid, you automatically get the quo. It’s remarkable that, as far as I know, nobody has made this point before.

But if Baker and other Western officials weren’t offering a quid pro quo to their Soviet counterparts, what were they doing exactly? I think it’s clear when you read their conversations that they were trying to convince the Soviets that it was in the Soviet Union’s interest to make sure the US would stay involved in Europe, which in turn required allowing Germany to remain in NATO after reunification. In other words, Baker wasn’t so much proposing a quid pro quo as making the argument that Germany would be less dangerous to the Soviet Union inside NATO than outside of it, because keeping Germany tied to Western security structures would reduce the risk that it might go back to nationalist policies after reunification. Thus, the question he put to Gorbachev was less the outline of a quid pro quo than a way to make that clear by presenting the choice that Moscow faced, now that it was clear that Germany’s reunification was inevitable. The assurance that NATO would not expand to the east was not meant as compensation for allowing Germany to stay in the Alliance, which as we have seen would make no sense, but as a way to make this outcome even more appealing than the alternative of a Germany unconstrained by membership to the West’s security structures by promising that NATO would not take advantage of Germany’s reunification to expand its footprint in that country. More generally, as we shall see shortly, Western officials did everything they could to reassure the Soviets that allowing that to happen would not fundamentally alter the balance of power in Europe. In fact, Baker’s argument was in essence that, by allowing Germany to stay in NATO, the Soviet Union would ensure that reunification would result in as little change to the status quo as was possible under the circumstances.

I think construing Baker’s argument as a quid pro quo is more misleading than helpful, but insofar as the deal outlined by Baker to Gorbachev in Moscow can be described as a quid pro quo, what the West was promising the Soviet Union in return for allowing Germany to stay in NATO after reunification was not that NATO would not expand to the east, but that it would collaborate with Moscow to create an inclusive post-Cold War European order in which the Soviet Union would have a place. As we have seen, the Soviet Union could in theory have ensured that NATO would not expand simply by insisting as a condition for reunification that Germany should leave the Alliance, but in practice things were more complicated. In order to do that, Moscow would have had to play hardball in the negotiations and, although some of his advisers urged him to do so, Gorbachev was not willing to do so because achieving his main policy goals, both internal and external, required that he maintain a cooperative stance with the West.93 Indeed, internally he couldn’t achieve his goal of modernizing the Soviet economy without the flow of Western capital and technology, while externally his goal of building a “common European home” could not be achieved if he adopted a hard line in the negotiations on Germany’s reunification. The internal and external goals were connected, because the Soviet Union could not reform its economy without significantly cutting military expenditures, which in turn was impossible unless tensions with the West were durably reduced. Gorbachev also needed money and knew that the FRG could be induced to provide generous aid if he played nice, but while people often emphasize that point, I think it was very much a secondary issue because the truth is that, even if he’d played hardball in the negotiations, the Germans would still have paid to reunify their country and in fact they would probably have paid a lot more than what they eventually gave.94 But again I think it’s a mistake to construe the deal outlined in Moscow as a quid pro quo. If Gorbachev refused to play hardball in the negotiations, it’s not really because he was consciously thinking about the negotiations as a quid pro quo by which he would get the West’s cooperation with his ultimate policy goals in exchange for allowing Germany to stay in NATO after reunification, but more fundamentally because he thought Western-Soviet relations had moved beyond that and it didn’t fit with his conception of what they should be like going forward. As we shall see shortly, Western officials realized this and they used that fact to their advantage during the negotiations, which ultimately allowed them to settle the German question on their terms.

1.2.2 The Two Plus Four process and the Treaty on the Final Settlement with Respect to Germany

Bush and the NSC feared that, if Gorbachev offered Kohl to allow reunification to proceed in exchange for Germany’s withdrawal from NATO or at least from the Alliance’s integrated military command, Kohl might find it difficult to resist such a deal since it would mean taking the risk of missing a historical opportunity to reunify his country. It’s very likely that if Gorbachev had played hardball and pursued this kind of strategy, he could indeed have driven a wedge between the FRG and its allies. As we have seen, Bush had sent a letter to Kohl right before he met with Gorbachev in Moscow to repudiate Baker’s assurance that NATO’s “jurisdiction” would not expand to the territory of the GDR and advise that instead Kohl suggest to Gorbachev the idea of a special military status for that territory after reunification, but Kohl had opted for a language closer to Baker’s proposal because he deemed it more conducive to obtaining a green light for reunification from Gorbachev, so Bush’s fear was certainly justified. However, as I also argued, Gorbachev couldn’t easily pursue this kind of strategy in the negotiations on Germany’s reunification because, to achieve his main policy goals, he needed to maintain a cooperative stance with the West. Bush and Brent Scowcroft, his National Security Advisor, understood that and, to make sure that Kohl did as well and that Bonn would remain on the same page as Washington during the Two Plus Four process, they decided to invite him to Camp David at the end of February to argue that Germany’s reunification could be secured without compromising on NATO and stiffen his back so he wouldn’t be tempted to cut a deal with Gorbachev to secure reunification on Soviet terms. Kohl and his advisers were convinced, committed to stand firm on NATO membership and kept their word during the rest of the negotiations.95 In particular, they agreed to drop Baker’s “jurisdiction” talk about the GDR in favor of Bush’s “special military status” formula, although it took some time for the West German foreign ministry to do the same.96 In exchange for the promise not to compromise on NATO, Kohl received Bush’s assurance that he would have Washington’s full backing during the Two Plus Four negotiations. On March 18, his position was further strengthened by the victory of the CDU in the East German elections, which showed that he also had the support of the GDR’s population.97

As we have seen, the Soviets eventually took note of those developments and, in the note he sent to Gorbachev in April, Falin not only noted this change of stance and described it as a hardening of the West’s position but also pointed out Genscher’s isolation. As we shall see, this is also how it’s described today by critics of Russia’s claim that NATO expansion to Central and Eastern Europe violated the assurances that were made during the negotiations on Germany’s reunification, which ironically plays an important role in their argument against that position. I will discuss their argument later, but for the moment, what matters is that while in a sense the abandonment of Baker’s “jurisdiction” talk in favor of Bush’s “special military status” formula did represent a shift to a more uncompromising position, as we have seen it was arguably more of a clarification of Baker’s proposal than a hardening of it and, as I will now explain, this was followed in subsequent months by a broadening of the assurances that Baker had made in Moscow. As I argued above, insofar as Baker’s proposal can be construed as a quid pro quo, the idea was not for the Soviets to allow Germany to stay in NATO after reunification in return for a pledge that the Alliance would not expand to the east, but that if they agreed to Germany’s continued membership in NATO the US and its allies would build a post-Cold War European order together with the Soviet Union that would be cooperative and accommodate Soviet concerns. Baker’s somewhat unclear assurance that, in that case, NATO would not move into the territory of the GDR was just meant to convince Soviet officials that Germany’s continued membership in NATO was not intended to fundamentally alter the European balance of power in the West’s favor, but to preserve stability on the continent by allowing the US to stay involved in Europe.

It’s true that after February, Western officials quickly moved away from Baker’s formulation of the restrictions he proposed to place on NATO’s expansion to the GDR and rephrased his proposal in a way that seemed to weaken it, but in other respects they also strengthened it. Indeed, over the next few months, they made a number of more general assurances that played the same function of convincing the Soviets that if the US and its allies insisted that NATO could not be dismantled despite the end of the Cold War and that Germany had to remain part of it, it wasn’t because they were trying to take advantage of the situation to strengthen the West’s geopolitical position in Europe, but because it was the only way to make sure that US troops would stay on the continent and that Germany would remain tied to Western political and military structures, which throughout the Two Plus Four process never ceased to be the rationale presented by the US and its allies for their position. As I argued above, although Baker’s assurance that NATO would not expand to the east referred to the territory of the GDR, the deal he outlined in February can only be understood as part of a more general understanding on the European security architecture that would emerge from the post-Cold War settlement.98 After the preliminary talks in Moscow, Western officials narrowed down the assurance about NATO’s future relationship with the territory of the GDR by abandoning Baker’s “jurisdiction” talk, but they also made clearer the general nature of the deal they were proposing by making a number of broad assurances that went beyond the status of this territory and concerned the post-Cold War European security order as a whole. Tom Blanton and Svetlana Savranskaya are therefore correct to argue that the assurances made by US and West German officials in February were not isolated statements but part of a “cascade of assurances” that Western officials made to the Soviets before, during and after the Two Plus Four process.99 Those assurances were instrumental in securing a settlement of the German question on Western terms, because they appealed to Gorbachev’s desire to end bloc thinking and start a new era of cooperation in Europe, yet were compatible with the preservation of Western political and military institutions.

Going into the Two Plus Four process, the Soviet goal was to obtain Germany’s neutrality and the dismantlement of the Cold War alliances, which eventually would be replaced by a new pan-European security architecture based on a beefed-up CSCE, which had produced the Helsinki Final Act in 1975.100 Gorbachev wanted to turn the CSCE, which unlike NATO and the European Economic Community (EEC) included the Soviet Union and its allies, into a permanent organization that would replace NATO and the Warsaw Pact to become the central institution of the post-Cold War European security order. Germany’s reunification could take place as soon as Bonn had recognized the Oder-Neisse border with Poland, but it would have to withdraw from NATO and the four powers would keep troops in the country and only relinquish their rights after this new pan-European structure was in place, at which point both NATO and the Warsaw Pact would be dissolved.101 While Gorbachev had not closed the door on Baker’s proposal during the February talks in Moscow, he continued to insist on Germany’s neutrality and the replacement of Cold War alliances by this kind of pan-European security architecture for several months after that. Some political leaders in other European countries on both sides of the Cold War line, such as the French President François Mitterrand, the West German Foreign Minister Genscher and the Czechoslovak President Václav Havel, expressed interest in a framework of that sort at one point or another, but US officials had no intention of dismantling NATO and were determined to make sure that it would continue to play a major role in Europe after the end of the Cold War. As we have seen, Bush had made sure that Kohl would not compromise on this point by inviting him to Camp David in February to make the case that it would not be necessary to secure Germany’s reunification, but US officials still needed some alternative package to convince the Soviets.

Their general strategy consisted in formally decoupling the negotiations on Germany’s reunification from the broader question of the post-Cold War European security architecture, while informally making assurances on the latter to help secure a settlement of the former on Western terms. Thus, since the Soviets were anxious that a post-Cold War European security order that included the Soviet Union and didn’t leave it on the outside be created after Germany’s reunification, Western officials made repeated assurances to convince their Soviet counterparts that Moscow would not be excluded from the post-Cold War European security architecture. They insisted that, while they agreed with the goal of building a pan-European security architecture, this would take time, whereas it was no longer realistic to delay Germany’s reunification. It was important that NATO be preserved to allow the US to keep a significant military presence in Europe, so as to make sure that in the meantime Germany would remain tied to Western security structures, which in turn would ensure the stability of the continent. Eventually, the CSCE would be strengthened to create the kind of pan-European security structure that Gorbachev had talked about, but it was not realistic to create such a structure before reunification occurred and not desirable to leave Germany untethered to Western security structures while that pan-European system was being worked out. The Soviet approach to the latter problem, which as we have seen consisted in allowing the four powers to keep troops in Germany after reunification until a pan-European security architecture had been created, was rejected on the grounds that the Germans wanted the Soviet troops out. However, in order to make the continued existence of NATO as non-threatening as possible, Western officials insisted that it would become more political.102 In short, Western officials used informal assurances to signal as much agreement as possible with their Soviet counterparts on the ultimate goal (they were merely proposing a different path to reach the same point), so the treaty by which the four powers would relinquish their rights over Germany and allow reunification to proceed would not be tied to the broader issues posed by the post-Cold War settlement in Europe and be narrowly focused on borders, troop withdrawal and the military status of the territory of the GDR.

Those repeated assurances eventually overcame Gorbachev’s reluctance to allow Germany to stay in NATO. The first significant crack appeared during the Washington Summit held from May 30 to June 3, where Bush tricked Gorbachev into admitting that Germany had a right to choose its own alliances, by noting that the Soviet Union had formally recognized this principle by signing the Helsinki Final Act in 1975.103 The Soviets tried to walk it back and continued to resist NATO membership for a while after that, but the truth is that, as I explained above, Gorbachev was not prepared to adopt the kind of hardball tactics that would have been necessary to obtain Germany’s withdrawal from the Alliance. Thus, he eventually relented and accepted the principle during a meeting with Kohl in July, shortly after he was reelected General Secretary at the 28^th Congress of the CPSU and NATO’s heads of state and government issued a declaration that publicly repeated some of the assurances Western officials had made during the Two Plus Four process on how the Alliance would be transformed with the end of the confrontation between the blocs.104 However, it was not over yet, for Gorbachev still insisted that restrictions be placed on NATO’s freedom of action in the territory of the GDR after reunification, so the last stretch of the negotiations was devoted to that question. Specifically, he wanted to bar foreign troops and nuclear weapons from being deployed over there, but while the treaty that was signed in Moscow on September 12 barred nuclear weapons from being deployed in that area it didn’t really forbid foreign troops from entering it.105 Indeed, while Article 5 of the treaty said that foreign troops “will not be stationed in that part of Germany or deployed there”, a written addendum that was included at the insistence of the US stipulated that Germany could decide how to interpret the word “deployed” as long as it did so “in a reasonable and responsible way taking into account the security interests of each contracting party”, which effectively allowed foreign troops to cross the Cold War line. As Bush had predicted to Kohl at Camp David in February, Germany had been reunified on Western terms, without having to jeopardize NATO’s future by taking the country out of it or accepting a French-style membership and withdrawing from the integrated military command.

Besides the assurances they made during the negotiations, which I will examine in more detail shortly, other factors allowed Western officials to secure Germany’s reunification on their terms. First, the FRG promised significant economic assistance to the Soviet Union, in the form of loans, grants and food aid. As early as May 1990, before the treaty was signed or Gorbachev had even accepted allowing Germany to stay in NATO after reunification, the West German government agreed to guarantee a loan to the Soviet Union extended by private banks to the tune of 5 billion DM. Kohl wrote a letter to Gorbachev to inform him of that decision in which he called it “a considerable political effort” on the part of his government and said that he assumed that Moscow would negotiate “in the same spirit” within the Two Plus Four framework to bring about “a constructive solution”.106 In September, to ensure the signature of the treaty, Kohl further agreed to pay 12 billion DM to finance the transfer of Soviet troops out of Germany as well as to extend a 3 billion DM loan.107 But as I already argued, economic assistance was not as important in securing the eventual deal as both contemporaries and historians have claimed, since Germany would almost certainly have paid much more if Gorbachev had played hardball. Another important factor in securing the eventual agreement was the fact that negotiations on the Treaty on Conventional Armed Forces in Europe (CFE), which resulted in a large reduction of conventional military forces in Europe, were conducted in parallel to the Two Plus Four process. Together with the assurance that NATO would take on a less military character and become more political in the post-Cold War environment, this contributed to reducing the Soviet perception of the threat posed by NATO, which in turn made it easier for Gorbachev to make the concessions that made a deal possible. Finally, the facts that Gorbachev’s program of reforms was approved by the 28th Congress of the CPSU in July and that he was reelected General Secretary of the Party had a similar effect, because it meant that he was less vulnerable to domestic criticisms.108 Indeed, NATO was already a very sensitive issue in the Soviet Union at the time and has remained so in Russia ever since, which made Gorbachev and every Russian leader after him wary of appearing weak in front of the Alliance.109 Thus, many factors played a role in the eventual agreement that allowed Germany to stay in NATO after reunification, but none of the other factors that contributed to that outcome diminish the importance of the assurances that Western officials made on the future of the European security architecture during the negotiations.

1.3 The case for the Russian position

As critics of the Russian position note, the Treaty on the Final Settlement with Respect to Germany did not say a word about NATO expansion to Central and Eastern Europe, but was exclusively concerned with the conditions under which Germany would reunify and the four powers would relinquish the legal rights they still held as a result of the occupation at the end of WWII. They argue that, whatever Soviet officials were told in February, it was superseded by the eventual agreement, so the US and its allies did not make any commitment not to expand NATO since the treaty didn’t include any provision about that. However, it doesn’t follow that the West did not have some obligation not to expand NATO to Central and Eastern Europe as a result of what transpired during the negotiations on Germany’s reunification, because the controversy was never about the content of the treaty but about what was said during those negotiations. Indeed, despite the fact that some people sometimes pretend otherwise, nobody is denying that the Treaty on the Final Settlement with Respect to Germany does not explicitly rule out NATO expansion. The debate is and always has been about whether the West made some kind of commitment during the negotiations that, while not written down in the treaty, was nevertheless incompatible with NATO expansion to Central and Eastern Europe against Russia’s wishes. It shouldn’t be controversial that one can’t settle this debate simply by examining the provisions of the treaty. Indeed, everybody agrees that money was part of the deal that allowed Germany to be reunified on Western terms, yet the treaty doesn’t say anything about financial assistance.110 I will argue that, in addition to promising the Soviets money to ensure that Germany could stay in NATO after reunification, the US and its allies made assurances during the negotiations and more broadly at the end of the Cold War that created some obligation not to expand NATO against Russia’s wishes, although reasonable people can disagree about how binding that commitment was.

Shifrinson argues that the assurances made in February were not superseded by subsequent developments and remained binding because, even if Western officials later changed their position, they had already obtained something from the Soviets in return for those assurances, namely Gorbachev’s assent to let Germany’s reunification proceed and to formally start the negotiations on the conditions under which it could take place, so they were obligated by the assurances made in Moscow.111 Since he also believes that the assurances made in Moscow by US and West German officials did not just concern the territory of the GDR but applied more broadly to Central and Eastern Europe, he concludes that the West had some obligation not to expand NATO against Russia’s wishes. Trachtenberg is somewhat more cautious about the scope of the assurances made in Moscow, but on the whole he agrees with Shifrinson that, in the case of Genscher and Baker at least, they applied to Central and Eastern Europe in general and not just the territory of the GDR. His argument that the assurances in question were not superseded by later developments is that, if those assurances pertained to Central and Eastern Europe as a whole and not just the territory of the GDR, then since the Two Plus Four agreement only concerned the latter, the further concessions subsequently made by the Soviet Union and enshrined in that treaty could only have superseded the part of the assurances that were about the GDR and left the rest intact.112

Thus, Shifrinson and Trachtenberg both argue that some kind of informal agreement was implicitly struck in Moscow when US and Western officials met with their Soviet counterparts in February, which did not cease to apply later because of what happened during the rest of the negotiations on Germany’s reunification and in particular was not superseded by the treaty signed a few months later.113 However, they acknowledge that it was not as binding as, for example, the informal agreement struck between Kennedy and Khrushchev by which the former agreed to withdraw Jupiter missiles from Turkey and pledged not to invade Cuba in return for the latter withdrawing Soviet missiles from the island, because the quid pro quo was less explicit. In any case, the argument hinges in both cases on the assumption that the assurances made in February were not just about the territory of the GDR, but about Central and Eastern Europe in general.114 However, as I argued above, the evidence supports the view defended by Kramer and most other scholars in the literature that, with the exception of Genscher, US and West German officials were only talking about Germany in Moscow. Therefore, neither Shifrinson’s argument nor Trachtenberg’s shows that Western countries were under some obligation not to expand NATO to Central and Eastern Europe against Russia’s wishes, at least as a result of the assurances made in February by US and West German officials.115 Yet despite what critics of the Russian position claim, it doesn’t follow that no such obligation existed, only that it couldn’t have been created by those assurances.

Critics of the Russian position argue that, when he met in February with US and West German officials, Gorbachev still hoped that he could obtain Germany’s withdrawal from NATO in return for the Soviet Union’s acquiescence to reunification and therefore didn’t take the deal that Baker floated in Moscow since it allowed Germany to stay in NATO. However, because the West subsequently took a harder stance on the question of NATO’s relationship with the territory of the GDR, he eventually had to sign a less favorable agreement. Baker had already made this argument in 1997, during the debate that preceded the Senate’s decision to approve NATO expansion to Poland, the Czech Republic and Hungary:

''I got off the word 'jurisdiction' very quickly,'' Mr. Baker said in a telephone interview. ''I do not recall using it with the Soviets. But let's assume I did use it once or twice. We quickly walked away from it. What defeats this whole argument is that we then insisted on the G.D.R. being in NATO, thereby moving NATO eastward.''116

In the scholarly literature on the controversy, it was made in a particularly clear way by Sarotte:

For a moment in February 1990, the Soviet Union could have struck a deal with the United States, but it did not. Obviously any agreement among the Americans, West Germans, and Russians would have needed alliance approval, but in the political climate of 1990 it would have been possible to secure it. Even a written press release would have helped the Soviet cause. But Gorbachev did not secure one, and the window closed. Germany united and NATO began to move eastward.117

In other words, for a commitment not to expand NATO to Central and Eastern Europe to have existed, Gorbachev should have taken the deal that was briefly floated in February, but since he didn’t and eventually signed a treaty that didn’t contain any provisions against NATO expansion to Warsaw Pact members and even authorized NATO to move across the Cold War line, no such commitment existed and therefore Western countries didn’t renege on any promise they had made when they expanded NATO a few years later.118

This argument portrays the deal that was floated by Baker in February 1990 as a quid pro quo by which the Soviet Union agreed to let Germany stay in NATO after reunification in return for restrictions on the kind of relationship the Alliance would have with the territory of the GDR for the Alliance. While in February, so the argument goes, US and West German officials had floated the idea that the restrictions in question might go as far as preventing NATO’s “jurisdiction” from extending to it, whatever this could have meant exactly, Western officials had subsequently revised the terms of the deal by proposing only a special military status for the territory of the GDR and the Soviet Union had ultimately agreed to that. However, as I argued above, this way of interpreting the deal that was first outlined in February and that eventually led to Germany’s reunification within NATO is wrong. The restrictions on the sort of relationship that NATO could have with the territory of the GDR was not something the Soviet Union would get in return for allowing Germany to stay in NATO, which would have made no sense since they could have prevented NATO from having any relationship whatsoever with that territory simply by demanding that Germany withdraw from the Alliance, it was something that was intended to make what it was giving less onerous. What the Soviet Union was supposed to get in return, to the extent that it even makes sense to construe the deal as a quid pro quo, was a general post-Cold War settlement that would be inclusive and not leave Moscow at the door. As I noted above, Western officials made repeated assurances after February to convince their Soviet counterparts that, although the settlement of the German question could not be tied to the more general issue of the post-Cold War European security architecture for both practical and principled reasons, a pan-European system broadly consistent with Gorbachev’s ideas on that issue would eventually be created. Even before the negotiations on that issue started, when they met off the coast of Malta in 1989, Bush had already tried to reassure Gorbachev that the US was not trying to exploit the Soviet Union’s difficulties at home and in Central and Eastern Europe, a point that Baker had stressed again in Moscow when he assured Gorbachev that “the President and I have made clear that we seek no unilateral advantage in [the German unification] process”.119 This effort to convince Soviet officials that Germany’s reunification and more generally the post-Cold War settlement would not be to Moscow’s detriment intensified after the February talks.

On March 2, during a meeting with Anatoly Adamishin (Deputy Foreign Minister of the Soviet Union), Kastrup assured him that Bonn “did not want to derive unilateral advantages from the changes in the East” and that it “wanted to create European security structures with the [Soviet Union], its friends and with [the FRG’s] friends, which would make alliances superfluous”. He also said that “NATO should have an increasingly strong political role”, that it would be “instrumental in developing cooperative security structures that would initially act as an umbrella over the two alliances”, adding that “the alliances would eventually be absorbed into these cooperative security structures”.120 On April 10, during a meeting with Gorbachev, who repeated his idea of a pan-European security structure that would replace Cold War institutions, Hurd assured him that in the future NATO would have “a more political role” and that the UK “recognized the importance of doing nothing to prejudice Soviet interests and dignity”.121 On May 4, according to a report he wrote for Bush on the meeting he had with the Soviet Foreign Minister, Baker told Shevardnadze that the US recognized “the need to adapt NATO, politically and militarily, and to develop CSCE” to reassure Shevardnadze that “the process would not yield winners and losers”, but would produce a European security order that is “inclusive, not exclusive”.122 On May 18, in a meeting with Gorbachev and Shevardnadze, Baker listed nine points that were supposed to illustrate how the US had taken into account the Soviet Union’s “completely legitimate concerns” in developing its policy on Germany. These included the assurances that “NATO will undergo an evolution to become more of a political organization” and that the US was making an effort to “ultimately transform the CSCE into a permanent institution that would become an important cornerstone of a new Europe”, though he argued that NATO was still needed to anchor Germany into Western institutions after reunification because creating a pan-European security structure based on the CSCE would take time and for the moment was “a wonderful dream, but just a dream”. Moreover, Baker told Gorbachev and Shevardnadze that he and Bush were “always emphasizing that [they were] not seeking unilateral advantages from the changes happening in the Soviet Union”, that US policy was “not aimed at separating Eastern Europe from the Soviet Union” and that Americans were “interested in building a stable Europe, and doing it together with you”.123

On May 25, in a meeting between Mitterrand and Gorbachev, not only did the French President tell his Soviet counterpart that the CSCE should have more weight in the post-Cold War European security architecture and compare his idea of a European confederation to Gorbachev’s concept of a “common European home”, but he also told him that he was “personally in favor of gradually dismantling the military blocs”. Furthermore, he acknowledged that it was necessary to “create security conditions for you, as well as European security as a whole” and claimed that “European security [was] impossible without the USSR”.124 On May 31, during his meeting with Gorbachev in Washington, Bush assured him that “we have no intention, even in our thoughts, to harm the Soviet Union in any fashion”.125 On June 8, according to a British memorandum summarizing their conversation, Thatcher told Gorbachev that “the CSCE could provide the umbrella for [various security issues they were discussing] as well as being the forum which brought the Soviet Union fully into discussion about the future of Europe” and that given the consequences of German unification “we must find ways to give the Soviet Union confidence that its security would be assured”.126 On July 6, NATO members issued a declaration at the end of the London Summit, in which among other statements aimed at diffusing Soviet fears they stated their intention “to enhance the political component of our Alliance” and said that the CSCE “should become more prominent in Europe's future” before making several proposals to institutionalize it.127 On July 16, in a speech in front of the Supreme Soviet of the USSR, NATO Secretary General Manfred Wörner also expressed his support for a larger role for the CSCE in the post-Cold War security order, denied that NATO sought a unilateral advantage from German reunification and claimed that it would not “shift the balance of power in Europe”.128

To be sure, those statements were vague and didn’t specifically mention NATO expansion (I will come back to this point later), but it doesn’t mean that it was unreasonable for the Russians to deem them misleading in light of what subsequently happened. On the contrary, I think that, based on what they were told during the negotiations on German reunification and more generally at the end of the Cold War, Russian officials have every reason to feel that the West didn’t hold up its end of the bargain. The assurances made during the Two Plus Four process were sufficiently vague that one could perhaps argue that each of them individually was respected, but it would be disingenuous to pretend that the Russian position has no basis in fact because in diplomacy the whole is not just the sum of the parts and collectively those assurances painted a picture of a post-Cold War European security architecture that was very different from the NATO-centric order that was created after the fall of the Soviet Union. Allowing Germany to remain in NATO after reunification, Soviet officials were told repeatedly, was not the first step toward creating a post-Cold War European security order dominated by NATO. The Alliance would become a more political organization and, in due time, a pan-European security architecture would be created in which a transformed CSCE would play the central role and NATO would merely be one pillar among others under that umbrella organization. But this would not happen overnight and, in the meantime, it was critical that NATO continue to play a role in Europe and that Germany remain part of it after reunification. Otherwise, the US would be unable to keep a significant military presence in Europe, NATO would become irrelevant or even cease to exist and the US would no longer have any political influence in Europe, which in turn would destabilize the continent.129 Allowing Germany to stay in NATO after reunification would not result, the Soviets were repeatedly assured, in a shift of the balance of power on the continent. On the contrary, it was the only way to preserve the status quo after German reunification and prevent this event from having a destabilizing effect, because a united Germany that was no longer constrained by Western security structures would take steps to ensure its security such as developing nuclear weapons and might go back to the kind of nationalism that caused so much damage in the 20^th century. It was in everyone’s interest, including and perhaps even especially the Soviet Union’s, that Germany be allowed to stay in NATO after reunification.

So it’s unsurprising that, when a few years later the US and its European allies built a NATO-centric European security architecture in which the CSCE played only a minor role (even though as we shall see it was turned into a permanent organization in 1995), from which Russia was largely excluded since it was not part of the Alliance and in which NATO not only remained primarily a military organization but even conducted military operations “out of area” for the first time in history without the authorization of the United Nations Security Council and against a historical Russian ally, Russian officials felt they had been misled. Indeed, not only is it not surprising that they felt that way after what they were told, but it would have been surprising if they had not felt that way. Western officials no doubt believed they had good reasons to subsequently make those decisions, and they may even be right about that (although I will argue that they are not), but this doesn’t make it any less true that, based on the assurances Soviet officials received at the end of the Cold War, the Russians had every reason to expect something very different. This argument doesn’t rest on the assumption that, when they made assurances about NATO expansion in February, US and West German officials were talking about more than the territory of the GDR. As we have seen, except for Genscher, they probably were not. Nor does it rest on the assumption that, at any point during the negotiations on Germany’s reunification, Western officials explicitly promised that if the Soviet Union agreed to let Germany stay in NATO the Alliance would not be expanded to Warsaw Pact members. As we shall see shortly, some of them did assure their Soviet counterparts that it wouldn’t happen a few months later, but this doesn’t seem to have happened during the negotiations on Germany’s reunification and in particular it didn’t happen during the February preliminary talks in Moscow because, with the exception of Genscher, US and West German officials were only talking about the territory of the GDR. Nevertheless, given the assurances Soviet officials received from their Western counterparts during the negotiations on Germany’s reunification and more generally at the end of the Cold War, the Russians had every reason to see the European security architecture that was created by the West in subsequent years and in particular NATO expansion as a form of betrayal. As Shifrinson argued, insofar as the West got something of value out of those assurances (which they did since as I have argued they played a key role in convincing the Soviets to allow Germany to stay in NATO after reunification and more generally to settle the Cold War in terms favorable to the West), they were to some extent binding.130 This is a relatively weak claim, whose import the Russians have no doubt exaggerated (especially in recent years), but that is not the same thing as saying they made up the whole thing and pretending they are is just adding insult to injury.

Indeed, it didn’t take long for the Soviets to start feeling they had been misled by those assurances, since they started to express concerns along those lines only a few months after the conclusion of the Two Plus Four process. During this period, the faction that thought Gorbachev had been too conciliatory with the West started to gain influence in the Soviet administration, which among other reasons led to the resignation of Shevardnadze in December 1990 and would eventually result in a coup against Gorbachev in August 1991.131 When John Major, who had replaced Thatcher as Prime Minister of the UK in November 1990, came to Moscow on 5 March 1991, Gorbachev complained to him that the assurances the Soviet Union had received during the negotiations on Germany’s reunification were already being violated:

Against the background of favorable processes in Europe, I suddenly start receiving information that certain circles intend to go on further strengthening NATO as the main security instrument in Europe. Previously they talked about changing the nature of NATO, about transformation of the existing military-political blocs into pan-European structures and security mechanisms. And now suddenly again [they are talking about] a special peace-keeping role of NATO. They are talking again about NATO as the cornerstone. This does not sound complementary to the common European home that we have started to build.132

Major replied that it was a misunderstanding and that Western countries were merely trying to increase coordination between NATO and the Western European Union, but Gorbachev clearly wasn’t convinced.133

According to Braithwaite, during Major’s visit, the Soviet Minister of Defense Dmitry Yazov also expressed his concern that Czechoslovakia, Poland and Hungary wanted to join NATO, but Major assured him that “nothing of the sort will happen”.134 On March 26, Hurd repeated this assurance a third time when he told Aleksandr Bessmertnykh, the Soviet Foreign Minister, that “there are no plans in NATO to include the countries of Eastern and Central Europe in NATO in one form or another”. This was arguably true at the time, though as we shall see some people in the Bush administration were already thinking about it, but it would not remain true for long.135 Finally, on June 16, Wörner declared that “granting NATO membership to former Warsaw Treaty members would be a serious obstacle to reaching mutual understanding with the Soviet Union”.136 Thus, while it’s true that, except for Genscher, Western officials didn’t explicitly rule out NATO expansion to Central and Eastern Europe during the negotiations on Germany’s reunification, some of them did a few months later. But as I have argued, in doing so, they were just making explicit a commitment that was already implicit in the assurances made during the negotiations on German reunification, which as the concerns voiced by Soviet officials in 1991 show was clearly interpreted by them as having that implication. To be sure, those statements made by Western officials in 1991 to the effect that NATO would not be expanded to Central and Eastern Europe did not constitute a quid pro quo of the sort that is often described by defenders of the Russian position, but rather assurances that no such plan existed. Again, this was true at the time, but as we shall see later this doesn’t mean that Western officials were not somewhat disingenuous in their attempts to mollify their Soviet counterparts.

I think there are several reasons why people resist this conclusion, even when they know about the assurances Western officials made to their Soviet counterparts at the end of the Cold War. First, in part because Russian officials have used their complaint that the West violated those assurances as a justification for some of their most controversial decisions in the past 30 years, people can’t help but conflate the question of whether this complaint has merit with the question of whether it justifies Russia’s subsequent actions. However, even if as I’m arguing the Russian claim that NATO expansion violated the assurances Western officials made at the end of the Cold War is not without merit, it doesn’t follow that Russia’s subsequent actions, such as the invasion of Ukraine, were thereby justified. One can acknowledge that the Russians have good reasons to feel they have been misled while criticizing their recent policy. Another reason is that, while the debate has focused on the narrow issue of NATO expansion, the problem with the West’s post-Cold War policy for Russia was not so much NATO expansion per se but Moscow’s exclusion from the European security architecture. As I will argue later, not only was NATO expansion but one aspect of that exclusion, but if the West had adopted a different approach it may have been possible to expand NATO without excluding Russia from the post-Cold War European security architecture, even though it made that more difficult. Now, the argument I have made above is that, to the extent that the assurances made by Western officials at the end of the Cold War created some obligation not to expand NATO, this obligation stemmed from the promise contained in those assurances to create a post-Cold War European security architecture in which Moscow would have a place and that would not be dominated by NATO. Thus, by focusing the debate more narrowly on the issue of NATO expansion, both critics and defenders of the Russian position have made it harder to see the relevance of those assurances. Finally, because they interpret Baker’s February assurance that NATO would not move “one inch to the east” as a reference to Central and Eastern Europe in general and not just the GDR, defenders of the Russian position and sometimes even Russian officials themselves argue that Western officials explicitly promised not to expand NATO if the Soviet Union allowed Germany to stay in NATO after reunification. Since this can be shown to be false, critics of the Russian position naturally concluded that Moscow had no case, but as I have argued the fact that Western assurances did not explicitly preclude NATO expansion to Central and Eastern Europe doesn’t mean that it wasn’t in tensions with those assurances. Thus, just because the obligation not to expand NATO stemmed from assurances that were less explicit than portrayed by the defenders of the Russian position, it doesn’t follow that Soviet and later Russian officials didn’t have good reasons to expect that NATO would not be expanded against their wishes based on those assurances and that as a result they didn’t have legitimate causes for complaint once the US and its allies started to take steps to expand NATO to Central and Eastern Europe despite Moscow’s opposition.

1.4 Other arguments made by critics of the Russian position

While they are wrong that every Western official who ever promised that NATO would not expand to the east was only talking about the territory of the GDR, since as we have seen Genscher was clearly talking about Central and Eastern Europe in general, critics of the Russian position are probably right that except for him, US and West German officials present in Moscow during the February talks were only talking about the territory of the GDR. In particular, even though out of context the language Baker used when he assured Gorbachev that if Germany was allowed to stay in NATO after reunification the Alliance would not move “one inch to the east” is more naturally interpreted as referring not just to the territory of the GDR but to Central and Eastern Europe in general, the context strongly suggests that he was in fact only talking about the former and there is evidence that it’s also how Soviet officials understood him. However, I have just argued that in spite of this fact, Western officials nevertheless made several broader assurances at the end of the Cold War that gave Soviet and later Russian officials every reason to expect that NATO would not be expanded to Central and Eastern Europe against their wishes and more generally that a very different European security architecture would emerge than the NATO-centric system the US and its allies eventually created. But in addition to arguing that US and West German officials were only talking about the territory of the GDR during the February talks in Moscow, critics of the Russian position have also made several other arguments that could also be used against the position I defended above, to which I now turn.

1.4.1 The treaty didn’t void the assurances made during the negotiations but arguably strengthened them

As we have seen, critics of the Russian position often argue that since the Treaty on the Final Settlement with Respect to Germany doesn’t preclude NATO expansion to Central and Eastern Europe, the West can’t have been under any obligation not to expand NATO.137 This argument implicitly assumes that, for the West to have contracted some obligation not to expand NATO to Central and Eastern Europe as a result of the conditions in which Germany was reunified, the treaty must have contained a provision that clearly prohibited it. In turn, critics of the Russian position often justify this assumption by implying that, unless assurances are explicit and written down in a legally binding treaty, they are not binding in any way. However, although agreements that are codified in a legally binding treaty obviously have more force, the view that informal and non-written assurances have no binding force whatsoever is completely untenable. As Shifrinson and Trachtenberg have pointed out, diplomats make informal assurances all the time and, if their interlocutors couldn’t assume that they have some binding force, diplomacy would become virtually impossible.138 Indeed, unless people could assume that what is said in top-level meetings has some political weight, and that people who take part in such meetings can put some trust in what they tell each other, verbal exchanges could not play anything like the role they do in actual diplomatic practice. Many diplomatic breakthroughs have only become possible because of conciliatory gestures that were not codified in any formal agreement, but created the trust necessary for a political solution to emerge, whether it was codified in a treaty or not. Clearly, if people didn’t assume that such gestures constrained the states on behalf of which they were issued to some extent, this would not be possible. Moreover, there are many cases where such a formal agreement is not possible for political reasons, so political leaders rely on informal and non-written assurances. Indeed, some of the worst diplomatic crises, such as the Cuban missile crisis, were resolved through informal agreements. This argument seems to rest entirely on the obvious fallacy that, since the assurances made by Western officials during that period were not codified in a treaty and therefore were not legally binding, they had no binding force whatsoever.

As Trachtenberg argued, precisely because the treaty was narrowly focused on the direct consequences of Germany’s reunification and in particular on the special military status that the territory of the GDR was to have after reunification, assurances made during the negotiations that were not strictly limited to that question cannot have been superseded by the treaty.139 Since as I have argued Western officials made broad assurances that created some obligation on the part of the US and its allies not to expand NATO to Central and Eastern Europe against Moscow’s wishes, the fact that the treaty did not explicitly prohibit this development can’t have abrogated that obligation, which remained intact. In fact, precisely because the treaty contained provisions that satisfied the Soviet condition that the territory of the GDR have a special military status (even if as we have seen it was somewhat watered down compared to their initial demand), the treaty arguably strengthened the obligation not to expand NATO to Central and Eastern Europe. Indeed, it would make absolutely no sense for the Russians to be very concerned about NATO expansion to the territory of the GDR and demand some kind of special status for it should they allow Germany to stay in NATO after reunification, yet not to care about NATO expansion to Central and Eastern European countries. Thus, the fact that the West consented to such a special military status for the territory of the GDR in the treaty, even if the restrictions on NATO’s freedom of action in that territory were more limited than what Soviet officials had initially envisioned, would naturally have strengthened the Russian expectation, created by the assurances made by Western officials at the end of the Cold War, that the US and its allies would not take the much more radical step of expanding NATO to Central and Eastern Europe a few years later. This made it even more natural for the Russians to feel betrayed when, after being told that we understood their concerns and giving them assurances that although Germany would stay in NATO after reunification there would be restrictions on what the Alliance could do on the territory of the GDR, they woke up a few years later to find that NATO had expanded to most of their former allies and was now on their borders. It’s true that, as Sarotte points out, the added minute of the treaty they eventually signed limited the scope of those restrictions by effectively allowing foreign NATO forces to be deployed on the territory of the GDR if the German government agreed, but it also stipulated that it should only be done “in a reasonable and responsible way taking into account the security interests of each contracting party” and that in any case no such forces should not be stationed over there. This clearly indicated that the FRG, the US, France and the UK acknowledged the security concerns of the Soviet Union and recognized that crossing the Cold War line should only be done in exceptional circumstances.

Thus, while it’s true that expanding NATO to Central and Eastern Europe didn’t violate the letter of the treaty, it’s hard to deny that, as Yeltsin told Clinton in the letter he sent him in September 1993 and as the Russians have consistently maintained since then, it violated its spirit and more generally the spirit that had presided over the multi-pronged negotiations that ended the Cold War peacefully. One cannot seriously claim that Soviet officials insisted that Germany would only be allowed to stay in NATO after reunification on the condition that restrictions be placed on what the Alliance could do in the territory of the GDR, but that had they known at the time that it would eventually happen, they would have been fine with NATO expansion to Central and Eastern Europe over their objections and without any similarly binding restrictions on what it could do over there.140 So the fact that NATO scrupulously respected the letter of the treaty that settled Germany’s status, which nobody denies, is of no relevance to the Russian argument that it violated its spirit. This argument is not only correct but obviously so and it certainly isn’t, as some have claimed, a “rewriting of history”.141 As we shall see, US officials responded to the Russian protest against NATO expansion a few years later by arguing that the GDR was a special case due to the legal rights that the Soviet Union still held over Germany as a result of World War II, but that Moscow had no such rights over Central and Eastern Europe and that unlike in the case of the GDR the US and its allies were therefore under no obligation to obtain Russia’s assent to expand in that area. However, while it’s true that the need for the Soviet Union’s assent in the case of Germany’s continued membership to NATO after reunification stemmed in part — though not only — from those legal rights, Russian officials were not for the most part making a legal argument and the negotiations record for the Treaty on the Final Settlement with Respect to Germany shows very clearly that Western officials not only understood that the Soviets saw NATO moving to the east as a security threat but also regarded this concern as legitimate and said as much to their Soviet counterparts on multiple occasions.142 Thus, it’s simply not true that, during the negotiations on the treaty, Western officials tied the legitimacy of the Soviet concerns that led to the special military status of the GDR exclusively to the legal rights Moscow held over Germany as a result of World War II.

1.4.2 The fact that Soviet officials didn’t raise the issue of NATO expansion to Central and Eastern Europe doesn’t mean the West didn’t have some obligation not to do it

Another common argument, which is related to the point that the treaty did not contain any prohibition against NATO expansion to Central and Eastern Europe, is that Soviet officials never raised the issue during the negotiations on Germany’s reunification, which critics of the Russian position seem to think implies that nothing Western officials said during those negotiations could have created any obligation not to expand NATO. This argument implicitly rests on the assumption not only that Soviet officials never raised the issue of NATO expansion to Central and Eastern Europe and demanded explicit guarantees against it during the negotiations, but also that unless Soviet officials did so the West can’t possibly have said anything that created some obligation not to expand NATO. Now, while the first assumption is mostly (though not entirely) true, the second is clearly false. Let’s start with the first point. It’s true that Soviet officials did not insist on specific guarantees against NATO expansion to Central and Eastern Europe during the Two Plus Four process and that discussions focused on whether Germany could stay in NATO after reunification, then on what restrictions the Alliance would face in the territory of the GDR after Gorbachev accepted the principle of Germany’s continued membership to NATO, but it’s not entirely true that they never brought up the issue and expressed their concerns about it. Indeed, during his conversation with Baker in Moscow on May 18, Gorbachev complained to him that he “had information that the goal of your policies [was] to separate Eastern European countries from the Soviet Union”.143 He mentioned this part of his conversation with Baker during his meeting with Mitterrand a few days later, when he said to the French president that he’d told Baker that he was aware of Washington’s “favorable attitude towards the intention expressed by a number of representatives of Eastern European countries to withdraw from the Warsaw Pact in order to subsequently join NATO”.144 However, those were pretty isolated comments, so the claim made by critics of the Russian position is mostly correct. Of course, even if they were concerned about the possibility that NATO might expand to Central and Eastern Europe (which they were), Soviet officials had good reasons not to explicitly raise the issue at the time.145 Indeed, one has to keep in mind that although Soviet officials realized that the Warsaw Pact would not survive in its current form, they were still hoping to save part of it. By insisting on guarantees against NATO expansion to Central and Eastern Europe, they would have effectively admitted that it was condemned, which would have made it harder to do that.

Now, if the Soviets had just avoided raising the issue of NATO expansion because it would have made their goal of saving what they could of the Warsaw Pact harder, it would have no bearing on the argument made by critics of the Russian position, because at the end of the day it’s still true that, except for a few isolated comments, they did not mention that issue and didn’t ask for guarantees against that possibility. However, if Soviet officials didn’t feel the need to raise the issue explicitly and demand specific guarantees against NATO expansion to Central and Eastern Europe, it’s also in part because as I have argued the assurances made by Western officials indirectly precluded that possibility. Indeed, despite what critics of the Russian position implicitly assume, the fact that Soviet officials almost never raised the issue during the negotiations and in any case didn’t insist on explicit guarantees against it doesn’t imply that Western officials can’t have said anything that created some obligation not to expand NATO against Moscow’s wishes and in fact they did. The fact that Soviet officials didn’t make such a demand doesn’t mean that, to the extent that assurances made by Western officials created some obligation not to expand NATO, the Russians wouldn’t remember them when NATO expanded anyway a few years later and wouldn’t be justified in retrospect to deem that Western countries had misled them during the negotiations on German reunification. In fact, as long as I’m right that assurances made by Western officials in 1990 created some obligation not to expand NATO to Central and Eastern Europe, it doesn’t even matter whether the Soviets were even thinking about that possibility in the moment. They would have been just as justified in deeming that Western countries had violated the assurances they made at the end of the Cold War even if that possibility had not even occurred to them at the time. Indeed, because people can’t think about every implication of the promises they make or receive, they often make promises that create obligations the recipients don’t think about at the time they are made, but it doesn’t release them from the obligations in question for all that. This is particularly true with vague assurances such as those made by Western officials during the negotiations on Germany’s reunification, which precisely because they’re vague have many implications people typically don’t think about, although their vagueness also makes them less binding. Finally, it’s worth pointing out again here that although Soviet officials didn’t insist on explicit guarantees against NATO expansion to Central and Eastern Europe during the negotiations on Germany’s reunification and only raised the issue on a few occasions, several Western officials did explicitly tell them it would not happen in 1991. Again, this wasn’t a pledge in the context of a quid pro quo, but it happened because Soviet officials expressed their concern that the West was going to expand NATO and made it clear that they thought it was incompatible with the assurances they had received a few months earlier, which Western officials did not dispute.

1.4.3 Moscow’s acceptance of the Helsinki principle didn’t void the obligation created by those assurances

Critics of the Russian position also argue that, even if the assurances made by Western officials created some obligation not to expand NATO, this obligation was superseded by the recognition by Gorbachev of the principle, enshrined in the Helsinki Final Act of the CSCE (which the Soviet Union signed in 1975), that sovereign countries are free to enter into alliances as they see fit.146 In May 1990, during the Washington Summit, Bush told Gorbachev that since he accepted that principle he also had to accept that if Germany wanted to stay in NATO after reunification he couldn’t object to it.147 At this point, to the surprise of US officials and the consternation of Gorbachev’s advisors, he reluctantly conceded the point.148 The Soviets immediately tried to walk it back, but soon after that Gorbachev stopped opposing Germany’s membership of NATO and started to focus on obtaining restrictions on the Alliance’s freedom of action on the territory of the GDR after reunification. This principle was once again endorsed implicitly by the Soviet Union in November 1990, when it signed the Charter of Paris just one month after the signature of the Two Plus Four Agreement. The Charter of Paris, which reaffirmed and expanded the principles stated in the Helsinki Final Act, would become a foundational document for the Organization Security and Co-operation in Europe (OSCE), which replaced the CSCE in 1995.149 So critics of the Russian position argue that Russia, which as the Soviet Union’s continuation state took on all its commitments, also has no grounds to object to the accession to NATO by Central and Eastern European countries a few years later. This argument, that Russia can’t object to NATO expansion because Central and Eastern European states are sovereign countries and therefore free to choose their own alliances, has become ubiquitous but it’s obviously fallacious. Indeed, while it’s true that Central and Eastern European states have a right to ask to join NATO, they don’t have a right to join NATO simply because they want to, for the simple and obvious reason that current members also have the right to refuse. Thus, if the Russians are right that Western countries were under some obligation not to expand NATO because of the assurances they had made during the negotiations on Germany’s reunification and the implicit recognition of the security concerns that NATO expansion posed to Russia contained in the Two Plus Four Agreement (which I have argued they are), it’s impossible to claim that Russia has no grounds to protest against NATO expansion simply because sovereign countries have a right to seek out alliances as they see fit, which is neither here nor there.

The key issue is therefore whether the Two Plus Four Agreement together with the assurances made during the negotiations that led to it created some obligation not to expand NATO against Russia’s opposition. If as I have argued, you think that is the case, then you cannot argue that Western countries get off the hook because of the Helsinki principle that sovereign states are free to seek whatever alliances they want. Not only does this not follow from that principle, as I have just explained, but even if that were not the case it would still not absolve them since they already adhered to that principle when they made those assurances and signed this treaty. Indeed, one can’t make assurances that will predictably and reasonably be interpreted by another party as creating some obligation toward them, only to argue later that no such obligation was created because it would be inconsistent with another commitment you had previously made. One can’t undertake commitments, even vague ones, but then ignore them on the ground that one wasn’t in a position to make them in the first place. If that is really the case, then one shouldn’t have undertaken that commitment to begin with. But again this doesn’t even matter ultimately, because as I have explained above, the Helsinki principle prevented neither Western countries from committing not to expand NATO nor Russia from opposing it. Voluntarily refraining from doing something that you might otherwise have done to ensure better relations with another state is a perfectly legitimate diplomatic practice in which statesmen engage all the time and that is just what I have argued that Western countries did by making the assurances they did and signing the treaty they did at the end of the Cold War. As we shall see, over the years, the idea that it would be illegitimate not to expand NATO because Russia was opposed to it on the ground that it would amount to giving Russia a “veto” on NATO’s policy became a common talking point. This confused argument therefore became the basis for NATO’s “open door policy”, which has since been elevated to the rank of dogma and, as I will argue later, had very unfortunate consequences. However, this is more a rhetorical device aiming at making something that, regardless of whether you think it’s a good idea, is within the normal confines of diplomacy look nefarious than a real argument.

I think part of the issue here is that many people have or pretend to have an overly irenic view of international relations. In particular, since NATO is a defensive alliance, they don’t think that its expansion to Central and Eastern Europe constitutes a security threat for Russia and they think that Moscow therefore has no business trying to prevent its neighbors from joining NATO if they want to. They may concede if pressed that, as I have just argued, the so-called Helsinki principle is not inconsistent with the idea that states can provide assurances to others they won’t seek to form alliances with third parties. But they clearly assume that there can’t be any legitimate reason for a country to be concerned about the fact that its neighbors want to form or join a defensive alliance with others and that any attempt to prevent it can only be explained by imperialism. However, as I will argue later in more detail, this is simply not true. Due to the anarchical nature of international relations, i.e. the fact that the international system lacks a superior authority that can resolve disputes (unlike in the domestic sphere where the state normally has a monopoly on the legitimate use of force), there is a sense in which states can only count on themselves to ensure their security and this makes them paranoid. In particular, they often care more about capabilities than intentions, because if at the moment they are confident that other states don’t mean them harm they can never be sure that this will always be the case. Furthermore, even if a state’s neighbors have no intention of attacking it, their membership even in a defensive alliance can weaken its position in international negotiations and make accidental conflicts more likely. Thus, while it’s true that Russia’s foreign policy elites are paranoid about NATO and that Russian fears about NATO expansion are exaggerated, those fears are not entirely irrational.

Moreover, when the shoe is on the other foot, nobody considers that states can’t have a legitimate interest in what alliances other sovereign countries enter into. As we shall see, the US in particular isn’t just concerned about what alliances its neighbors want to join, but it regularly applies various kinds of pressure on countries that are thousands of kilometers from its borders to ensure they will not enter security arrangements Washington deems harmful to its interests. The real difference between the US and Russia in that respect isn’t so much that the US respects the sovereignty of other countries whereas Russia doesn't, but that the US is strong enough to usually get its way without having to start a war or engage in histrionic behavior and Russia isn't. The fact that many people don’t recognize this reality leads them to adopt a very strong interpretation of the Helsinki principle, which as formulated in the Final Act of the CSCE and the Charter of Paris is very vague. Indeed, the former just says that states have “the right to be or not to be a party to treaties of alliance”, while the latter just says that the signatories “fully recognize the freedom of States to choose their own security arrangements”.150 On the strong interpretation many critics of the Russian position implicitly adopt, states are not allowed to do anything to prevent the constitution of alliances they deem harmful to their interests, but as we have seen while people may sometimes talk as if they believed that in practice no state behaves in a way that is consistent with that interpretation. The truth about the Helsinki principle is that on such a strong interpretation it can’t be a realistic basis for international relations, while on a weaker interpretation, it’s not clear exactly what constraints it places on states.

1.4.4 Western officials were not trying to mislead their Soviet counterparts when they made broad assurances at the end of the Cold War

Another common argument made by critics of the Russian position is that, regardless of what assurances Western officials made to their Soviet counterparts at the end of the Cold War exactly, they were not trying to mislead them. In other words, even if you agree with the argument I have made that Western countries had some obligation not to expand NATO to Central and Eastern Europe against Russia’s wishes because of those assurances, critics of the Russian position insist that Western officials made them in good faith and in particular were not planning to expand NATO at the time. However, not only is the claim that Western officials were not being disingenuous when they made broad assurances to their Soviet counterparts at the end of the Cold War not entirely true, but it has no bearing on whether those assurances created some obligation not to expand NATO which the West subsequently violated and whether the Russians are justified in blaming them for doing so. Indeed, even if Western officials had every intention to uphold the assurances they made at the end of the Cold War at the time they made it, the Russians would still have a legitimate grievance against the West if they or their successors later violated them. Now, the evidence is clear that Western officials had no definite plans to expand NATO to Central and Eastern Europe at the time, but that is not to say that all of them were fully honest with the Soviets. First, beyond the specific issue of NATO expansion, Western and in particular US officials were privately saying things at odds with the assurances they were making to Soviet officials about the future of the European security architecture. Moreover, while nobody had made the decision to expand NATO, some US officials were already thinking about it. Finally, not only were they already thinking about it, but as we shall see there is evidence that it affected their stance during the negotiations on German reunification. However, they were careful not to mention it to Soviet officials in the conversations they had with them during that period, because they knew that it would be anathema to them. This would have prevented the US from achieving several goals it cared deeply about, such as ensuring that Germany would stay in NATO after reunification and bringing to a successful conclusion the arms control negotiations that were taking place in parallel to the Two Plus Four process, because Moscow’s cooperation was indispensable for that.

The decision to expand NATO to Central and Eastern Europe was not made until 1994, during the Clinton administration, but as we shall see many officials in the Bush administration had already reached the conclusion that it would have to happen eventually by 1992 and several US officials were already thinking about it during the negotiations on German reunification.151 For instance, in his book on NATO expansion, US diplomat Ronald Asmus wrote:

Such nuanced diplomatic points aside [he is referring to the arguments exchanged by US and Russian officials in the mid-1990s about whether NATO expansion was consistent with the assurances made in 1990 and the Two Plus Four agreement], the reality was that no one in either Washington or Moscow was thinking about further NATO expansion in the spring and fall of 1990. Indeed, the issue had not yet been raised by the Central and East Europeans. These countries would not embrace that goal for another two years.152

However, although critics of the Russian position often make a similar claim, every single part of that claim is demonstrably false. First, as we have seen, there is conclusive evidence that Soviet officials were in fact thinking about that possibility in the spring of 1990. It’s also not true that Central and Eastern Europeans hadn’t raised the issue yet at the time, even if Asmus is right that no Warsaw Pact state had formally adopted that goal yet.153 In fact, Asmus himself wrote that, during a conference organized by RAND in June 1990 and attended by both US and Polish officials, the Poles asked whether they could join NATO and a Polish general even asked during a panel discussion co-chaired by Asmus if US forces could be stationed in Poland.154 Over the summer, after NATO announced that it would open “liaison offices” in Warsaw Pact states, one of Havel’s aides also quietly asked the Americans “how NATO would respond if Czechoslovakia applied for membership”.155 Thus, it is unsurprising that, contrary to what Asmus claims, US officials also were thinking about NATO expansion and not ruling it out, even if again they hadn’t decided to do it yet and wouldn’t for a while.

For instance, in July 1990, Baker argued during a meeting of Bush’s foreign policy team that Washington needed to craft “a ‘half-way house’ for governments who want out of the Warsaw Pact and CEMA but can’t join NATO and EC (yet)”.156 Yet just one month and a half earlier, during his meeting with Gorbachev in Moscow, the same Baker was assuring the Soviet leader that although the US thought that NATO was important in the short and medium term, in the long run they would create a pan-European security architecture. Indeed, while it’s true that he had called the idea of immediately replacing the Cold War alliances by such a system “a wonderful dream, but just a dream”, he quickly realized that it was a blunder and clarified later in the conversation that he didn’t mean that a pan-European security architecture of the sort advocated by Gorbachev shouldn’t be created later and reiterated the argument that allowing Germany to stay in NATO was not a way for the US to upset the balance of power in Europe but to preserve the status quo:

And one more thing. I said that pan-European security is a dream. What I meant is that it is a dream today. We made concrete proposals on how to build its structures in order for it to become a reality. In the meantime, we consider it important for Germany to be firmly anchored in security institutions, so it is not tempted to create some kind of security structure of its own.157

It’s hard to argue that Baker’s statement was not somewhat disingenuous when, around the same time, he was considering expanding NATO to Warsaw Pact states. This is especially true given that, as we shall see, in the same conversation he also ignored Gorbachev’s repeated suggestion that the Soviet Union might ask to join NATO. For US officials, who assured their Soviet interlocutors that Washington had no intention of excluding Moscow from the post-Cold War European security architecture, creating a “half-way house” for Central and Eastern European countries pending NATO membership was a serious option but there was no question of doing that for the Soviet Union.

Not only were some US officials already thinking about expanding NATO to Central and Eastern Europe during the Two Plus Four process, but there is evidence that it had a significant impact on the negotiations toward the end, which almost derailed the deal at the last minute. Indeed, during the last meeting on the eve of the scheduled signing ceremony in Moscow, a difficulty arose in connection to NATO’s freedom of movement in the territory of the GDR.158 Genscher and Shevardnadze reached a deal according to which the treaty would prohibit both the stationing and deployment of foreign troops in that territory after the departure of the Soviet troops from Germany, but the German government would have the option, at its discretion, to authorize foreign troops to cross the line temporarily as long as it was not a large number, in order to make it possible for NATO to hold exercises in that part of Germany. This deal would not be part of the treaty, but it would just be stated orally to the other foreign ministers and confirmed orally during the press conference for the signing. However, the US and even more so the UK insisted that oral assurances would not do and demanded that it be written down in the treaty (the French delegation seems to have been closer to the German position and played a role of mediator), which led the Soviets to threaten to call off the signature at the last minute. As no agreement could be found, Baker went to bed, but discussions continued between the Western delegations. Genscher was absolutely beside himself at the British for putting reunification at risk just when a deal had appeared imminent and insisted that US diplomat Robert Zoellick wake up Baker in the middle of the night so he could talk to him. They eventually came up with a solution with the help of Roland Dumas, the French Foreign Minister, in the morning. The Western allies proposed that a so-called “agreed minute” be added to the treaty stipulating that, although the deployment of foreign troops in the territory of the GDR was prohibited under the terms of the treaty, the German government would be able to determine what counted as “deployment” as long as it did so “in a reasonable and responsible way taking into account the security interests of each contracting party”. As Sarotte notes, the Western allies insisted that all parties sign under this “agreed minute” as well as under the treaty, which showed how important it was to them.

Why did the US and the UK jeopardize the deal for something that, in comparison to Germany’s reunification, seemed rather unimportant? Part of the explanation is probably that they wanted to avoid any permanent constraint that, in case of war, could have made the defense of Germany more difficult for NATO.159 But it’s unlikely to have been the only explanation, since even US officials acknowledged at the time that a prohibition on the deployment of foreign troops in the territory of the GDR, even if absolute, would not prevent the US from fulfilling its obligation to defend Germany.160 Similarly, the Western allies were very concerned about not agreeing to anything that would suggest Germany’s sovereignty was less than total and may have opposed the Soviet position for that reason, but since the Germans themselves were okay with it that is also not a very plausible explanation.161 According to Zoellick, another reason was that he was already thinking about the possibility that NATO might expand and wanted to make sure that NATO troops would be able to cross into Poland from Germany, though he also said that few US officials were thinking about that. Moreover, while for a long time German officials blamed the British (who they suspected of having tried to prevent reunification at the last minute because Thatcher had never been able to reconcile herself with it), Zoellick also claimed recently that, although the British had taken the lead, it was the US who had really been responsible for that position:

The night before the signing of the treaty, a difficult question appeared. It was about the possibility of stationing non-German NATO forces on the territory of East Germany, theoretically made impossible by the Treaty. Still, we needed to secure that possibility because, if Poland were eventually to join NATO in a second step, we wanted American forces to be able to cross East Germany on their way to be stationed in Poland. The Germans blamed the Brits for holding up the signature with that late-night request, whereas we were in essence responsible for that position. We finally agreed that NATO troops would not be present in East Germany until the Soviets left. And that we could cross East Germany if Poland joined. Remember that today we have battalions in the Baltics! Germans were upset because they felt the British wanted to undermine the signature. It is true that the British had the same idea as us. But as usual, they were aggressive and rash, contrary to us, whom Germans considered as more trustworthy. It was basically considered as a mere provocation. We all agreed to drop a note stating that “this question will be for the Germans to decide.” Meanwhile, an infuriated Genscher had called Dumas and Baker. … Coming back to the further enlargement of NATO, Scowcroft would later say that he did not even think of it. I can say that I certainly did. It was necessary and inevitable. But it is true that in Ukraine and Georgia, in 2007, we went too far. This is another story.162

While to my knowledge Zoellick’s claim is not supported by any contemporary document, and even he doesn’t claim that the possibility of NATO expansion was not the only reason for the American position, it’s nevertheless very plausible for several reasons.163 First, although no contemporary document supports Zoellick’s claim, he has been making it consistently for more than 20 years. Moreover, as we have seen above, Central and Eastern Europeans had already been probing Washington about the possibility of joining NATO and internally US officials had not ruled it out. Finally, according to the British account of the end of the Two Plus Four process, during the last minute dispute that led to the addition of the “agreed minute” to the treaty Baker in particular insisted that “something in writing would be necessary about the provisions on military activity in the former GDR after Soviet troop departure” and that “oral assurances would not do”.164 He didn’t mention the possibility of NATO expansion and instead claimed that it was because otherwise the Senate would not ratify the treaty, but this is not very plausible and, in any case, it at least supports Zoellick’s claim that despite the German impression that London was behind the roadblock the US was actually instrumental to it.165

It was not just on the possibility that NATO might be expanded to Central and Eastern Europe that US officials were, to put things mildly, less than fully transparent with the Soviets, but more generally on their thinking about the future of the European security order. While they were showering Soviet officials in assurances that, despite their insistence on the importance of not dissolving NATO, they agreed with the long-term goal of building a pan-European security architecture and only wanted Germany to stay in NATO to preserve the status quo, privately they were voicing skepticism about pan-European security and were already planning to expand the US footprint in Europe to fill the vacuum left by the Soviet Union’s retreat from Central and Eastern Europe.166 Their motivations were complex and will be discussed in more detail later, but they are not relevant to the question of whether US officials tried to mislead the Soviets during the negotiations on German reunification, so for now I just want to highlight the gap between what they were telling Soviet officials about the future of the European security architecture and what they were saying in private. In that respect, what Scowcroft reportedly told Jacques Attali, one of Mitterrand’s closest advisors, in March 1990 is particularly revealing:

We want to find a forum where we can talk with Europeans, to contribute to the political stability of the continent. The CSCE can be used in technical areas (such as the environment), but the USSR is a member, so it's difficult to talk about the security of democracies in it. [emphasis is mine] The best solution would be NATO, but the danger would be to make it an exclusively political body, which would remove any justification for the presence of our troops in Germany. Another option would be to extend NATO to the countries of Eastern Europe, but the danger would be to isolate the USSR. We could also use the OECD to involve Japan (a potentially very dangerous country in that it has never been and still isn't a democracy), but there are too many people. We could also think of the G7. The best thing would be to set up the "Directoire à Quatre" proposed by de Gaulle in his day.167

This statement by Scowcroft is yet another piece of evidence that US officials were thinking about NATO expansion from very early on, even if at this point they still opposed it, but it also makes it clear that, despite their assurances to the Soviets that eventually the CSCE would become the main institution in a pan-European security system, they were not actually prepared to deal with important security issues in a forum where Moscow would have an equal voice. As we shall see, this attitude would persist in subsequent years as Western states increasingly made their most important security decisions within NATO and the CSCE/OSCE was gradually relegated to less important issues, which explains a lot about the degradation of Russian-Western relations since the end of the Cold War.

To be fair, though as we have seen US officials led their Soviet counterparts to believe that eventually a pan-European security architecture would be created, they occasionally let their skepticism about pan-European security come out in conversations with them. For instance, when he met Shevardnadze in April 1990, here is what Bush told him:

We are working with you and the other European countries to strengthen the CSCE. CSCE will be an important pillar in the new Europe but we should not try to make it do more than it can. Those of us who remember when Europe was not at peace do not want to return to notions of collective security that almost inevitably fail.168

Nevertheless, immediately after he said that, he added that the US willingness to stay in Europe was “in the interest of a Europe whole and free and the common European home as you call it”. In doing so, he was equating his concept of a “Europe whole and free” with Gorbachev’s concept of a “common European home”, but they were very different and US officials were perfectly aware of it. For instance, in a recent testimony about the Two Plus Four process, here is how Zelikow explained why US officials opposed Gorbachev’s “common European home”:

Gorbachev thought that his vision, known as “the common European House,” would be welcomed because it was close to Mitterrand’s vision of a confederation, with the background melody of De Gaulle’s Europe from the Atlantic to the Urals. Of course, we did not favor these views. Not only because obviously there was no place for the US in this “house,” but also because the common house concept avoided expressing and voicing the final win of a system against the other one. It only alleviated the military tensions.169

As this passage shows, not only did US officials oppose Gorbachev’s “common European home” concept because it would have reduced the US influence in Europe, but also because they wanted the outcome of the Cold War to make clear who had won. As we shall see, this way of thinking goes a long way to explain how the Bush administration dealt with the end of the Cold War, which may have prevented a different settlement that would have made the subsequent deterioration of the relations between Russia and the West less likely.

So did Western officials try to mislead their Soviet counterparts during the negotiations on German reunification by making broad assurances they intended to violate? Not if by that you mean that, when they assured Soviet officials that eventually a pan-European security architecture would be created even if for the moment it was impossible and NATO was still necessary to ensure stability on the continent, they were already planning to expand NATO to Warsaw Pact states. This is not the case, but at the same time, some US officials were already thinking about NATO expansion and not ruling it out in the future, even though such a policy was hard to reconcile with the assurances they were making to Soviet officials about the future of the European security order. Beyond the specific issue of NATO expansion, for a number of reasons I will discuss in more detail later, US officials never truly believed in pan-European security and always intended NATO to remain the main security institution in Europe. Yet they still led the Soviets to believe that in time a transformed CSCE would play this role, as Gorbachev was proposing. It would be unfair to the Americans to say that they openly lied to the Soviets, but it’s hard to deny that their behavior involved a measure of deception. To be sure, it’s a form of deception that is common in diplomacy, but that doesn’t make it any less deceptive and knowing that is unlikely to make the Russians feel better about it. Finally, it’s worth reiterating that even if Western officials had been completely forthcoming about their intentions and had not been trying to mislead their Soviet counterparts in any way, the Russians would still have a legitimate grievance against the West insofar as, regardless of whether it was premeditated or not, Western countries ended up violating the assurances they made at the end of the Cold War.

1.4.5 Gorbachev never denied that the West had violated its obligations

It is sometimes claimed that even Gorbachev, who obviously would know how Soviet officials interpreted the statements made by Western officials in February 1990, denied that a pledge not to expand NATO had been made at the time. Indeed, back in 2014, Gorbachev told a journalist interviewing him for Russia Beyond the Headlines, a state-owned Russian media outlet that the issue of NATO expansion to Central and Eastern European countries wasn’t even discussed in February 1990 and basically endorsed Kramer’s position:

RBTH: One of the key issues that has arisen in connection with the events in Ukraine is NATO expansion into the East. Do you get the feeling that your Western partners lied to you when they were developing their future plans in Eastern Europe? Why didn’t you insist that the promises made to you – particularly U.S. Secretary of State James Baker’s promise that NATO would not expand into the East – be legally encoded? I will quote Baker: “NATO will not move one inch further east.”

M.G.: The topic of “NATO expansion” was not discussed at all, and it wasn’t brought up in those years. I say this with full responsibility. Not a single Eastern European country raised the issue, not even after the Warsaw Pact ceased to exist in 1991. Western leaders didn’t bring it up, either. Another issue we brought up was discussed: making sure that NATO’s military structures would not advance and that additional armed forces from the alliance would not be deployed on the territory of the then-GDR after German reunification. Baker’s statement, mentioned in your question, was made in that context. Kohl and [German Vice Chancellor Hans-Dietrich] Genscher talked about it.170

If even Gorbachev says there was no promise, so the argument goes, then surely the Russians are dissembling when they claim to have interpreted Baker’s statements as a pledge that NATO would not be expanded.

The problem is that Gorbachev said that in 2014 and, as what he says immediately after this passage makes clear, he was trying to defend himself against the accusation, which has become very common in Russia and isn’t without justification, that he had been played by the West in 1990:

Everything that could have been and needed to be done to solidify that political obligation was done. And fulfilled. The agreement on a final settlement with Germany said that no new military structures would be created in the eastern part of the country; no additional troops would be deployed; no weapons of mass destruction would be placed there. It has been observed all these years. So don’t portray Gorbachev and the then-Soviet authorities as naïve people who were wrapped around the West’s finger. If there was naïveté, it was later, when the issue arose. Russia at first did not object.171

As you can see, he is talking about himself in the third person and is trying to deflect the accusation that he failed to safeguard Russia’s interests because he was naïve, in a context where this accusation had become widespread in Russia. Indeed, the interview took place immediately after the US and Europe had taken harsh sanctions in response to the annexation of Crimea and the shooting of MH17 over Ukraine, so his statements have to be interpreted in light of that context.

In particular, one can’t infer from this interview that Gorbachev didn’t believe that, as I have argued above, statements made by Western officials before and during the negotiations on German reunification created some obligation not to expand NATO over Russia’s objections. In fact, not only did he not say anything in this interview that implies that he did not believe that it was the case, but he explicitly said that he did:

The decision for the U.S. and its allies to expand NATO into the east was decisively made in 1993. I called this a big mistake from the very beginning. It was definitely a violation of the spirit of the statements and assurances made to us in 1990. [emphasis is mine] With regards to Germany, they were legally enshrined and are being observed.172

In this passage, Gorbachev clearly distinguishes between the assurances that were made by Western officials about Germany, which were put in writing in the treaty and were subsequently observed, from the assurances that were made about other topics, which were implicit and subsequently violated. Since he also claimed that NATO expansion to Central and Eastern Europe was never discussed during the negotiations on Germany’s reunification, his argument must be that it didn’t have to be discussed explicitly for the assurances made at the time to create some obligation not to do it, which is precisely the argument I have made above.

1.5 Many Western officials also interpreted, both at the time and later, the assurances made at the end of the Cold War as ruling out NATO expansion

I have argued that, in light of the assurances made by Western leaders during the negotiations on Germany’s reunification, Russian officials were justified in having the expectation that NATO would not expand to the east against their wishes, but there is also evidence that some Western officials at the time similarly interpreted those assurances as ruling out that possibility. Indeed, Shifrinson recently discovered in the British archives notes from a meeting that took place on 7 March 1991 between high-level officials of the US, France, the UK and Germany about security in Central and Eastern Europe showing that even some Western officials thought the Soviet Union had been promised that NATO would not expand in the area.173 The document starts by noting that there is "general agreement that membership of NATO" would be unacceptable. Of course, this could have been unrelated to any promise Western officials felt they had made to the Soviet Union, but Jürgen Chrobog, Political Director of the German Foreign Office, is noted to have said that "we had made it clear during the 2+4 negotiations that we would not extend NATO beyond the Elbe" and that "we could not therefore offer membership of NATO to Poland and the others".174 Another document recently discovered in the German archives by Klaus Wiegrefe, which summarizes the same meeting, talks about "the understanding already expressed in the 2 + 4 process that the withdrawal of Soviet troops must not be used by the West for its own advantage".175 This is somewhat less specific, but on the other hand this remark is attributed to both Chrobog and Raymond Seitz, the US Assistant Secretary of State for Europe and Canada. Thus, according to German diplomatic archives, at least one top-level US official also felt at the time that during the negotiations on Germany's reunification a commitment had been made not to take advantage of the Soviet Union's withdrawal from Central and Eastern Europe and that this precluded the expansion of NATO in that area.176 Yet today Western officials insist that claims to that effect by Russian officials are nothing but a figment of their imagination.

Nor is the March 7 meeting the only evidence that some Western officials continued to regard themselves bound by assurances they'd given to the Soviets in 1990 even after the Treaty on the Final Settlement with Germany was signed. Wiegrefe also found in German diplomatic archives that Genscher, Chrobog's boss at the German Foreign Office, had also told other Western officials they couldn't expand NATO to Central and Eastern Europe because of the commitments they'd made during the negotiations about Germany's reunification on three other occasions in 1991. He made that point to US officials on March 1, to his Greek counterpart on April 18 and finally to his French and Spanish counterparts on October 11. Thus, Genscher explicitly ruled out NATO expansion to Central and Eastern Europe before, during and after the negotiations on Germany's reunification. This is inconsistent with the view defended by critics of the Russian position, who not only argue that no Western official ever mentioned NATO expansion to Central and Eastern Europe before the negotiations on Germany's reunification started and that any assurances given during those negotiations concerned only the territory of the GDR, but that in any case even those more limited — according to them — assurances were later superseded by the final settlement eventually reached. Indeed, if this were true, then how could Western officials have still felt bound by assurances given in 1990 not to expand NATO to Central and Eastern Europe several months after the signature of the Treaty on the Final Settlement with Germany? One could argue that, since many Western officials still opposed NATO expansion to Central and Eastern Europe in 1991 (in part because they feared it might cause a backlash in the Soviet Union), some of them fabricated a promise not to expand it that was allegedly made during the negotiations on Germany's reunification, but this seems far-fetched, especially since everyone they talked to apparently agreed that NATO shouldn't be expanded. It would mean that some Western officials just happened to come up with the same lie — according to critics of the Russian position — back then as Russian officials today in order to convince people who in most cases also didn't want to expand NATO to Central and Eastern Europe anyway. It's much more likely that, even after the Treaty on the Final Settlement with Germany had been signed, they really felt that expanding NATO would violate assurances that had been made during the negotiations of that agreement.177

This interpretation is also supported by statements that Western officials involved in the negotiations about Germany’s reunification made later. William Burns, who in 1990 was deputy director of the Policy Planning Staff in the State Department and therefore well positioned to know what had transpired during the negotiations on Germany’s reunification, wrote in his recently published memoirs:

With President Bush’s support, Baker sold the [Two Plus Four negotiations format that associated the FRG and the GDR to the US, France, the Soviet Union and the UK] to German chancellor Helmut Kohl and foreign minister Hans-Dietrich Genscher in early February, agreeing to use Two Plus Four negotiations to press for rapid German unification and full NATO membership, while reassuring the Soviets that NATO would not be extended any farther to the east, and would be transformed to reflect the end of the Cold War and potential partnership with the Soviet Union.

In meetings a few days later with Shevardnadze and Gorbachev in Moscow, Baker won their initial support, and began the effort to ease their resistance to membership of a unified Germany in NATO. Baker maintained that Soviet interests would be more secure with a united Germany wrapped up in NATO, rather than a Germany untied to NATO and perhaps eventually with its own nuclear weapons. He also said that there would be no extension of NATO’s jurisdiction or forces “one inch to the east” of the borders of a reunified Germany. The Russians took him at his word and would feel betrayed by NATO enlargement in the years that followed, even though the pledge was never formalized and was made before the breakup of the Soviet Union. It was an episode that would be relitigated for many years to come.178

In this passage, Burns talks about “the borders of reunified Germany” when he mentions the pledge not to expand NATO eastward, which is inconsistent with the view that it only applied to the territory of the GDR. In case that was not clear enough, later in his memoirs, he also writes that “Yeltsin and the Russian elite assumed, with considerable justification [emphasis is mine], that Jim Baker’s assurances during the negotiation of German reunification in 1990—that NATO would not extend its reach ‘one inch’ farther east—would continue to apply after the breakup of the Soviet Union”.179

Of course, Burns’s understanding may not reflect the thinking of the rest of the Bush administration and indeed we have seen that other Western officials who played a role at the time disagree, but he is hardly alone among the participants in the diplomatic activity that took place at the end of the Cold War in thinking that Western officials had, if not outright pledged that NATO would not be expanded, at least led the Soviets to believe that it would not happen. For instance, here is what Robert Gates said when, after the first round of NATO expansion was carried out, he was asked what he thought Bush would have done if he had been reelected in 1992:

That would have been a tough fight within the administration. I’m not sure where people would have come down, but I think that the Bush administration would have, at the end of the day, kept our focus on our priorities, which in my view are, if you don’t get it right with Russia and China, none of the rest matters. And at a time of a special humiliation and difficulty for Russia, pressing ahead with expansion of NATO eastward, when Gorbachev and others were led to believe that wouldn’t happen [emphasis is mine], at least in no time soon, I think probably has not only aggravated the relationship between the United States and Russia but made it much more difficult to do constructive business with them. I think between that and the bombing of Belgrade we have really antagonized the Russians in a major way and I think those are two things that the Bush administration would not have done, when all is said and done.180

He makes a weaker claim than Burns, one that is more in line with the view I defended above, but they both think that the Russians have good reasons to be upset by the decision to expand NATO to Central and Eastern Europe given what they were told at the end of the Cold War.

Seitz is another high-ranking member of the Bush administration who is on the record saying that, although US officials never gave the Soviets any explicit guarantee that NATO would not expand to Central and Eastern Europe, Russia’s feeling of betrayal is justified:

At the time, there was never an explicit guarantee that NATO would not expand eastward. But the Russians felt later that there was an implicit commitment by the West, even if it was never part of any written agreement. The decision to expand NATO was taken later, by the Clinton administration. The idea was to stabilize eastern Europe. That was a very good argument in terms of expanding the EU. But I thought that it was not a good argument for expanding NATO. All the liberated Eastern countries definitely wanted NATO membership, but the Soviets did not believe we would agree. They had the impression that we would not take advantage of their weakened position. But at the end, we did, and it's understandable that they feel double-crossed.181

Dobbins, who in 1990 was Seitz’s deputy, also testified recently that he thought the Russian complaint was justified:

The 2+4 was signed, and with it started the ambiguity on Eastern Europe in NATO, in a context of the USSR falling apart. … It would be untrue to say that at that time there was a written doctrine on the enlargement of NATO. I personally opposed it in 1990. Again, at that moment, there was no intention of moving on very quickly. It became an issue only with the Clinton administration. I think that the “you betrayed us” of the Russians is a valid objection. There is no doubt that Genscher and Baker said that the West had no intention to enlarge. But Gorbachev did not ask for it in writing. The Warsaw Pact was operating, troops were in the barracks. The Soviets never imagined that things would change so fast. It was just unimaginable! It was their fault for not taking the assurances and formalizing the deal.182

Seitz’s view is very similar to mine, whereas Dobbins makes a stronger claim, but again they both validate the Russian position in some way.183

Nor are former Bush administration officials the only Western officials who think the Russian position has merit. Elbe, who in 1990 was the head of the West German delegation during the Two Plus Four process, passed away in 2022, but a few years before that, he had endorsed the Russian position in the strongest possible terms:

Under Clinton, the enlargement of NATO started. Having been a negotiator in the 2+4 process in 1990, I can say that there was no mis-understanding that the USA had given a stringent verbal commitment to the Soviet Union on February 8, 1990 that, “NATO should never expand by one inch to the East.” Secretary of State Jim Baker had discussed the issue with Gorbachev. In a letter that Baker wrote on the day he left Moscow, he informed Chancellor Kohl of his discussions, in particular on the deal of integrating united Germany into NATO. The clarity of the American position was instrumental in triggering the decision by Gorbachev a day later that the Soviet Union would not oppose reunification of Germany.184

Mallaby, who at the time was the British Ambassador to Moscow (like Elbe he died in 2022), also said in 2020 that several Western officials pledged not to expand NATO to Central and Eastern Europe:

I think there was a mistake by three Western countries. Baker, for the United States, and Kohl, for the Federal Republic, as well as John Major, for the UK, told the Soviet Union informally that NATO would not be extended further east than the eastern frontier of the united Germany. Later NATO was extended further east and the Soviets were understandably furious. I confess that I am glad about this extension of NATO because it gave those new members of NATO security in the face of Putin's subsequent policies.185

Here it seems that Mallaby is not just referring to the February 1990 talks in Moscow, but also to Major’s visit in March 1991, when he did assure the Soviet Minister of Defense who complained that NATO was thinking about inviting Warsaw Pact states to join that “nothing of the sort will happen”.

I don’t quote those Western officials because I necessarily think what they say is true. In fact, many of them argue that Baker explicitly promised Gorbachev that NATO wouldn’t be expanded to Central and Eastern Europe, but as we have seen, critics of the Russian position are right that he was only talking about the territory of the GDR. However, the fact that many Western officials who were directly involved in the negotiations on German reunification at the highest level, both immediately after the end of the Cold War and more recently, interpreted the assurances made at the time as ruling out NATO expansion to Warsaw Pact states is still relevant to the controversy about whether the Russian complaint has merit. First, although many incorrectly claim that Baker’s infamous February assurance was about Central and Eastern Europe in general (as opposed to the territory of the GDR), some make a weaker claim that is very similar to the view I have defended above. In other words, they don’t claim that Western officials explicitly promised not to expand NATO to Central and Eastern Europe, but that they nevertheless made broad assurances that implicitly were hard to reconcile with such a policy, so it wasn’t unreasonable for the Russians to feel they had been misled when it happened. Moreover, if even some Western officials who played a role in the negotiations with the Soviet Union at the time feel that NATO expansion violated the assurances made at the end of the Cold War, it seems hard to argue that it’s unreasonable for the Russians to feel the same thing.186 Indeed, the argument I have made is that Western officials mollified the Soviets by showering them in assurances that, although individually vague enough for Western officials to be able to claim — probably in good faith — that NATO expansion did not violate them, collectively were such as to create expectations that were clearly incompatible with it. Again, this doesn’t mean that Western officials deliberately tried to mislead the Soviets in that way (although we have seen that US officials were less than entirely forthcoming about their intentions), but it doesn’t make it any less true that it was justified for the Russians to deem those assurances misleading in retrospect.

1.6 Hell hath no fury like a Russian scorned

So what should we make of this controversy that has been plaguing the relations between Russia and the West since the end of the Cold War? As usual, the answer is complicated, but some things are pretty clear. First, despite what is often claimed or suggested (including sometimes by Russian officials), Baker was almost certainly talking only about the territory of the GDR when he infamously told Gorbachev that if the Soviet Union consented to let Germany stay in NATO after reunification, the Alliance’s “jurisdiction” would not move “one inch to the east”, whatever this could have meant exactly. It’s understandable that people think he was talking about Central and Eastern Europe, because this interpretation of the words he used is very natural, but a close reading of the conversation as a whole makes it very unlikely that he actually was.187 Genscher on the other hand, who repeatedly made a similar assurance at the beginning of 1990, was clearly talking about Central and Eastern Europe in general and not just the territory of the GDR, contrary to what critics of the Russian position allege. As we have seen, at some point even Baker was ambiguous on this point, though his ambiguity may have been calculated. In any case, Genscher’s early statements on the issue were informed by his views on what a post-Cold War European security architecture should look like, which though close to Gorbachev’s “common European home” concept were not shared by the FRG’s allies and in particular the US or even by the more Atlantist Kohl. As a result, his position was quickly marginalized among Western officials and he eventually had to fall behind the Bush-Kohl position, a fact that as we have seen the Soviets did not fail to notice. Thus, critics of the Russian position are still correct that no obligation to not expand NATO to Central and Eastern Europe arose from the statements made by Western officials in February, but as they like to point out the negotiations on Germany’s reunification didn’t stop in February.

During the Two Plus Four process, Western officials hardened their position on the GDR by abandoning Baker’s “jurisdiction” talk (whose precise meaning had never been clear anyway) in favor of Bush’s “special military status” formula, but they also made a number of broader assurances on what the post-Cold War European security order would eventually look like. As I have argued, the assurance that NATO would not expand to the east had never been meant as compensation for allowing Germany to stay in the Alliance, but as a way to make this outcome even more appealing than the alternative of a Germany unconstrained by membership to the West’s security structures by promising that NATO would not take advantage of Germany’s reunification to expand its footprint. The broader assurances made after February were similarly meant to make the prospect of a reunified Germany in NATO more appealing by promising that although it couldn’t happen overnight eventually a pan-European security system would be created. Those broad assurances, I have argued, were clearly in tension with NATO expansion and created some obligation that it would not happen against Russia’s wishes. Indeed, as we have seen, this was explicitly confirmed by Major to Soviet officials in 1991 during a visit in Moscow and contemporary documentary evidence shows that at least some Western officials agreed this was implied by the assurances made in 1990. Western and in particular US officials were not actively planning to violate any of the assurances they made during this process, but they kept them deliberately vague and papered over their disagreements with Moscow about the future of the European security order, so the Soviets would hear what they wanted to hear. Their main concern was the immediate goal of securing Germany’s continued membership of NATO, with as few restrictions as possible, after the reunification and in the pursuit of that goal they made broad assurances to mollify the Soviets without giving much thought about what problems could arise in the future as a result of how they interpreted those assurances. It worked and, when later the Russians complained because they had interpreted them more expansively than what US officials intended, but still very reasonably given what they had actually been told, they accused them of making things up. It’s not hard, provided that one is capable of a modicum of empathy, to understand why the Russians are upset.

There are in principle three different things people could mean when they say that Western officials pledged to the Soviet Union that NATO would not expand at the end of the Cold War:

A.    they signed a legally binding treaty that ruled out NATO expansion to Central and Eastern Europe;

B.     they made a verbal pledge not to expand NATO to Central and Eastern Europe in exchange for allowing Germany to stay in NATO after reunification;

C.     they gave other, more vague verbal assurances that it would nevertheless have been natural for the Russians to interpret as ruling out NATO expansion to Central and Eastern Europe.

Everybody in that debate agrees that A is false, even if critics of the Russian position sometimes make it sound as if Moscow were making that claim, but the debate has been focusing on B, while C has mostly been ignored even though unlike B it’s very well supported and still vindicates the Russian position to a large extent. Indeed, in light of the assurances Soviet officials received at the end of the Cold War, the Russians have very good reasons to regard NATO expansion and more generally their exclusion from the post-Cold War European security architecture as a betrayal, even if they sometimes overstate their case by suggesting that something like B is true.

One could acknowledge that, as I have argued, the assurances made by Western officials at the end of the Cold War created some obligation for the West not to expand NATO against Russia’s wishes, but insist that it’s Russia’s fault if NATO was expanded anyway and more generally if the sort of pan-European security architecture that Soviet officials were assured would eventually be created never materialized. On this view, to the extent that assurances made at the end of the Cold War create some obligation not to expand NATO to Central and Eastern Europe against Russia’s wishes, this obligation was defeasible and depended on Russia’s behavior, but after 1991 Moscow acted in a way that made it impossible to create the kind of pan-European security system that Western officials promised would emerge in due time during the negotiations on Germany’s reunification. I don’t think anyone has actually proposed this line of thinking, which is not surprising since critics of the Russian position deny that any obligation not to expand NATO resulted from the assurances made at the end of the Cold War, but it’s a logically possible position and as we shall see many people in the 1990s did argue that Russia’s actions made NATO expansion necessary. It’s certainly true that, if only because it wasn’t part of a legally binding treaty, the obligation not to expand NATO and to include Moscow in the post-Cold War European security order could not have been unconditional and that creating the kind of pan-European security architecture promised at the end of the Cold War required Moscow’s cooperation. Thus, in principle it could be that although NATO was expanded to Central and Eastern Europe and Moscow was de facto excluded from the post-Cold War European security architecture, it was Russia’s fault and the West can’t be blamed for not fulfilling the assurances made at the end of the Cold War. However, while it’s true that Russia has a share of the responsibility for the degradation of the relations with the West in subsequent years, I will argue that the bulk of it falls on the US and its allies. The truth is that US officials in particular were never able to shake the deep suspicion of Russia inherited from the Cold War, which effectively made it impossible to find a place for Russia in the post-Cold War European security order.

People generally assume, more or less implicitly, that whether the West violated some kind of obligation it had contracted at the end of the Cold War by expanding NATO to Central and Eastern Europe is independent from the question of whether NATO expansion was a good policy, but I don’t think that is quite true.188 Indeed, since that obligation was not codified by a treaty or even in a non-legally binding written document but resulted from broad verbal assurances, there is no objective fact of the matter about how strong it was exactly and therefore I don’t think those issues can be so neatly distinguished. I think that everyone should agree that some obligation not to expand NATO was created by those assurances, and that wasn’t as strong as it would have been if Baker had made a clear and explicit quid pro quo with Gorbachev (in the way Khrushchev agreed with Kennedy to remove Soviet missiles from Cuba in exchange for the withdrawal of Jupiter missiles from Turkey), but that beyond that there is plenty of room for reasonable people to disagree about how strong it was. Since there is no objective fact of the matter about how binding that obligation was, it’s natural and legitimate that how bad one thinks the fact that NATO was expanded to Central and Eastern Europe in spite of those assurances was depends in part on one’s view about whether it was a good idea to expand NATO. One can argue that although the assurances in question created some obligation not to expand NATO, this obligation was outweighed by other considerations, but that is not the same thing as denying that some obligation not to do it existed. Unfortunately, because participants in the debate have focused on B while mostly ignoring C, this is what critics of the Russian position have done. Meanwhile, people sympathetic to that position have implausibly tried to argue that B was true instead of focusing their efforts on defending C, which is on much firmer grounds.189 To be fair, Russian officials are partly to blame for this situation, because as we have seen in their desire to cast NATO expansion as illegitimate they have encouraged the confusion between B and C by throwing the kitchen sink at their opponents. However, this is not a reason to ignore C and deny that, after what they were told at the end of the Cold War, Russian officials had legitimate reasons to feel they had been misled when the process to expand NATO to Central and Eastern Europe was launched and a NATO-centric European security order started to take shape.

Indeed, what most people miss in this debate is that it’s not so much NATO expansion per se that Russia resents as much as its exclusion from the post-Cold War European security order. NATO expansion to Central and Eastern Europe was merely the main concrete manifestation of this exclusion. Western officials could have acknowledged that, in light of the assurances they had made at the end of the Cold War, they understood that Russia felt betrayed, but argued that circumstances had changed in ways that were impossible to foresee at the time and could only be dealt by expanding NATO. As I will argue when I discuss how the decision to expand NATO was made, I don’t think there is a good case to be made for this view, but at least by presenting that argument Western officials would have acknowledged that NATO expansion was in tension with the assurances made at the end of the Cold War, even if ultimately they thought that other considerations trumped the obligation that resulted from those assurances. Instead, they just denied that any such tension existed and replied to Russia’s complaints by making irrelevant points such as the observation that the Treaty on the Final Settlement with Respect to Germany did not contain any provision that precluded NATO expansion to Central and Eastern Europe (which the Russians never claimed) or the claim that Central and Eastern European states have a right to seek admission into NATO (which is beside the point), thereby adding insult to injury. The truth is that when those assurances were made the Soviet Union was still perceived as a force to be reckoned with and the West needed Moscow’s cooperation to solve a number of issues it cared a great deal about. This was no longer true a few years later as people came to realize how weak Russia was, so Western officials ignored the assurances they had previously made or at least interpreted them in the way that suited them best, which is often how things go in international relations. As unfair as it was, the Russians would have been better served to accept all of this and focus on building a stronger economy, but they didn’t and more recently the West has, or rather should have, learned that ignoring Russia’s wounded pride was not as costless as it once seemed, because even in this weakened state it can still make a mess.

2 The road not taken at the end of the Cold War

I have spent a lot of time on the controversy about the assurances that were made during the negotiations on Germany’s reunification, not only because it has become a focal point in the dispute between Russia and the West in subsequent years (especially since the invasion of Ukraine), but also because a deep-dive into that controversy was the occasion to identify some important facts about the end of the Cold War that are crucial to understand what happened later. In particular, this discussion showed that if US officials were so insistent that Germany had to stay in NATO after reunification, it’s because they knew that if Germany left NATO the US would no longer be able to keep a significant military presence in Europe, which in turn is directly related to what I call the “original sin” of the West’s post-Cold War policy toward Russia. Indeed, this “original sin” is not the fact that the US and its allies made broad assurances at the end of the Cold War that, as I have argued, it later violated by expanding NATO to Central and Eastern Europe a few years later. The “original sin” is the decision to solidify NATO at the end of the Cold War, rather than create a new European security architecture that would have included Russia. This decision put the world on the path that led to the events of 2014 and eventually the invasion of Ukraine in 2022. While it did not make this trajectory inevitable, as we shall see there were numerous occasions after that when it could have been avoided if Russia, Ukraine and the West had made other decisions, it made the degradation of relations between the West and Russia, as well as a conflict between Russia and Ukraine, much more likely. In this section, I will explain why that is the case, but also why the US and its allies nevertheless made the fateful decision to enshrine NATO’s primacy instead of trying to create a truly inclusive European security order to replace the Cold War alliances as the conflict that structured international relations for 40 years came to an end.

2.1 NATO’s primacy and the logic of exclusion

NATO was created in 1949, at the beginning of the Cold War, to protect Western Europe against a possible attack by the Soviet Union. By committing the US to defend Western European countries if they were attacked, it represented a major break in US foreign policy, which so far had avoided this kind of commitment in Europe and had been strongly influenced by isolationism. The logic behind the decision to create NATO, from the US point of view, was to avoid another costly intervention in Europe by deterring the Soviet Union from attacking Western Europe, which US officials thought would inevitably force Washington to enter the war.190 However, this commitment did not initially translate into a large build-up of US troops on the continent, which didn’t happen until North Korea invaded South Korea in 1950, prompting fear that the Soviet Union might attempt a takeover of Western Europe.191 The Soviet Union created the Warsaw Pact in 1955 as a counterweight to NATO and in order to ensure the “stability” of the Eastern bloc. Since NATO had been created to counter the threat posed by the Soviet Union at the beginning of the Cold War, it would prima facie have been logical to dissolve it at the end of the Cold War and create a new European security architecture that would have included the Soviet Union and Central and Eastern European countries, but this did not happen. For reasons I will discuss shortly, although some Western officials briefly explored the possibility of creating a pan-European security system along the lines proposed by Gorbachev, Western countries in general and the US in particular refused to countenance the dissolution of NATO and were determined to ensure that it would continue to be the main forum where they would discuss hard security issues. Before I explain what led them to make this choice, I want to explain why it made the exclusion of Russia from the European security architecture, the subsequent deterioration of relations between Russia and the West and a conflict between Russia and Ukraine more likely.

The logic that made this kind of outcome more likely is pretty straightforward. As long as NATO continued to exist but no pan-European security organization was created, Central and Eastern European countries would find themselves in a security vacuum and, having recently shaken off the Soviet yoke, would inevitably be pulled toward NATO and try to join the Alliance to obtain security guarantees against the possibility that Russia might recover and try to subjugate them again. As Sarotte noted, declassified materials show that US officials understood that point and privately made it as early as the summer of 1990, when Central and Eastern Europeans started to probe Washington more forcefully about the possibility of joining NATO:

Havel’s security advisor asked (as the NSC expert on Eastern Europe, Robert Hutchings, reported to Scowcroft on August 16, 1990), “how NATO would respond if Czechoslovakia applied for membership.” Hutchings understood why they were asking. If the “East Europeans want out of the Warsaw Pact but cannot join NATO,” he asked, “where do they find their security in the Europe of the future?” Genscher and Mitterrand had previously tried to answer that question by proposing that some kind of pan-European entity replace both military pacts, but the American preference for NATO had prevailed.192

Unless Russia also joined NATO, it would find itself completely excluded from the European security architecture, as important security issues would be discussed and decisions made in NATO, where it doesn’t have a vote. In turn, this would cause a bad reaction in Russia as NATO moved closer to its borders and it found itself increasingly isolated, until finally a point was reached where the US and its allies would no longer be willing to risk a conflict with Russia by expanding NATO further, leaving some countries in the former Eastern bloc facing an aggravated Russia on their own.

The only way to prevent that would have been for Russia to join NATO, but for both objective and subjective reasons, it was unlikely to ever happen. First, although it was not as large as the Soviet Union, Russia was still a gigantic country that spanned the whole Eurasian landmass with very large armed forces. So in the event that it joined NATO, not only would the Alliance suddenly find itself committed to a vast geographic area going far beyond the Euro-Atlantic zone, but it would also have to make a very difficult and costly effort to modernize the Russian armed forces and make them interoperable with NATO forces. Russia would also have had to carry out enough democratic reforms to qualify for membership, though how much was enough would ultimately have been a political decision, so this criterion was only partly objective. In addition to such objective factors, there were also subjective factors that made Russia’s membership in NATO unlikely. Indeed, despite the end of the Cold War, people in the West remained suspicious of Russia and feared that if Moscow joined the Alliance it would turn into an ineffective collective security system because Russia’s interests were insufficiently aligned with the West’s.193 This suspicion would only increase as Central and Eastern European countries, with their deep-seated fear of Russia, started to join the Alliance. Baker himself emphasized the importance of that factor a decade later when he argued that NATO should invite Russia to join the Alliance:

If Russia wishes to join NATO and at some point in the future satisfies all of the criteria for membership, what is the problem? The problem is, of course, that many participants in this process impose one more condition for membership that Russia can never satisfy—namely, that the candidate not be Russia. To some extent, this sentiment reflects historical bias. NATO was formed to resist the threat of Soviet aggression against the West. Russia essentially controlled the Soviet Union. QED. Why admit to the alliance the very adversary against whom the alliance was formed?194

Subjective factors were equally present on the Russian side, where the legacy of the Cold War also loomed large and NATO was still perceived as an anti-Russia organization, making it politically difficult for any Russian government to apply for membership. That is not to say that it was impossible for Russia to join NATO, as we shall see it was seriously considered both in Russia and in the West for a while, but it would have required to overcome those obstacles and that was unlikely to happen.

As long as it did not happen and NATO moved toward Russia’s borders while Moscow was left on the outside, it was bound to perceive this process with increased suspicion, until eventually things came to a head and caused a clash. On the other hand, once NATO expansion had started, it was always going to be difficult to stop the process because this would have amounted to redrawing a line in Europe and implicitly admitting that Russia had a veto over NATO expansion. No Western leader was likely to explicitly close the door to further expansion because, only a few years after the end of the Cold War, the idea that Russia should have a say on Western policy was clearly going to be a political nonstarter, especially in the US. Clinton explained this logic very clearly to Yeltsin in 1997, when the Russian President tried to convince him that, if NATO expansion was to continue, they should at least make a “gentlemen’s agreement” that former Soviet republics would not be invited to join:

"Good," said Clinton. "But I want you to imagine something else. If we were to agree that no members of the former Soviet Union could enter NATO, that would be a bad thing for our attempt to build a new NATO. It would also be a bad thing for your attempt to build a new Russia. I am not naïve. I understand you have an interest in who gets into NATO and when. We need to make sure that all these are subjects that we can consult about as we move forward. 'Consult' means making sure that we're aware of your concerns, and that you understand our decisions and our positions and our thinking. But consider what a terrible message it would send if we were to make the kind of supposedly secret deal you're suggesting. First, there are no secrets in this world. Second, the message would be, 'We're still organized against Russia-but there's a line across which we won't go.' In other words, instead of creating a new NATO that helps move toward an integrated, undivided Europe, we'd have a larger NATO that's just sitting there waiting for Russia to do something bad.

"Here's why what you are proposing is bad for Russia. Russia would be saying, 'We've still got an empire, but it just can't reach as far west as it used to when we had the Warsaw Pact.' Second, it would create exactly the fear among the Baltics and others that you're trying to allay and that you're denying is justified.

…

"I've been repeating that I'd leave open the possibility of Russia in NATO and in any event of having a steadily improving partnership between NATO and Russia. I think we'll have to continue to work this issue, but we should concentrate on practical matters. However, under no circumstances should we send a signal out of this meeting that it's the same old European politics of the cold war and we're just moving the lines around a bit."195

Thus, while initially there was little appetite for NATO expansion in the West, once it started the idea that it should continue indefinitely became a dogma, which predictably further aroused Russian suspicions. Indeed, no matter how often US officials insisted that NATO expansion wasn’t directed against Russia, that NATO had rejected the logic of blocs and that even Moscow could join eventually, the Russians were not buying it and as we shall see they had excellent reasons not to. While Clinton himself was mostly sincere, for most people in the West and Central and Eastern Europe, creating a “a larger NATO that's just sitting there waiting for Russia to do something bad” was precisely the point.

Therefore, from the moment the West decided to keep NATO in place instead of creating a pan-European security architecture to replace the Cold War alliances, the expansion of NATO, the exclusion of Russia and the deterioration of relations with the West had already become likely. It was clear that, with the disintegration of the Warsaw Pact (which had been imposed by force in the first place and could not survive once Gorbachev had made the decision not to use force to keep the Soviet Union’s satellites in line), Central and Eastern European countries would find themselves in a security vacuum and NATO would become a magnet for them. This would eventually result in the exclusion of Russia from the European security architecture unless it could also join NATO, which though not impossible was unlikely for both objective and subjective reasons. Again, that is not to say that this outcome was inevitable, but the tracks that led to it had been laid and it would have taken unusually competent statesmanship to avoid it. This was entirely predictable and, indeed, it was predicted by several people at the time. In particular, it’s in part because he foresaw that development that Mitterrand proposed his project of a European confederation at the end of 1989, which by including the Soviet Union could have prevented Moscow’s exclusion from the post-Cold War European political order.196 This project initially attracted support in both the West and the East, but it eventually failed and never came to anything, in part because it was shot down by the US for reasons to be discussed shortly.

Another person who foresaw this outcome, perhaps more clearly than anyone else, was Gorbachev himself. In May 1990, after trying and failing to convince Baker that Germany should be neutral after reunification, he told him this:

Let us continue this conversation in Washington. And if none of my arguments convince you, then I’ll suggest to the President and announce publicly that we want to join NATO too. After all, you say that NATO is not directed against us, that it is just a security structure that is adapting to the new reality. So we will propose to join NATO.197

He added immediately after that it was not “a purely hypothetical question” or “some absurdity” and came back to the issue later in the conversation by saying that “our potential membership in NATO is not such a wild fantasy”, but Baker ignored him and changed the subject in both cases. While at the beginning of the year, Gorbachev still harbored some hope that part of the Warsaw Pact could be preserved, by then he knew that it was condemned and understood that if NATO was not dissolved, it would result in Moscow’s isolation unless it could also join. As Baker later acknowledged, in a paper where he argued that Russia should be invited to join NATO (precisely to avoid the kind of process I’m arguing was made likely by the decision not to create a pan-European security architecture at the end of the Cold War), Gorbachev “was more serious than we gave him credit for at the time” and was “looking several moves ahead”.198 A very similar scene happened in March 1991 during Major’s visit to Moscow. As we have seen, during that visit, Gorbachev complained that, despite the assurances the Soviet Union had received during the negotiations on German reunification, NATO was taking steps to expand to Central and Eastern Europe. Major assured him that nothing of the sort would happen, but Gorbachev was not convinced, and he said that in that case the Soviet Union would apply for membership in NATO, to which Major just replied that he should apply to the EEC instead and then changed the subject. Gorbachev understood that, despite the West’s protests to the contrary, the Soviet Union was still perceived as a threat. He also understood that, in time, this would result in Moscow’s isolation and threaten Europe’s stability.

2.2 Anatomy of a decision

I have argued that keeping NATO in place as the main European security institution, instead of creating a pan-European security architecture, was a mistake, but it’s important to understand why that decision was made. While the US did not make that decision alone, as the leader of NATO, it was by far the main player. As we have seen, in February 1990, Bush essentially made a deal with Kohl. The US would support German reunification and, in exchange, Bonn would resist the temptation to secure Moscow’s assent to it by giving up NATO membership. Thatcher did not like German reunification, but she understood that she could not prevent it and wanted to preserve NATO, while France was interested in pan-European security but knew that it couldn’t impose such a solution once the US, the UK and Germany had agreed to maintain NATO’s primacy and therefore had no choice but to join the bandwagon. The motivations of US officials in adopting this position were complex and should not be caricatured or judged without taking into account what people knew at the time. Although I’m critical of the Bush administration’s decision to maintain NATO’s supremacy after the end of the Cold War instead of supporting the creation of a pan-European security architecture that would have included Moscow as well the Central and Eastern European countries, I think it had by far the best foreign policy team of any US administration since then and that it made that decision in large part based on solid reasoning that deserves to be fairly characterized. In short, US officials in Bush’s foreign policy team were no fools, so it’s important to understand why they adopted that position despite the drawbacks I have identified. Of course, that is not to say their policy cannot be criticized and again I think ultimately it was a bad policy, but to criticize it one has to understand not just the outcome but also the decision-making process in context.

First, Bush was a very cautious man and so were his main foreign policy advisers, which played a major role in how the US managed the end of the Cold War. In general, this caution was perhaps his greatest virtue, but unfortunately in this case it proved to be a liability and led to a suboptimal outcome. However, once you take into account the context in which the decision to maintain NATO’s supremacy was made, it becomes harder to blame Bush for making not only that decision but other controversial choices he made at the time. By 1988, Reagan was fully convinced that Gorbachev was serious about reform and was actively collaborating with him to bring about the end of the Cold War and assist in the transformation of the Soviet Union, but his Vice-President was not. When Bush became President in 1989, he immediately ordered a pause in US-Soviet relations, because he and Scowcroft did not believe that Gorbachev was sincerely interested in reforms. They suspected that he was just a smarter kind of Soviet leader who was just pretending to be a democrat to further traditional Soviet aims and were afraid that Reagan had been too hasty in beginning a rapprochement with the Soviet Union.199 They did not become convinced that Gorbachev was the real deal until several months later, when as promised a year earlier, he did not use force to stop the revolutions that were overthrowing communist regimes everywhere in Central and Eastern Europe. Some people argue that, had Bush trusted Gorbachev and collaborated with him from the outset instead of ordering the pause, perestroika might have succeeded, but I find that rather dubious.200 However, that is not to say that the pause ordered by Bush at the beginning of his administration did not have serious consequences, if only because as a result the US and the Soviet Union lost precious time to move forward on arms-control agreements. By the time Bush was ready to resume the negotiations, Gorbachev was less capable of delivering, so it’s plausible that had Bush never ordered the pause more ambitious arms-control agreements would have been signed. But more importantly, Bush did not abandon his cautious approach toward the processes then underway in the Soviet Union and Central and Eastern Europe, it simply took a different form and had far-reaching consequences on subsequent relations between Russia and the West.

Indeed, even after he became convinced that Gorbachev was a genuine reformer and that he really wanted to end the Cold War, Bush still kept him at arm’s length compared to Reagan and was much less willing than his predecessor to entertain a complete transformation of the nature of the relationship with the Soviet Union. That’s because he and his foreign policy team made a distinction between Gorbachev and the Soviet Union and they continued to regard the latter as a geopolitical adversary of the US. In their view, the Soviet Union had been weakened and for the moment was cooperating with the US, but it could nevertheless be counted on to eventually revert to a more hostile position. In particular, Bush and his advisers thought it was likely that, sooner or later, Gorbachev would be overthrown and that hardliners would take his place. Their strategy was therefore to lock in as many advantages as they could until the window of opportunity opened by Gorbachev closed and the Soviet Union reverted to a more adversarial position.201 The decision not to undermine NATO’s effectiveness by embarking on a very uncertain process to create a pan-European security architecture that would include Moscow was in part a hedge against that scenario. While it can and, in my opinion, should be criticized as lacking vision at a critical moment history, this is easier to say now with the benefit of hindsight and one must acknowledge that it was not crazy at the time. In any case, since Gorbachev was delivering on several key US goals, such as the reunification of Germany, the CFE Treaty and the liberation of Central and Eastern Europe, the Bush administration endeavored to prop up Gorbachev as much as possible to reduce the probability that he would be overthrown or at least get more time to extract more irreversible advantages from him before he was replaced. In particular, this meant supporting the center against the republics (with the exception of the Baltics whose annexation by the Soviet Union had never recognized), many of which had already started to agitate for their independence by the end of 1989.202 Thus, despite a widespread belief to the contrary, the US policy at the time was not to encourage the dissolution of the Soviet Union but on the contrary to support the center against the republics.

The most famous illustration of that policy is the speech Bush made in Kiev on his way back from Moscow on August 1, 1991, in which he clearly took side for Gorbachev and the center against Ukrainian nationalism:

Freedom cannot survive if we let despots flourish or permit seemingly minor restrictions to multiply until they form chains, until they form shackles. … Yet freedom is not the same as independence. America will not support those who seek independence in order to replace a far-off tyranny with a local despotism. They will not aid those who promote a suicidal nationalism based upon ethnic hatred. We will support those who want to build democracy.203

Three weeks later, there was a coup attempt in Moscow against Gorbachev, which quickly failed but undermined his authority and made the disintegration of the Soviet Union inevitable. This was precisely the kind of thing that US officials feared would happen, and what had motivated the policy of supporting the center against the republics to avoid undermining Gorbachev in the first place, but since the coup failed that policy has been described as inept and short-sighted by most commentators ever since. In particular, after the coup had failed The New York Times columnist William Safire sarcastically dubbed Bush’s speech the “Chicken Kiev speech” for his failure to espouse the cause of Ukrainian nationalism against the center, a moniker that has stuck to this day.204 But this was easy to say after the coup had failed and it had become clear that Gorbachev and the Union would not survive for long. Bush on the other hand did not have the benefit of hindsight and, while one could perhaps criticize his exact choice of words, I think his policy was the prudent thing to do under the circumstances. It’s not just that Gorbachev was delivering on US long-standing objectives, it’s also that US officials were rightly afraid that the disintegration of the Soviet Union could lead to widespread violence. This had already begun to happen in Yugoslavia, but the prospect of a civil war in the Soviet Union was far scarier, if only because a huge number of nuclear weapons were disseminated across the country. The fear of such a scenario, which Baker famously described as “Yugoslavia with nukes”, was another important factor in Washington’s policy toward the Soviet Union during that period.205

Nevertheless, the Bush administration support for the center against the republics remained purely tactical and the US continued to regard the Soviet Union as a geopolitical adversary, which had been weakened and was temporarily collaborating with the US because of Gorbachev, but would eventually reverse to a more hostile stance. Bush and other US officials made that view of the Soviet Union very clear in discussions with allies at the time:

At a meeting with the German delegation in London [during the Summit of the G7 in 1991], Bush disliked French agnosticism about NATO’s future mission. “Let’s be sure we don’t reopen things,” he said. “As long as Soviet missiles are aimed at the United States, I know who the enemy is.”206

This attitude was recently summarized by Jacques Blot, then a high-level French official directly involved in the negotiations on German reunification in the Ministry of Foreign Affairs, in connection with the failure of Mitterrand’s European confederation project:

On December 12 in Berlin, James Baker had sketched out another vision, that of an Atlanticist Europe, with a reformed and perhaps more political NATO as its linchpin. A reunified Germany would naturally find its place in it. This conception was an extension of the logic of the blocs. Negotiations with the USSR led to appeasement and a reduction in forces and arsenals, but the aim was to organize a balance that would maintain the counterweight of the American superpower against what remained the imposing and indisputable Soviet military power. Of course, during this period, President Bush, like the French President, was concerned not to do anything that might destabilize Gorbachev, but American diplomacy never ceased to push its advantage against a country that was always seen as an adversary to be weakened and its influence reduced, and not, as had been the case with Germany after the war, treated as a partner to be helped to recover.207

As Sarotte argued, despite what is often claimed (especially by former US officials), the evidence doesn’t support the view that “integration of the Soviets into new or existing institutions was dominant” in US policy at the end of the Cold War but rather that “the goal was to get the Soviets out”.208

As Vladislav Zubok put it, despite being urged by some people to base US policy on a more ambitious vision, Bush “continued to pursue the short-term vision of locking in the Cold War gains”.209 This attitude sometimes bewildered US allies, who found it short-sighted and tone-deaf, but the US was the undisputed leader of the Alliance and, as this example from July 1991 illustrates, they had no choice but to follow:

At the opening session of the G-7, which took place in Gorbachev’s absence, Bush shared the advice he had received from the scholars gathered at Kennebunkport. He also raised other points against the “Grand Bargain”: the Soviet state continued to spend a lot of money on its military needs, including modernization of strategic missiles, and Gorbachev was still unwilling to set the Baltic republics free. The US President also pitched the priority of helping the Eastern European states, rather than the Soviet Union. Some Western leaders in London felt embarrassed. The American team was finding every possible reason not to help Gorbachev. The Prime Minister of Canada, Brian Mulroney, made a moving speech. In 1985, he said, then Vice-President Bush had attended Chernenko’s funeral in Moscow. What would Bush have done then if, meeting him after the funeral, Gorbachev had said: I will free Eastern Europe, I will dismantle the Warsaw Pact, a united Germany will join NATO, a UN force will start a war against Iraq, the USSR will sign the CFE and START agreements, there will be elections and democracy, I will develop personal ties with America, and economic ties with the West will grow. “If Gorbachev had said these things in 1985,” Mulroney concluded, “I would have hurried in with a check.” Andreotti reminded the group of what Reagan had said in 1985: “I don’t know if Gorbachev will succeed, but none of us should have on our conscience the responsibility of not helping.” Yet nobody at the G-7 wanted to question American leadership. And nobody wanted to hold out the money bag.210

But adopting the kind of ambitious vision that many people urged him to embrace simply wasn’t consistent with Bush’s character. While Reagan was a visionary, his successor was first and foremost a pragmatist, who was risk averse and preferred certain but limited advantages to uncertain but more desirable outcomes. Again, this was generally a virtue, but in this case it proved to be a major flaw and we are still paying the price today.

However, Bush’s risk aversion was not the only reason why he did not support the creation of a pan-European security architecture that would have included Moscow and instead did everything he could to ensure NATO’s continued primacy in the post-Cold War era, he also had ulterior geopolitical motives. Indeed, thanks to the efforts of historians and the declassification of many documents from the US archives, we now have a very good idea of how US officials were thinking at the time and what sort of considerations guided their decision-making about strategy. Analysis of these documents shows that, in addition to the fear that Gorbachev would be overthrown by hardliners and that Moscow would revert to a more adversarial position vis-à-vis the US, officials in the Bush administration were also guided by the goal of sustaining and even reinforcing US supremacy in Europe in the post-Cold War era.211 In order to understand why that goal was the highest priority for the US in 1989-1991, and why this led Washington to do everything it could to ensure that nothing would threaten NATO’s primacy on European security, it’s crucial to understand the relationship between US power and its role as a European power. The European alliance is a major power multiplier for the US because, while individually the economic, political and military power of each European states is negligible relative to that of the US, collectively it’s formidable and the ability of the US to harness it greatly enhances its ability to further its goals. For instance, Washington’s ability to impose a cost to adversaries like China by denying them access to certain technologies through export controls — regardless of whether such a policy is a good idea — relies on cooperation with economically advanced allies in Europe, because in the long-run unilateral export controls will only harm US companies by reducing their market shares while US adversaries will be able to source the products and services they need from European companies. Similarly, while the US has by far the most powerful military in the world, the support of European countries still enhances the ability of the US to conduct military interventions abroad — again regardless of whether such interventions are a good idea — not only because they can provide troops but also and perhaps mainly because conducting these kinds of operations at the head of a large coalition rather than unilaterally is perceived as more legitimate both domestically and internationally.

But NATO is a major channel through which the US is able to exert political influence in Europe, because as long as European states depend on the US for their defense, they can’t easily ignore requests to cooperate with US policy. Of course, it’s not the only reason why they often agree to cooperate with the US (the community of values between the US and Europe, for instance, also plays a role), but it’s a major and arguably the main factor. As recently declassified materials show, this is partly why during that period US officials in the Bush administration were determined to ensure that new security arrangements emerging from the settlement of the Cold War would not threaten NATO’s primacy on security in Europe, because they otherwise would have undermined one of Washington’s main levers of influence on the continent and thereby also its ability to, as one US official put it, “harness European power” to further its interests or at least what it perceived as such.212 Just as Gorbachev foresaw that NATO’s continued primacy would eventually result in Moscow’s exclusion from the European security architecture, US officials predicted that unless it could be preserved in the post-Cold War era, the US would, if not be excluded from Europe altogether, at least lose some of the influence it had on the continent and with it the ability to use the European alliance as a power multiplier. Moreover, influence over Europe was not just a way for the US to channel European power to further its foreign policy goals, but also important to ensure that European states would not endanger US-led economic globalization. Indeed, as they felt more secure and less reliant on the US security umbrella with the end of the Cold War and became more able to coordinate with the progress of European integration, there was a risk that Europeans would become more reluctant to collaborate with the US to build such a globalized economic order and might even be tempted to endanger it by turning toward protectionism, which US officials believed would harm not only the US but also Europe and the rest of the world.213

As this last point makes clear, US officials did not see their desire to cement US dominance over Europe in the post-Cold War era as a kind of end in itself, but saw the US as a benevolent hegemon and thought that Washington’s continued influence over Europe would be good for everyone, including for Europeans. In their view, it wasn’t just that by preserving the US ability to influence decision-making in Europe they would be protecting US economic interests, but also that Europeans had to be protected from themselves, lest without the benevolent guiding hand of the US they should drift left and reject the neoliberal economic order Washington promoted at the time. Similarly, in the domain of security, US officials saw the preservation of NATO’s primacy not just as a way for the US to continue to be able to harness European power to further its foreign policy goals, but also to preserve stability in Europe and preempt the need for another costly military US intervention in Europe because in the absence of a US military presence the continent slid back into old habits:

Against this background, the White House also understood that U.S. policy towards Western, Central, and Eastern Europe was tightly linked together. “Our future as a European power will depend in large measure on how well we grasp these new opportunities,” the NSC wrote to the President at the beginning of 1990. Germany was quickly filling the European power vacuum in Central Europe. Soon, the only powers with “real influence” in the region were going to be Germany and the Soviet Union, the NSC argued. On the one hand, this was “not the architecture of a stable European security order,” American officials believed, as such a framework invited a return to the “cyclical pattern of Russo-German conflict and condominium that bedeviled Europe from 1870 to 1945.” On the other hand, U.S. engagement would help shape German reunification, manage “an eastward drift” in Germany’s policy, and strengthen the “future position” of the United States within European security and economic affairs. Thus, the NSC resolved that the United States had to remain engaged in Central Europe “between Germany and Russia.”214

While I have no doubt that US officials sincerely viewed things in that way, I don’t think this should completely absolve them of the charge of imperialism, because throughout history imperialists have always come up with good reasons and self-serving beliefs to justify their domination but it doesn’t mean that the will to power was not part of what motivated them. What this shows is rather that the motivations of US officials were complex and combined altruistic motives with more imperialist inclinations.

Indeed, the desire to preserve NATO’s primacy in Europe didn’t just lead the Bush administration to shoot down pan-European alternatives such as Mitterrand’s European confederation project or Gorbachev’s “common European home” concept, but also to start thinking about NATO expansion to Central and Eastern Europe as a way to prevent the emergence of a common European defense policy within the EU that would make it strategically independent from the US, which is harder to explain in non-imperialist terms. Up until recently, the consensus was that NATO expansion had never even been seriously considered by the Bush administration, which had been content to establish loose partnerships between NATO and Central and Eastern European countries. It was thought that the US only started to think seriously about NATO expansion under the Clinton administration and that the decision to expand it wasn’t made until 1994.215 However, as I have already noted above, US officials in the Bush administration had already started to consider NATO expansion to Central and Eastern Europe as early as 1990, during the negotiations on German reunification. In fact, recent work has shown that, by the end of Bush’s presidency, a consensus had emerged in the administration that NATO would have to be expanded, although the decision to do so had not formally been made yet and the timeline remained unclear.216 It’s very clear, based on declassified materials, that the desire to keep Europe dependent on the US and preserve Washington’s ability to influence decision-making in Europe played a key role in the thinking of US officials at the time:

Over the subsequent months, Washington planners assessed that Central Europe’s yearning for American involvement offered both challenges to and unique opportunities for cementing U.S. influence in Europe—influence that was deemed necessary for achieving broader U.S. political and economic goals. In terms of challenges, the Central Europeans’ fears—rekindled by recent instability and American hesitancy—ranked high. U.S. analysts concluded if the U.S. government blocked NATO expansion, the “new democracies” would lose interest in a transatlantic bond and seek entrance into Western European security structures. Nobody in the region believed seeking such shelter was “ideal.” In America’s absence, there was “little military teeth” in European frameworks. And yet, within a decade, the Europeans, led by Paris, would merge security and economic institutions, with the Community thus becoming the “de facto keeper of European peace.” As a consequence, NATO would be marginalized. Washington would retain “indirect and implied” security responsibilities due to the linkage between the Alliance and Western European security structures. Equally, the U.S. government would have “little to say” about European decision-making.217

With the end of the Cold War and the dislocation of the Soviet empire, the US was now the only superpower and intended to keep things that way, which required keeping allies dependent on the US and preventing the emergence of alternative power centers. It also had no intention of undertaking security responsibilities in Europe without any way to influence European decision-making. As we shall see, by the mid-2000s this goal had been reached in the case of the EU’s Common Security and Defense Policy (CSDP), whose potential for disrupting American leadership over Europe was effectively neutralized by ensuring that it would be tightly linked to NATO and would not compete with it.218

Another factor that cannot be neglected in the decision to preserve NATO’s primacy and reject pan-European security, though it would be wrong to say it was the only one, is the triumphalism that prevailed in the West in general and in the US in particular at the end of the Cold War. In the eyes of Bush and his foreign policy team, the US had won the Cold War and the Soviet Union had lost it, so the post-Cold War settlement had to reflect this fact. This was already clear when he met Kohl at Camp David in February 1990:

Having said that, the Soviets are not in a position to dictate Germany's relationship with NATO. What worries me is talk that Germany must not stay in NATO. To hell with that. We prevailed and they didn't. We can't let the Soviets clutch victory from the jaws of defeat.219

To be fair, because he wanted to protect Gorbachev against domestic attacks, Bush was careful not to put this triumphalism on display in public. But after Gorbachev resigned and the Soviet Union had ceased to exist, he let his true feelings be known publicly in his 1992 State of the Union address:

Even as President, with the most fascinating possible vantage point, there were times when I was so busy managing progress and helping to lead change that I didn't always show the joy that was in my heart. But the biggest thing that has happened in the world in my life, in our lives, is this: By the grace of God, America won the Cold War. [emphasis is mine]

I mean to speak this evening of the changes that can take place in our country, now that we can stop making the sacrifices we had to make when we had an avowed enemy that was a superpower. Now we can look homeward even more and move to set right what needs to be set right.

I will speak of those things. But let me tell you something I've been thinking these past few months. It's a kind of rollcall of honor. For the Cold War didn't end; it was won. [emphasis is mine] And I think of those who won it, in places like Korea and Vietnam. And some of them didn't come back. Back then they were heroes, but this year they were victors.220

Such feelings were understandable under the circumstances, especially for someone of Bush’s generation, but it would be naïve to imagine that they didn’t have some impact on the decisions that were made at the time about the post-Cold War settlement and in particular about how Moscow would be treated. As we have seen, Zelikow recently admitted as much, saying that in addition to the fact that pan-European security arrangements such as Gorbachev’s “common European home” and Mitterrand’s European confederation had no place for the US, the Bush administration rejected them because they “avoided expressing and voicing the final win of a system against the other one”.221

The problem is that the Russians had, and still do, a very different interpretation of how the Cold War ended. In their view, they didn’t lose the Cold War, but just stopped fighting it because they came to reject what Vladislav Zubok has called the “imperial-revolutionary” ideological foundation of Soviet foreign policy.222 There is a sense in which both the American and the Russian interpretations are correct. On the one hand, it’s true that the communist ideology championed by Moscow during the Cold War was defeated because a centrally planned economy just isn’t efficient, so the Soviet Union could only sustain the competition with the US by allocating a huge share of its GDP to the military, thereby keeping consumption low and being forced to resort to coercion to quell discontent. On the other hand, as Zubok showed in more recent work, the Soviet Union was not forced to stop fighting the Cold War.223 This couldn’t have lasted forever, because any political-economic system so at odds with human nature would eventually have collapsed, but it probably could have sustained the competition with the US for a while by continuing to allocate a large share of its GDP to the military at the expense of the well-being of its citizens while resorting to violence to keep them in check. This didn’t happen because Gorbachev wanted to improve living standards in the Soviet Union, which required that military expenditure be diminished radically, and in turn this was impossible without reducing tensions with the West, but in the process he came to reject the whole ideological edifice on which the Soviet Union was built and in particular to regard Soviet foreign policy as immoral. While his reforms ended up destroying the Soviet Union, which he had not intended, the fact that the Cold War ended when and how it did was very much the result of a choice on his part. Again, both interpretations are correct, but they focus on different aspects of the end of the Cold War.

This difference of interpretation regarding how the Cold War ended goes some way toward explaining the deterioration of relations between Russia and the West in subsequent years. Indeed, from the Russian point of view, not only did they freely choose to stop fighting the Cold War after rejecting the communist ideology and not because they were literally forced to so (in the way Germany or Japan were by total military defeat in WWII), but they collaborated with the US and its allies in settling that conflict in a way that was favorable to Western interests in part because they were promised that Moscow would have a place in the post-Cold War European security architecture. As we have seen, this view is the main reason why they felt they had been misled by the West, with more than a little justification, after NATO started to expand to Central and Eastern Europe and Russia was gradually excluded from the European security order. For the West in general and the US in particular, on the other hand, Russia was seen first and foremost as the loser of the Cold War and anything done by the West to accommodate it was essentially supererogatory. That is not to say that one should caricature the way in which the West treated Russia after the end of the Cold War and the dislocation of the Soviet Union, it’s not as if the US and its allies didn’t try to accommodate Moscow’s sensitivities to some extent, but as we shall see they did so mostly with trinkets that could not make up for the fact that, despite what they had been led to believe at the end of the Cold War, it was still treated as a potential adversary and increasingly excluded from the European security architecture. This point was made very clearly by none other than Baker a decade after the dissolution of the Soviet Union:

One sad lesson of the twentieth century is that refusing to form alliances with defeated adversaries is more dangerous than forming such alliances. The Treaty of Versailles solved the "German problem" in exactly the wrong way— by sealing the defeat with territorial and economic retribution. This resolution certainly demoralized Germany in the immediate aftermath of war, but the resulting grievances fueled something unimaginably worse and more dangerous. One hesitates to stretch historical analogies too far, but little imagination is required to see something similar happening in Russia. The end of the Cold War was certainly not settled by anything like Versailles, but from Russia's perspective, the results have been much the same—a humiliating loss of territory, prestige, and power. Russia's economic and political problems have stemmed largely from Russia's own failure to implement the necessary reforms to encourage the creation of a market economy, but it would not be difficult—indeed, it has not been difficult—for Russian demagogues to blame the West for the troubles ordinary Russian men and women have experienced during the past decade. Couple all these developments with the expansion of NATO up to Russia's doorstep and the use of NATO as an offensive (as opposed to defensive) alliance to bomb Russian allies, and you have a recipe for disaster. By continuing to treat Russia like a potential adversary, we may encourage it to become our enemy, the very thing we fear. The best way to find an enemy is to look for one, and I worry that that is what we are doing when we try to isolate Russia. The same principle would apply, of course, if we were to embark on a policy to try to isolate China.224

Sadly, his warning was not heeded, and today anyone who says the same thing would immediately be accused of being a Putin apologist or some such nonsense. But as I have argued in this section, this was to some extent predictable at the end of the Cold War, when Baker and the rest of the Bush administration’s foreign policy team made the decision to ensure NATO’s continued primacy in the post-Cold War era and shot down pan-European security alternatives. This made the subsequent trajectory of the relations between Russia, the West and Ukraine, which ultimately led to the Russo-Ukrainian war, much more likely.225 However, it wasn’t inevitable, and for the worst to happen, many other mistakes had to be made on both sides. In the rest of this work, I will explain what they are and why I think they could have been avoided, even if the choices made at the end of the Cold War made that more difficult.

• America’s elite university students are more demographically diverse than the general population, but more politically divided along lines of race, gender, sexuality, and religion.

• Minority and female students are far more liberal on campus than in the general population, whereas straight white Christian men are somewhat more conservative on campus than in the general population. Current trends portend a politics in which elite women, minorities, gays, and the nonreligious are more left-leaning while elite whites, males, and Christians remain relatively conservative.

• White Christians tend to cluster in red state flagship universities, which are the most politically balanced in the country and have similar shares of liberal and conservative students. Yet many flagship universities in flyover states with conservative reputations actually have more liberal than conservative students.

• A quarter of students are LGBT, and there are roughly equal shares of Christian and nonreligious students. LGBT, Nonreligious, and Christians are set to become more important political groups among America’s future leaders.

• Liberal arts colleges are the least politically diverse. Many have almost no conservatives, and thus very low viewpoint diversity. But they have high sexual diversity, at nearly 40 percent LGBT.

• Ivy League schools average 10-15 percent conservative and 60-75 percent liberal. Across 150 leading schools, there are nearly 2.5 liberals for every conservative.

• Democrats outnumber Republicans by a 55-23 margin on campus, and liberals outnumber conservatives 53-21. Elite students are thus two-thirds more Democratic and twice as liberal as the American population.

• Among elite students, there is a 15-point gender gap in political ideology and party identification between men and women. This is 3 to 5 times larger than the gender gap in the general population. It is also 2 to 3 times larger than the gender gap among either the 18-25 or college-educated general population. The campus gender gap has grown steadily since 2004.

• The university with the highest viewpoint diversity ranking is the University of Arkansas, whose students are 35% conservative, 37% liberal, 36% Republican, and 41% Democratic. The least diverse is Smith College, at 81% liberal, 1% conservative, 78% Democratic, and 2% Republican.

• Thirty percent of students and nearly the same share of academic staff in Social Sciences and Humanities (SSH) fields identify as the furthest left point on a 7-point conservative-liberal scale. For Sciences, Technology, Engineering, and Math (STEM) subjects, a smaller share – 20% of students and 10% of staff – identify as far left.

• Self-identified Jews make up only 3% of elite students and just 7% of Ivy League students, suggesting a considerable decline since the early 2000s.

• Homeschooled and parochial schooled undergraduates are as or more likely to identify as LGBT or non-binary as those from public or private school backgrounds.

Note: An interactive version of the table in the appendix that allows users to search by University and sort by characteristic can be found at http://www.sneps.net/viewpoint-diversity-ranking-of-leading-us-colleges.

Click “download" for a PDF

5.71MB ∙ PDF file

Download

Download

Introduction

The most elite fifth of undergraduates in the country will supply most of the nation’s future leaders. They are also an important leading indicator of emerging social and political trends. The Foundation for Individual Rights in Education (FIRE)’s campus expression surveys sampled 57,000 undergraduates across 159 of the top universities in the country, allowing for an unprecedented opportunity to learn more about this important group.

Method

This report makes use of FIRE’s distinctive 2020 and 2021 student free speech surveys, drawn from College Pulse’s panel of over 500,000 verified undergraduates.1 This contains some 57,000 respondents and a rich range of political questions to explore emerging trends among elite students. The FIRE survey focuses on students at research-intensive, selective institutions. In 2021, 75% of FIRE respondents attended one of the 113 Research 1 (R1) institutions, and all attended one of the approximately 335 R1, R2, or R3 doctoral-granting institutions or a prestigious liberal arts college. This means the data represents about 3 in 4 R1 institutions, just over 10% of R2 schools, and very few R3 or lower-ranked institutions. The colleges in the survey educate around 1 in 5 of the nation’s undergraduate students, focusing very much on the intellectually influential top tier.

Across the country, R1 schools, including those not in the FIRE dataset, enroll around 1 in 7 undergraduates. By comparison, the Higher Education Research Institute (HERI) Freshman Survey samples a representative stratified draw of around 100,000 freshmen pupils from roughly 100 institutions selected out of a possible 1,427 4-year degree-granting colleges.2 This report should therefore be viewed as a portrait of the political views and demographics of the approximately 15% of American undergraduates who attend top institutions. The FIRE sample is weighted during analysis by College Pulse data weights based on official data submitted by each university to the Integrated Postsecondary Education Data System (IPEDS) database and other sources.3

Student Demographics

Race

I begin with the basic demography of the 57,000 college students surveyed by FIRE. The survey aimed at a reasonably representative sample of the student body of research-intensive universities and elite colleges. In the combined 2020-21 sample of approximately 57,000 students, 52% of respondents are white, 19% are Asian, 11% Hispanic, 7% black, and 7% multiracial. With survey weights applied, the white share rises to 61%, which is about 10 points whiter than Gen-Z as a whole. The HERI survey, based on a far less exclusive sample of colleges, shows a more racially diverse student population, with 50% white, 11% Latino, and 10% black. However, the Asian share was lower in the HERI data, at 12%, and the multiracial component higher, at 16%.

In the FIRE data, only six colleges were more than 80% white, though all were at least 10% minority: BYU, Maine, UNH, Vermont, Wyoming, and Utah State. A larger number of institutions were less than 20% white, with three under 10% white, including Howard (0%), Cal State Los Angeles (6%), and UC Riverside (7%). These figures approximate to actual reported numbers from the universities themselves.

The share of white students in Ivy League universities averaged 41% in the survey data, with Asians at 27%, both of which tally with the university figures.4 Asian students comprised 19% of the FIRE student sample overall, well above this group’s share of the national population of 18-24-year-olds, which is around 5%.5 In the FIRE data, the colleges with the highest Asian share were Carnegie-Mellon, Stony Brook, and the University of Hawaii. These campuses’ official Asian component stands at 33%, 27%, and 37%, respectively. At the other end of the scale, Asians made up under 3% of students in the FIRE sample at six institutions (BYU, Kansas, Kent State, New Mexico State, Utah State, and the University of Maine).

The African American share in the dataset was 7%, ranging from less than 1% at four schools (BYU, Colorado School of Mines, Oregon State, University of Idaho) to 88% at Howard University, a historically black institution. Latinos made up nearly 11% of students, ranging from over half the student body at four colleges (Cal State Fresno, Cal State Los Angeles, Florida International, and New Mexico State) to between 1 and 2% at the University of Vermont, University of Maine, Howard University, and Ohio University. Official figures place the Hispanic share of the student population at around 55% at Cal State Fresno and 70% in the other three colleges.

Gender and Sexual Orientation

In terms of gender, 61% of the unweighted sample are female and 36% male, with 3% gender nonconforming. With survey weights applied the numbers are 50% female, 48% male, and 2% gender nonconforming. The unweighted figures are not far off the national student figures, in which around 60% of students are female. Nine institutions are more than 75% female in the data. This is accurate for mainly female colleges, though in some cases the samples overstate a particular university’s actual sex ratio. Clearly there is a certain amount of measurement error at the college level given an average enrollment of 20,000 undergraduate students and a sample size of around 250 students per school. This produces a margin of error of 6% for the typical college in the dataset.

Twenty-three percent of students identify as LGBTQ: 16% of men, 28% of women, and virtually all non-binary respondents. The bisexual and ‘questioning’ sexual orientation categories account for 15% of females but only 6% of males, while the gay/lesbian share is around twice as high among male students (6.1%) as females (2.8%). Non-binary students comprise 3.6% of the 2021 sample, adjusted to 1.1% in the weighted data. In 2022, the question was broadened to 7 categories, leading to 3% of the weighted sample identifying as non-binary. In the HERI data, a lower share of students, 15%, identify as LGBTQ.

Much of this pattern can be accounted for by the share of LGBTQ respondents among young people, with 21% of Gen-Z identifying as other than heterosexual in the most recent Gallup survey. The unweighted Cooperative Congressional Election Study (CCES) finds no difference between students and non-students in LGBTQ share. Regardless, the substantial LGBTQ quotient among the young is a trend that is independent of the effects of college, although LGBTQ individuals appear to be self-selecting into elite higher education institutions in somewhat larger numbers compared to lower-ranked institutions.

Across 159 campuses sampled, three schools contained a weighted majority of students identifying as LGBTQ – Oberlin (51%), Wellesley (61%), and Smith College (70%). Given the 6% margin of error in the data for these colleges and possible noise from any sampling biases, these figures should be treated as indicative rather than definitive estimates. On the other hand, the 38% share of LGBTQ students among the more than 4,000 students in the 23 liberal arts colleges sampled is accurate to within a 1% margin of error. This suggests several liberal arts colleges, such as those listed above, could be majority LGBTQ. At the other end of the scale, LGBTQ students made up no more than 15% of students in the weighted sample at BYU (10%), Utah State (11%), Bucknell (13%), University of Miami (14%), Notre Dame (14%), and Clemson (14%). Importantly, no institution had fewer than 10% LGBTQ students.

There were minimal differences between Ivy League, R1, R2, and R3 schools on this measure, but, as Figure 1 shows, the LGBTQ share across the 23 liberal arts colleges was substantially higher, at 38%. Part of this effect is linked to the strongly liberal ideology at these colleges, where ‘very liberal’ students are 37% LGBTQ compared to just 5% of ‘very conservative’ students. Liberal arts colleges sport a relatively female (56%) and liberal (72%) student body, though these factors alone do not fully explain their exceptionalism on this measure.

Religion

Turning to religion, 47% of students identify with a Christian denomination, with 38% stating they have no religion. Among whites, 48% are Christian and 39% have no religion. FIRE surveys do not ask about attendance, belief, or fine-grained sect. Students ticking the Protestant, Catholic, or ‘Just Christian’ box may be only nominally religious. It is unclear how many are evangelical or mainline Protestants, or whether Jews are Orthodox, Reform, or Conservative. Those identifying as ‘Just Christian,’ for instance, are no more conservative or Republican than those who ticked Protestant as their denomination.

Twenty-four campuses had students who were majority nonreligious (identifying as atheist, agnostic, or ‘nothing in particular’), many of which are liberal arts colleges or part of the California state system. At the extreme, 6 in 10 students at Colorado College and 57% at Smith College reported no religious affiliation. On the other side of the ledger, just seven schools had a student body where at least 70% identified with a Christian denomination. These were Pepperdine (71%), Auburn (72%), Mississippi State (74%), Notre Dame (77%), Baylor (78%), Utah State (79%), BYU (89%), and Hillsdale (93%).

Jews averaged 3% of students in the sample, but comprised 15% at Tulane University, 18% at Barnard College, and 39% at Brandeis. All three schools are known to have significant Jewish populations, though these figures suggest the share identifying as Jewish, at least in terms of religious affiliation, is less than some believe. Muslim students made up 2.8% of the total but formed around 10% of the student body at two campuses: the University of Illinois at Chicago (14%) and the University of Texas at Dallas (9%). Again, these figures should be treated as indicative rather than point estimates, given the 6% margin of error in the sample.

Aggregation by institutional type, which is at a higher level than the college, offers more precision, with a margin of error in the 2% range. Figure 2 presents the broad religious patterns by institutional type. Nonreligious students are a near-majority of students in liberal arts colleges and a 43% plurality in the Ivy League, where they outnumber Christians 43% to 37%. In R1 universities, which comprise most of the sample (and include the Ivy League), Christians are only slightly more numerous than those without religion. In R2 and R3 institutions, however, Christians are a majority. In fact, in R3 institutions, the nonreligious share is just half as large as the Christian share. The proportion Protestant (including ‘Just Christians’), though not shown in Figure 2, is also much higher in R3 institutions. Among white students, Protestants outnumber Catholics 4:1 in R3 institutions but only between 1:1 and 2:1 in the more elite R1 and R2 institutions.

Finally, Jews make up a mere 3% of R1 students in the sample – little above their share of the population – and only 7% in the Ivy League. This figure is, ironically, lower than the discriminatory quota applied by Ivy League universities in the early-to-mid twentieth century.6 The share of Ivy League undergraduates identifying their religious affiliation as Jewish stood at around 20% as recently as the early 2000s but has since declined significantly. These figures confirm similar data collected by universities, but reveal just how advanced this trend is.7 It is possible that a larger share of ethnic Jews are choosing to tick ‘no religion’ rather than Jewish, though one would expect ethnic affiliation to prompt them to select the Jewish box more so than ex-Christians. Precisely which combination of secularization, lower entry qualifications, or the overall decline in the white share of students at top institutions accounts for the lower-than-expected share of Jews is a question for future research.

Differences by Discipline, and Between Faculty and Students

We also see important distinctions by discipline when we put students’ majors into two broad categories, Social Sciences and Humanities (SSH) and Science, Technology, Engineering, and Mathematics (STEM). Figure 3 compiles data from the FIRE surveys as well as a smaller faculty convenience sample I gathered in 2020. The latter is not a random sample, but its political composition matches that of studies that use complete samples and voter registration data, and is thus a reasonable representation of the demographics and political beliefs of faculty in the leading 100 US universities, and also comparable in scope to the FIRE sample.8

SSH students are about 3 points whiter, 13 points more LGBTQ, 11 points more far-left, and 7 points less conservative than STEM students. The proportion Christian and No Religion is similar between both disciplines. So-called ‘Studies’ majors (Women’s, African American, Native American, Latino, Ethnic) are 75% non-white and 58% LGBTQ. Interestingly, students in these fields are no less Christian than others, and fewer are nonreligious. They are more far-left (39%) than SSH students are (30%), while being considerably less conservative (5%) than SSH (14%) or STEM (21%) students.

In line with existing findings, faculty are considerably less racially diverse than students. 85% of SSH faculty and 79% of STEM faculty are white, according to my faculty sample. This compares to 60 and 57%, respectively, among SSH and STEM students. The Jewish share is also substantially higher among faculty – 13% in SSH and 10% in STEM – than among students (5% and 3% for SSH students and STEM students, respectively). As with students, the far-left share among faculty is higher among SSH (28%) than STEM (12%). Conservatives are also more numerous in STEM (10%) than SSH (5%) among academics. Disciplinary differences are thereby somewhat more pronounced among faculty than students.

Finally, the similarities between staff and students should be noted: conservatives and Christians are greatly underrepresented among both in comparison to the population, while those without religion and far-leftists are heavily overrepresented (except for far-leftism among STEM faculty, which is around the national average). In all cases, the discrepancy between SSH and STEM is more pronounced for faculty than students. In addition, minorities are underrepresented, and Jews overrepresented, among faculty, but both groups come close to their national share among students. Finally, the share of conservatives is 2 to 4 times higher among students compared to their professors. These figures indicate that the culture of the university is more exaggerated among the professoriate than the students, even as both share a distinct religious and political profile in relation to the wider US population.

The Politics of Students

American students lean left, but not as heavily as some might believe. Political ideology in the two waves of the FIRE survey revealed 53% to be liberal, 26% moderate or unsure, and 21% conservative. Party identification is 55% Democrat, 22% Independent, and 23% Republican.9 This means there are about 2.5 liberals or Democrats for every conservative or Republican among students.

This is similar to other surveys and shows that students differ a good deal from the national average. The 2016 American National Election Study (ANES) reports that in the wider population, there is a roughly even Democrat-to-Republican party identification ratio and a 40-33 conservative-to-liberal lean.10 Among all Americans with degrees in the ANES, the partisan ratio is 44-34 Democratic-to-Republican and 41-38 liberal-to-conservative. So American students are considerably more liberal and Democratic than the average college-educated American.

But American students are less exceptional when we take their age into account. In the 2020 presidential election, Gen-Z broke 65-31 for Biden, according to the exit polls.11 For party identification, the large-sample CCES shows that among Americans under 25, the (survey-weighted) partisan split in the US population is 30-19 in favor of the Democrats against the Republicans. For Americans under 25 with a degree, the partisan balance in the CCES tilts to 46-20 in favor of the Democrats. This is within 10 points of what we see in the FIRE student surveys and mirrors the 2020 exit polls. Triangulating with other sources, the FIRE student figures match other estimates and indicate that the student body largely, but not fully, mirrors the wider Gen-Z population from which it is drawn.12 Results from these external election surveys suggest that both age and education skew the D:R ratio from 1:1 in the population to 2:1 in the student population to 2.5:1 in the elite student population, with age a more important influence than education level in explaining students’ Democratic tilt.

Moving from party identification to ideology, the FIRE student sample (53% liberal, 26% conservative, 21% moderate) is more liberal than the wider US student average as reported by HERI, which is 37% liberal, 20% conservative, and 43% ‘middle of the road.’13 Though the moderate share is much larger in the HERI than the FIRE data (43 vs. 26), a 7-point question similar to FIRE’s without the ‘middle of the road’ phrasing may have drawn more leaners to choose one side or the other.

More importantly, the elite nature of the FIRE sample as compared to HERI helps explain the discrepancy. Bracketing differences between the size of the moderate group in both surveys, in HERI the liberal-to-conservative ratio is a bit less than 2-to-1, and in the FIRE surveys it is 2.5 to 1. The fact that R1 schools in the FIRE sample score over 3 to 1 in favor of liberals while R2 and R3 institutions lean just 2.2 to 1 indicates that the elite nature of the FIRE sample likely accounts for most of the discrepancy. Overall, there is more similarity than difference between the two surveys.

The HERI 2-to-1 ratio reflects the young population more generally. Thus 2020 data from the Survey Center on American Life (SCAL) shows that Gen-Z overall leans 42% liberal, 19% conservative, and 39% moderate or other – very similar to the HERI data that includes low-ranking colleges.14 The ANES shows more of a difference between well-educated and less-educated Gen-Z respondents: the under-25 population with at least some college is 51% liberal and 29% conservative, while the wider under-25 population, including those without any college education, is 40% liberal and 34% conservative. SCAL and ANES have small Gen-Z samples, but they suggest that FIRE’s elite sample is somewhat more liberal than the wider student average and, in addition, that elite university students are somewhat more liberal than young people who choose not to attend college.

Segregation

A growing literature points to the geographic sorting of the American electorate, with fewer Americans living in politically mixed places and more in ‘landslide counties’ where one party wins each election by a large margin. Educated people in particular, who in today’s political climate lean left, are far more concentrated in major metropolitan areas than in 1970.15 The more urban a locale, the more liberal and Democratic, with the reverse holding true in rural areas.16 In addition to geographic segregation, partisans are increasingly segregated in their social networks, even if they reside in an area not dominated by their political tribe. Thus, around 55% of Republicans and Democrats socialize exclusively with those sharing their partisan inclinations.17 Highly-educated Democrats are least likely to have politically mixed networks compared to the less-educated or highly-educated Republicans.18 This hyper-segregation among well-educated Democrats may be relevant for elite American students, though the FIRE study did not specifically ask about this.

Figure 4 uses a standard measure of segregation known as the Index of Dissimilarity (ID) to assess how segregated groups of students are compared to what would occur if students of various characteristics were randomly allocated across universities. The ID in this case is based on comparing each group to those who are not members of the group. The proportion for each group represents the share of each that would have to move university in order to achieve a uniform distribution of the group across all campuses.

The most important form of segregation in the FIRE data is ethnic segregation, with college-level IDs for various groups in the mid-30s. This is followed by political and religious segregation in the .15-.25 range. Conservatives cluster more in certain colleges than liberals or far-leftists do. Sexual orientation and gender are only weakly segregated, with IDs around .12-.13, exhibiting less variation from one campus to the next than ethnic, religious, or ideological characteristics.

Most ethnic groups have a college-level ID in the mid-30s, from .33 for African Americans to .38 for Hispanics. On the one hand, these figures are a fair bit lower than national levels of residential segregation between racial groups at the level of the census tract (geographic units averaging 1,000 to 8,000 people) and would be considered low-to-medium segregation on the Massey-Denton classification scale.19 For instance, the geographic census tract ID in 2020 is 55 for African Americans, 45 for Hispanics, and 40 for Asians. This means 40 to 55% of these groups would have to move census tracts to achieve a random distribution of groups across tracts, eliminating segregation.20 Interestingly, whereas blacks are the most segregated group in residential terms, when it comes to elite colleges, the white segregation level of .35 is slightly above that of blacks (.33), an inversion of the residential pattern. One takeaway from this is that African Americans are much less segregated between research-intensive universities than they are between American neighborhoods.

Whites’ relatively high level of segregation compared to national residential patterns could be the result of whites choosing to attend relatively white schools, but, since they are a large group, may also arise from where they live or the choices made by other groups. For instance, focusing only on the 13 colleges of California in the dataset, where white students are only the third-largest group (23%) after Asians (34%) and Hispanics (26%), shows that the white level of segregation of .20 is considerably lower than that of Asians at .39 or Hispanics at .35. This does not necessarily indicate that whites self-segregate less than other groups, though, because more California whites, compared to other groups, may have opted for colleges outside the state, while more Hispanic students are concentrated in less prestigious large public colleges that are more closely tied to local residential patterns. White students in the state are relatively evenly distributed across the larger public colleges, but the white share is substantially higher in certain smaller institutions like Pepperdine or Claremont, which don’t affect the ID scores much.

While collegiate segregation is lower than residential segregation, IDs in the mid-30s do not necessarily indicate that students self-segregate less when choosing a college than adults do when choosing where to live. After all, neighborhoods often reflect inherited patterns among settled communities while a student’s choice of college represents a fresh start, and student bodies are constantly turning over so they might be expected to be less segregated. We might further assume that the most elite 15% of colleges (those comprising the FIRE dataset), many of which are residential, would limit the impact of local residential segregation in shaping the student body. The fact that the racial group ID of leading American universities is in the 30s could thereby be viewed as evidence for a relatively high level of inter-collegiate segregation.

In truth, the figures reflect a mix of residential segregation patterns, educational segregation – as people to some extent choose schools that resemble themselves – and de-segregating forces linked to prestige or, in some cases, affirmative action policies. Thus, as Figure 5 shows, students of various backgrounds are fairly evenly distributed between the eight Ivy League schools, a marked contrast to the national picture among (mainly R1) universities in the survey.

While this could simply mean that all Ivy League universities have similarly high levels of segregation (they are, after all, highly liberal outliers with few conservatives and Christians), the fact there are few differences among them is telling. It indicates that local residential segregation in Ivy League communities like Ithaca, NY or New Haven, CT has less of an impact on the composition of the student body than is true of less prestigious institutions. Prestige factors and a high share of out-of-state students thereby override the influence of local geographic patterns. Common affirmative action policies may play a role in reducing racial disparities in representation between Ivy League schools but cannot account for the limited ideological and religious clustering that is also observable between Ivy League institutions. These results suggest that segregation among lower-ranked colleges outside the FIRE dataset is greater, reflecting differences in local residential patterns.

How much does geography matter for various forms of inter-collegiate segregation? The degree to which variation in a state’s ethnic, ideological, or religious composition affects that of its colleges differs widely depending on the group in question. For instance, in Figure 6, the dot for each college is labeled with the name of the state in which it is located. While there is as much as 50 points variation in white share between universities within states, some states like West Virginia and Hawaii have only one college in the dataset. Moreover, the racial makeup of a state really matters. No campus in Wisconsin is under, and none in California over, 60% white. Sparsely populated, largely white states like Nebraska and Wyoming are represented by one or two flagship universities with more homogeneous university populations. Nearly half (.43) the variation in white share across colleges is explained by the state in which a college is located.21

Using the same method for other characteristics in Figure 7 shows that geographic variation underlies a great deal of the inter-collegiate segregation we find for Hispanics, Whites, and Christians but explains less about the institutional clustering of Blacks, Asians, and Jews. This suggests institutional choice plays considerably more of a role than geographic location in accounting for the clustering of the latter. Notice as well that state-level variation seems to account for more of the variation in White Christian share from one university to another than it does for variation in the overall Christian share.

How can this be the case? We might expect that subgroups like minority conservatives would exhibit a level of segregation midway between the larger minority and conservative main groups. Since minorities and whites have a segregation ID of .34 and conservatives are at .25, we should expect something around .29 for both white and minority conservatives. In fact, we see sharp differences compared to what we would expect. Examining subgroups within the main racial, religious, and ideological groups in Figure 8 shows something interesting. Namely, that white Christians, coded in red, are more segregated than non-white Christians, coded in brown, by a margin of .33 to .23, while white conservatives are twice as segregated as minority conservatives, by a .33 to .17 margin.

On the other hand, minority liberals (.29) and far-leftists (.28) are more segregated than white liberals (.21) and far-leftists (.22). It seems that ideological and religious attachments, especially being Christian or conservative, reduce the level of segregation for minorities but do not do so for whites who are Christian or conservative. Liberalism and far-leftism reduce segregation a lot for whites and somewhat for minorities. All of this could be interpreted as indicating that religion and ideology reinforce racial group identity for white conservatives and white Christians but crosscut it for non-white conservatives, non-white Christians, and especially for white liberals. The result is that white conservatives and white Christians cluster more in certain institutions – often those in less populated states like Wyoming or Utah – than minority conservatives and Christians, or white liberals. They appear less willing to move to colleges in more diverse states than their white liberal co-ethnics.

The net result is that the typical white Christian attends a college that is 34% white Christian though the student sample is just 26% white Christian. By contrast, 30% of white liberals’ classmates are white liberals, even as this group forms 27% of the elite student total. White conservatives’ fellow students, meanwhile, average 22% white conservative though this group comprises only 14% of elite students. Here we have evidence that white Christians and white conservatives are behaving in a more self-conscious manner than whites without religion, white liberals, or non-white conservatives and Christians. That is, their ideology and religion do not moderate their ethnic segregation much, if at all.

Trends Over Time

The large-scale HERI freshman surveys allow us to track change in the ideological composition of undergraduates across a range of institutions representing all 4-year undergraduates over time. As noted, 37% of students in the HERI data in 2019 identify as liberal and 20% as conservative. The figures for the FIRE survey are 53% liberal and 21% conservative, reflecting the smaller share of moderates in the R1 institutions which make up the FIRE sample.

HERI survey trend data in Figure 9 reveals that the freshman population was very liberal in 1970, after the student revolts of the late 1960s. It grew more conservative during the 1970s. By 1981, slightly more students described themselves as conservative or far right than liberal or far left. Since then, the share of liberals has increased steadily, albeit with fluctuation, from 20% in 1980 to 37% in 2019 (not shown in the chart). The proportion of conservatives has, on the other hand, remained flat at around 20% during a period in American politics when ideology became more meaningful and aligned with partisanship. Given national trends, we would have expected a rise in conservatism among students as well.22 The lack of such a rise means American students are nearly twice as liberal as they were in 1980.

A portion of this may be explained by demographic shifts in the student population, from 20% minority in 1976 to 27% minority in 1997 to 47% today.23 The share of women has also risen substantially, from 42% in 1970 to 52% in 1985 to nearly 60% in 2021. I estimate that around one point of the 10-point liberal shift since the mid-1980s can be accounted for purely by changes in gender composition. The conservative share, assuming women moved against conservatism in proportion to the rate they moved toward liberalism, would be similar.

Estimates are more difficult for minorities as their doubling since the 1980s has been accompanied by a shift from more liberal black students to somewhat less liberal Hispanic and Asian students. Using the FIRE data, showing that minority students are about 10% more liberal and 50% less conservative than whites, indicates that rising racial diversity barely affects the 10-point liberal increase since the mid-1980s. However, a near-doubling of minority share since 1985 is likely to have reduced the conservative proportion by 1.5 points. Thus, the combined increase in female and minority students may account for 3.5 points of the 15-point increase in the liberal-over-conservative gap since the mid-1980s, nearly a quarter of the effect. While this is important, demographic shifts are not the main driver of student political change. Instead, a left-wing drift caused by the rising prominence of cultural issues in party politics that have a stronger age, gender, sexual orientation, and education gradient may explain the trend.24

The Gender Gap

Alongside the trend of greater student liberalism over time is a widening gender gap: males have inched up a few points in liberalism since the mid-1980s, but females have risen from around 25% liberal in the mid-1980s to 42% in 2016. The big increase appears to have begun from when George W. Bush took office in 2004. What has emerged is a large gender gap among students, especially elite college students, that is about twice as large as that in the general population.

Student trends appear to be a more intense variation on wider developments in American politics. The gender gap in presidential voting among those of all ages has steadily increased, from zero in 1976 to 11-13 points during 2016-20.25 However, the gap in self-professed ideology has not widened as much. American National Election Study (ANES) data presented in Figure 10, using the entire under-30 population as my comparator pool to increase sample size, shows only a small increase in the gender gap among young people for ideology over the past few decades.

Between 1988 and 1996, the discrepancy changed from women favoring conservatism over liberalism by 5 points to their favoring liberalism by 15 points. Thereafter, the difference among women narrowed to just 5 points in favor of liberalism, holding steady through 2016.26 The ideological gender gap therefore appears to be three times as large within the student population as it is outside it.

Excluding female colleges and taking the share who identify as liberal minus the proportion conservative gives an average gender gap in the FIRE data of 29 points between men (who lean liberal by 44 to 28, or 16 points) and women (who tilt 61-16 liberal, that is, by 45 points). Democrats likewise have an advantage over Republicans of 42 points among women and 13 points among men, again a 29 point difference. The concentrated effects of the new cultural cleavages in American politics on young educated women (a substantial share of whom are LGBTQ) may explain the large gender gap among students.

Going college-by-college, there is a wide range of gender partisan gaps, from Georgia State, where women lean Republican more than men by 8 points, to the University of Maine, where women are 33 points more Democratic than men. These figures are generally subject to a margin of error of around 6% and sampling bias is also a potential issue, so it is difficult to speak precisely about any one institution.

We can have high confidence in the broader patterns, however. Figure 11 plots the gender gap in partisanship by proportion white, since white share appears to be the most important correlate of the gender gap. The gender gap is significantly lower in colleges with fewer white students. This arguably reflects the lower level of political polarization and political mobilization around gender found among minorities.27 The partisan gender gap is not associated with a college’s sex ratio or its political slant.

The American Student Voter

The large sample size in the FIRE surveys allows us to answer a classic question in electoral studies: which characteristics best predict a student’s partisanship and ideology? In statistical models, race, gender, sexual orientation, and religion account for 24% of the variation in students’ 7-point liberal-conservative ideological identification and 19% of the variation in student responses to a 7-point Democrat-to-Republican party identification scale. Logistic regression models using the same parameters have a model fit (pseudo-R2) of .21 for predicting a Republican student and .19 for predicting a conservative. These are strong demographic models for political science: the same specification in the 2020 CCES, with a very similar sample size of around 60,000, predicts 12 to 13% of the variation in ideology and partisanship in the general US population. In ANES, albeit with a much smaller national sample, the model predicts just 6 to 10% of the variation. Thus race, gender, sexuality, and the religion/nonreligion distinction seem to matter more for the politics of students than the general population.

The most important sociodemographic predictor of a student’s political ideology and party identity is whether they affiliate with a Christian denomination or have no religion. These are two relatively evenly sized groups in the student body at elite colleges, with 44% indicating a Christian denominational affiliation (as Catholic, Protestant, Orthodox, Mormon, or ‘Just Christian’) and 38% saying that they have no religion. Note that those who identify as Christian may be only nominal Christians who rarely attend church and may not hold religious beliefs. Likewise, some who say they are not religious may hold religious beliefs and even attend a church.28 All else being equal, Christian students are 2.8 times more likely to be Republican than the average. Jewish, Hindu, and Muslim students are, respectively, 2.2, 1.8, and 1.6 times more likely to be Democrats than the typical student.

Race and gender have similar explanatory weight. Whites, who form half the student body, are 2.2 times more likely to be Republican than the typical student, and men are 1.8 times more likely to be so. LGBTQ individuals form a quarter of the student body, an important demographic that renders the heterosexual student population a meaningful category. Straight students are 1.5 times more likely to identify as Republican than the average student.

Seventy-seven percent of students identify as heterosexual, and just 4.6% as gay or lesbian. Fully 10.5% are bisexual, with the remaining 8% roughly spread among pansexual, queer, questioning, or ‘other.’ Heterosexuals and gays/lesbians account for 90% of men but only 75% of women, showing how prominent the bisexual or other category is among women. When it comes to partisanship, 21% of heterosexual females are Republican compared to under 4% of lesbian, bisexual, or other women. Among men, the share of Republicans is 32% for heterosexuals, 11% for bisexual/other, and 8% for gay men. This indicates that the sexual orientation gap is 5:1 among women and 3:1 among men, and thus accounts for a larger share of partisan variation among women than men. Together with the fact that nearly 30% of women are LGBTQ, this means – in the context of Gen-Z having twice the share of LGBTQ as Millennials, five times that of Gen-X, and eight times that of Boomers29 – that the female sexually-nonconforming group is an important factor that may help explain the leftward drift in the female student population over time.

White Christians are an important subgroup of white students that stand out as especially Republican, considerably more so than either non-white Christians or whites with no religion. Finally, whether a person is in the 62% of students receiving financial aid or the 35% who are not only matters a little. Those not receiving aid are 1.33 times as likely to be Republican as the average student.30

Figure 12 shows how partisan identity differs by various demographic and social characteristics. Students without religion or who are LGBTQ break around 70-8 in favor of the Democrats. Christians and men are more Republican, but still lean 44-33 to the Democrats. The only Republican-leaning groups are two tiny demographics, those who have been homeschooled (1% of students), who lean 44-25 for the GOP, and men in fraternities (3%), who are finely balanced at 33-31.

In terms of schooling, results suggest that attending a private or public high school makes little difference for a student’s ideology and party identity, while even among parochial, i.e., religious, high school students (over half of them Catholic), Democrats outnumber Republicans 45-33. Interestingly, around a fifth each of homeschool and parochial school students identified as nonreligious, suggesting considerable religious loss among both groups in young adulthood.31 One further curious finding is that those who attended parochial school or were homeschooled were at least as likely in 2021,and more likely in 2022, to identify as LGBT. In 2022, for instance, female students from a parochial school background were 11 points more likely, and those from a homeschool background 3 points more likely, to identify as non-heterosexual compared to those from public or private school backgrounds. Non-binary and other forms of unconventional gender identity were also higher among homeschooled and parochial schooled undergraduates.

Majors play some role: among white students, social sciences and humanities (SSH) majors lean 59-19 for the Democrats against the Republicans compared to 48-30 for science, technology, engineering, and mathematics (STEM) majors. There is almost no disciplinary gap among minority students, who lean about 64-11 in favor of the Democrats for both sets of majors. The biggest disciplinary difference is the share of those with far-left ideology: overall, 32% of SSH majors identify as ‘very liberal,’ the furthest left point on a 7-point ideology scale, compared to 19% of STEM majors. There is a smaller difference when it comes to the share of conservatives (20% in STEM, 13% in SSH). This reflects a similar pattern to what we find among SSH and STEM academics, where the main difference pertains to the share identifying as far left rather than the proportion conservative.32

Institutions, not just groups of individuals, tend to vary substantially in political composition. Figure 13 reveals that private universities lean more Democratic (62-17) than public ones (51-26). Ivy League campuses are even more left-leaning, with 68% of students identifying as Democrats and just 12% as Republican. By contrast, the less prestigious R2 colleges are a slightly more balanced 48-27. FIRE’s free speech rankings code those with good free speech policies green and those with censorious policies red, with those in between labeled yellow. This is associated with some differences in viewpoint diversity: green campuses break 52-25 for the Democrats while red ones are a more pronounced 60-17.

Regionally, southern colleges are more conservative, though their students still lean Democratic by a 49-29 margin. Northeastern universities are the least politically diverse, with just 17% of students identifying as Republican. While I have focused on differences, it is also worth noting that there is at least some political diversity in each type of institution. Students at Ivy League schools, for example, are 14% conservative and 12% Republican-leaning. This means that close to 10,000 Republican undergraduate students graduate from Ivy League colleges each year. Alumni include Republican elites such as Josh Hawley, Ted Cruz, Brett Kavanaugh, and Tom Cotton.

At the extreme, we can use statistical models to construct subtypes that illustrate the political poles in the student population. For instance, in Figure 14, just 2% of the 425 female LGBTQ Ivy League students are Republican, with 85% identifying as Democrats. Only 2% of 1,586 minority females at Ivy League institutions support the GOP, with 76% identifying as Democratic. At the other extreme, nearly 6 in 10 white heterosexual male Christian students are Republicans, and just 21% are Democrats. The most finely balanced group is white males, who favor the Democrats over the Republicans by a narrow 40 to 36.

Social Identity Gaps in the Elite Student Population

The pattern that emerges from comparing national election data and FIRE data on elite students is twofold. First, that students are considerably more diverse in terms of sexual orientation and religion than the American public. Second, that gender, and to a lesser degree race, are more important determinants of ideology and party identification among elite students than among the wider public. Religion and sexual orientation have a similar predictive power for both the public and students, but the fact there is a larger share of nonreligious and LGBTQ people among students elevates their importance for predicting the total variation in political attitudes.

In terms of race, what is noticeable when comparing students to the public is how few conservatives there are among minority students compared to the wider minority public. Figure 15 shows that 36% of minorities in the CCES (i.e., the public) identify as conservative compared to just 13% of minority students in the FIRE sample, a nearly threefold difference. By comparison, 26% of white students identify as conservative compared to 47% of white adults, less than a twofold discrepancy.

Comparing students with their corresponding 18-25 age group in the wider population shows that minority elite students are markedly more liberal (24 points more liberal and 24 points less conservative) than their non-elite age cohort while white elite students differ less (20 points more liberal, 10 points less conservative). Moreover, there is only a small difference between white elite students and the white public sample of under-25s in the CCES but a large difference of 13-19 points between elite minority students and young, degree-holding members of the minority public. Minority elite students appear to be distinctively liberal compared to both their own classmates and their racial peers.

Race is also an important variable because sexual orientation, religion, and to some extent gender, tend to split the white population politically more than they do the minority population. Figures 16 and 17 set out the pattern for elite students and the general public respectively. Gaps represent the average of the difference between the conservatism plus the liberalism of contrasting identity groups.

In Figure 16, which examines the student population, white men are 16 points more conservative and less liberal than women, and minority men are 13 points more so than minority women. This is only a modest difference. However, consider the next two sets of bars which show that identity-political dynamics are more intense within the white population. On sexuality, straight whites are 31 points more conservative and less liberal than LGBTQ whites, while the corresponding gap among minority LGBTQ and heterosexuals is 12 points. On religion, white Christians lean right by 37 points more than whites without religion. For minorities, the difference between Christians and those with no religion is only half as large, at 17 points.

Comparing elite students with the wider public in Figure 17 shows one major difference: the gender gap in ideology is a negligible 3 points between men and women, whether white or non-white. Men and women in general do not differ much in their liberalism or conservatism in the wider public. This contrasts with the 16-point difference among white students and 13-point gap among non-white students in Figure 16. Clearly gender is a more salient political divide on campus than off it.

By contrast, the other sets of bars look relatively similar between students and the public. Identity divides around sexuality and religion loom much larger among whites than among minorities, as with the student figures. LGBTQ members of the public who are white are 25 points more liberal than heterosexual whites, and minority LGBTQ citizens are 10 points more liberal than their heterosexual minority counterparts. On religion we find a 25-point gap between white Christians and whites without religion, and a 9-point gap between minority Christians and minority members of the public without religion. These ideological differences are around 10 points smaller among the public than among elite students, but what really stands out in the public sample in Figure 17, compared to students, is how muted the gender gap is.

Could it be that the educated public resembles elite students? Or that things are different with party identification as opposed to ideology? Turning to party identification in Figure 18 and comparing elite students with members of the public that have a degree shows some interesting patterns. First, there is no difference in the partisan race gap between students and the graduate public. In both demographics, whites are around 15 points more Republican and less Democratic than non-whites.

The same is true of religion, the rightmost set of bars, where Christian students and the Christian educated public are around 25 points more Republican and less Democratic than their nonreligious counterparts. LGBTQ citizens differ from straight citizens more among the public (18 points) than among students (13 points), which may reflect the fact that the LGBTQ share is a smaller and perhaps more self-conscious group among the public than among students, where they brook larger. What really stands out, however, is that the 15-point partisan gap between male and female elite students is much larger than the 8-point difference in the educated public. If we compare with the wider public rather than the educated public, the partisan gender gap declines to 4 points. The under-25 public has a gender gap of 6 points. If we compare with the college-educated public under age 25, the partisan gap rises to 10 points.

Figure 19 shows the same pattern with ideology gaps, though this time students show a larger gap around sexuality and race. So college education and youth are both associated with a larger partisan gender gap, but this appears to be supercharged in elite colleges to produce a student gap of 15 points for both partisanship and ideology. Something about young, elite, educated environments accentuates the importance of gender, and to a lesser degree sexuality and race, for politics in a way less true for older generations of college graduates. This echoes what we saw in the preceding two charts, where the gender gap in ideology was about five times higher among students than the general public.

Women and minorities are especially likely to lean left on elite campuses compared to their counterparts in the general public. Figure 20 shows that female students are 33 points more liberal and 27 points less conservative than women in the population. Even minority women under 25 with a degree in the public are almost twice as likely as elite student women to identify as conservative. Powerful selection effects, combined with the possible impact of collegiate socialization in elite networks, seem to result in women and minorities in top universities positioning themselves sharply to the left of the public.

The same holds for LGBTQ students, who are 30 points more liberal than the LGBTQ population and 14 points more liberal than the LGBTQ under-25 college-educated population. Those without religion are 30 points more liberal than the nonreligious public, though they are only 8 points more liberal than the young, college-educated, nonreligious population.

This does not simply represent a blanket effect of elite college status acting uniformly on all groups via selection or socialization effects, for elite university students who are from stereotypically conservative demographics differ less from the public. By way of illustration, Figure 21 shows that the Republican-Democrat partisanship ratio is relatively similar among white and Christian students compared to their counterparts with degrees off-campus. Indeed, White Christian male elite students are slightly more Republican than the college-educated population of white Christian males. Thus, it is only liberal-leaning demographic groups that seem to exhibit an unusually high liberal Democrat skew on campus – especially women and, with respect to liberalism, minorities.

Do Economic Factors Matter?

A notable feature of elite student opinion is the relative dominance of group identity in predicting ideology and party affiliation. By contrast, economic factors play a relatively modest role. Figure 22 shows that those not on financial aid (35% of the sample) are 5 points more likely than the 62% of students on financial aid to be Republican identifiers. Those from a private school background are 2 points more Republican than those from public high schools. Those who describe their family’s financial situation as upper class or upper-middle class are 6 points more Republican than those from a working-class or lower-class background. On the other hand, those attending more expensive private universities are 6 points less likely to identify as Republican than those attending public universities. Regardless of whether there is a consistent effect, these variables explain only about half a percent of the variation in a 7-point scale of party identification. This is about the same-sized effect as family income in predicting party identification among the wider public, suggesting that material factors tell us very little about partisanship and ideology, whether on- or off-campus. Identity, associated more with cultural than economic cleavages, very much dominates sociodemographic models of political belief.

Are Students Influenced by their Peers’ Politics?

Given the substantial left skew among students and professors alike, a substantial literature has emerged seeking to ascertain whether university influences the ideology and politics of students. Some claim a university education makes students more liberal while others argue it has no effect.33 The FIRE data are purely cross-sectional, derived from single snapshots in time. As such, they do not permit us to see whether individual students’ views change over their time at college. We can look at different graduating classes to see if those that have been at university longer are more liberal or Democratic. Here the data show no significant difference in liberalism or Democratic voting between older and younger students, indicating that the university experience is probably not having much effect on students’ political beliefs. It could be the case that students’ politics are liberalized very early in their college experience, or that aging and university socialization counteract each other. Still, these findings seem to reinforce the existing literature that finds little effect of university on students’ political attitudes.34

Even if the average effect appears minimal, it may be that concentrations of liberal or conservative students exert a peer effect on individual student attitudes. The FIRE data show apparent evidence for a small effect in which conservative students shift rightward in more conservative universities whereas black, Asian, and LGBTQ students resist moving right in such environments. Dividing colleges into four equal-sized quartiles based on students’ average ideology gives three quartiles of left-leaning colleges and a centrist quartile where colleges lean slightly left or right of center. Figure 23 shows that in centrist colleges, net of other key identity variables, political moderates who are heterosexual (left chart) or white (right chart) are around 10-12 points more likely to identify as Republican. By contrast, LGBTQ and black moderates on centrist campuses are about as likely to identify as Republican on leftist and centrist campuses. For Hispanic and Asian moderates, centrist campuses are associated with a slightly higher likelihood of identifying as Republican.

We can similarly observe the relationship between campus political climate and ideology. Figure 24 illustrates how white independents are about 6 points more likely to identify as conservative on a centrist campus than on a highly Democratic one while black independents react to a more right-leaning climate by shifting 4 points away from conservatism. In the right-hand chart, Christian students are about 5 points more likely to be conservatives on centrist campuses than if they attend a college in one of the left-leaning three quartiles. The fact the sample is restricted to independents in this chart and moderates in the previous one, as well as the fact that white, heterosexual, and Christian students display different effects from LGBTQ, nonreligious, and minority students in centrist political environments, suggest that these patterns are more likely to be the result of campus climates than self-selection effects.35

Attitudes and Partisanship

In the FIRE data, a 7-point left-right scale accounts for 59% of the variation in a 7-point Strong Democrat to Strong Republican partisanship scale. Using a 5-point ideology scale by collapsing the 7-point scale still predicts 56% of the variation in partisanship. Among the general public, a 5-point liberal to conservative scale predicts 45% of the variation in the same 7-point scale in the CCES overall, but this rises to 55% among the university-educated. Among those who are 25 and under with a degree, ideology explains just 42% of the variation, 14 points less than among elite students in the FIRE survey. The CCES data shows that young people are less likely than their elders to sort ‘correctly’ by pinning the tail of ideology onto the donkey of partisanship. All of which throws the strong association between ideology and partisanship among elite students into relief and indicates a relatively politicized context.

If I examine both group identities and a suite of attitudes toward no-platforming controversial left- and right-wing speakers, as well as attitudes to obstructing speech, it becomes clear that these are powerful predictors of partisanship. Controversial left speakers included those arguing that “Religious liberty is used as an excuse to discriminate against gays,” “White people are collectively responsible for structural racism and use it to protect their privilege,” “The police should be abolished because they are racist,” and “Looting is a justifiable form of protest.” Controversial right-wing speakers were those claiming “Black Lives Matter is a hate group,” “Transgender people have a mental disorder,” and “All abortion should be illegal.” Responses were combined into separate indices through principal components analysis. On speech obstruction, questions on whether it is acceptable to shout down, block or use violence against a speaker were combined into a similar index.

Figure 25 compares the effects from two models. The first contains identity characteristics and the average partisan lean on an individual’s campus. This accounts for 22% of the variation (18% without average partisan lean included). The three indices predict 28% of the variation, rising to 35% when identity characteristics and party lean are added to the model. Support for banning right-wing speakers has by far the strongest predictive power, with a standardized effect size (beta) of over .4 in predicting a Democratic rather than Republican lean. Those who support banning left-wing speakers are significantly more likely to lean toward the GOP, but this parameter is only half as predictive of partisanship as support for banning right-wing speakers. Those who oppose shoutdowns, blocking, and violence lean right, but this is less predictive than the no-platforming variables. This indicates that those on the right have an element of bipartisan free speech consistency while those on the left are somewhat bipartisan in their relatively restrictive stance on speech, even for controversial left-wing speakers.

Comparing the identity and attitudes models shows that adding attitudes to speech restriction to the model greatly erodes the impact of gender. This is largely because women are substantially more restrictive than men on the left no-platforming scales, though much less so on the obstruction of speakers scale. LGBTQ students, by contrast, are much more restrictive than heterosexuals in terms of speech obstruction but less so on no-platforming. The net effect, however, is that the effect size on the sexual minority parameter drops considerably once the speech attitudes are added to the model. White Christians are the most important pro-Republican group, and this is the most important identity variable for predicting partisanship. It retains much of its predictive power even when speech attitudes are included in the groups plus attitudes model.

Viewpoint Diversity

The foregoing discussion of the political composition of colleges showed that three of four quartiles of elite universities lean left and Democratic. Only the fourth quartile of colleges has a relatively even balance of liberals and conservatives, Democrats and Republicans. Figure 26 plots all 159 colleges by average ideology on a 7-point scale, noting their share of liberals in blue and conservatives in red. Each college has two vertically aligned dots. Were it not for variation in the average share of moderates per college, the two lines would be perfect mirror images of each other.

The first point to note is the strong liberal tilt of most colleges. At the extreme, Wesleyan is less than 1% conservative and 86% liberal. A number of liberal arts colleges are in this extreme range of political homogeneity, with nine institutions containing fewer than 5% conservatives. At the other extreme, Hillsdale College is the only clearly conservative university in the data, at 76% conservative and 11% liberal.

Only 14 institutions contain more conservative than liberal students, a similar number to the colleges in the dataset that contain less than 5% conservatives. Moving from left to right, we encounter a number of Ivy League schools, with Brown the most liberal and Princeton the least – though even Princeton is 64% liberal and just 15% conservative. Despite stereotypes, the University of Kentucky has 2.5 liberals for every conservative student while Arizona State leans liberal by almost a 2-to-1 margin. Only when we arrive at a number of flagship universities in the South or inland West do we encounter politically balanced schools like Texas Tech, the University of Arkansas, Baylor, or the University of Mississippi. Finally, at the right of the chart, we see the only two colleges with anything close to a conservative majority, Utah State (49% conservative), Brigham Young University (BYU) (48%), and Hillsdale College (76%). Plots by party identification show a very similar pattern: thus, only two colleges have a Republican majority: Utah State (51%) and Hillsdale (72%).

A list of the ten most right-leaning and left-leaning student bodies appears in figures 27 and 28. The dominance of mainly New England liberal arts colleges among the most liberal institutions at the left of the chart, and of universities in the South and inland West on the right, is evident.

To derive a measure of viewpoint diversity, I draw on two questions in the FIRE survey data. First, the 7-point ‘very liberal’ to ‘very conservative’ ideology question. Second, the 7-point ‘Strong Democrat’ to ‘Strong Republican’ party identification question. The mid-point in both is 4 (moderate, Independent). My Index of Viewpoint Diversity (IVD) is calculated by adding together the two indices, then subtracting the sum from 8, which represents a sum of the mid-points of the two scales. The absolute value of this figure yields a range from 0 to 6, where the most diverse institutions score a 0, the least diverse a 6. The results, plotted in Figure 29, show that most colleges have at least some viewpoint diversity insofar as their students are not 100% clustered at either the ‘very liberal’/Strong Democrat or ‘very conservative’/Strong Republican pole.

However, it is also evident that high viewpoint diversity scores are generally found in institutions in the highest quartile of conservative students, with those at the optimum score tending to have 3 in 10 students conservative, 3 in 10 liberal, and 4 in 10 moderate. An index score above 2 indicates that students are, on average, more doctrinaire than party leaners or the ‘slightly’ ideological. Using this cutoff point, Hillsdale is the only low viewpoint diversity institution that leans right. Among the 14 right-leaning institutions, all but Hillsdale have high viewpoint diversity: an IVD of 1 or below.

At the other end of the index, 63 left-leaning colleges have low viewpoint diversity (IVD above 2), with quite a few liberal arts colleges in the extreme 3 to 4 range. Twenty-nine colleges with more left- than right-leaning students have a low IVD below 1, but fully 116 have an IVD above 1 – compared to just one of the 14 right-leaning colleges. Importantly, one-third of low viewpoint diversity colleges are lower-ranked R2 or R3 schools and none are liberal arts colleges. By contrast, 83% of very low viewpoint diversity schools (IVD above 3) are liberal arts colleges, and none are R2 or R3 schools. In fact, all but one (Trinity College) of the 23 liberal arts colleges in the survey data have an IVD above 2, indicating low viewpoint diversity.

Based on the index, the top 10 institutions for viewpoint diversity are given in table 1, and the bottom 10 in table 2. Flagship state universities in the South dominate the top rankings while the bottom 10 consist entirely of liberal arts colleges. The full set of rankings appears in the appendix.

Table 1. Top 10 Colleges for Highest Viewpoint Diversity

Table 2. Bottom 10 Colleges for Viewpoint Diversity

As of writing, the 2022 data has just been released. The rankings look very similar, even as modest sampling error would be expected to result in some shuffling between institutions. The University of Alabama, Texas Tech, University of Tennessee, University of Arkansas, Texas A&M, Montana State, Bowling Green, Purdue, Penn State and Boise State form the top ten for viewpoint diversity in 2022. Grinnell, Wheaton, Georgia State, Colorado College, Bard College, Hillsdale, NYU, Vassar, Knox and Colorado College score lowest. The data weights make it difficult to achieve pinpoint accuracy in the rankings because the raw partisanship and ideology percentages are adjusted to account for a college’s race, gender, voter registration and financial aid composition.

Regardless, the broad patterns by college type are crystal clear. These rankings suggest that those seeking to experience a wide range of political opinions are best advised to attend a flagship state university, R2 or R3 school, and avoid liberal arts colleges. Ivy League schools lean heavily liberal but, in contrast to liberal arts colleges, contain a 10-15 percent conservative minority – many of whom do not have conservative parents. In a future report, I will examine the important link between viewpoint diversity and expressive freedom on campus.36

Conclusion

This report has examined the demographic and political composition of 159 leading universities based on surveys of 57,000 undergraduate students. It shows that elite students lean liberal and Democratic by a margin of 2.5 to 1. This places them to the left of the median voter, the young and college-educated public, and students attending lower-ranked colleges.

Today’s elite students are substantially more diverse in terms of race, sexuality, and religiosity than elite students of the 1980s and 90s. A quarter identify as LGBTQ, and the share with no religion is nearly as large as the proportion who are Christian. In addition, these identity characteristics matter more for students’ political opinions than they do for the public. Consequently, women, minorities, and LGBTQ students at leading universities lean considerably further left than members of their identity groups in the wider population, even compared to those that are young and college-educated.

Only a quarter of students are white, heterosexual, and identify as Christian. White straight Christian men make up little more than 10% of the total. White Christians, white conservatives, and white males, therefore, appear to be increasingly self-conscious minorities on today’s liberal and secular elite campuses. The level of segregation of white conservatives and Christians within the elite college system is notably higher than the clustering of white liberals, minority conservatives, and minority Christians among elite universities. In contrast, white Christian, white male, and white heterosexual elite students differ much less from members of their group in the wider public than female, minority, and LGBTQ elite students. Young liberals, especially minority liberals, seem to cluster towards elite universities but are not institutionally segregated among them. Conservatives, especially white conservatives, do not cluster toward elite schools but are relatively institutionally segregated among them.

In the liberal-dominated elite college system, viewpoint diversity is strongly associated with colleges that recruit more conservative students. State flagship universities in the South or inland West have the highest levels of viewpoint diversity. Liberal arts colleges, many in New England, have the lowest levels of viewpoint diversity. Whether the former can capitalize on their political diversity, and the latter take steps to improve theirs, turns on whether university leaders consider viewpoint diversity as important as more conventional forms of diversity.

Far from being a monolithic bloc, students in today’s leading universities are diverse and divided. That is, there is more sexual, religious, racial, and gender diversity among elite students than in the past, and these identities are more strongly connected to ideology and partisanship. The gap in party identification between male and female students, for instance, has risen steadily since the early 2000s to the point where female elite students are 18 points more likely to be Democrats and 12 points less likely to be Republicans than their male classmates.

Elite students will supply many of the leaders of tomorrow’s America, so it is vital to understand trends among them. The country has polarized substantially since the 1980s, and the profile of cultural issues around identity – race, gender, and sexuality – has risen in the media and national politics. All this as the decline of religion among young people reconfigures society and elevates religious ‘nones’ to prominence as a political constituency. Elite students furnish leading indicators of what may be emerging and present a more intense, magnified version of national political trends. This portends a politics in which elite women, minorities, gays, and the nonreligious are more left-leaning while elite whites, males, and Christians remain relatively conservative.

Appendix

One can find an interactive version of the table below at http://www.sneps.net/viewpoint-diversity-ranking-of-leading-us-colleges.

References

• The last decade has seen a precipitous rise in the share of Americans identifying as LGBT, particularly among the youngest adults. Today, among those under 30, a wide range of surveys converge on a number of around 20%.

• Government data from Canada and the UK indicate that surveys might be overestimating the extent of the rise in LGBT identity. This caveat must be kept in mind in understanding this report. Nonetheless, these government sources indicate that the trend is real, even if less reliable surveys might exaggerate it. The UK’s Office for National Statistics finds that 7.6% of those 16-24 identify as LGBT, which can be taken as a low-end estimate for that country.

• The most popular LGBT identity is bisexual, which is significantly more common among women than men.

• When we look at homosexual behavior, we find that it has grown much less rapidly than LGBT identification. Men and women under 30 who reported a sexual partner in the last five years dropped from around 96% exclusively heterosexual in the 1990s to 92% exclusively heterosexual in 2021. Whereas in 2008 attitudes and behavior were similar, by 2021 LGBT identification was running at twice the rate of LGBT sexual behavior.

• The author provides a high-point estimate of an 11-point increase in LGBT identity between 2008 and 2021 among Americans under 30. Of that, around 4 points can be explained by an increase in same-sex behavior. The majority of the increase in LGBT identity can be traced to how those who only engage in heterosexual behavior describe themselves.

• Very liberal ideology is associated with identifying as LGBT among those with heterosexual behavior, especially women. It seems that an underlying psychological disposition is inclining people with heterosexual behavior to identify both as LGBT and very liberal. The most liberal respondents have moved from 10-15% non-heterosexual identification in 2016 to 33% in 2021. Other ideological groups are more stable.

• Very liberal ideology and LGBT identification are associated with anxiety and depression in young people. Very liberal young Americans are twice as likely as others to experience these problems. 27% of young Americans with anxiety or depression were LGBT in 2021. This relationship appears to have strengthened since 2010.

• Among young people, mental health problems, liberal ideology, and LGBT identity are strongly correlated. Using factor analysis in two different studies shows that assuming one common variable between all three traits explains 40-50% of the variation.

• Because the rise in LGBT identity is so heavily concentrated on the political left, its influence on the balance of power between the two parties is likely to be limited.

• College students majoring in the social sciences and humanities are about 10 points more LGBT than those in STEM. Meanwhile, 52% of students taking highly political majors such as race or gender studies identify as LGBT, compared to 25% among students overall.

• Various data sources indicate that gender nonconformity – trans and non-binary identity – reached its peak in the last few years and has started to decline.

• What kind of high school or college a young person attends poorly predicts their likelihood of identifying as LGBT. The one exception is Liberal Arts colleges, where 38% of students describe themselves in this way. This indicates that schooling might not have a large effect on changes in LGBT identity.

• Overall, the data suggest that while there has been an increase in same-sex behavior in recent years, sociopolitical factors likely explain most of the rise in LGBT identity.

Click "download" for a PDF

1.31MB ∙ PDF file

Download

Download

Introduction

In April, Ross Douthat observed that the share of young Americans identifying as lesbian, gay, bisexual, or transgender had risen precipitously over the last decade. This, he argued, was the context behind what he termed the new “LGBTQ Culture War” in politics and the media. Douthat set out three possible explanations for the phenomenon:

1. LGBT, like left-handedness, is a true expression of the range of human physiological diversity, which until now had been repressed by conservative social mores.

2. The rise in LGBT represents a form of youthful exploration, and many will revert back to heterosexual and cisgender identification in later life.

3. Alternative sexual identities are a form of social contagion, incubated online and by educational and medical institutions.1

The battle over Florida’s Parental Rights in Education Bill, dubbed the ‘Don’t Say Gay’ bill by progressives, is emblematic of this new culture war.2 Likewise, a claimed 1,000% rise in transgender identity since 2010 has been hotly debated.3 Is this a result of socialization and imitation, or does it spring from the lifting of social barriers? This report finds evidence for both the first and third propositions above. However, it posits a vital distinction between LGBT identity and behavior. The youthful surge is mainly about LGBT identity, with considerably less change in sexual behavior. The rise is greatest for bisexuality, especially among females, with less change for gays and lesbians. The growth in LGBT identification shows no signs of slowing down among the young, but there is compelling evidence that gender nonconformity peaked around 2020 and declined in 2021. It appears less prevalent among teenagers than those in their early twenties.

In addition, there is an important political dimension to the rise. This report finds that much of the LGBT rise has occurred among very liberal or far left-wing young people. Is this because such people are more liberated, or because LGBT is now a kind of political identity? The evidence reviewed here indicates that an explicitly political motivation may be at work among whites and the university educated, but this is less likely to account for the rise in LGBT identity among young minorities and those without a college education.4 The data suggest that while progressives form the core of the movement, there has also been a liberal shift in the broader culture that has affected a wider range of people. This in turn raises the question of the political implications of the LGBT surge. If, as the data show, sexual orientation is highly correlated with political beliefs, then might we expect to see Democrats benefit from this shift in the future? This report suggests perhaps not, as the LGBT rise is taking place largely within the liberal voting bloc, limiting its potential to shift the partisan balance.

Sexual Orientation: Nature or Nurture?

There is an established literature on sexual orientation and the degree to which it is heritable, shaped by early life experience, or by broader cultural trends. Twin studies show that identical twins can have different sexual orientations, suggesting that both genes and environment play a role.5 In terms of psychometrics, an early measure of sexual behavior was the Heterosexual-Homosexual Rating Scale, known as the Kinsey Scale after the pioneering sexuality researcher Alfred Kinsey who first introduced it in Sexual Behavior in the Human Male (1948). The scale ranges from 0, completely heterosexual, to 6, completely homosexual, with 3 representing equal sexual attraction to men and women.

In the 1950s, Kinsey claimed that 37% of men and 11% of women had a same-sex experience, though his sample has been criticized for having a selection bias. Subsequent research based on physiological measures of arousal finds that many more people experience “incidental homosexual feelings and contacts” than “persistent and strong feelings and frequent same-sex experiences.” Research also finds that male arousal is more bimodal and female arousal more spectral. This links to accumulated survey findings in recent decades showing that women (87%) are less likely than men (93%) to identify as ‘completely heterosexual’ on the Kinsey Scale.6

These findings indicate that there is physiological variation in patterns of sexual attraction. Society creates rules that govern when we express emotions, such as when it is appropriate to laugh or cry.7 Likewise, we might expect that culture and incentives can play a role in repressing or encouraging people to identify with their feelings, however fleeting. The evidence here appears to show that people with strong attractions, especially liberals, no longer feel repressed when it comes to identifying as non-heterosexual. However, it also indicates that very liberal young people with incidental homosexual feelings are increasingly identifying as LGBT. I posit that a common psychological orientation underlies both strong political liberalism and non-heterosexual identification. This has interacted with a more modernist, transgressive youth culture. Subgroups that are equipped with the right psychological predispositions are pushed towards identifying as both LGBT and very liberal.

The Rise of the LGBT Population?

There has been a dramatic rise in the LGBT share of the US population since 2012, and especially since 2017. Gallup’s 2021 annual survey of more than 12,000 American adults found a pronounced rise, from 5.6% in 2020 to 7.1% in 2021. In fact, this was twice as high as the share in 2012, driven largely by rapid LGBT growth among Gen Z and Millennials. The General Social Survey (GSS) fields a smaller sample than Gallup of around 2,000 people, but is known as a very high-quality, nationally representative survey. The 2021 GSS, like the Gallup data, saw LGBT share tick up to 7.6% from 6% in 2018 and 4% in 2012.

It is important to recognize that young people who are high in psychological openness or other characteristics associated with non-heterosexual identity may be more likely to complete surveys, and this may be biasing the data. In Britain, YouGov finds that a quarter of the 5,407 18-20-year-olds in its Profiles panel in 2022 identified as LGBTQ, with no clear difference by age.8 However, the 2019 official Office of National Statistics (ONS) figures, using a similar question on a 320,000 national sample, finds that only 7.6 percent of 16-24 year-olds identified as LGBTQ in its most recent release.9 This suggests the share of sexual minorities could be overstated in surveys by a factor of 2 or 3. And if we compare the 0.8 percent transgender and non-binary share of Gen-Z in the 2021 Canadian census with Gallup’s 2.1 percent transgender Gen-Z estimate for the United States in the same year, it seems plausible that the gender nonconforming share is less than half of what surveys suggest. While Canadian census and UK government data suggest that surveys are overestimating the share of LGBT-identifying individuals, they also indicate that the rise in LGBT identity over time is real. Nonetheless, in reviewing the data in this paper, it is important to keep in mind that the top-line results on LGBT identity and behavior might be exaggerated.

As Figure 1 reveals, the sharpest increase in LGBT share according to Gallup was among Gen Z, born between 1997 and 2012 (i.e., now under 25), with the LGBT proportion doubling from 10.5% in 2017 to 20.8% in 2021. The share also rose impressively among Millennials (born 1981-1996, now aged 26-41), from 7.8 to 10.5% of the total. The trends in other generations, by contrast, were relatively stable, with a mere 4.2% of Gen X (now 42-57) and 2.6% of Boomers (now 58-76) identifying as anything other than heterosexual.

The GSS tells a similar generational story. Gen Z’s LGBT share rose from 9.4% in 2016 to 12.1% in 2018 to 19.8% in 2021. Millennials increased less, from 6.9% in 2012 to 10.8% in 2018 to 13% in 2021. Among Gen X the numbers were more modest: 5.1% in 2012, 4.4% in 2018, and 8.8% in 2021. For Boomers the figures showed the lowest rise, from 2.1% in 2012 to 2.7% in 2018 and 4% in 2021. It is important to note that GSS used an online rather than face-to-face survey in 2021, which may have boosted the LGBT share if a stigma is still attached to identifying as LGBT. Even so, we see a pattern similar to Gallup: a substantial rise between 2017-18 and 2021. In absolute terms, this is most pronounced among Gen Z, but in the GSS it encompasses virtually all generations.

Since generational analysis is confounded somewhat by age, a clearer picture appears when examining trends over time by discrete age groups. Figure 2 shows the results by fixed age group for each year of the GSS. The age groups contain a similar number of people each year, and this permits us to notice a marked change within each age band over time. The share of LGBT individuals among Americans under 30 jumped from 4.8% in 2010 to 16.3% in 2021 but also rose substantially among those aged 30 to 44 and, with some noise, within the 45 to 64 group. While the raw point increase is greatest for those under 30, almost all age groups experienced a similar percentage increase, between 100 and 200%.

Youth LGBT Identification in Other Large Datasets

For further comparison, the Foundation for Individual Rights in Education (FIRE)’s 2020 and 2021 surveys of a combined sample of 57,000 undergraduate students at top US universities showed a statistically significant increase in non-heterosexual identification from 18.6% in 2020 to 25.7% in 2021. Age made no statistically significant difference, but students in 2021 were considerably more likely to identify as LGBT than those in 2020. Though FIRE expanded its coverage from the leading 50 colleges to 150 (mainly elite) schools between the two years, this does not account for the difference. When I examine the change within the original 50 colleges surveyed in 2020, there is a rise in LGBT share from 18.6% in 2020 to 24.5% in 2021.

The Cooperative Congressional Election Study (CCES), which polls over 60,000 individuals each year, finds that the LGBT share in the (weighted) data among those aged 18 to 25 rose from 21.3% in 2016 to 27.9% in 2020. While the figures in the FIRE and CCES are somewhat higher than in the Gallup or GSS, we see the same broad trend: a substantial increase in non-heterosexual identity from the mid-2010s to the present, with over 20% of young people identifying as LGBT.

The Distinctive, but Not Exclusive, Rise of Bisexuality

Figure 3 breaks out the main components of LGBT identification across four surveys.10 The numbers broadly show that bisexuality is between 50 and 300% more prevalent among women than men. By contrast, homosexuality is 50 to 200% more common among men than women. The ‘other’ category varies a great deal and is affected by question wording, as FIRE has the richest set of options among the survey institutions, including pansexual, queer, questioning, and other. Whether students are more likely to select one of these, or question wording plays a key role, is difficult to ascertain.

Finally, the gender nonconforming figures show a great deal of variability, with just a lone individual under 30 identifying as such in the entire GSS in 2021 to nearly 800 individuals (2.4%) among young people surveyed by the CCES in 2020. The CCES is the least representative in terms of sampling frame, even as data are weighted on major demographic variables. Thus we should take the 6% figure for trans identification with caution. The overall figure for non-heterosexuals among those under 30 is 16% in the GSS, 21% in Gallup, 23% in FIRE, and 25% in the CCES. Factoring in government and census data indicating that surveys may exaggerate the relevant numbers, I am confident that somewhere between 10-20% of young Americans today identify as LGBT.

Which identities have grown the most? Decomposing the change by subcategory in the Gallup data in Figure 4 shows that the most popular LGBT category is bisexuality, which stands at 15% among Gen Z, 6% for Millennials, and just 1.7% for the middle-aged Gen X. By contrast, gay and lesbian numbers are much lower. Among Gen Z, the gay share is only 2.5% and the lesbian share 2%, less even than transgender at 2.1%.

The share of bisexuals is 13 points higher among Gen Z compared to Gen X, but the combined share of gays and lesbians among Zoomers is only 2.6 points higher. In proportional terms, there is a generational difference across all LGBT categories, but the absolute point difference between newer and older generations is substantially higher for bisexuality than others. Once again, Gen Z stands out, with Millennials also appearing to embrace new sexualities, albeit to a lesser degree. Gen X and older generations continue to manifest low LGBT shares.

Breaking down the over-time trends among people under 30 by gender in the GSS in Figure 5, we see that the largest category in 2021 is bisexual female at 12.1%, followed by bisexual male at 10.3%, gay males at 7.6%, and lesbians at 5%. Trend lines have risen for all four categories since 2008, with bisexuality showing the highest raw point growth. The GSS, however, suffers from low sample sizes for under-30 sexual minorities when they are subdivided by year and age. Thus it is vital to smooth the volatility in the lines over time. Of the trends, the rise in female bisexuality is the most statistically significant, peaking in 2018.

The CCES is less nationally representative but has a much higher sample of young people than the GSS. It shows a bigger gap between male and female bisexuality among the under-30s: 15.5% of women under 30, but just 7.2% of men in 2020. This represents an increase over 2016 for women, who were then 10.8% bisexual. Men are unchanged between the two periods. Gays and lesbians combined form approximately 5% of the under-30 sample in both 2016 and 2020. Thus CCES seems to indicate growth taking place mainly in the female bisexual group within the young population. This comports with statistical analysis of the GSS, where female bisexual growth also stands out. In the FIRE student samples, focusing only on the schools sampled in both years, 15.5% of women in 2021 are bisexual, up from 12.6% in 2020. The lesbian share rose to 2.9% from 2.1%, and the gay share to 6.7% from 5.8%. The male bisexual share increased as well, to 6% from 4% in 2020.

What of trans or non-binary identification? In the GSS, such individuals make up only a tiny fraction of the sample, at just 0.25% compared to 1.1% in the 2021 FIRE student data and 2.1% in the 2021 Gallup data among Gen Z. In the FIRE data, the non-binary share of students declined in these years, from 1.5% in 2020 to 0.9% in 2021. This is a statistically significant drop that will be discussed in a later section on trends in gender nonconformity, where I examine evidence that the phenomenon has peaked and is in decline. The takeaway from the FIRE data is that bisexuals account for slightly under half the LGBT share and have registered increases among both women and men. The gay and lesbian component also increased during 2020-2021 but appears to have risen more slowly than bisexuality in percentage point terms. Overall, the various surveys show that for young people there has been significant growth across all LGBT categories since 2018. That said, the growth since 2010 is greatest for female bisexuality and slower – in percentage point terms – for gays, lesbians, and nonbinary individuals. Yet the rate of growth over 2012 is probably highest for the non-binary group, despite their small share, even as this rise appears to have peaked around 2020.

Sexual Behavior

As noted, prior studies show a degree of flexibility around sexual attraction, especially among women, who have a less sexually bimodal arousal pattern than men. There has been a large rise in non-heterosexual identification since 2017, especially among younger Americans. The question this raises is whether this dramatic change is primarily about psychological affiliation or whether it involves a concrete change in sexual behavior. Fortunately, the GSS contains a number of longstanding questions about sexual behavior that have run since the late 1980s. These can be compared to the LGBT identity questions in the GSS from 2008 onward to examine whether the change is primarily psychological or behavioral and how the two trends have interacted.

The GSS asks respondents, “Have your sex partners in the last 12 months been: a) exclusively male, b) both male and female, or c) exclusively female.” The question was also asked for partners in the previous five years, with the answers very highly correlated with the 12-month responses. In all analyses in this paper, I exclude those who report not having any partner in the relevant time frame, which does not change the results much.11 Due to the small number of same-sex attracted people, there is considerable noise in the sample of young people over time, so I also compare with the total sample, which includes all ages.

In 2021, 12.7% of men under 30 reported same-sex (11.4%) or mixed-sex (1.3%) partners, while in 2018, just 4% of men under 30 reported same-sex or mixed-sex partners. This appears to be a large increase, but the sample size of young men reporting same-sex partnering is just 99 in 2018 and 79 in 2021, so the findings are based on a small count. For women, the proportion reporting same-sex or mixed-sex partners dropped in the same period from 10% in 2018 to 7.1% in 2021. However, this is again based on a small sample. To screen out statistical noise, I present data for the entire range of age groups to minimize error, followed by the smaller sample of under-30s only, across four time periods.

The trend, summarized in Figure 6, shows a significant decline in heterosexual behavior between 1972-92 and 1993-2003 among both men and women. The probability of a woman being heterosexual in behavior dropped from 99% to 97%, while men fell from 97.5 to 96%. After a period of stability, there is a further shift between the 2004-16 and the 2017-21 periods of around 1 point, with women and men converging at around 95% heterosexual behavior.

Moving to look at the under-30s, and bearing in mind the noisier data due to small samples for young people, Figure 7 reproduces Figure 6 for under-30s only. It shows a similar pattern of change between 1972-92 and 1993-2003. In this period, female heterosexual behavior fell from 98.5 to 96% while male sexual activity with women only fell from 97.5 to about 95.5% of total sexual behavior. Between 2004-16 and 2017-21 women dropped from 94.5 to 91.5% opposite-sex -only partnering while men declined from 96 to 92%. Overall, among young people, female heterosexual-only partnering declined 7 points between 1972-92 and 2017-21 while male heterosexual-only behavior declined 5.5 points in the same period. The change in sexual behavior is real, though I cannot discount the possibility that it either became more socially acceptable to report such behavior in a survey over this period or, because the 2021 GSS was conducted online rather than in person due to covid, increased anonymity between 2018 and 2021 may have increased the incidence of reporting same-sex encounters.12

Across all ages, when it comes to sexual behavior women shift 4 points away from heterosexuality and men 2.5 points during 2008-21. These figures show that there has been an increase in homosexuality, but that the biggest change took place at the turn of the century. Even so, there was a genuine increase in reported same-sex behavior between the mid-2010s and early 2020s, albeit of smaller magnitude. Among those under 30 there is more change than among older people, but the share of young people reporting that all their partners were of the opposite sex remains around 92% in the 2017-21 period. For men in 2017-21, only 1 point of the 8-point homosexual share consists of men with unconventional sexual behavior – that is, those not exclusively heterosexual – having partners of both sexes. For women with unconventional sexual behavior, those reporting sexual activity with both sexes is 2.7% compared to 5.8% with women only. The female ratio thus contains more evidence of bisexual behavior than is true for non-heterosexual men, who are mainly same-sex attracted.

LGBT identification is higher among women than men, regardless of sexual behavior. 98% of men under 30 who reported having only female sex partners in the past year identified as heterosexual. Among women who only had male partners, just 94% did so. Meanwhile, 85% of men and 80% of women reporting same-sex or mixed-sex partners identified as non-heterosexual.

The upshot of all this, as Figure 8 shows, is a growing divergence between sexual behavior and identity among Americans under 30. Whereas in 2008 attitudes and behavior were similar, by 2021 LGBT identification was running at twice the rate of LGBT sexual behavior.

These trends are especially noticeable among women. In 2008, 3.1% of women under 30 identified as lesbian and a further 4.2% as bisexual. This total of 7.3% roughly matched the 7.4% (4.6% same-sex, 2.8% both-sex) who reported having a female sexual partner in the past year. By 2021, a very different picture had emerged. Twelve percent of young women identified as bisexual and 5% as lesbian, but in terms of behavior, just 7% had a same-sex experience. Nearly 6 in 10 women who identified as lesbian or bisexual reported having only male sex partners in the past year.

Among men identifying as gay or bisexual, the share reporting only opposite-sex partners was considerably lower than for LGBT women, at 38%. Even among men, however, there appears to be a divergence between identification and behavior, with the share of young gay- or bisexual-identifying men reporting female-only sex partners rising from 20% in the 2008-14 period to 32% in 2016-21. Though the sample only consists of 7-17 individuals per year, and thus error bars are wide, Figure 9 illustrates the statistically significant finding that the share of bisexual women who report having only had male sexual partners over the past five years has risen since 2012. A rising share of women with heterosexual behavior is choosing to identify as bisexual. The data for men are inconclusive.

My analysis of GSS data tells us that the trend toward LGBT identity, as distinct from behavior, is stronger in women even as it encompasses both sexes. While the big change has been in LGBT identification, it is important to note that there has also been a modest rise in same-sex behavior. While I cannot rule out the possibility that the rise in same-sex behavior is due to less stigma in reporting, at least some of it seems to represent a genuine change in behavior. In both cases, the trend appears more pronounced among women than men.

The Role of Ideology and Partisanship

Left-wing Americans are significantly more likely to be LGBT than those on the right. Among Americans under 30 in the GSS, 25% of people who identify as liberal are LGBT compared to 9% of conservatives. In the CCES, the corresponding figures are 36% LGBT among liberals and 21% for conservatives. Among undergraduate students at top universities in the FIRE data, 33% of liberals identify as LGBT compared to 6% of conservatives.

Yet what is apparent is that very liberal respondents stand out from moderate liberals in having high LGBT identification. Figure 10 illustrates how the sexual orientation of respondents in three surveys varies along a 5-point self-identified ideology scale. In the FIRE data, 49% of ‘very liberal’ students on a 5-point scale call themselves LGBT compared to 5% of ‘very conservative’ students. The corresponding LGBT numbers for ideological extremes among under-30s in the CCES 2020 are 43% and 13%, and in the 2021 GSS they stand at 34% and 14%. This indicates that young people with conservative views are only around a third as likely to be LGBT as strong liberals, with the gap especially pronounced among elite college students.

Another important pattern emerges from the data. Generally speaking, moving left to right, there is a sharp difference in LGBT identification between strong liberals and weak liberals. In the GSS and FIRE data, the gap between the far left and center left is several times larger than the difference between the center left and moderates. Moderates under age 30 are more like conservatives than liberals, while there is no consistent difference between slight and strong conservatives.

Strong liberals are the most distinctive ideological group. Looking at those under 30, in the CCES, 42% of strong liberals identify as LGBT compared to 19% of moderates and 13% of conservatives. In the FIRE data, 18% of moderate students are LGBT compared to the numbers mentioned above of 5% for the most conservative and 49% of very liberal students. In the 2021 GSS, 34% of strong liberals are LGBT compared to 10% of moderates and 9% of conservatives. Far-left views correspond strongly to higher LGBT identification while far-right views differ much less from the center.

The pattern over time among the under-30s shows that the largest change in sexual identification has taken place among those identifying as very liberal. While other ideological segments have also experienced some increase in LGBT identification, Figure 11 shows that the most liberal respondents have moved from 85-90% heterosexual identification in 2016 to just 66% heterosexual identity in 2021. Very liberal people experience a much larger ‘multiplier effect’ from the new transgressive modernist culture than other segments of the population.

What explains the pattern? It could be that LGBT identifiers gravitate to strong liberal self-identification. Or perhaps those with strong liberal views or self-conceptions are more inclined to identify as non-heterosexual. While it is not possible to definitively discern from the data sources whether sexual identity shapes political beliefs or vice-versa – or both are related to some other variable – there is evidence that something beyond political issues or partisanship, such as psychological orientations that underlie ideology, is shaping sexual identity.13

First, it is not readily apparent why being non-heterosexual would have such a powerful impact in moving someone from the center left to the far left, but so little effect in shifting them from the right to the center or center to the moderate left. Comparing answers to the sexual identity question with views on homosexuality and gay marriage shows an important difference. For LGBT issues like attitudes towards gay marriage, strong liberals and conservatives are equally distant from the center. However, when it comes to predicting sexual identity, strong liberals stand out from the rest. When controlling for views on gay marriage and the acceptability of homosexuality, the same pattern – of strong liberals but not conservatives standing apart from the center – obtains. Tolerance of homosexuality is thus unlikely to explain the ideology-LGBT connection. Controlling for the basic liberal-conservative divide, a very liberal self-description is significantly associated with LGBT identity but not with views on LGBT identity.

Second, controlling for LGBT identification, LGBT sexual behavior over the past five years is not significantly associated with a person’s ideology. If same-sex sexual activity was the principal driver of LGBT identification, which in turn causes an increase in liberalism, we would expect LGBT identifiers engaging in same-sex behavior to be more liberal than LGBT identifiers who have only had heterosexual partners. Instead, we find no significant differences in ideology between LGBT identifiers engaging in LGBT sexual behavior and those who have only had conventional sex partners. When compared to all heterosexuals, those engaging in same-sex behavior, especially men, are of course more likely to identify as LGBT. Yet same-sex behavior itself has only an indirect link to being more liberal, via its effect on increasing the likelihood of identifying as LGBT. Something in the psychology of identifying with, rather than behaving as, LGBT is linked to the psychology of identifying as very liberal.

This said, the precise mechanism whereby ideology and sexual identity influence each other needs to be specified. Figure 12 is based on a statistical model that tries to predict whether someone identifies as heterosexual or not. Here liberal respondents under age 30 who have had a same-sex experience in the past 12 months almost all identify as LGBT. Moderates and conservatives who report same-sex experiences are significantly less likely to do so, with moderates who had a same-sex experience having a 30% chance of identifying as heterosexual and conservatives a 40 to 50% chance of doing so. For men and women across all age groups we find a similar pattern, with moderates and conservatives differing by just 10 percentage points. Either conservatism tends to lead many same-sex attracted people to identify as heterosexual, or an underlying psychological orientation among some same-sex attracted people inclines them to identify as both conservative and heterosexual. This suggests that the rise in LGBT identification is in part because of a lower reticence among same-sex attracted people to identify as LGBT. This is the first of Douthat’s hypotheses, endorsed by progressives.

But a decline in reticence among those with same-sex experience does not account for the majority of the LGBT shift. The leftmost part of the top line in Figure 12 shows that among the most liberal respondents who say they have had no same-sex experiences in the past five years, there is a significantly lower level of heterosexual identification than among heterosexual-behaving slight liberals and moderates. Thus in 2021, 13% of very liberal (on a 5-point scale) young women reporting only heterosexual behavior in the past five years identified as LGBT. This drops to 5% among moderate and conservative women under 30 with the same sexual history. Among men, the equivalent figures for those who only had conventional sex partners are 9% for strong liberals and 5% for those in the middle or on the right. What this suggests is that very liberal ideology is associated with identifying as LGBT among individuals engaging in exclusively heterosexual behavior – especially women. It seems that an underlying psychological disposition is inclining people with heterosexual behavior to identify both as LGBT and as very liberal. This likely reflects those that Bailey and Moore termed incidentally-attracted rather than people with strong and sustained same-sex attraction.14

A third reason pointing to psychological aspects of ideology affecting sexual identity is that issue positions themselves are not clearly connected to LGBT identification. For instance, blacks have a different interpretation of ideology from whites.15 Figure 13, based on a statistical model which tries to predict whether someone identifies as LGBT, shows that views on whether the police make a person feel safe are only correlated with sexual identity among whites, but not among blacks. A similar, if less dramatic, racial difference in correlation holds for abortion and a series of other issues. Hispanics and Asians are somewhat intermediate in that while their issue positions are correlated with the chance of being LGBT, the relationship is weaker than for whites. That is, political issues appear more correlated with sexual identity among whites than blacks, suggesting that LGBT identity is more political among very liberal whites than among very liberal minorities, especially blacks.

Given the less constrained pattern of issue attitudes underlying their ideology, it is perhaps surprising that among blacks under 30, those identifying as ‘very liberal’ are substantially more likely to call themselves LGBT than black weak liberals or centrists. Something underlying the choice to identify as ‘very liberal,’ beyond issue positions, seems to be associated with a higher likelihood of identifying as LGBT. This speaks to the possibility that the relationship between liberalism and LGBT identity can be explained by psychological openness to experience, conscientiousness, or another underlying disposition like proneness towards conforming to media messaging, rather than issue positions and interests.16

A fourth piece of evidence is that party identity, which is arguably more visible and closer to current events than ideology, is more weakly correlated with sexual orientation than the more abstract notion of ideology. Though ideology and party identification are correlated, Figure 14 shows a different sexuality profile for party identification than for ideology. Focusing on the larger-sample CCES and FIRE surveys, we find little substantial difference in the frequency of LGBT identification between strong Democrats, weak Democrats, and moderates. Many LGBT people are Independents rather than Strong Democrats. Unlike with ideology, the far left does not stand out.

These factors all suggest that the association between very liberal ideology and being LGBT is not simply about homosexual people gravitating to ideologies and parties that will defend their interests. Rather, the psychological factors behind very liberal ideology likely predispose an individual with some same-sex feelings to identify as LGBT.

Left-Wing Politics Matters for Educated Whites

While there is a broader modernist cultural atmosphere favoring the expression of difference, there is also evidence that egalitarian politics matters – at least for white college students. To wit, left-wing attitudes toward the protection of minorities predict higher LGBT identification. Consider Figure 15. White female students in leading US universities who identify as very liberal and support shouting down speakers to prevent them from uttering harmful speech have a nearly 7 in 10 chance of identifying as LGBT. For very liberal minority female students with the same speech attitudes, the probability of identifying as LGBT drops to around .55. At the most pro-free speech end of the scale, very liberal white female students who say shoutdowns are never acceptable are only around half as likely to identify as LGBT as their counterparts who strongly oppose free speech.

It is worth noting that a person’s views on shoutdowns are nearly as important as their ideology in predicting LGBT identification. While it is likely that non-heterosexual identifiers are more sensitive to speech they consider harmful to their own interests, it is noteworthy that minority and female students, who might be similarly sensitive to discrimination, are much less supportive of shoutdowns than LGBT students. The relatively large and invisible bisexual group opposes shoutdowns as much as gay men, who would be expected to be more likely to experience discrimination. In addition, adding a control for speech attitudes reduces the size of the ideological effect on LGBT identification by a modest amount, suggesting that to some extent, speech attitudes inform ideology or spring from the same underlying inclination.

Two other indicators of the impact of left-wing politics concern the subject matter in which students are enrolled. Students in social sciences and humanities (SSH) fields are about 10 points more LGBT than those in the less politicized science, technology, engineering, and math (STEM) fields. This holds for males and females, whites, and minorities. Meanwhile, 52% of students taking highly political majors such as race or gender studies identify as LGBT, including 64% of the 109 students in the FIRE survey majoring in gender studies and 47% of 285 students majoring in ethnic and racial studies. This compares to 25% LGBT identification among students overall. Choice of major significantly predicts LGBT identification even when controlling for ideology, gender, race, and religion.

More generally, Figure 16 illustrates how LGBT identification rises as I adjust the profile of students based on key predictors of LGBT identity in the 57,000-strong FIRE dataset of undergraduates at 150 leading US universities.

Sexual Liberalism and the Wider Culture

Ideology is clearly important in LGBT identification, and very liberal respondents have increasingly identified as LGBT. But the share of very liberal respondents in the GSS has not significantly risen since 2008 among Americans under 30. The fact that age and year matter for LGBT identification while race and university education do not suggests that ideas once considered the preserve of the cultural left have influenced the wider culture within which less politicized individuals live. Acceptance of homosexuality and gay marriage has risen especially quickly since 2004, when similar numbers of people thought homosexuality should be accepted and discouraged. By 2017, the numbers stood at 70% for acceptance and 24% for discouragement.17 This, I would argue, has permitted those in same-sex partnerships to identify as LGBT more openly. Yet those engaging in LGBT sexual behavior are a small group.

Arguably more important is the way the new liberal culture has encouraged individuals with conventional heterosexual behavior to identify as LGBT. This has made the biggest impact on young, female, and very liberal people. This could be because their sexual feelings are more fluid and open than those of others due to underlying psychological factors, or because they are more influenced by the education system and media. Nonetheless, the rise of LGBT is not a phenomenon limited by race or class. It encompasses both politicized groups like college-educated whites and less politicized groups such as black males and Hispanic immigrants without a degree. Post-Trump polarization and the Great Awokening among educated white liberals seem to have had an impact on sexual identity among elite whites, but likely cannot explain why minorities and those without degrees have experienced a similar trajectory.18

Polarization along issue and party lines tends to be more intense among white and university-educated Americans, who are more likely to sort issues into (often inconsistent) ideological packages and then into party platforms. Minorities vote at lower rates (as with Hispanics or Asians) or tend not to attach ideology to party as much (as with blacks).19 Yet the LGBT component, while more polarized by ideology, is not higher among white or college-educated Americans than among minorities or those with low education. LGBT identity is relatively high among all young people, whether white or nonwhite, college-educated or non-college educated.

In the CCES, there is no significant difference in LGBT share between white and minority Americans under 30, with around 25% of both groups identifying this way. In FIRE, the difference between whites and minorities is just 1 percentage point, and in the GSS only 1.5 points. Minority faiths in the CCES and FIRE – especially Jews, Hindus, and Buddhists – register higher LGBT shares than Christians. Moreover, in the CCES, just 81% of black men and 67% of black women under 30 without a college degree identify as heterosexual.20 The gender difference in LGBT between black men and women under 30 is noteworthy: young black women are similar to young women of other races, but young black men have a lower LGBT share than other men.

For university education, which is commonly associated with aligning issues of ideology to partisanship, the CCES shows that 19% of college graduates under 30 are LGBT compared to 26% of those without a college degree.21 In the GSS there is no significant difference between the two groups. The more politicized campus experience and exposure to political ideas at university does not lead to a wider difference in LGBT identification between very liberal and very conservative individuals. Young people with and without college degrees are identical in this respect. The LGBT rise may be influenced by identity politics, but is much broader than a woke phenomenon.

What Led to Change?

As noted, we should not overstate the impact of increased toleration of homosexual behavior in accounting for the substantial rise in LGBT identification among the young in recent years. Even if everyone under 30 who had a same-sex experience identified as LGBT in 2021, this would account for no more than 4 points of the 11-point rise in LGBT identity in the GSS among under-30s between 2008 and 2021. Recall that in 2021, 6 in 10 young female LGBT identifiers and nearly 4 in 10 male LGBT young people only had partners of the opposite sex in the previous 12 months. With over 92% of survey respondents reporting only heterosexual behavior, such individuals are driving the bulk of the trend.

Overall, three phenomena have influenced the rising LGBT trend:

1. Rising same-sex behavior.

2. An erosion of the inhibition to identify as LGBT among same-sex attracted people.

3. The rise of ‘ideological’ LGBT identification among those with conventional sexual behavior.

In terms of a), the share reporting same-sex behavior increased 4 points among young people over 2008-21. For b), the proportion of young people engaging in same-sex behavior who self-identify as LGBT has approximately doubled, translating into a 2-point increase in LGBT identification. Given that the same-sex partnered share has increased by 4 points, we can assume that, at most, half of this effect on LGBT identification might have been suppressed in 2008. This would render a) and b) similarly important.

Regardless of the mix of a) and b), the two forces together only account for 4 points of the 11-point rise in LGBT identity between 2008 and 2021. These 4 points should be viewed as a maximum, because it excludes young people who reported no sexual partners in the past year and assumes that greater tolerance of homosexuality between 2008 and 2021, and the switch to an online survey in 2021, played no part in people reporting more same-sex behavior. Thus c), higher LGBT identification among those exhibiting heterosexual behavior impacts on a much larger group of people (the 92% engaging in heterosexual partnering) and therefore accounts for no less than two-thirds – 7 of the 11 points – of the post-2008 rise.

What might have caused the rise in the share of heterosexual-behaving people opting for LGBT identification? The decline of religion among Millennials and Zoomers is a well-documented trend. Close to 40% in these generations are religious ‘nones.’22 In the GSS, the share of under-30s reporting no religion rose from 11% in 1990 to 28% in 2008 to 46% in 2021. Religiosity is connected to sexual identity, as Figure 17 shows. In statistical models, individuals with no religion or who never attend church are over twice as likely to identify as LGBT as those who are Christian or attend regularly.

This relationship appears to hold over time as well, as in Figure 18. In years with a higher proportion of young non-religious respondents (typically more recent years), there is a higher share of young LGBT people. However, this is based on just seven aggregated annual data points. While the fit is better than with other variables such as proportion very liberal, I can’t be sure that religious decline is the key factor. It is also the case that religiosity has been declining among young people since the 1980s and fairly steadily since 2004. LGBT identification has also increased in longstanding low-religion societies such as Britain.23 While there is a spurt of non-religiosity in the 2021 data that corresponds to the surge in LGBT identification, there is also a rise in the proportion of very liberal and anxious/depressed people between the 2018 and 2021 surveys and a change in survey methodology. It is therefore unclear how much of the time trend of LGBT growth is associated with religious decline as compared to, for example, a more general increase of transgressive modernism in the culture.

The Mental Health-Sexuality-Liberalism Nexus

In a recent essay in The Atlantic, Derek Thompson remarked on the recent surge in mental health problems among American teens.24 The centerpiece of the article was Figure 19 below, which uses Center for Disease Control (CDC) data to show a marked rise in emotional problems among youth since 2015. The highest levels are among LGBT respondents, with 76% of them reporting feeling persistently sad or hopeless in 2021. Women also experienced a larger jump in emotional distress than men during the pandemic.

The staggering rate of mental health problems among LGBT identifiers in the chart was not addressed by Thompson, who focused on trends affecting all groups over time. But there is evidence that non-heterosexuals have been hit hardest by the post-2010 crisis of youth mental health. While the CDC only breaks out the figures from 2015, a major study of nearly 40,000 teens in Wisconsin shows that LGBT mental health deteriorated significantly faster than heterosexual mental health between 2012 and 2018. Between these years, the share of heterosexual young people reporting anxiety the previous month rose from 32% in 2012 to 35% in 2015 to 41% in 2018. However, among young gay and bisexual teens it soared from about 55% in 2012 to 65% in 2015 to 72% in 2018.25 All this during a time of rising toleration of LGBT lifestyles.

The GSS has asked whether people in its survey are happy since 2010. Respondents can answer ‘very happy,’ ‘pretty happy,’ or ‘not too happy.’ Figure 20 plots the trends in the GSS since 2010. These broadly confirm those from Thompson for the post-2015 period. Though the sample size is small, there is some indication that the sexuality gap in youth happiness widened from 2010 to 2018, even while the pandemic hit both groups hard in 2021. This is also visible in data aggregating across all age groups. LGBT identity seems to be acting as a multiplier for the forces boosting mental illness.

Very liberal and LGBT people are more anxious and depressed than others across multiple surveys. The GSS introduced a question in 2018 asking how often a person had felt anxious, depressed, or irritable in the previous seven days. I have also previously conducted two Qualtrics surveys during 2020, evenly balanced between white and black for a combined sample of 1,770 individuals. Here I asked, “How often would you say you are sad or anxious?” Those who answered ‘Most of the time’ or ‘Always’ were classified as anxious or depressed. The Qualtrics data only contains data for those aged 25-29, with no figures for the under 25s.

Figure 21 compares depression and anxiety levels across the two different datasets, with their alternative question wording and differing samples. Despite the differences, the broad pattern is similar. Very liberal and non-heterosexual respondents are considerably more likely to be anxious or depressed. In both the GSS and Qualtrics, very liberal young people are 12 points more likely than somewhat liberal young people to have mental health problems. LGBT respondents are 32 points more likely than heterosexuals to have emotional problems in the GSS and 20 points more likely in Qualtrics.

Is there a connection between ideology and sexuality here, in line with other findings in this report? CSPI Fellow Zach Goldberg, using 2020 Pew data, showed that very liberal whites were far more likely than others to say they had been diagnosed with a mental health condition. Very liberal whites had a 46% chance of answering that question in the affirmative compared to 33% of somewhat liberal whites, 29% of moderates, and around 20% of conservative whites.26 We also know that very liberal Americans are substantially more likely to identify as LGBT than slight liberals, moderates, or conservatives. And in the datasets here, ideology predicts mental health problems even when controlling for sexual orientation, race, gender, and other demographic characteristics.

Examining relationships between questions on sexual orientation, ideology, and anxiety/depression shows a powerful set of correlations. In the GSS, factor analysis reveals that one underlying factor accounts for almost half (49%) the variation in answers to these three questions among young people. In Qualtrics, this common factor explains 43% of the variation in the three variables. The ‘very liberal’ portion of the ideological spectrum stands out as being both considerably more LGBT and unusually prone to mental health problems. However, just as liberalism has not risen among youth at anywhere near the rate of LGBT increase, so too I find that the aggregate level of anxiety and depression rose more slowly than LGBT share prior to the pandemic. Between 2010 and 2014, 11-13% of all young people had mental health problems. In 2016 and 2018 this stood at 16%. Only during the pandemic in 2021 was there a surge to 30% of all youth.

It therefore appears that the rise in LGBT young people did not cause a rise in mental health issues but, as with liberalism, the LGBT share disproportionately expanded within a subset of the young population – in this case those with anxiety and depression. Here we see another example of the multiplier effect mentioned earlier, in which wider forces disproportionately affect subgroups with particular psychological predispositions. Thus in 2010 and 2012, only 7% of unhappy young people were LGBT while 4% of very happy people were LGBT. However, by 2018-21, 21% of all unhappy young people and 27% of those with anxiety or depression identified as LGBT. By contrast, just 9% of very happy young people and 7% of those who rarely feel anxious or depressed were LGBT during 2018-21. In the context of limited increases in total youth unhappiness prior to the pandemic, this indicates that the LGBT rise has been disproportionately concentrated within the overlapping ‘very liberal’ and anxious/depressed part of the young population.

It is possible that the groups which report worse mental health – such as young LGBT and very liberal people – have disproportionately shaped, and been shaped by, a new left-modernist culture. This zeitgeist values transgressing social boundaries while valorizing vulnerability and victimhood. Figure 22 shows that in the years 2010 and 2012 (combined for reasons of sample size), when it comes to the number of days in the past month a person said their “mental health was not good…which includes stress, depression, and problems with emotions,” there were few differences between very liberal young people and those who were slightly liberal, moderate, or conservative. LGBT and heterosexuals also reported similar levels of poor mental health. For the pre-pandemic years 2014, 2016, and 2018 (combined), the picture changes dramatically, with very liberal and LGBT young people now reporting worse mental health while others reported similar levels as in 2010-12. Given that prejudice against LGBT individuals was decreasing during that time period, something else was going on through the 2010s to increase the association between despair and identifying as something other than heterosexual.   

The data for a somewhat different question on whether someone is ‘not too happy,’ as opposed to ‘pretty’ or ‘very’ happy, show a somewhat similar pattern in which very liberal and LGBT young people have experienced a faster decline in happiness than others since 2014. This is a period in which a social media-fueled culture of fragility arguably expanded, a phenomenon which parallels the Great Awokening among liberal white Americans – evident in media and campus activism – yet extends beyond it.

In the GSS data, social media use significantly predicts strong liberalism and LGBT identification, as well as unhappiness, anxiety, and depression. Yet social media consumption only explains a modest portion of the ideology-sexuality-mental health nexus.27 The underlying common driver likely involves psychological traits such as high openness to experience, which increases receptivity to a new culture of boundary-transgressing modernism.

Does School Play a Role?

The politics of schools has become more prominent of late, with parents increasingly concerned that far-left ideas about race, gender, and sexuality are being taught to their children. Some schools openly challenge conventional understandings of these topics, such as the Nova School in Seattle, which has 80% LGBT students and describes its pedagogical approach as “decentering whiteness, patriarchy, [and] hetero- and cis-normativity.”28 On March 28, Florida Governor Ron DeSantis signed the Parental Rights in Education Bill, which prohibits discussion of gender identity and sexuality in K-3 classrooms.29 The discussion surrounding this bill and similar controversies raises the question: do schools play a role in the rise of LGBT identification and behavior? If they do, does the type of high school or university someone attends affect their propensity to identify as LGBT?

The answer to the second question, as Figure 24 illustrates, is that whether a college student attended a public, private, or parochial school, or is homeschooled, appears to make little difference to their propensity to identify as LGBT. It may be that top colleges disproportionately attract LGBT students from parochial or homeschooled students, but this would assume that non-LGBT students select away from top colleges in large numbers. It is also worth remarking that the LGBT share is no higher among college students than non-students in the CCES. Of equal note is that the LGBT share is similar among various categories of college, from the more elite Ivy League to the less prestigious R2 colleges. The major exception is Liberal Arts Colleges, where 38% of students identify as LGBT, a significantly higher share than among the rest, a finding that holds even with controls for gender, race, and ideology.

Attending a college with a high LGBT share, controlling for other predictors, does mean it is more likely a student will be LGBT. The effect is as strong as gender, suggesting that LGBT-rich contexts may be playing a small role in increasing the likelihood of identifying as non-heterosexual. However, this could also be the result of personal characteristics not captured in the FIRE surveys that induce LGBT people to cluster toward certain colleges, rather than because of the impact of social context on identity choice. It is also worth mentioning that the FIRE data show only limited clustering of LGBT students in particular colleges, with considerably lower segregation than for race or religion. More broadly, GSS and CCES data show no difference between rural areas and cities in the chance of a young person identifying as LGBT once demographics and ideology are held constant. This suggests that local social context may have only a limited effect on a person’s sexual identity.

It would be incorrect to interpret the data as suggesting that the education system has played no role in the rise of LGBT. Though there is little difference between various kinds of schools in the extent to which their student bodies identify as LGBT, it could still be the case that there is a common pro-LGBT influence across all educational institutions, with the homeschooled sample skewed by LGBT-identifying homeschooled respondents who disproportionately select into top universities. The same could be happening with parochial schools. If this is the case, it would not appear in this data. Moreover, as can be seen in Figure 24, the samples for individuals who as high schoolers who were homeschooled or in parochial schools are quite small. The results for those categories should therefore be interpreted with caution.

Has Gender Nonconformity Peaked?

Most of this discussion has focused on the rise in bisexuality and, to a lesser extent, homosexuality. But what about gender nonconformity? Coverage of trans issues has surged since 2014, focusing particularly on young girls’ use of puberty blockers and surgery to transition to become boys, and on older men who identify as female, raising questions such as who should have access to women’s spaces and sports competitions. Between 2010 and 2020, the US recorded a 1,000% jump in the share of teenagers identifying as trans.30 In 2020, the CCES asked, “Have you ever undergone any part of a process (including any thought or action) to change your gender / perceived gender from the one you were assigned at birth? This may include steps such as changing the type of clothes you wear, name you are known by, or undergoing surgery.” 2.4% of over 60,000 respondents said they were transgender, with a further 2.8% replying ‘prefer not to say.’ Among the under-30s, 5.8% identified as transgender.

CCES is not a nationally representative sample, even with data weights applied. The figure of 5.8% is far higher than the 2.1% transgender recorded by Gallup for those aged 18 to 25.31 It is also considerably larger than the 0.85% trans and nonbinary share recorded for the 20-24 population (0.3% overall) by the 2021 Canadian census, which provides perhaps the best approximation to the actual US number.32 In terms of sex, 61% of trans identifiers under age 30 in the CCES responded that they were male and 39% female. For the 18-24s, the ratio is 64-36. This changes to 54-46 for the 25-29 group. This dovetails with the Canadian census data showing a preponderance of trans men over trans women in the 15 to 24 age group, reaching parity in the 25-29 group.

The FIRE figures are interesting because they cover two years, with a sample of nearly 20,000 students in 2020 and over 13,000 students from the same schools in 2021, as well as a further 24,000 students from other schools in that year. When asking respondents their gender, FIRE has the choices of male, female, and non-binary. It has no transgender option. Given the overlap between trans and non-binary identity, we may take the non-binary number in the FIRE data as indicative of gender nonconformity. The FIRE survey reports that 1.5% of American undergraduates from the leading 50 universities identified as non-binary in 2020. This declined to 0.85% in 2021. The wider sample of 150 schools in 2021 shows a slightly higher non-binary share of 0.95%, but this is still well below the 1.5% recorded in 2020. Figure 25 below charts this decline. Moreover, in 2021, freshman and sophomore students were less likely to report being non-binary than older students. Younger age and the more recent survey year of 2021 are both significantly associated with lower gender nonconformity. This suggests that non-binary identity, and perhaps transgenderism, have peaked and are declining.

This conclusion triangulates with two international datasets. First, there is the Canadian census, which shows that the share of trans and non-binary people is lowest among the elderly, rises to a peak of 0.85% among the 20 to 24 population, but then declines to 0.73% for those aged 15 to 19. Another data source is referrals for transgender surgery at Britain’s leading Tavistock Clinic. These show that referrals rose from 136 in 2010-11 to a peak of around 2,745 in both 2018-19 and 2019-20, subsequently dropping to 2,383 in 2020-21.33 The FIRE results are consistent with this pattern of a 2020 peak followed by a substantial decline, with younger people less likely to identify as gender nonconforming.

The decline of non-binary identification among students is all the more interesting because it took place at the same time as a substantial rise in gay, lesbian, bisexual, and other non-traditional sexual identifications, which increased from 18.6% in 2020 to 24.5% in 2021 in the same dataset. Same-sex attraction and gender identity are ostensibly distinct, yet there is an important relationship. The data suggest that a considerable number of people who indicate that they are non-binary also identify as gay, lesbian, bisexual, pansexual, or queer (LGBQ). What seems to have occurred is a shift of many from non-binary-plus-LGBQ identification to LGBQ-only identification between 2020 and 2021. FIRE data in Figure 26 shows that nearly 4 in 10 non-binary students identified as either gay, lesbian, or bisexual. Only 11% ticked the ‘other’ sexuality box and just 14% identified as heterosexual.34 Between 2020 and 2021, the number of non-binary individuals declined within both the heterosexual and LGBQ groups, with two-thirds of the loss occurring within the latter.

Gender nonconformity is associated with identifying as non-heterosexual, being younger, and not being religious. The three parameters are of approximately equal predictive power. In the CCES, being liberal – especially very liberal – predicts trans identification even after controlling for sexual orientation. In the FIRE data, very liberal ideology strongly predicts non-binary identification, but this effect disappears when accounting for sexual orientation. The FIRE sample shows that 18-19-year-olds are significantly less likely to be non-binary than 21-22-year-olds. Despite some differences, the correlates of gender nonconformity – ideology, religion, age – are broadly similar to those that predict bisexuality and homosexuality.

Political Implications

What are the political implications of the rise in LGBT identification? Figure 27 displays Higher Education Research Institute (HERI) Freshman surveys of a representative sample of some 100,000 students per year across all types of colleges. It shows a substantial rise in the proportion of female students identifying as liberal or far-left (dark blue bars), from 27% in 2003 to 42% by 2016 (I don’t have access to gender breakdowns for more recent data). Men (light blue bars), by contrast, have remained relatively similar in their degree of liberalism.

What accounts for rising liberalism among young women since 2012? GSS data would indicate that religious decline is more closely associated with liberalism than sexual identity is. At an individual level, having no religion, along with a person’s level of religious attendance, explain about 2 to 3 times more of the variation in ideology than sexual orientation does, and this holds for both men and women. However, at the individual level, LGBT identity remains a significant predictor of being liberal, even when controlling for religiosity.

If the rise in LGBT share were only taking place among liberals, there would be no impact on aggregate liberalism among young people. Figure 28 summarizes the data for females for the period where we have comparable indicators. The religiosity-liberalism relationship at the aggregate level is tighter than that between LGBT share and liberal share. This indicates that the rise in LGBT identification over time is relatively independent of liberalism and has largely taken place within the liberal population, rather than the rise in bisexuality and homosexuality producing more liberals.

Party identification is heavily associated with ideology. Thus, the rise in liberalism among young people should have translated into an increase in support for the Democrats and a decline in support for the Republicans. However, an alternative possibility is that more left-leaning moderates are now calling themselves liberal and more right-leaning moderates are identifying as conservative due to polarization. Indeed, Figure 29 shows that partisanship among young Americans in the GSS has been relatively stable, with young voters gravitating to the Democrats during the Obama years, then somewhat away from them in the Clinton period, then back toward the Democrats after 2018. The American National Election Study shows a similar trend in the period up to 2019. All this during an era in which both homosexual identification and secularism were on the rise.

The surge towards the Democrats in the GSS data between 2018 and 2021 may look significant but should be taken with caution as it is one data point based on a small sample and straddles a change in methodology from in-person to online between 2018 and 2021. Other surveys, such as the Harvard Youth Poll, show a more modest Democrat to Republican advantage among under-30s: 38-23 in 2021, narrowing to 38-25 in 2022.35 In short, there is little evidence that the rise in LGBT identification among young people since 2010 has had much effect on the partisan balance. This suggests that while LGBT identifiers are much more liberal and more Democratic than average, most of the growth in LGBT identity has occurred among young people who were already liberal and has not translated into significant switching toward the Democrats.

In addition, this provides further evidence for the proposition that a common psychological substrate influences ideology and sexual identity, rather than LGBT identity serving as an independent force that inclines people toward liberalism.

Conclusion

This report considers the remarkable phenomenon of the tripling of LGBT identification among young Americans between 2012 and 2021. I hypothesize that a more sexually liberal and modernist culture, one which values difference, best accounts for the new trend. This culture has its greatest impact on the most sexually fluid and perhaps easily influenced groups: the young, the very liberal, and women. Bisexuality, especially among women, accounts for nearly half the LGBT total among young people. While government data from the UK and Canada indicate that the rise in LGBT identity might be significantly exaggerated in other surveys, the underlying trend is certainly real.

Much of the rise in LGBT identification has occurred among those in the most liberal fifth of the political spectrum. While there appears to be some rise in LGBT sexual behavior, at least two-thirds of the increase in LGBT identity is among those whose sexual behavior is heterosexual. The one category that displays a different dynamic is gender nonconformity. Like gays, lesbians, and bisexuals, trans and non-binary individuals are more likely to be young and very liberal. However, while other LGBT categories continue their rise among younger age groups and have arguably yet to reach their zenith, gender nonconformity shows a peak in 2020 followed by a decline in 2021. It is highest among those aged 20 to 24 and less popular among those 19 and under.

Strong political beliefs, such as hostility to the police and offensive speakers, are an important predictor of LGBT identity among white college-educated Americans but are less likely to explain the relatively high LGBT share among minorities and those with only a high school education. The Great Awokening probably cannot account for the modest rise in LGBT identification among older age groups and conservatives. It is a broader phenomenon. While LGBT identifiers lean heavily liberal and relatively Democratic, the sharp increase in their share among those under 30 has not translated into higher Democratic party identification. When considered alongside other evidence in this report, this indicates that the rise in non-heterosexual identity largely unfolded within the liberal Democratic bloc, blunting its impact on the wider partisan balance.

References

• A recent study claimed small cash transfers to the parents of newborns improved their babies’ brain activity. The study was lauded in the media and by D.C. policymakers who argued its results supported redistributive policies, most notably the child tax credit.

• Closer inspection of the baby brainwaves study reveals its claims to be wildly overstated and its methodology to be suspect, as is common for policy relevant social science.

• Authors Troller-Renfree et al. committed numerous bad research practices, including:

  • Deviating from their analysis plan, and justifying it by referencing studies that did not support and even contradicted their results.

  • Highlighting results obtained through typically unimportant methodological decisions that weren’t preregistered.

  • Ignoring that larger and more prolonged interventions typically generate smaller effects than they found, meaning their results were likely due to chance or error.

• Looking beyond the study in question, the theoretical and empirical basis it claims to build upon is largely a house of sand. The evidence that is purported to show a link between brain waves and cognitive outcomes is extremely weak, and most interventions seeking to improve cognitive ability in children have small to nil effects.

• Despite its inconclusiveness, the study was portrayed as relevant to child tax credit policy, vigorously promoted in media outlets like Vox and The New York Times, and championed by thinktanks and policymakers.

• These issues are not unique to the baby brainwaves study. Social scientists frequently engage in questionable research practices and exaggerate the strength and implications of their findings. Policymakers capitalize on low-quality social science research to justify their own agendas. This is particularly true when it comes to research that claims to improve cognitive and behavioral outcomes.

• Thus, we ought to be skeptical of social scientists’ ability to reliably inform public policy, and of policymakers’ ability to evaluate social-scientific research objectively, particularly when it is “policy relevant.”

Click "download" for a PDF

796KB ∙ PDF file

Download

Download

For the executive summary, click here. Code is available for all original analyses.1

Policy, Politics, and Social Science

Social science often informs, guides, and justifies public policy. Theories and empirical findings in economics, sociology, psychology, and political science can inspire and legitimize new programs such as behavioral interventions to improve public health, organizational and administrative schemes to boost diversity, and efforts to combat inequality by redistributing income. Given the credibility of social science with policymakers and the public, one might be inclined to believe that its findings and recommendations are generally useful and accurate. Yet much of the social science informing policy is low quality, and policymakers often benefit from uncritically citing second-rate studies that support their preferred policies.

Political actors aiming to promote a certain outcome frequently use the language and methods of social science to provide evidence that policies they want are effective. Some, like Austrian economists or advocates of Modern Monetary Theory, wield theories as their weapons in the war over policy. Others stick to empirical evidence, which can often be manipulated into supporting any number of narratives. The former typically contort facts to fit with theories and present the beneficial conclusions of their theories as a result, thus justifying policies that follow from their theories. The latter overfit, distort, omit, and sculpt evidence to fit a narrative, however convoluted, and present their work as an affirmation of said narrative and their expectations of it. While both types of actors do a disservice to science, this report is focused on the latter school of empirical policy-centric social science. It focuses on one particular study, the research that it supposedly builds upon, and literatures related to it to demonstrate how far away the social sciences are from being able to reliably inform policy debates.

To give an example, consider the concept of food deserts: areas in which people are believed to have poor access to healthy foods. They have been the subjects of intensive investigation; studies correlating access measures, poverty, obesity, and various other outcomes all seem to confirm that food deserts cause poor health and drive disparities between rich and poor. Yet, the research basis for food deserts has been almost entirely uninformative. It has largely consisted of research designs supporting the conclusion that food deserts are related to bad outcomes, correlating those outcomes with variables like poverty, obesity, and the proximity of supermarkets. However, there are methods to assess the causal impacts of food deserts, not just their associations. The establishment of government-subsidized supermarkets, supermarkets opening due to tax incentives, or a variety of small food store interventions have all been fruitfully leveraged to gauge the effects of changes in the food environment on all the important correlates of living in a food desert.2 They have, at best, delivered effects on ill-defined measures of ‘knowledge,’ like which foods are fattier, and modestly increased the amounts of fruit and vegetables people eat. The ‘improvements’ seen in these studies are not clearly related to health at all, and studies have not supported improvements in objectively measured health, as indicated by obesity rates or other health measures.3 Taken seriously, the causal studies suggest that socioeconomic disparities transcend differences in access to healthy food.

The academics, policymakers, and activists who in the past promoted food desert studies that were purely associational never produced sufficient evidence for their policy guidance. They contributed to a now-widespread belief that disparities in access to healthy food drive socioeconomic differences in health when, to date, the evidence to that effect is extraordinarily weak at best. This naïve analysis has been used to justify many public and corporate policies, including the Obama administration’s nationwide program to end food deserts, Lyft’s Grocery Access Program in D.C., Walmart opening more than 200 stores in areas with food deserts, and the National Institute of Health issuing millions of dollars in grants to research food deserts.4 And yet, the best evidence suggests that these kinds of policies won’t improve health, because food deserts aren’t causing their inhabitants to be unhealthy.

There are many other examples of weak social science being used to questionable ends: ineffective school management interventions foisted on hundreds of Indian schools, decades of wasteful “Drug Abuse Resistance Education,” and lucrative grants for fraudulent nutrition science, not to mention the plethora of ineffective “nudging” policies and practices inspired by social psychology and behavioral science.5 It is intuitive that research will influence popular political will, and, with qualifications, it probably should. But when research is weak or, in the worst cases, based on fraud, we might suffer from its political effects. A bad research paper can hurt an academic’s career, but a bad policy can drastically harm individuals, communities, and even countries. Thus, those with the relevant technical training – scientists and hard-nosed policy analysts, for example – should be openly and vocally critical of bad social science, especially when it is policy relevant.

Cash and Baby Brainwaves

On January 24th, 2022, The New York Times published a “Breaking News” piece entitled “Cash Aid to Poor Mothers Increases Brain Activity in Babies, Study Finds.6 The article discussed Troller-Renfree et al. (2022), a study published in the Proceedings of the National Academy of Science (PNAS) that purported to show a monthly subsidy of $333 yielded considerable improvements in children’s brainwave activity.7 Vox also covered the story, opining that “we can be reasonably confident the cash [parents received] is a primary cause of these changes in babies’ brains. And we can be reasonably confident it will be a causative factor in whatever future outcomes the… researchers find.”8 They weren’t alone: think tanks, medical news aggregators, and other mainstream outlets like NBC and Forbes all praised the study and the alleged psychological benefits of brainwave improvements.9

Laudatory coverage was accompanied by claims that the study had special relevance to expanding child tax credits. The authors even put out a pro-child tax credit press release on their website: “This study’s findings on infant brain activity… really speak to how anti-poverty policies – including the types of expanded child tax credits being debated in the U.S. – can and should be viewed as investments in children.”10

High praise and lofty implications aside, a closer examination of the study reveals its results to be dubious and its methods severely flawed. While the authors’ public statements portrayed the study as strong, reliable, and policy relevant, none of these claims hold up to scrutiny. The study and its portrayal stand as excellent examples of what can go wrong when political desire supersedes scientific reasoning, and why we ought to be skeptical of social scientists’ ability to reliably inform public policy with their research.

Troller-Renfree et al. (2022) was an analysis of an ongoing project known as the Baby’s First Years study. It comprised a total sample of 1,000 low-income (approximately $20,000 per year) mothers and their children, randomly divided into a treatment group in which parents were given monthly subsidies of $333, and a control group that received $20 a month. The cash benefits were considerable in the treatment group, boosting their incomes by around 20% (compared to 1% in the control group). The point of the project was to “provide the first definitive understanding of the extent to which income plays a causal role in determining early child cognitive, socio-emotional and brain development among low-income families.”

The study looked at differences in babies’ brainwaves between control and treatment groups. Brainwaves are fluctuating electrical pulses produced by the actions of neurons in the brain that can be assessed by electroencephalography (EEG). Brainwaves have been used to measure neural activity and, as the authors described, are usually measured along two dimensions: frequency and power. Frequency is the spectrum along which activity is measured, in Hz, and power is the amount of activity at certain frequencies. Power is often subdivided into absolute and relative power, with the difference being the amount of activity at a given frequency versus the absolute power as a fraction of absolute power across a certain range of frequencies. These frequencies are subdivided into ranges or bands which are given Greek-lettered names, like delta (0.5 – 4 Hz), theta (4 – 8 Hz), alpha (8 – 13 Hz), beta (13 – 35 Hz), and gamma (>35 Hz), giving us terms like “alpha power” or “delta power” to denote power within those ranges. The ranges for these specific brainwaves vary slightly by source, but the order I’ve listed them in places them in their consistent relative positions from lowest to highest frequency. Power in some of them has been related to a variety of outcomes.

To establish the importance of brainwaves for their analysis, Troller-Renfree et al. cited several studies that found relationships between EEG power and important cognitive and developmental outcomes. They then described the treatment effect of cash gifts on specific brainwave bands using regressions to control for several variables in the part of their study that had EEG data. This is how they worded their results:  

[I]nfants whose mothers were randomized at the time of their birth to receive a large monthly unconditional cash transfer showed greater mid- to high-frequency absolute EEG power in the alpha-, beta-, and gamma-bands, compared with infants whose mothers were randomized to receive a nominal monthly unconditional cash transfer.11

Several academics and public intellectuals pointed out flaws in Troller-Renfree et al.’s study shortly after it was released. Psychologist Stuart Ritchie noted the article’s potentially-dubious peer review status, the disconnect between observations made by the study’s authors and outcomes that matter, the potential for fade-out effects, deviations from preregistered methodology, and the nonsignificance of the reported effects of cash transfers on any of the study’s main outcomes.12 Statistician Andrew Gelman found that the significance of results was not robust and that the graphs produced by the authors could be replicated even if the children in the sample were randomized into artificial treatment and control groups.13 Finally, psychiatrist Scott Alexander offered further commentary on the article to the effect that the results seemed unlikely for various reasons, alongside a summary of what other people had said.14

All three critical commentators agreed: Not only did the study authors make sloppy methodological choices, but they also went out of their way to portray their findings as more policy relevant than they were. Moreover, even if their findings were wholly accurate, which is highly unlikely, their policy suggestions would not follow from their results.

Problems With the Literature Linking EEGs to Psychological Outcomes

For the results of the study to be important when taken at face value, changes in EEG power must either be causally linked to changes in their psychological correlates – such as IQ or language skills – or they must reflect a common cause for change in both. In either case, the results must transfer to the phenomena that EEG power has been linked to in the past. The study’s authors cited several pieces of research that linked EEG power to their desired cognitive outcomes. These studies included Benasich et al. (2008), Gou, Choudhury and Benasich (2011), Brito et al. (2016), Maguire and Schneider (2019), Williams et al. (2012), and Brito et al. (2019).15 Cumulatively, Troller-Renfree et al. argued that these studies linked absolute power in mid-to-high frequency bands to linguistic, cognitive, and social-emotional development, while simultaneously arguing that low-frequency band power was linked to behavioral and learning issues.

These studies suffered from several issues, five of which I will address. First, all had small sample sizes ranging from 13 to 129 participants. Second, the studies they cited were frequently inconsistent, with results from different studies pointing in different directions. Third, these studies’ associations with socioeconomic status were almost all confounded with other characteristics like genes and family environment.16 Fourth, the study samples were often highly unrepresentative of the general population. Finally, many did not address the crucial issue of multiple comparisons, which means it is likely their policy relevant results were due to chance or p-hacking.17

Small Sample Sizes

Small sample sizes are a serious problem in scientific research because they lack statistical power: the probability that a statistical test detects a real effect of a given size at some level of significance. If I have a sample size of 20 and I need to find a correlation between X and Y that has a p-value below 0.05, the lowest significant correlation would be Pearson’s r = 0.44. If I quintuple my sample size, the lowest significant correlation is 0.20. This does not mean that a small sample cannot yield small correlations, only that, when it does, they will not be significant.

When small samples are combined with reviewers and editors being biased towards results that are statistically significant, effect sizes will tend to be exaggerated. Think of it this way: to get published, I need a result with a p-value below 0.05. In terms of the typical level of statistical power people in this area seek (80%) and the typical p-value cutoff they use (0.05), I would have 80% power to detect a correlation of r = 0.58 with a sample size of 20, and 80% power to detect a correlation of 0.28 if I increased my sample size to 100. If I’m stuck with a small study, I must have a big effect or else what I find will not be significant. If I get an effect, it is probably going to be much larger than it would if I had more power to detect it.18

We can think of this another way, by comparing the mean significant effect versus the actual effect in a comparison of two group means. At 100% power, the mean significant effect is the same size as the actual effect. However, at 80% power, the mean significant effect is 20% larger than the actual effect, and at 30% power, the mean significant effect is 70% larger than the actual one. The extent of statistical exaggeration is almost twice as large for correlations.  

So, how were the studies that purported to link EEG activity to cognitive outcomes? Their relevant descriptive statistics are in Table 1. Though I am not discussing them in this section, the studies linking EEG activity to SES are covered in Table 2.

Table 1. Study Descriptives for Cited Cognitive Ability-EEG Studies19

Table 2. Study Descriptives for Cited SES-EEG Studies20

­

With sample sizes ranging from 13 to 129, the sizes of the effects in this literature were almost certainly greatly exaggerated. If we take the minimum significant r as the mean effect and assume the real effect is like the relationship between nerve conduction velocity and intelligence,21 at r = 0.10, the best-powered study of the bunch (Brito et al., 2019) still presented an effect that was 1.7 times larger than it should have been, but that study did not find significant relationships between any EEG parameters and measured cognitive ability, only with a parent-reported measure of the child’s social and emotional development. The next best-powered study in that list would have exaggerated the size of the correlation by 2.1 times. If we calculate effect size exaggeration from power rather than the critical r, the results are even less favorable.

As stated in their supplement, Troller-Renfree et al. was sufficiently powered to detect a treatment effect of 0.21 SDs if they had retained 80% of their initial 1,000 participants. With sufficient power to detect an effect of 0.21 at a sample size of 800,22 they ended up with a total sample size of 435.

Inconsistent Foundations

Troller-Renfree et al.’s study claimed that prior work generally supported positive associations between power at higher frequencies and better cognitive outcomes, whereas power at lower frequencies was associated with the reverse. Yet these studies – and another they cited but did not discuss – failed to follow that pattern. The results from Benasich et al. (2008) were consistent in showing positive associations between absolute gamma power and their various cognitive measures. Likewise, Harmony et al.’s (1990) results went in the same direction, and so did Gou, Choudhury and Benasich’s (2011), Williams et al.’s (2012), and Brito et al.’s (2016), with inconsistent significance. But Brito et al.’s (2019) and Maguire and Schneider’s (2019) studies were not supportive. The former offered nonsignificant results for their cognitive ability measures despite having the largest sample of the group, and the latter only had one significant cognitive ability-EEG power association: a positive one between low-frequency theta power and working memory. Since theta power is a low-frequency band that the authors argued to be deleterious at higher levels, this finding is striking.

Looking at all the studies they cited, an important and undiscussed one was Begus and Bonawitz’s (2020) paper The rhythm of learning: Theta oscillations as an index of active learning in infancy.23 This paper came with a plenitude of citations relating theta power to cognitive test outcomes. Several of these suggested greater theta power during memory-related tasks, and that the degree of increase in power and its level are related to greater cognitive performance. Some – in the citations of their citations – even suggested alpha activity during tasks was negatively related to performance, and yet more, that theta activity increased while alpha fell when tasks were difficult.

While these outcomes involved mostly temporal and not resting-state theta or alpha relationships with cognitive performance, they are a cause for concern about how theta and alpha were characterized. The reason for this is simple: many modern theories of intelligence emphasize a potential crucial role of working memory, and even resting-state theta is sometimes positively related with it.24 Since we do not have temporal EEG information for Troller-Renfree et al.’s study, this concern cannot yet be addressed.

Unfortunately, all these studies suffered from the same kinds of sample size issues as those Troller-Renfree et al. cited positively. As a result, they were similarly uninformative. In fact, all the studies Troller-Renfree et al. cited alongside those mentioned by Begus and Bonawitz suffered from another major, unaddressed issue of potentially being reverse-causal and reflecting behavior rather than causing it or being related due to common causes. Mendelian Randomization or similar methods could address this issue, but the relevant data does not currently exist.

Troller-Renfree et al. needed a systematic meta-analysis of each of the types of EEG-cognitive ability associations they examined to support the patterns they wanted to observe being beneficial. Moreover, they needed a systematic meta-analysis to assess magnitudes, because if magnitudes of associations are too small, there is no necessary transitivity between EEG enhancements and improved cognitive function. Transitivity is the entire basis of the potential importance of their study; that is, their results might have importance because the changes in EEG power they claimed they observed caused enhanced cognitive ability.25 If there was no transfer from EEG changes to cognitive ability, then the results were purely of theoretical interest rather than social relevance since no one is concerned with EEG activity on its own.

Confounded Associations

All these studies were uninterpretable for Troller-Renfree et al.’s purposes for another reason: they dealt with cross-sectional associations. In the real world, associations occur for many reasons, and when it comes to neural variables, they are often related to cognitive performance because of genes shared between the neural phenotype and performance. In a study that has been replicated numerous times, Posthuma et al. (2002) found that the positive relationship between general intelligence and brain white and gray matter volumes was accounted for by shared genes, rather than environments.26 Similarly, the relationship between socioeconomic status and cognitive ability has been found to have a substantial genetic component, and the link between them is substantially severed when genetics are accounted for.27 Even EEG power is heritable, with most relationships among EEG power at different frequencies attributable to shared genes.28

As a result of substantial genetic influence and other potential forms of confounding, how are we to generalize either from experimental effects on EEG parameters to cross-sectional samples or vice versa? Troller-Renfree et al. were at least somewhat aware of these facts, as evidenced by their citation of Wax (2017), who argued that “the so-called neuroscience of deprivation has no unique practical payoff… Because this research does not, and generally cannot, distinguish between innate versus environmental causes of brain characteristics, it cannot predict whether neurological and behavioral deficits can be addressed by reducing social deprivation.”29

Unrepresentative Samples

Though genetic confounding loomed large, it was not the only sampling issue. These studies generally sampled populations that were in some way unusual or limited with respect to the full ranges of cognitive ability and socioeconomic status. For example, Benasich et al. (2008) used a sample within which 85% of parents were in the highest socioeconomic strata and the other 15% were middle-class, and nearly a third of the sample came from families with histories of language impairment. In Maguire and Schneider (2019), the low-income sample had an IQ (99.2) that was indistinguishable from the population average (100). The sample used by Gou, Choudhury and Benasich (2011) had an average IQ of 113 – a standard deviation above the general population! Williams et al.’s (2012) sample may have had lower than average cognitive development because of the direct effects of the participants’ congenital heart disease, or because of genes shared between low cognitive ability and heart disease. In the UK Biobank dataset, these phenotypes are related such that angina, chronic ischemic heart disease, major coronary heart disease events, and a broadly defined ischemic heart disease are all significantly negatively genetically correlated with fluid intelligence scores. Other reviews have also highlighted frequent cognitive deficits among individuals with congenital heart disease, while noting that phenomena such as enhanced mosaicism in trisomy 21 and greater numbers of copy-number variants, commonly found in congenital heart disease sufferers, were related to worse cognitive outcomes.30 Some studies even involved individuals with IQs below 90 – only two-thirds of a standard deviation from the mean – being removed from the samples used in their analyses. Given that the studies cited by Troller-Renfree et al. had problematic and unusual sampling, we have little reason to believe they serve as a solid evidentiary basis for theorizing about or predicting population outcomes.

Multiple Comparisons

Finally, only three of the nine studies corrected for multiple comparisons, and of the three that did correct for them, only one (Williams et al. 2012) did so in a standard and verifiable way. The problem of multiple comparisons is important, and amenable to graphical explanation. Figure 1 illustrates the relationship between the number of statistical tests and the probability that at least one of the results is significant with a significance threshold of 0.05 if there is no effect.

  The issue with small sample sizes is also open to graphical explanation. Figure 2 illustrates the relationship between sample sizes and the average sample correlation conditioned on reaching a 5% significance threshold (the mean significant correlation) in simulated data with different true underlying correlations. The mean sample sizes in Tables 1 and 2 were 89 and 67, respectively. If the true correlation is typical of the correlations observed in large, reliable studies in neuroscience (closer to 0.10), correlations in those studies were exaggerated by 2.6 to 3 times, assuming balanced samples, a lack of measurement error, and no heteroskedasticity, high-leverage outliers, or nonnormality, and therefore overestimates power while underestimating effect size exaggeration.

If we perform five tests with a p-value cutoff of 0.05 and there are really no differences, the probability that at least one p-value is below 0.05 is 1 – (1 – 0.05)5 = 23%. With ten tests and no correction for multiple comparisons, the likelihood of a false-positive is 40%. The interpretation of a p-value as the probability of a more extreme result given the truth of the null relationship drifts as more tests are conducted. All the studies without corrections for multiple comparisons in Table 1 had more than ten tests, so the results should be considered to have been run with unacceptably liberal p-value cutoffs.

The only study that can realistically be generalized from to support the patterns Troller-Renfree et al. wanted was by Marshall et al. (2008). That study featured comparisons of absolute and relative alpha and relative beta power between groups of Romanian children who were randomized between institutional and foster care settings. The p-values were Bonferroni-corrected, the sample included 41 institutionalized children and 49 who were placed into foster care, and the authors observed no significant differences in power measures between a group of 22 placed into foster care before two years old, a group of 27 placed after two years of age, and the 41 institutionalized children. Placement age likewise was not significantly related to absolute theta, alpha, or beta, nor to relative theta or beta, though it was significantly negatively related to relative alpha power. None of the mean differences in absolute or relative power between the institutionalized or foster care groups were significant. It is hard to infer anything regarding EEG patterns of socioeconomically meaningful interventions from this study.

The studies that Troller-Renfree et al. proffered to support a strong interpretation of their results were weak. They utilized small samples, had inconsistent findings, and were sampled in ways that make generalization difficult. Even if they had found significant effects of their intervention on EEG power parameters, there was little to no reason to think these would elicit a causal impact on behavior.

Problems With the Study Itself

Once these methodological issues are considered, it becomes clear that Troller-Renfree et al. misled their readers about the ironclad nature of their results. In fact, their intervention yielded no reliable statistically significant effects. They observed no significant differences in group means of the absolute or relative power parameters and, after including many covariates, they arrived at treatment effects ranging in absolute value from 0.02 to 0.26. Prior to correction for multiple comparisons, the only significant effects were found for absolute beta and gamma power. With p-values of 0.02 and 0.04, respectively, the effects were at best marginal, and as might be expected given a marginal result among many tests, they stopped being significant when the authors corrected for multiple comparisons.

With no significant results, the authors still positively interpreted their study. They claimed that their “findings underscore the importance of shifting the conversation to focus more attention on whether or how income transfer policies promote children’s development,” but this claim was not supported by their results.31 Moreover, these findings probably ought to be given less prominence because they did not stick to a preregistration plan in the way claimed by the authors.32

The Unregistered Tests May Have Contradicted the Authors

The largest effect Troller-Renfree et al. observed – for absolute beta – was not preregistered. They attempted to justify assessing effects on beta in their supplement by saying that studies at the time of preregistration did not link income to EEG power, but studies since then had indicated that income was in fact linked to beta activity. They cited two studies in support of that view.

The first study, Brito et al. (2020), featured a sample of 94 infants and reported that maternal education and the income-to-needs ratio were significantly associated with EEG beta power (p’s = 0.025 and 0.048); however, neither survived the authors’ correction for multiple comparisons. Furthermore, in their regressions with adult word count, conversational turns, and child vocalizations, there was one significant relationship in 15 regressions, and these same regressions included no significant relationships between the income-to-needs ratio and beta in any region or whole-brain gamma power. Infant auditory comprehension and expressive communication were significantly related to one another, but not to either measure of income or maternal education. Other relationships with those measures were unclear because the authors’ supplement was improperly uploaded and is currently just a .zip file containing a .xml file that includes nothing.33

The second study, Jensen et al. (2021), took 210 6-month-old children from the “Cryptosporidium Burden Study,” and another 210 36-month-old children from the “Performance of Rotavirus and Oral Polio Vaccines in Developing Countries” study. After attrition, their sample sizes totaled 160 and 187, respectively. Among the 6-month-olds, wealth, maternal education, family care, and maternal stress were not significantly related to any EEG parameters; among 36-month-olds, maternal education and family care were unrelated to EEG parameters, while maternal stress was significantly positively related to frontal and central theta, and wealth was significantly negatively related to all beta and gamma power parameters. In their regressions, there were no significant absolute or relative EEG relationships with maternal education, wealth, family care, or maternal stress in the 6-month-old group. In the 36-month-old group, the regressions yielded the same results for relative power, but there were significant positive relationships between absolute theta and maternal stress, and significant negative relationships between wealth and absolute beta and gamma power.

The inconsistencies of both results with those of Troller-Renfree et al. are obvious: one study produced nulls and may have done the same with cognitive measures, and the other produced effects in the opposite of the direction of Troller-Renfree et al.’s intervention effects, alongside a bundle of null associations. These hardly serve to justify their unregistered analyses, and thanks to the specific results in these citations and their generally larger size than the other studies Troller-Renfree et al. cited, they further reinforce the inconsistency of this literature with their results and the potential incommensurability of experimental and cross-sectional EEG associations.

The Study was Reliant on Typically Unimportant Analytic Decisions

Two additional unregistered methodological choices were critical to Troller-Renfree et al.’s results. The first revolved around the fact that they produced their estimates using robust regressions. However, a glance at their code revealed that the regressions were not quite robust. They used what are known as HC1 standard errors – the default in the program they used to analyze their data, Stata. These standard errors fare poorly under scenarios where certain participants exert high leverage on estimates, and with even moderate sample sizes like theirs, the false-positive rate of HC1 robust regressions is considerably above the nominal rate of 5%. Rerunning their analyses using HC5 standard errors – which perform better with small samples, high leverage, nonnormality, and severe heteroskedasticity – I found no significant results even before accounting for multiple comparisons.

The second methodological decision involved how they computed their effect sizes. In evaluating RCTs, it is common to divide the effect by the standard deviation (SD) from the control group. It is also common to divide the effect by the pooled SDs from either group, weighted or unweighted by the sizes of the samples. The authors did not offer any principled justification for one effect size calculation method over any other, but they did choose the only one that would deliver them significant results. In choosing to divide by the SD from the control group, they bought themselves significant p-values for the absolute beta and gamma power effects, where any other method would have rendered their effects nonsignificant. In the case of absolute alpha power, had they used either SD pooling method, the effect would have been negligibly larger and still nonsignificant.

Though not a methodological decision, the authors also included an interpretive one that was biased towards a positive interpretation of their results. The authors were inconsistent about the interpretation of results that were marginal. Twice in the main study, they noted that effects were “at the margins of statistical significance,” and they did the same three times in their supplement. So-called “marginally significant results” are often those that are just nearly above the p-value cutoff. Researchers will frequently interpret these, but they seem to nearly universally neglect that the “marginal” label must be two-sided: if a result is marginal at a p-value above the researcher’s chosen cutoff, it is equally marginal at the same distance below that cutoff. These authors called p-values as high as 0.10 marginal; if we were to give those findings a significant interpretation, they ought to consider at least anything greater than 0.01 to also be marginal. This turns every significance test they ran at best marginal, and more realistically, nonsignificant. They cannot have it both ways: effects cannot be marginally significant if they are not, on the other end, also marginally nonsignificant.

These decisions are normally unimportant, and they possibly represent the study’s authors using defaults without giving them any thought. For a study that produced borderline results, these decisions were critical. For a well-powered RCT, this sort of methodological decision would not usually affect results, but when they determine the result, the study must be considered suspect.

In their supplement, Troller-Renfree et al. produced a series of region-specific EEG effects. Six of the thirty-two region-specific associations were marginally significant (p-values between 0.01 and 0.04). These associations did not survive correction for multiple comparisons when I did them correctly (i.e., across bands), but three of them narrowly remained under 0.05 when they were done incorrectly (i.e., only within frequency bands, unlike the analyses in the body of their paper). The reason this result is notable is because the study’s senior author, Kimberly Noble, claimed that this post hoc vindicated their results in an update to the Vox story on this study.34 She “argues that the results are robust because of what is known as ‘regional analysis’ – analysis that looks at where in the brain differences between the high-cash and low-cash children showed up” and is quoted as saying that “if they had come from, say, parts of the brain that support vision, we might have been very skeptical that we were seeing something meaningful. But they came from parts of the brain that are critical for supporting higher-order thinking.”

There are four major problems with this kind of thinking. First, according to their studies and reasoning for assessing beta associations (noted above), the effects they observed went in the wrong directions for the beta power band, and potentially also for the gamma band. Second, there are no validity studies that support their proposed region-specific effects with a reasonable degree of certainty. Third, this multiplies the number of tests, and we ought to take it for granted that, were they all significant together, they would have put them up front and center, since they now seem to be doing something similar with more restricted analyses in order to protect their study from criticism. And finally, the reasoning is plainly faulty: claimed region-specific associations with brain variables ought to almost never be trusted. The reason this is the case is simple and has to do with sample sizes and power. When you have low power, your effect sizes must be exaggerated. When you have low power and you find a region-specific association, it is almost certainly because you had one region with an extreme and unrealistic result, while the others were more realistic, and the study was not powered to detect if those differences existed at all. Many region-specific associations with fMRI have been like this: they are the results of low-power studies being exploited to capture chance phenomena.35

Other EEG Analyses

The authors went on to find that there was a significant treatment effect (p = 0.02) on summed high-frequency bands. But they likely didn’t expect this to happen per their own cited beta (and maybe gamma) validity studies. This test was not corrected for the many comparisons it was featured among and all it could do was recapitulate group differences in the high frequency bands, with additional power to reject a lack of difference between groups because summing them reduces measurement error, while pulling many assumptions into the analysis and removing analytic clarity.

In another analysis, they performed their focal analyses across all power bands but with log-transformed absolute and relative powers, because some people in psychophysiology had argued for logging variables prior to analysis. This rendered all their estimates nonsignificant prior to multiple comparison correction. Because of how the authors are now backtracking, it is reasonable to suspect they would have both justified and presented log-transformed results if they were more significant than their untransformed results.

Researcher Degrees of Freedom

The idea that researchers would only present their significant results and methods that supported their desired interpretations while moving the rest of their results to their supplement or out of the paper entirely is sometimes dubbed “researcher degrees of freedom.” Preregistration reduces the degree to which authors can positively misreport or chop up their findings, but there is usually still leeway to selectively report and highlight results. The proposition that splitting data in different ways would validate nonsignificant findings because some number of the subset effects are significant is a spurious one. The result of doing this is more comparisons and an effectively lower significance threshold, which means more exploitation of chance.

This search for significant results in subsets of data has led to the popular repetition of the phrase “the difference between ‘significant’ and ‘not significant’ is not itself statistically significant.”36 This phenomenon appears in situations such as when an effect is significant in one group but not another and authors interpret that observation as evidence of a difference in the size of the effect by group. But the difference in the effects may not be significant; the study may simply be underpowered for the effect to be significant in one group while the other may have had a more extreme result than they should have by chance alone.37 The proper way to test this difference is to test for an interaction in a singular model or to directly compute the significance of the difference in effects. One reason this sort of thing is not done more often may be because it requires much larger sample sizes.38

Many papers have dealt with this kind of data abuse to wring out significant results. For example, Miller and Sanchez-Craig (1996) gave tongue-in-cheek advice to researchers to do things like “use liberal definitions of success” and “always declare victory regardless of findings.”39 In more recent times, Simmons, Nelson and Simonsohn (2011) have noted how “flexibility in data collection, analysis, and reporting dramatically increases actual false-positive rates. In many cases, a researcher is more likely to falsely find evidence that an effect exists than to correctly find evidence that it does not.”40

Wicherts et al.’s (2016) checklist of researchers’ degrees of freedom has considerable relevance to Troller-Renfree et al.’s paper and their attempt to salvage nonsignificant main results with reference to exploratory secondary ones.41 Examples they listed include:

• Studying a vague hypothesis that fails to specify the direction of the effect

• Creating multiple manipulated independent variables and conditions

• Measuring the same dependent variable in several alternative ways

• Failing to conduct a well-founded power analysis

• Measuring additional constructs that could potentially act as primary outcomes

• Selecting the dependent variable out of several alternative measures of the same construct

• Trying out different ways to score the chosen primary dependent variable

• Operationalizing manipulated independent variables in different ways (e.g., by discarding or combining levels of factors)

• Choosing between different statistical models

• Choosing the estimation method, software package, and computation of standard errors

• Failing to report so-called “failed studies” that were originally deemed relevant to the research question

• Misreporting results and p-values

• Presenting exploratory results as confirmatory (Hypothesizing After Results are Known, HARKing)

All of these can be fine in specific scenarios, but they are also pathways towards allowing the misuse of statistics and the misrepresentation of results. The post hoc presentation of region-specific analyses with dubious expectations for those results is one such abuse example.

Did the Intervention Affect Cognitive Development?

Troller-Renfree et al. also tested the effects of cash transfers on a questionnaire about language milestones and its relationships with EEG parameters. Their language milestone assessment, the Ages and Stages Questionnaire 3 (ASQ-3) is a maternal report instrument used to assess linguistic development in young children.42 The authors noted that this instrument had been strongly correlated with another development screener, the Battelle Developmental Inventory, and that it had also been found adequately reliable.

In the subset of the study with EEG data, there was a significant treatment effect on the ASQ-3 (p = 0.03), but it was marginal, per the authors’ implicit definition. If coupled with other tests, it could not survive correction for multiple comparisons. The authors went on to test the treatment effect again, but this time using the full sample of 900. When they did this, the effect shrank enough to become nonsignificant (p = 0.21). The difference between these coefficients was not significant, but the authors did not test to see if that was the case.

The authors attempted to explain their nonsignificant effect in a few ways and did note that some of the explanations were not especially plausible. First, they explained the finding as the result of a possible difference response in patterns between in-person and over-the-phone responders. They rendered this less likely by noting that ASQ-3 scores did not differ by collection method (p = 0.67). Second, they suggested differences between the control and treatment samples with respect to levels of language input could have given rise to a reduced effect because the pandemic equalized language input due to increased time at home for everyone regardless of group. And third, systematic differences between the EEG and total samples that interacted with the cash gift may have negatively affected the total sample relationship. They reduced the probability of this explanation by noting that their non-response weight analysis better matched the EEG sample to the total one and the same pattern of EEG effects from before the weights were applied was still observed.

These explanations did not apply to the same things despite being listed together. The first and third explanations had to do with the difference between the EEG and total samples, the second, with the existence of the effect. This is relevant because it is peculiar, and more importantly, because all these explanations are, in a sense, immediately testable by the study’s authors. But even though they released their data with the study, we cannot test it.

The way to test all these explanations for the differences is to assess something known as measurement invariance. Measurement invariance is the condition where a questionnaire or assessment is unbiased when used in different groups.43 Without bias, differences between groups reflect the construct the instrument is supposed to measure.

If all the utilized ASQ-3 items were provided, we could assess the first and third explanations by assessing measurement invariance between the EEG and total samples. If they were correct, measurement invariance would be violated. The second explanation is testable by assessing measurement invariance between treatment and control groups – if it (or another explanation) was true invariance would be violated.

The method used to test the second explanation must be applied in general. To assess intervention effects on psychological outcomes and interpret the effects as alterations of a particular psychological construct requires invariance between the treatment and control groups. To interpret changes over time in terms of a given construct target requires longitudinal invariance, and to interpret differential changes in a construct, invariance over time needs to be assessed across groups.

If the effects of treatment are heterogeneous, this will cause a violation of the strict phase of testing invariance, in which the equality of causes of differences aside from the construct are tested. This would yield an important note: if different influences existed between groups, the group with the greater variances would have additional influences on responding. Failing to support this level of invariance might make some of the intervention effect meaningless.

Consider intelligence. An RCT with an effect on all the scores from a test, in which strict invariance was violated, means something besides intelligence was affected, either alone or in addition to effects on intelligence. In intelligence research, it is a stylized fact that most – and often virtually all – of the predictiveness of IQ tests is due to the general factor of intelligence, g. If your intervention affects g and question-specific performance, your intervention will affect things g leads to, like higher rates of learning on the job or a greater capacity to solve a variety of complex puzzles. When it affects performance specific to a certain test, this is not as likely to affect external outcomes or to yield success in the future. If you get very good at addition, that by no means makes you better at public speaking; if you improve your intelligence, that should. If I give a sample all the answers to an IQ test and I make the control group take the test normally, the treatment (giving out answers) does not affect intelligence, it only affects the scores, and that does not mean the kids with the answer key became more intelligent. Psychological RCTs always need to be measured with this in mind: measurement matters, and if it is not done, the results are ambiguous and changes in measured outcomes might be useless.44

Finally, the authors threatened the interpretation of their study by correlating ASQ-3 scores with the whole gamut of absolute and relative EEG power estimates. The p-values ranged from 0.39 to 0.95: they were all very nonsignificant, despite being much larger than the studies from Table 1. They offered more measurement-based explanations and the possibility that the links would grow with development. But if this were the case, and they were asking us to expect totally different effects on cognitive outcomes at older ages, why not also have totally different patterns and magnitudes of EEG effects?

Addressing Attrition and Missingness

Troller-Renfree et al. also tested what inverse probability weighting (a technique to allow the computation of effects when the samples are more equal in terms of baseline traits) and non-response weights (intended to address missingness issues) did to their estimates. These weighting schemes made their results universally nonsignificant, so they must have known before presenting their results that it was wrong to present any of them as significant when they were not robust to reasonable and informative analytic choices like these.

Did the Study Show Parents Spent the Money Well?

Per the New York Times’ coverage, economist Lisa Gennetian – one of the study’s co-authors – said “the results indicated the parents could be trusted to make good decisions” with free money. The study as presented did not and could not show that money given to parents was spent well. The lack of effects means we cannot tell if giving people money has improved anything for their kids. More importantly, the controls likely removed from the analysis much of the tendency of parents to misbehave. Among the long list of controls were maternal mental health, maternal smoking and alcohol consumption during pregnancy, and a variety of other life history speed and misbehavior correlates. When parental misbehavior is treated as a construct that can be measured by things that indicate it, it is apparent that the authors overcontrolled by controlling for at least part – and probably a large part – of the variance in parental misbehavior. At the very least, they overcontrolled if the goal was to show money was spent well. Perhaps parental misbehavior was an important source of heterogeneity! Because the effects were so nonsignificant, the present data cannot be used to test that suggestion.

Should We Expect Effects to Grow?

Vox quoted psychology professor Allyson Mackey on Troller-Renfree et al., “My prediction is that the brain effects of cash transfer will grow as kids grow up.”45 If this idea is relegated to effects like those on EEG parameters, then, as we have seen, it may well be useless. If, however, it also includes downstream behavioral effects from those brain ones, then we have an existing evidence base that contradicts it.

A nigh-universal phenomenon known as the fadeout effect plagues interventions designed to improve people’s educational (and other) outcomes. In a 2015 review, John Protzko showed that interventions as diverse as intensive preschool programs and vitamin A supplementation, which initially appeared to elicit effects, all faded in their effects at subsequent assessments.46 Bailey et al. (2020) also conducted an excellent and accessible review of this phenomenon.47

In addition to the fadeout effect, there is a phenomenon in intelligence research48 known as the Wilson Effect, by which systematic environmental effects fade and genetic effects increase with age.49 These alone are enough to throw the idea that the intervention effects should grow into disarray. If people are arguing that continued exposure to higher income will accumulate larger effects, then they have a gigantic burden of evidence, and, hopefully, they will assess fadeout after the intervention inevitably ends.

The New York Times quoted Representative Suzan DelBene (D-Wa.) as saying that Troller-Renfree et al. showed that “investing in our children has incredible long-term benefits.”50 The evidence for this claim is nonexistent. The study covered one year, and there was no reason to think effects would persist or generalize such that they would become meaningful by modifying participants’ actual behavior and cognitive abilities. Long-term benefits can never be taken for granted.

It’s a “Contribution”

PNAS has two submission categories: the typical one and “Contributions.”51 Contributions are available as an option for National Academy of Science members, and they have been controversial since they have been alleged to make peer review easier.52 This kind of publication involves the article submitter selecting their own reviewers and gathering up their own reviews. Troller-Renfree et al.’s article was submitted by Greg Duncan, and he selected Martha Farah and Joan Luby to review.

These reviewers had prior publications with some of the coauthors of Troller-Renfree et al. Martha Farah has published at least six articles53 with Kimberly Noble, and Joan Luby54 published an article55 with Troller-Renfree and Nathan Fox. Normally, this would not be concerning. Fields are sometimes small and subject matter experts are often few, but it is an issue when you get to choose your reviewers and they are personal friends and colleagues. The freedom of the review process from bias in this situation is hard to maintain.

Most Interventions Have Small to Nil Effects

To increase the believability of their large and probably exaggerated effects, Troller-Renfree et al. stated that “the observed effects in the alpha-, beta-, and gamma-bands are similar in magnitude to those reported in other large-scale environmental interventions. For example, a meta-analysis of 747 randomized control trials of educational interventions targeting standardized achievement outcomes found an average size of 0.16 SDs.” This was a great overstatement. The study (Kraft, 2019) was interpreted as charitably as possible to their own views.56 The effect of 0.16 the authors noted was nonsignificant and was likely to result from small study biases like those mentioned above. In fact, small studies did have much larger effects than larger studies in this meta-analysis, and there were fewer studies the larger the samples got. This was also consistent with the observation that the median effect sizes were lower than the mean ones and their distance decreased with increasing sample size. The meta-analytic effect size weighted by study sample sizes was 0.04, and not significantly different from zero. Because small study effects still exist even when studies are weighted, this ultimate effect is still probably upwardly biased. When looking at math and reading separately, the same phenomenon occurred, and the same result happened. These educational interventions were not what they were cracked up to be.

Educational Interventions

Kraft’s 2019 study was not the only one on educational interventions. The 2010 Head Start Impact Study conducted by the Office of Planning, Research and Evaluation looked at effects elicited by the nationally-implemented, well-funded, and widely celebrated Head Start program for three- and four-year-old children.57 Using a p-value cutoff of 0.10, they found significant positive effects on eight of fourteen cognitive tests given in the year of the program for the three-year-old children, and this reduced to two of fourteen by the next year, at age 4. In kindergarten there were nineteen tests administered, with one significant positive effect, and one significant negative one, while, in the 1st grade, only one effect of twenty-two remained significant. Because this was not significantly affected during Head Start and its effect only became significant amongst a pile of nonsignificant effects, this was probably noise. For the four-year-old group, seven of fourteen scores were significantly positively affected in the Head Start year, but effects disappeared in kindergarten, and one effect became significant again in first grade. In a particularly informative analysis by Lortie-Forgues and Inglis (2019), they found that the typical educational intervention had a minute effect (mean = 0.06 SDs) and was unconvincing in Bayesian terms.58 Though their analysis included fewer studies than Kraft’s, their analysis was probably better because it mitigated publication bias by only using studies funded by the Education Endowment Foundation and the National Center for Educational Evaluation and Regional Assistance, both of which require all trials they fund to be published and to follow standardized reporting practices.59

As shown by Duncan and Magnuson (2013), the effect sizes associated with early childcare programs declined over time.60 Repurposing their data, I found a correlation of r = -0.37 between a study’s precision and its effect size, meaning that the more power the study had, the smaller its effect. Adjustment for publication bias rendered the meta-analytic effect nonsignificantly different from zero, and this was true even though the authors recoded outliers via Winsorization. This was qualitatively like the result of a Brookings report from 2018 that found state-level changes in prekindergarten enrollment were not associated with changes in state-level achievement scores.61

Cash Transfer Effects

Although Troller-Renfree et al.’s study was the first to assess the effects of unconditional cash transfers on EEG parameters, it was not the first to assess effects on cognitive ability. Baird et al. (2013) found that unconditional cash transfers did not have a significant effect on test scores (0.04, 95% CI: -0.04 – 0.12), and neither did conditional cash transfers (0.08, 0 – 0.16), but together, they yielded a marginally significant effect (0.06, 0.01 – 0.12). On the other hand, both types of cash transfer had considerable effects on enrollment and attendance, but there was substantial evidence of publication bias that the authors did not compute the effects of. Luckily, since the data were available, I was able to find that the results did not survive weighting studies by the square root of each study’s sample size. To explain away the nulls, one could argue that these interventions did not offer large enough cash subsidies or that they were too small, or enrollment was range restricted. But to the extent that is true, it just increases our uncertainty about cash transfer effects since they have not been found to have meaningful effects on test scores, much less psychometrically meaningful variables like intelligence, memory, or verbal ability.

The World Bank also evaluated several other cash transfer programs.62 The studies in their Table 3.1 ranged in timeframes from one to ten years. Three of them were unconditional, while the other five were conditional cash transfers, with sample sizes ranging from 106 communities to 2,069 kids and were all either RCTs or randomized phase-ins. Unlike the unconditional transfer studies, the conditional ones supplemented cash transfers with schooling and health check-ups. In Paxson and Schady (2010), language, fine motor function, and socioemotional skills were not affected, and there was a positive effect on one of the many Woodcock-Johnson subtests, long-term memory (0.18 SDs). In Fernald and Hidrobo (2011), there was a significant effect on the MacArthur Language Test (0.15 SDs) for rural kids, but no effect on urban ones. In Lopez, Boo and Creamer (2018), ASQ-3 scores were improved by 0.13 SDs and communication was improved by 0.18 SDs. Barham, Macours and Maluccio (2013) found 0.15 SDs of cognitive effects and no significant socioemotional effects. Fernald, Gertler and Neufeld (2008) found effects of 0.11, 0.10, and 0.09 on long- and short-term memory and visual integration when the cash transfer was doubled, and they also found a 0.18 SD effect on vocabulary, but no significant effects on fine motor function. Fernald, Gertler and Neufeld (2009) found no significant effects on cognitive ability or language skills, but they did observe a 0.14 SD reduction in behavioral problems. The only study in the batch that involved multiple follow-ups (Macours, Schady and Vakis, 2012), showcased fadeout. At kids’ first follow-ups, there were significant effects on short-term memory (0.15 SDs), vocabulary (0.23 SDs), and other language parameters (0.14 SDs), alongside a 0.13 SD socioemotional effect, and no significant effects on fine motor function. At the second follow-up, the short-term memory effect declined to 0.09 SDs, the effect on vocabulary was not measured, the language effect dropped to 0.09 SDs, and the socioemotional effect dropped to 0.01 SDs, while, inconsistently, effects on fine motor function became significant at 0.16 SDs.

The same World Bank report also covered three cash transfer studies in greater depth. These studies were conducted in Niger, Mexico, and Colombia, and the Colombian study was also covered by Baird et al. (2013). Initial sample sizes ranged from 1,420 to 6,856 children. The Colombian study compared three treatment groups to a control group, who received conditional cash transfers. The first treatment group received psychosocial stimulation, the second received micronutrient supplementation for anemia, and the third received both. There were no significant differences in any of the treatment group comparisons with the control group in height, weight, socioemotional development, and gross and fine motor function. There was a significant effect of treatment relative to the cash group for the Bayley-III cognitive test and receptive language at the first follow-up in the comparison of the first treatment and the control group. None of these effects remained at follow-up. The Mexican trial compared two treatment groups to a control group. The control group received money and the treatment groups received money plus available (group 1) or available and encouraged (group 2) parenting group sessions. There were no significant differences between group 1 and the control group, but there were considerable impacts on memory, full-scale IQ, and verbal scores for the second treatment group. There was no follow-up. The Niger study compared a control group given unconditional cash to a treatment group given conditional cash plus parent training about nutrition, psychological stimulation, health topics, and sanitation. There were no significant effects on cognitive development, but children’s socioemotional problem index and sociality scores increased. There was considerably less self-reported illness, more handwashing, a greater likelihood of receiving all recommended vaccinations, and iron levels improved.

Nutritional Interventions

Though they usually have nothing to do with cash transfers, it can be informative to look at other large intervention evaluations to see what sorts of things elicit effects and in what magnitudes. In Nepal, Dulal et al. (2018) ran a double-blind RCT where groups of mothers were given prenatal multiple micronutrient supplementation in the treatment group, and iron and folic acid only in the control group.63 At twelve years of age, there were no effects on the children’s cognitive ability. Behrens et al. (2020) reviewed whether vitamin B affected rates of cognitive decline and found no effects, despite vitamin B mechanisms supposedly being well-defined and the effects plausible at the time of writing.64 Similarly, a large-scale follow-up of two double-blinded, placebo-controlled, cluster-randomized trials of vitamin A effects on prenatal and newborn children by Ali et al. (2017) yielded nonsignificant effects across the board.65

Deworming

Deworming charities have been a cause célèbre among effective altruists for some time now. Welch et al. (2019) reported the results of a systematic review of deworming effects on a variety of outcomes including cognitive ability, height, and weight.66 Despite parasitic worms often being debilitating and the mechanisms through which they should impede development being abundantly clear, there were no significant meta-analytic effects on weight, height, or cognitive ability. Welch et al. compared their results to earlier systematic reviews by Taylor-Robinson et al. (2015) and Welch et al. (2016) that, altogether, found one marginally significant meta-analytic effect on weight and none for height or cognitive ability.67

Brain Training

Brain training interventions have ebbed and flowed in popularity in the intelligence testing literature. Methods to improve working memory, such as the Dual N-Back, were at one time very popular and, based on published results, even seemed to raise intelligence.68 However, the case of the Dual N-Back is like those of other methods in that, once greater rigor was applied, the effects stopped generalizing or disappeared.69 A large second-order meta-analysis of brain training studies (Sala et al., 2019) came to similar conclusions.70

Will the Effects be Larger with Greater Cash Transfers?

One may suspect that cash transfers larger than those distributed by Troller-Renfree et al. might have a desired effect, thus putting a more positive gloss on the results of their paper. In general, however, interventions cannot simply magnify their effects by increasing the amount of stimulus provided. For example, if we set up a daycare program, it is not as though adding twelve hours of daycare supervision would improve children’s cognitive outcomes even if a study showed that one hour was associated with better outcomes. Similarly, if children were deprived of iodine, an intervention that gave children enough to poison them would not yield improvements over providing the recommended daily allowance.

Certain designs can appear to be scaled-up versions of the present interventions. For example, the Moving to Opportunity program that relocates people to high-quality neighborhoods has a staggering economic value. And yet, Ludwig et al. (2013) found that Moving to Opportunity had no effect on a mathematics and reading assessment, attained education, or almost any other outcome.71

An even more impactful form of intervention is adoption. The largest adoption studies afford modest effects72 – on the order of one-third of a standard deviation – despite very large environmental improvements.73 But effects like this are only a fraction of the apparent relationships with socioeconomic status found within nonadoptive families, in which siblings and parents share genes rather than just environments. To understand this, take the following plot, derived from Odenstad et al. (2008).74 The plot includes three groups in Sweden: Korean adoptees, non-Korean adoptees, and biological children. These groupings are relevant because, uniquely for adoptions, among Korean adoptees, would-be parents cannot choose the child they receive, so selection effects are removed and, consistent with this, the effect of age at adoption on attained IQ disappeared in the Korean – but not the non-Korean – samples. Despite considerable adoption effects on attainment, the effects were not large enough to generate a systematic relationship between adoptive family socioeconomic status and cognitive ability.

This plot clarifies that adoption effects, as large as they may be, are a fraction of the realized differences related to socioeconomic status in the general population when the link between family socioeconomic status is confounded with genes. Additionally, there is just no reason presently to suggest the effects will grow considerably past some likely already near enough point, or that they will ever reach the level expected based on cross-sectional outcome-socioeconomic status associations. Extensive intervention fadeout at all ages might also signal a long-term lack of efficacy after the intervention. In any case, Troller-Renfree et al. and commentators in this area should not generalize from biologically, or otherwise considerably, confounded associations.

A potential counterargument is that effects earlier in life will better persist and have larger effects than ones based on later interventions. That is just an argument for a lack of age-related generalization across studies, which virtually nullifies any theoretical basis this study might have ever had to consider its results meaningful in the first place. This is a common belief, but the evidence for it – popularly known as the Heckman Curve – is scant.75 Regardless, if assessment of scaling is ever done, we can only hope that the analysis is well-powered, properly designed, and psychometrically sound.

Conclusion

Troller-Renfree et al. sought to assess how unconditional cash transfers affected brainwave activity in infants. The results they observed were null, but discussed as though they were positive, theory-confirming, and, most importantly, policy relevant. Their results were ambiguous, and, because their cited literature was so poorly powered and inconsistent, their theory formation should not be taken seriously.

The policy relevance of Troller-Renfree et al.’s null results was founded on the belief that they found effects; that those effects were beneficial; and that they are, or will be, large enough to generate positive behavioral and cognitive change in affected children. The evidence mustered in favor of the syllogism that would make these results meaningful was poor, contradicted by aspects of the study itself, and, at present, not worth serious consideration.

The contemporary obsession with the scientific justification of policy gives rise to politically motivated and shallow research, as well as the use of questionable research practices in the pursuit of a researcher’s extra-scientific goals. University press departments and mainstream media will often credulously report such research, seemingly without any awareness of its weaknesses. The press departments won’t raise a single criticism – they will write what they are told by a study’s authors. The mainstream press tends to take the studies’ authors too seriously when they have reached personally agreeable results, regardless of the quality of the work. Troller-Renfree et al. should serve as an example of low-quality review at every step, from peer review at PNAS to its reception by The New York Times, Vox, Forbes, and other outlets, and parts of the broader public.

It is difficult to say who is more at fault for popularizing what were essentially null results. Does the blame lie with the journalists, who had all the opportunity in the world to look at the research, read the citations, and realize it did not hold up? Or is it the fault of the authors, who pushed their paper through PNAS’ easy track for submission and chose to promote an untenable interpretation of their findings to the wider world? It might be the fault of readers, who believed any of them. At every step, people acted irresponsibly. Scientists have a duty to correct misconceptions about their research, to prevent people from portraying it as more than it was, and to curb public excesses regarding the interpretation of their work.

The public arena is at least somewhat self-correcting, in that the findings in Troller-Renfree et al. were quickly critiqued by a handful of bloggers and scientists. The bigger problem, however, is that if it had not gotten a write up in the New York Times accompanied by a news alert, and if the study’s topic and author’s framing had not been so palpably absurd, the paper may have never been carefully checked and the results may have turned into conventional wisdom. Moreover, the handful of articles that criticized the study were likely seen by fewer people than the many articles that portrayed the study positively and uncritically. Bad science is the norm; correcting it is not.

Cases like this ought to make us reconsider the role social science plays in our public policy debates. Political bias and confirmation bias are heuristics that plague us all, and we have no reason to think that social scientists or policymakers are immune. Flashy results that support a popular policy are almost always untrustworthy, and large effects are usually exaggerated, p-hacked, or due to chance. The reality is that most social interventions and policies have a negligible impact when it comes to improving cognitive ability or behavior. Until researchers and the educated public come to grips with this fact, we should be skeptical of policymakers’ ability to evaluate research objectively and social scientists’ ability to reliably inform public policy with their work.

References

• The National Science Foundation (NSF) is the main governmental scientific grant distributing body in the United States, with an annual budget of over $8 billion.

• This report uses natural language processing to analyze the abstracts of successful grants from 1990 to 2020 in the seven fields of Biological Sciences, Computer & Information Science & Engineering, Education & Human Resources, Engineering, Geosciences, Mathematical & Physical Sciences, and Social, Behavioral & Economic Sciences.

• The frequency of documents containing highly politicized terms has been increasing consistently over the last three decades. As of 2020, 30.4% of all grants had one of the following politicized terms: “equity,” “diversity,” “inclusion,” “gender,” “marginalize,” “underrepresented,” or “disparity.” This is up from 2.9% in 1990. The most politicized field is Education & Human Resources (53.8% in 2020, up from 4.3% in 1990). The least are Mathematical & Physical Sciences (22.6%, up from 0.9%) and Computer & Information Science & Engineering (24.9%, up from 1.5%), although even they are significantly more politicized than any field was in 1990.

• At the same time, abstracts in most directorates have been becoming more similar to each other over time. This arguably shows that there is less diversity in the kinds of ideas that are getting funded. This effect is particularly strong in the last few years, but the trend is clear over the last three decades when a technique based on word similarity, rather than the matching of exact terms, is used.

• Taken together, the results imply that there has been a politicization of scientific funding in the US in recent years and a decrease in the diversity of ideas supported, indicating a possible decline in the quality of research and the potential for decreased trust towards scientific institutions among the general public.

Click "download" for a PDF

2.47MB ∙ PDF file

Download

Download

Code and data for replication are available here.

Introduction

Richard Feynman introduced a concept he called “cargo cult science” during a commencement speech at Caltech in 1974.1 In the Second World War, Allied and Japanese airbases sprung up on islands in the South Pacific that were home to pre-industrial cultures that previously had little contact with the modern world. The soldiers on many of these bases would trade manufactured clothing, medicine, canned food, and other goods with the natives, most of which arrived by airdrop. After the war ended and the soldiers left, the native populations on some of the islands began to create replicas of things like airstrips, airplanes, and flight control towers. They even made mock radios and headphones out of coconuts and straw. The natives believed that by recreating the conditions under which the airplanes came and dropped goods, they could get the airdrops to resume.

There were entire areas of academic study that Feynman called “cargo cult science” – mostly in the fields of social science and education. These areas of inquiry see the success that the scientific method delivers in disciplines like physics, chemistry, and medicine, and produce superficial replicas of scientific practices. They miss something essential, however, and as Feynman says, “the planes never land.” So, what is this missing element in “cargo cult science?” It is “a kind of scientific integrity, a principle of scientific thought that corresponds to a kind of utter honesty.” Science requires a willingness to relentlessly assail one’s assumptions and a capacity to bend over backwards to try and disprove ideas even if one passionately wants to believe they are true. I would love to be able to revisit this critique of Feynman’s and have a good laugh at how strange and backwards academic institutions and scientific agencies must have been in the 1970s, but unfortunately, this critique feels as salient today as it must have when Feynman first expressed it almost 60 years ago.

Ideally, the “cargo cult sciences” should start to recede as their practitioners start to notice that “the planes aren't landing.” What if, however, we have in place institutions that have degraded to a point where they subsidize and reward practices that are not actual science, but a kind of science-like interpretive dance? Patrick Collison and Michael Nielsen find that there has been a precipitous increase in the number of science publications, PhD students graduating in STEM fields, and government spending on National Institute of Health (NIH) and National Science Foundation (NSF) grants since the 1960s.2 Their same work, however, shows that when scientists were surveyed about the importance of Nobel Prize winning discoveries between the 1920s and 1980s, results indicated anywhere from a decline to a general stagnation in the impact of science over that time period – nothing anywhere near the output we might expect given the tremendous amount of time and energy now invested. One theory posited by Collison, as well as economist Tyler Cowen and others, is that science might just be getting harder; we have gotten to all the low-hanging fruit and now are in the territory of diminishing returns.

Science may be getting harder, but it seems unlikely that this is the sole, or perhaps even largest, cause of decreasing productivity. In addition to the previously mentioned increase in funding to scientific research, the costs of many of the important inputs to research – such as computing power, gene sequencing, and various types of lab equipment – have been declining exponentially. With a rise in funding and a decline in the costs of many inputs, we might expect to be able to generate increased scientific output despite potentially increasing difficulty.

Although a fuller explanation for technological and scientific stagnation is beyond the scope of this work, here I analyze the abstracts of successful NSF applications and find two reasons to believe that something has gone wrong with the culture of science, particularly in the last few decades. The first of these is increasing politicization. If paying lip service to fashionable political ideas becomes an important criterion for successful grant applications, this will certainly detract from the importance of other more vital criteria – namely those related to the quality and importance of the proposed research. When the process of deciding what research projects get funded comes to be based on a political litmus test, the scientific endeavor suffers. Additionally, the more that scientific institutions come to be viewed as conduits for promulgating ideology, the less capable they will be of swaying public opinion on important issues. We may be starting to see the harmful effects of this process in the current epistemic crisis regarding public health. The growing view of science as a vehicle for activism detracts from its more vital role of being a dispassionate referee that adjudicates the validity of empirical claims.

The second major result in this work is the constriction of the space of ideas within NSF award abstracts. The number of NSF awards given and the total amount of taxpayer money spent by the NSF have increased consistently since 1990, and yet this work provides evidence suggesting that the breadth of ideas within NSF award abstracts has been contracting. In different contexts, bureaucracies can become positive feedback chambers reinforcing and amplifying favored ideas while excluding others. Recent work by Johan Chu and James Evans supports this view by showing that the larger a scientific field becomes, the more it tends to stagnate, with more reliance on established works in citation patterns and fewer fundamental breakthroughs.3 The NSF is ostensibly an organization meant to stimulate scientific progress for the benefit of the nation, but the way in which it has become entangled with academia and established institutions may make it seem more like a professional guild representing the interests of its members. Such factors could explain the stagnation we see in the ability of the NSF to identify and support novelty.

NSF Awards

The NSF, an independent federal agency, has a stated mission “to promote the progress of science; to advance the national health, prosperity, and welfare; to secure the national defense.”4 It has an annual budget of around $8.5 billion and funds approximately a quarter of all federally funded basic research at colleges and universities in the US. The NSF provides public access to archives of awards granted in every fiscal year since 1967.5 The data available in these archives do not include the full proposal texts, which are often lengthy and detailed documents, but only the abstracts of the proposals, which are generally only several hundred words in length. Additionally, all awards prior to 1986 do not have abstracts, and most awards do not have abstracts until 1988. Given these limitations of the data, this work only analyzes awards from 1990-2020. Other information provided in these archives for each grant are the identities and institutions of the primary investigators, the award amount, the title of the proposal, and the NSF directorate that granted the award. There are seven directorates that are responsible for dispersing the majority of NSF funding: Biological Sciences; Computer & Information Science & Engineering; Education & Human Resources; Engineering; Geosciences; Mathematical & Physical Sciences; and Social, Behavioral & Economic Sciences. All analysis conducted in this work is broken up by these seven directorates.

Methods

This section intends only to explain the processes used in analyzing the NSF award data at a higher level. These processes are explained in greater depth in Appendix IV. Additionally, the code used for the analyses in this work is publicly available to facilitate easy replication of the results presented here.6

Term Counting

This method simply counts the number of documents containing some form of a given term. For example, for “inclusion” terms we look for documents containing any of the following forms: “inclusive,” “inclusivity,” and “inclusion.” Even if a document contains multiple instances of different forms of a given term, it will only be counted once. The percentage of documents containing a particular term over the total number of documents in each NSF directorate in a given year is reported.

Word Frequency Average Cosine Distance

This method collects all the words from all the abstracts in a given directorate and year and creates a list of each word and the frequency with which it occurs. The frequency distributions of words in individual abstracts are then compared with the overall frequency distribution of words in the directorate and year to which the abstract belongs. A mathematical technique called cosine distance is used to determine how different these two distributions are. A cosine distance of 0 means that the two distributions are exactly the same, whereas a cosine distance of 1 means that the distributions are maximally different. The cosine distances of the word distributions of every abstract from the distribution of its directorate and year are collected and averaged. This average cosine distance for a directorate and year provides a measurement of how similar or different from each other the abstracts in that group are based on word frequency.

Word Embedding Vectors Average Cosine Distance

Word frequency analysis only deals with language in a purely lexical sense. Using this tool, we can say certain things about the amount of variety in a corpus of text in terms of word frequency, but we cannot say anything definitive about the variety of ideas or meaning. It seems reasonable to expect that variety of words correlates with variety of meaning and ideas, but there may be certain cases where this does not hold true. In order to remedy this shortcoming of word frequency analysis, a second related method is applied.

Word embedding vectors are numerical encodings of words or symbols meant to capture their semantic and syntactic nature. Statistical machine learning techniques are used to generate these encodings by processing a tremendous amount of text. If words have similar meanings and are used similarly in natural language, then their word embedding vectors (numerical encodings) should be very similar. The word embedding vectors used in this work consist of 300 numerical values per token (word or symbol) to represent its semantic and syntactic properties. These encodings were generated using machine learning algorithms that analyzed approximately 600 billion words of text.7

With these word embedding vectors, we can estimate the aggregate semantic and syntactic properties of a corpus of text by simply adding together the vectors of all the words in the corpus. We can then apply the same cosine distance technique to determine the difference (in terms of semantic and syntactic properties of words) between abstracts in a given directorate and year in the same way we have with the word frequency average cosine distance technique. This method puts forth a way of reinforcing the results of the word frequency method by providing an additional measure of document similarity.

Results

Politicization of Awards

The following figures demonstrate a considerable rise in the frequency of award abstracts that contain selected politicized terms over the past 30 years. Additionally, the amount of funding granted to awards containing any of these terms is shown. To supplement these results, Appendix I gives general statistics on award abstracts, Appendix II offers a more detailed breakdown of the increase in incidence of individual terms in abstracts, and Appendix III provides data on the frequency of awards specifically containing two of the terms “diversity,” “equity” and “inclusion.”

As of 2020, across all fields 30.4% of successful grant abstracts contained at least one of the terms “equity,” “diversity,” “inclusion,” “gender,” “marginalize,” “underrepresented,” or “disparity.” This is up from 2.9% in 1990 (Figure 2). This increase is seen in every field. As of 2020, the two most politicized fields seem to be Education & Human Resources (53.8%, up from 4.3% in 1990) and Biological Sciences (43.8%, up from 6.6%), although “diversity” may sometimes have non-political connotations in the latter. Even the fields that should be most disconnected from politics have seen a massive jump in these terms: Mathematical & Physical Sciences went from 0.9% to 22.6%, and Engineering from 1.6% to 25.4%.

Note that word counts is a somewhat crude way of measuring politicization. Bias, particularly in the social sciences, is often subtle, and can apply to the kinds of questions that get asked and the standard of evidence used to accept or reject a hypothesis. Thus, the fact that so many grants contain terms that are in most contexts clearly associated with left-wing political causes likely underestimates the degree of politicization in science funding.

Measures of Document Similarity

The following figures demonstrate the decline in lexical and semantic distinctiveness of award abstracts over time. The techniques used to obtain these results are detailed more precisely in Appendix IV, along with the method for sampling words to correct for a correlation between average cosine distance and average abstract length.

When word frequency measures are used (Figure 6 and Figure 7), we see an increase in similarity between documents that is particularly pronounced beginning in 2017. When I use word embedding vectors (Figure 8 and Figure 9), which consider different words with similar meanings, the effect is present throughout the entire period of the dataset.

This report presents direct evidence that scientific funding at the federal level has become more politicized and less supportive of novel ideas since 1990. While the findings do not allow us to extrapolate back to periods of faster economic and scientific progress before the last few decades of the twentieth century, they certainly are suggestive and may be the latest iterations of longer-term trends. Future research should use other methods and datasets to measure the politicization of the scientific establishment and how supportive it has been of novel research ideas and approaches across time in order to provide more insights into the causes of scientific and technological stagnation.

Appendix I: General Statistics

This section graphically explores general properties of the dataset of NSF Awards from 1990 to 2020 analyzed in this work.

Appendix II: Individual Term Counts

This section presents the frequency of documents containing a number of terms associated with political activism.

Appendix III: DEI Awards

“Diversity,” “inclusion,” and “equity” are terms that tend to cluster together and are associated with a particular set of political programs. This section presents graphs showing the frequency of awards containing at least one of the forms of two of three of these terms.

Appendix IV: Technical Details

Parsing: Individual award data is obtained in archives by year from the NSF website.8 The data are available in XML format. The parser reads all of these and takes the award title, the award ID, the abstract narration, the start date, the award amount, the NSF directorate, and the primary investigator’s first name. The parser then repackages these data into large collections of award data aggregated by year and directorate.

Stop Words: In natural language processing removing “stop words” is a commonly used technique. Stop words are words that are very common in natural language but generally do not carry much meaning. All the words from a commonly used stop word list are removed from abstract titles and abstract narrations before any analysis is done.9

Stemming: Stemming is another technique used in natural language processing that reduces words to their root form. This work uses NLTK (a publicly available natural language toolkit for the Python programming language) to stem words in the abstract titles and narration text. Stemming is used for word frequency distribution analysis but not for word embedding analysis.10

Word Embeddings: Word embeddings are a way of representing individual words as vectors of real numbers for use in text analysis in natural language processing applications. Word embeddings are generally high dimensional vectors – the ones used in this work are 300 dimensions. If two words have a similar meaning, they will have vector encodings that are close together (measured in Cartesian distance.) Word embeddings can be obtained through several different machine learning techniques. A large amount of text will be used as input to a learning algorithm that will find good high-dimensional numerical encodings for each word such that words with similar meanings that are used similarly in writing will have similar encodings and words that have different meanings and usages will have encodings that are more distant from each other. This work uses a set of word embeddings that was trained on over 600 billion words of English text scraped from the web.11

Word Frequency Cosine Difference Technique: To calculate the average cosine distance of word frequency distributions of individual abstracts to the overall word frequency distribution of all abstracts in the same year and directorate, the following method is used:

All stop words are removed from the title and abstract text of all awards. The remaining words are stemmed using the NLTK stemmer. The total number of instances of individual word stems in all awards in a given year and directorate are counted. Additionally, the total number of tokens (word stems) that occur in all the award texts in a given year and directorate are counted. The frequency of all individual stems is calculated by the ratio of occurrences of a given stem over the total number of tokens in all awards in that year and directorate. This can be thought of as a large vector of frequencies of words.

The word frequency distributions are similarly calculated for individual awards. The frequency of an individual word stem is the number of times that stem appears in the award text divided by the total number of words in the award text. If A is the word frequency distribution for an award and D is the word frequency distribution for the entire directorate the award belongs to, the cosine similarity is calculated by the following formula:

Because all the values of word frequencies are positive, Cosine Similarity can only have a value between 0 and 1, and Cosine Distance is merely 1 - Cosine Similarity. The Average Cosine Distance is then computed by taking the average of the computed Cosine Distance for all articles in a given year and directorate.

Aggregate Word Embedding Cosine Distance Technique: The set of word embeddings (Facebook Open Source, 2021) maps words to vectors of 300 real numbers. These vectors can be summed up to represent the aggregate semantic content of all the text in an entire award or all of the text in all awards in a given directorate in a given year.

The cosine distance of the aggregate word embedding vector of an individual award to the aggregate word embedding vector of all the text in all awards in the directorate and year to which it belongs can be calculated as follows:

The Average Cosine Distance of Aggregate Word Embeddings from the Mean is then computed by taking the average of the computed Cosine Distance for all articles in a given year and directorate.

There is a correlation coefficient of -0.49 between the average word length of award abstracts and the average cosine distance of word frequency distributions within each directorate and year. This effect occurs when all of the words in each individual award abstract are analyzed. To correct for this effect (which made it seem like cosine distance was decreasing more dramatically), instead of using all the words in the award text, 100 words were randomly sampled from each award and these words were used in the processes described in this Appendix to produce the document similarity findings shown in the Results section.

References

• Academics, experts, and laypeople often assume stereotypes about groups are inaccurate. This assumption is used to justify policies meant to reduce or eliminate such beliefs.

• Most stereotypes that have been studied have been shown to be approximately correct. Usually, stereotype accuracy correlations exceed .50, making them some of the largest relationships ever found in social psychology.

• Even when people hold true stereotypes, they have little effect on how people judge or treat individuals about whom they have other, individualized information.

• Unlike most findings in social psychology that are small and flimsy, the results noted above are clear, large, reliable, and untouched by the replication crisis.

• The field of stereotype accuracy casts doubt on the usefulness of programs meant to reduce stereotypes in education, government, and business as a way to achieve equality.

Click "download" for a PDF

294KB ∙ PDF file

Download

Download

Introduction

A recent paper titled The Misperception of Racial and Economic Inequality has gotten a lot of attention. It has received over 40 citations, despite being out for less than two years. The work was highlighted at a major professional social psychology conference,1 and was even featured on Twitter by a prominent member of the science reform movement in psychology.2 Why? It has shown quite clearly that many people, including those surveyed in a large representative sample, drastically underestimate the Black/White wealth gap. It shows this by comparing people’s perceptions of the gap to federal data on the actual gap.

The paper makes an interesting point: People’s perceptions of racial wealth inequality are largely inaccurate. As such, the finding is doubly interesting: It is consistent with the longstanding claim in the social sciences that stereotypes are inaccurate, yet it is inconsistent with the longstanding claim that stereotypes typically exaggerate real differences. In this case, they underestimate real differences.

But this paper raises some additional issues. Are stereotypes usually inaccurate, or are perceptions of wealth an outlier? Perhaps even more important, it shows that a claim that is often informed by ideological agendas and activism – that stereotypes are inaccurate – is actually transformable into an empirical question that is readily answerable by social science. The recipe is simple:

1. Assess people’s beliefs about one or more groups.

2. Identify credible criteria for what those groups are actually like.

3. Compare the two.

Once the door is open to addressing some question scientifically, sometimes, the answers may not be what people expect or want. Of course, there are good reasons for stereotypes’ bad reputation: some stereotypes are malevolent and destructive, and have been exploited for propaganda purposes for generations. Each of these were once common and some can still be found today:

• Women as fit for nothing but child-rearing and homemaking.

• Arabs and Muslims as nothing but bloodthirsty terrorists.

• Jews as grasping hook-nosed Nazis perpetrating genocide on innocent Palestinian babies.

Such characterizations are inaccurate, immoral, and repulsive, to say the least. As malevolent as these images may be, they say nothing about the accuracy of what everyday people think of women, Jews, Muslims, or any other group. Here is a secret which, if you go by their behavior, many social scientists and academic intellectuals seem to either not know or forget when the topic of stereotypes comes up: If you want to know what people think about something, you cannot go by television commercials, political cartoons, or characters in movies; it is far more useful to ask them. This is not a silver bullet, because answers depend on how the questions are asked, and people are indeed subject to many biases and distortions. However true this may be, things that cannot possibly reflect their personal beliefs, such as movies or political cartoons, are pretty obviously absurd substitutes for figuring out what people believe. Which raises the question: what do people actually believe about groups, and are those beliefs inaccurate?

A Simple Test

Before continuing, therefore, please take the following quiz:

1. Which group is most likely to commit murder?

A) Men

B) Women

2. Older people are generally more __________ and less __________ than adolescents.

A) Conscientious; open to new experiences

B) Neurotic; agreeable

3. In which ethnic/racial group in the U.S. are you likely to find the highest proportion of people who supported Democratic presidential candidates in 2008 and 2012?

A) Whites

B) African Americans

4. People in the U.S. strongly identifying themselves as ___________ are most likely to attend church on Sunday.

A) Conservative

B) Liberal

5. On December 24, 2004, a dad and his three kids wandered around New York City around 7 pm, looking for a restaurant, but found most places closed or closing. At the same time, his wife (their mom) performed a slew of chores around the house. This family is most likely:

A) Catholic

B) Baptist

C) Jewish

D) Pagan/Animist

Answers appear at the bottom of this paragraph. If you got three or more right, congratulations – your stereotypes assessed here were quite accurate. On the other hand, don’t be too impressed with yourself. Lots of people hold stereotypes about as accurate as yours. And yet, most of us have had it beaten into our heads that ‘stereotypes are inaccurate.’ Why is that? (The answers are: A, A, B, A, C.)

Activism Versus Evidence

It is very hard to have a serious, scientific, evidence-based discussion about stereotypes because the term itself is imbued with malice and delusion. People generally presume that stereotypes are irrational, invalid, and practically synonymous with prejudice. It’s even hard to have a serious discussion about stereotypes with scientists. Many have been deeply concerned with combating oppression – anti-Semitism in the immediate aftermath of the Second World War, racism and sexism following the civil rights and women’s movements in the ’60s, and, more recently, various movements for social justice for all sorts of marginalized groups. Many scientists erroneously believe that since stereotypes have been used to incite hatred, ostracism, and even genocide, they must also be false (though this doesn’t follow logically).

As a result, stereotypes have a terrible reputation. If you make almost any claim about almost any group, the quickest way to have your claim dismissed and delegitimized is for someone to declare: ‘That’s just a stereotype!’ This ‘works’ because of the widespread assumption that stereotypes are inaccurate, and the unstated insinuation that you are probably a bigot. Thus, any evaluation of the validity of your claim is short-circuited by implied impugnment of your moral character as depraved.

Still, high moral purpose (or lack of it) does not translate to scientific truth (or its absence). We do not usually presume people making other generalizations – say, about the weather in Anchorage or the taste of cherries– are doing something bad and inaccurate. Despite this obvious point, the notion of generalizations about people as inherently bad and inaccurate has long been baked into the science without much scientific support.

The Black Hole at the Bottom

One of us (Jussim) began research on stereotypes in the early 1980s. At the time, there were no reasons to doubt the widespread belief in stereotype inaccuracy. As a grad student, I (Jussim) sought to track down the evidence supporting those claims – not to refute them, but to promote them and proclaim to the world the hard scientific data showing that stereotypes were wrong. So, when some published article cited some source as evidence that stereotypes were inaccurate, I would track down the source hoping to get the evidence.

And, slowly, over many years, I made a disturbing discovery. There was just about no there there. Claims of stereotype inaccuracy were often literally based on nothing. For example, a classic paper from 1977 describing research by social psychologists Mark Snyder, Elizabeth Tanke, and Ellen Berscheid stated: “Stereotypes are often inaccurate.”3 Ok, but scientific articles are usually required to support such claims, typically via a citation to a source providing the evidence. There was no source there. Obviously, it could still be based on something unarticulated, but scientists cannot be in the business of making claims up out of whole cloth. If no evidence or citation to evidence is provided, the article literally provides no evidence to support that claim.

This pattern is pervasive in the scholarly literature on stereotypes, but it comes in two flavors.  Therefore, we have given it two names, for slightly different manifestations of essentially the same idea. The black hole at the bottom of declarations of stereotype inaccuracy refers to the following pattern that almost anyone can see for themselves when reading the scholarly literature on stereotypes. When an article declares a stereotype to be inaccurate, it often provides no scientific citation whatsoever (it’s an evidence-free declaration). It’s just a black hole. Try it. Next time you read a social science article declaring stereotypes to be inaccurate, be alert – do they report original evidence, or, if not, do they cite another article as providing that evidence? If not, then you have found one of the stereotype inaccuracy black holes.

Idea Laundering

The second flavor of claims unhinged from actual evidence is “idea laundering” (a term we did not coin).4 It refers to a process that can create the appearance of scientific legitimacy on the basis of little or no empirical evidence supporting the claim. It works like this: some academic makes some claim with little or even no evidence in an academic journal. It might even be flagged as speculation. Nonetheless, another academic finds the idea useful and cites that article. Now there are two published articles touting the idea. Rinse and repeat for scores or even hundreds or sometimes thousands of articles, and you have a “scholarly consensus” spring up around an entirely speculative or unproven idea giving it the false appearance of “fact” or “knowledge.” Wikipedia calls the same phenomenon the “Woozle effect,” named after Winnie-the-Pooh’s quest to find a non-existent animal by following tracks that he was unaware of having made himself previously.5

Many researchers cite social psychologist Gordon Allport’s book, The Nature of Prejudice (1954) in support of the claim that stereotypes are inaccurate, or, at least, exaggerations of real differences. And Allport did declare that stereotypes exaggerated real differences. But, aside from an anecdote or two, which is hardly scientific evidence, he presented no evidence that they actually did so. Famous psychologists declaring stereotypes inaccurate without a citation or with a citation to a source that itself provided no evidence meant that anyone could do likewise, creating an illusion that pervasive stereotype inaccuracy was ‘settled science.’ Of course, if “settled science” means “most scientists believe this is true,” it is settled science in a social sense, but not because it is actually true. Only if one looked for the empirical research underlying such claims can one discover that there is nothing there.

There is, however, another alternative which, at first glance, might appear to “save” the ability to declare stereotypes to be inaccurate. Perhaps those who make this declaration are not stating an empirical description of the world. Instead, perhaps they are simply defining stereotypes as inaccurate. Anyone can define their terms pretty much how they choose. If a unicorn is defined as a supernatural animal that looks a lot like a horse with a horn, that is not a description of the empirical state of the world (which has no unicorns); it is a definition of the word “unicorn.”

But there are two problems with defining stereotypes as inaccurate. The first is that if all beliefs about groups are stereotypes, and all stereotypes are defined as inaccurate, then all beliefs about groups are inaccurate. It is, however, logically impossible for all beliefs about groups to be inaccurate. This would make it ‘inaccurate’ to believe either that two groups differ or that they do not differ, and both cannot possibly be inaccurate. The idea that ‘all beliefs about groups are stereotypes and all are inaccurate’ can be summarily dismissed as logically incoherent.

There is, however, a second way to define stereotypes as inaccurate that solves this incoherence problem but brings in another.  Perhaps stereotypes are the subset of beliefs that are inaccurate. In this case, only inaccurate beliefs are stereotypes; accurate beliefs about groups may exist, but they are not stereotypes. This solves the incoherence problem that comes from “all beliefs about groups are inaccurate” but produces a new one.

If stereotypes are the subset of beliefs about groups that are inaccurate, before declaring some belief to be a stereotype one would need to first empirically establish that that belief is inaccurate – otherwise, one could not know that it is a stereotype. The logic is inexorable, as can be seen from any example outside of stereotypes. If one declares that Covid is caused by a novel coronavirus, one cannot assume a patient with pneumonia has Covid without testing for Covid (the person might have a bacterial infection or an infection from a different virus). In the case of Covid, you need a Covid test. Similarly, one can define a “fatal transmission failure” as a car failing to move even though the engine is on due to clutch slippage. However, just because a car can’t move does not mean there has been a transmission/clutch failure. One needs to evaluate whether the clutch is actually slipping, because “not moving” can come from several things besides clutch failure. This is all obvious.

Amazingly, what is obvious with infection-testing and car failure all-of-a-sudden ceases to be obvious to many people, including many with PhDs, when it comes to stereotypes, even though the logic is identical. One might think a college education—particularly an advanced, graduate school education—would provide training in logic and inference that would ensure this would be obvious. If one thought this, one would be obviously wrong. If one uses this meaning of stereotype inaccuracy, one needs to show that a particular belief about a group is inaccurate before one can refer to it as a stereotype. Absent evidence of inaccuracy, one cannot know the belief is inaccurate; therefore, one cannot know it is a stereotype.

Vanishingly few social scientists performing research that refers to stereotypes ever present such a test. If this is the definition, most claims about “stereotypes,” including most social science scholarship, would need to be dismissed because almost none of it actually first showed the beliefs in question to be inaccurate. The need to dismiss the work would stem from the fact that whenever a belief is not shown to be inaccurate, it cannot be known to be a stereotype (if one uses this definition). Of course, all of these problems can be easily solved by using a definition that is neutral with respect to inaccuracy. Our definition of “stereotype” is simply “a belief about a group.” It may be right, wrong, partially right, moderately right, or any other combination or complex pattern of right, wrong, close, etc. On the other hand, a neutral definition removes social justice-oriented activists' justification for presuming that stereotypes are inaccurate, and this may be a price many are unwilling to pay for mere logical coherence.

The Evidence on the Accuracy of People’s Beliefs about Groups is Clear

When stereotypes are defined neutrally, accuracy is an empirical question. If one wanted to know the accuracy of the weather forecast, say, with respect to predicting tomorrow’s high temperature, one would:

1. Identify the prediction.

2. Measure tomorrow’s high temperature.

3. Compare the two.

One can do this in many ways. One could literally do it for tomorrow by computing a discrepancy score.  If a person’s prediction is that the high will be 65 degrees, it either is or it is not. We call these personal discrepancies because they assess the accuracy of a single person. But what if it’s 64 degrees? Is this one-degree discrepancy ‘wrong’ or is it ‘close enough’?  This is a matter of individual judgment and probably depends on the context. For most casual purposes, we suspect most people would consider 64 “close enough.”

Another thing one can do is compare predictions across many days; one can correlate a person’s predictions with the daily highs over many “tomorrows.” The higher the correlation, the more that variations in predictions correspond to variations in the actual temperatures. We call these personal correlations because they constitute the correlational accuracy of a single person. One can also assess the accuracy of consensual beliefs about daily high temperatures; these constitute the beliefs held by many people together.

One could ask many people to predict them, and then compute the average prediction. One can then compute a consensual discrepancy score (by comparing the average prediction to the actual temperatures) or a consensual correlation, by correlating the average predictions with actual high temperatures over many “tomorrows.”

Now let’s return to stereotypes. These empirical tests accomplish different things, depending on what one means when one declares stereotypes to be inaccurate.  If this is an empirical claim, then the empirical evidence can falsify or support this claim.  If one defines stereotypes as the subset of beliefs about groups that are inaccurate, then the empirical evidence indicates whether any particular belief is (if it is inaccurate) or is not (if it is accurate) a stereotype.

So what do the data say? Over 50 studies have now been performed assessing the accuracy of people’s beliefs about demographic, national, political, and other groups. The evidence is clear.6 Based on rigorous criteria, laypeople’s beliefs about groups correspond well with what those groups are really like. This correspondence is one of the largest and most replicable effects in all of social psychology. Stereotype accuracy has been obtained and replicated by multiple independent researchers studying different stereotypes and using different methods all over the world. Table 1 is based on our 2016 review of all studies of stereotype accuracy we could find up to that time.7

Table 1 shows that stereotypes are more accurate than most social psychological hypotheses. We can confidently say this, because a large meta-analysis of just about the entire literature in social psychology up until about 2000 found that, across all topics, the average effect size in social psychology was about r=.20.8  As shown in Table 1, only about 24% of effects in social psychology exceed r=.30, and only 5% exceed r=.50.  In contrast, most stereotype accuracy correlations exceed .50, making them some of the largest relationships ever found in social psychology.

For example, way back in 1978, in a study reported in the Journal of Personality and Social Psychology, social psychologists Clark McCauley and Christopher Stitt first obtained U.S. census data comparing African Americans and other Americans on the likelihood of completing high school or college, becoming an unwed mother or unemployed, and on having a family with four or more children.9 They then asked people from various walks of life – college and high school students, union members, a church choir, master’s of social work students, and caseworkers in a social service agency – about their beliefs about the percentages of Americans in general, and African Americans in particular, with these characteristics. People’s estimated percentages were close to the census data, and correlated extremely highly with the actual differences.

People are also quite good at perceiving many gender differences. By the 1990s, several meta-analyses of sex differences had been published. Much as McCauley and Stitt started with U.S. census data, social psychologist Janet Swim, in two studies reported in a 1994 issue of Journal of Personality and Social Psychology, took this meta-analytic research as her starting point – as criteria for real sex differences.10 The meta-analyses showed, for example, that males outperform females on math tests, and are more restless and aggressive, whereas females are more influenced by group pressure and are more skilled at decoding nonverbal cues. She then asked people to estimate the size of these differences; again, people were quite good, and their estimates again were highly correlated with the actual differences. Similar results have been found for all sorts of other stereotypes, including those about ethnic groups, age, occupational groups, college majors, and sororities.

Despite the frequency with which researchers proclaim the inaccuracy of stereotypes, we are regularly accused of debunking a straw man argument when we contest this claim. In reviews of articles submitted for publication, we were regularly told variants of “Modern scientists do not broadly declare stereotypes to be inaccurate.” So we did a small-scale study, a content analysis of how several famous textbooks and influential books treat stereotypes. Results are summarized in Table 2. In our home discipline of social psychology, stereotypes are not always declared inaccurate, but a slew of famous canonical texts either do so or emphasize inaccuracy without ever mentioning accuracy.

Some Exceptions and Qualifications

Inasmuch as over 50 studies of stereotype accuracy have been performed, it is fair to say that, in general, they found that the correlations between people’s stereotypes and criteria were some of the largest relationships ever found in the social sciences, ranging from about .4 to .9 (the average correlation in social psychology is about .2). This is surely because, whatever their limitations, motivations, and biases, people have some sensitivity to actual realities, at least most of the time, and at least if those realities are relatively apparent.

There are, however, two important limitations to this work. First, some stereotypes have been found to have very little accuracy, although they are probably not among the first things that come to mind when most people think about “stereotypes.” Political stereotypes tend to have high correlational accuracy because people have a good general sense of the direction of differences between Democrats and Republicans, or liberals and conservatives. However, they also consistently exaggerate the real differences. This probably occurs for several reasons, but the most crucial ingredient seems to be that people caricature their opponents. Colloquially speaking, the further people are on the left, the more they see those on the right as fascists and Nazis, and the further people are on the right, the more they see those on the left as Marxists and communists. It may not always be quite that extreme, but the point is that people exaggerate real political differences primarily by distorting their opponents as being more extreme than they really are.11

There is one type of stereotype, however, that the bulk of the research shows to be inaccurate – national stereotypes of personality. A slew of studies have administered standard personality inventories to people on every continent, except Antarctica, and those results were then used as the criterion against which to compare people’s beliefs about, for example, the openness or agreeability of people in those countries. The most common finding is that people’s perceptions are almost completely unrelated to the actual personality scores. Studies have focused mainly on assessing levels of accuracy rather than explaining inaccuracy, so we do not really know why this occurs. It does seem plausible, though, that few people have extensive experiences with large swaths of people from other cultures and countries. If stereotypes are, at least sometimes, based on realities, but people have little or no access to those realities, there is no reason to expect much accuracy in such stereotypes.

Last, as we described in the introduction, Americans’ beliefs about differences in wealth between White and Africa Americans wildly underestimate real wealth differences. We suspect that this is mostly because people do not understand that most wealth comes from home ownership.

So, even though most of the research finds stereotypes of the people studied are fairly accurate, there are some notable exceptions. In addition, there are other reasons not to run screaming to the world that “all stereotypes are accurate!” (over and above that not being true to the evidence). All studies are limited. All stereotypes have not been studied. All people’s stereotypes have not been studied. There are no good criteria against which to compare many stereotypes. Therefore, it is possible that, over the coming decades, more and more evidence of inaccurate stereotypes will emerge. Or perhaps a particular belief is one that has never been evaluated for accuracy.

Another important limitation is that some stereotypic beliefs have no accuracy standard. For example, sometimes, people consider role prescriptions to be stereotypes (“children should be seen and not heard”).  What people “should” do is a moral question, and there are no accuracy criteria identifying what the right morals for people to hold are. Therefore, one cannot evaluate the accuracy of prescriptive stereotypes.

People also may hold stereotype beliefs about group differences that have never been studied. Are New Yorkers really louder than other people? We doubt there is any good data on this. Absent such data, it is impossible to evaluate the accuracy of someone who believes New Yorkers are unusually loud. Of course, this cuts two ways. We certainly cannot say such beliefs are accurate, but nor can we declare them to be inaccurate, and any person who does is making an unjustified claim.

Another qualification to this line of work stems from confounding accuracy in perceptions of some characteristic with their explanation for that difference. Let’s say some people believe men are more interested in STEM careers than are women. This is either true or not at any given point of time for any particular sample of men and women. Determining whether this belief is correct provides no information whatsoever about the explanations for this difference or nondifference. This should be obvious. Establishing that something is true is very different from establishing why it might be true.

What About Biases in Evaluating Individuals?

One common form of pushback to this line of argument is that while people’s beliefs about groups may not be completely out of touch with reality, they may inaccurately judge individuals. It is true that, in general, in the absence of lots of detailed and relevant information about a person, people’s stereotypes do bias their judgments. Of course, any Bayesian, unless afraid of political blowback, will tell you that this is completely rational – “priors” (also known as beliefs and expectations) should influence judgments under uncertainty, in the absence of evidence to the contrary. So, the more important question is, ‘When contrary information becomes available, do people rigidly stick to their stereotypes or do they adjust their perceptions and judgments accordingly?’

The answer is clear, though with a bit of nuance. Scores of studies show a consistent pattern: People generally judge others on their merits and the extent to which they do this is very powerful. Reliance on individuating information averages about r=.70, which is also one of the largest effects in social psychology. Judging people on their merits is a pattern so powerful that a 1996 review and meta-analysis by social psychologist Ziva Kunda and cognitive scientist Paul Thagard described those effects as “massive.”12

Of course, just because people almost entirely judge others on their merits does not mean that stereotypes have no biasing effects. This is where some nuance is required. Even in the presence of relevant individuating information, stereotype biases average quite small, often around zero. Table 3 presents the average bias effects found in a slew of meta-analyses of studies of the role of stereotypes in person perception, and is based on data presented in Social Perception and Social Reality: Why Accuracy Dominates Bias and Self-Fulfilling Prophecy (Jussim, 2012). The simple average for the overall effect of stereotypes on person perception, across nearly all the studies of stereotyping that have been performed, is a correlation (between target group label and perceiver judgment of an individual) of .10. Furthermore, that is the simple, unweighted average, so is probably an overestimate, because the correlation of the bias effect with the number of studies included in each meta-analysis shown in Table 3 is -.39. The more studies in the meta-analysis, the smaller the average biasing effect of stereotypes. This suggests the existence of bias in favor of publishing studies demonstrating bias, which declines as more studies get published.

Why biases sometimes do versus do not occur has not been well-established by scientific research (there are lots of different explanations and theories out there that are beyond the scope of this report).  Interestingly, despite the current craze over implicit biases, work in our lab shows essentially the same pattern: even for implicit perceptions (those measured by reaction time measures rather than questionnaires), people rely heavily on individuating information, and far less, or not at all, on their own stereotypes.13

Why Should Anyone Care?

There are several reasons why we think people should care about this state of affairs, ranging from scientific to political to redressing social problems.

The Priority of Truth-Seeking

Our view is that science’s first goal should be to find out things that are actually true. Prioritizing anything else is not science. This may be difficult, and there are often wide swaths of uncertainty surrounding much work on social science, especially on controversial topics.  Therefore, scientists should exercise due caution with respect to canonizing conclusions. Study X may find something, but most likely, we should not be singing our findings to the world as “confirmed scientific fact” until a skeptical community of scientists independent of the original team find the same thing. With all those as qualifiers, our primary goal should still be truth.

In this context, the conclusions currently justified, based on the overwhelming weight of the evidence are that:

• The overwhelming majority of studies of demographic stereotypes find moderate to high accuracy.

• Most studies of national character stereotypes have found them to be inaccurate.

• No general tendency of stereotypes to exaggerate real differences between groups has yet been discovered.

• There is a general tendency, confirmed by multiple independent teams, for political stereotypes to exaggerate real differences.

Distinguishing Accurate from Inaccurate Beliefs

On the one hand, one cannot simply declare all beliefs about groups to be inaccurate. On the other, not all stereotypes are accurate. Therefore, in order, to identify when inaccurate stereotypes may or may not be contributing to social problems, one needs to assess their accuracy. Sometimes, this will produce evidence of inaccurate stereotypes that could contribute to social problems; other times, it will show the stereotypes to be relatively accurate.

Political Dysfunctions in Academia

Demonstrating bias, at least when there are accuracy criteria, is relatively straightforward: One must show that beliefs systematically deviate from the criteria. It is not enough to show simply that people perceive differences between groups or individuals; those differences may be valid. This is completely obvious in many situations outside of stereotypes. And yet, when people perceive differences between groups, this is routinely attributed to “bias” even without bothering to assess for accuracy. That laypeople routinely do this is bad enough, but it’s common among academics as well. Why might this be?

As Jonathan Haidt argued in The Righteous Mind, ideology “binds and blinds.” If the only possible explanation for a group difference one has ever considered is “bias,” well, then, of course it would never occur to one to even consider accuracy. Furthermore, for socially valued characteristics, group differences mean that one group has more than the other. Even if social processes produced those differences, the mere acknowledgment of their existence may be viewed as “blaming the victim” – something taboo in politically left and academic circles. It is so taboo that, even if one is not blaming the victim, many may decide that the risk of social costs of considering non-bias explanations for differences are not worth incurring, often because there is a heightened risk of others misrepresenting one as having blamed the victim.

Policy and Social Problems

Policies are not likely to be effective if they target the wrong problem.  For example, scientists once believed that “bad humors” caused fevers, so they would bleed sick people to cure them. They also once believed that stress caused ulcers, so a whole industry was built around addressing stress (for the uninitiated, Barry Marshall received the Nobel for showing that bacteria, not stress, caused ulcers).

In the same spirit, targeting changing stereotypes to solve inequality and injustice is not likely to be very effective if most of the stereotypes are fairly accurate. For example, we have argued that whatever is captured by measures usually referred to as “implicit bias,” using methods such as the implicit association test, often, in substantial part, reflect social realities.14 This may help explain why interventions designed to change implicit biases have virtually no effect on discriminatory behavior.

In contrast, let’s say some sort of social process caused some sort of unjust inequality. Purely hypothetically, let’s say there are two ethnic groups, A and B. In 2021, although there is little individual-level discrimination, A’s live in rural areas and B’s live in suburban areas. Social and political processes over the prior 100 years have produced better suburban than rural public schools. Thus, B’s do better in school, and are more likely to go to college and get good jobs.

Thus, there clearly is an inequality, it is socially valued and important, and we may consider it to be unjust. The solution here starts with acknowledging, and being able to publicly discuss, the A-B difference in life outcomes without fear of being denounced and blacklisted. But that is just the start. If one leaped to “discrimination” as the source in this hypothetical, one would be wrong because it does not exist.

With that in mind, we make the following recommendations:

1. For those who consider stereotypes to be any beliefs about groups, stop declaring them to be inaccurate.

2. For those who consider stereotypes to be “the subset of beliefs about groups that are inaccurate,” feel free to describe national character stereotypes, and American’s beliefs about wealth differences between Black and White people as inaccurate. There are no other stereotypes that, by this definition, one can currently refer to as inaccurate.

3. No one should be (as have many social scientists) declaring that “stereotypes exaggerate real differences.” This is, itself, wildly exaggerated because, though sometimes stereotypes exaggerate real differences, just as often, or more so, they underestimate real differences.

4. Evaluate claims on their scientific merits and validity, not on their usefulness for advancing political agendas.

Why the Deep Disconnect Between the Evidence and the Canon?

We know that academic perspectives emphasizing stereotype accuracy are wrong. And they are not wrong in random ways. Perspectives relentlessly emphasizing stereotypes are systematically biased in the sense that they ignore evidence of accuracy and rational and unbiased judgments regarding individuals. So having established bias, now we can ask, “Why are the academics so biased?”

Of course, we cannot know for sure, because no study has directly linked researchers’ personal characteristics to the biased conclusions they reach in their scholarship.  Nonetheless, Occam’s razor – keeping it as simple as possible – suggests that the short answer is that many academics study topics to correct what they see as societal ills; that is, they prioritize activism and their view of social justice over truth. Given the massive evidence showing a massive left skew of academia, especially in the social sciences and humanities (the areas most likely to address politicized topics), academic activism is essentially an intellectual form of leftwing activism.

This can be seen in a myriad of ways. CSPI’s new report shows a large portion of academics surveyed in the social sciences and humanities in the U.S., Canada, and the U.K. see themselves as activists, radicals, or both.15 This group is most likely to endorse attempts to sanction colleagues who express dissenting views (e.g., by getting them fired), and discriminating against people whose views they oppose. Although it was not addressed in the report, we speculate that the idea that many stereotypes are accurate would be anathema to this group, and that no amount of evidence could convince them otherwise (we doubt many would even consider the evidence).

Political biases can produce distorted bodies of scholarship through a variety of routes. It impacts who becomes an academic or scientist. The more academia is (rightly) viewed as a hotbed of left-wing activism, the less appealing it may become for scholars who hold different views. Notable percentages of faculty indicate explicit willingness to discriminate against those in their ideological outgroups, and ideological minorities report experiences of hostility, stigmatization, and a lack of belonging. It should not come as a surprise, then, that non-liberals – or, really, anyone whose work seems to contest narratives that are sacred to left and far-left faculty – are overlooked in hiring decisions, or self-select out of pursuing academic careers.

Political bias also manifests in the questions researchers ask, how they measure what they’re studying, and how they interpret findings. Researchers generally have large latitude in the topics they choose to study and investigate. Surely, however, radicals and activists will tend to ask different questions than centrists, libertarians, conservatives, and the apolitical.

Measurement and interpretation may also fall victim to political biases for similar reasons, as concepts and findings are filtered through one’s personal worldview. Ideas that are contested in the wider society (value of affirmative action, the importance of microaggressions), but widely shared by those on the left may become normative and treated as “well-established scientific facts” among those in the bubble that is the academic left, even in the absence of solid or rigorous evidence. As such, asking skeptical questions about or drawing conclusions that conflict with these social norms may be inconvenient because such actions draw the ire of one’s colleagues and may result in condemnation or denunciation. Academia operates as a social-reputational system whereby an individual’s success hinges largely on the favorable evaluations of colleagues. As such, there are strong incentives for doing work that will lead to social approval from others (and, especially, avoiding work that will garner disapproval from peers). If colleagues might reject or vehemently disagree with one’s findings, there may be a strong incentive for self-suppression of those findings out of fear of punishment. If one elects not to self-suppress, one risks being denounced, ostracized and possibly even punished (through forced retraction, deplatforming, or even firing). Although more severe punishments, such as firing, are rare, we suspect that even the public shaming that comes with a forced retraction or deplatforming will often be sufficient to persuade a great many scholars to never again go public with a similar idea or finding.

Furthermore, this problem of self-censorship of findings that make sense in the wider world but are anathema to leftwing academics, especially the large minority of activists, creates further problems downstream from research. That is, if researchers conduct a study that could produce a finding unpopular in academia but which would be welcome elsewhere (e.g., that preferential selection forms of affirmative action are ineffective), the researchers are now in an ethical bind.  They have two choices, both bad: 1. Try to publish the study and risk denunciation, protracted battles to get the paper published and not retracted, and possibly even firing; or 2. Suppress it, which would be scientifically unethical, even though it would be politically safe. There is a simple solution: Do not study such topics. This entirely eliminates the ethical quandary from the researchers’ standpoint but is obviously no way to conduct a society that prefers to evaluate its practices, policies, and interventions based on scientific evidence rather than political popularity.

Finally, research may be subject to political bias in the form of selective citations. The easiest way for researchers to ignore inconvenient or disliked findings is to not cite or discuss them. Bias can occur when work is cited more frequently based on its political content rather than scientific quality. Work that is highly cited will often enter the canon – the rarefied list of findings deemed to be “true with great certainty” and which get highlighted in major reviews and textbooks. Of course, work that is ignored never makes it into the canon. This is science working well when work is ignored because it has been refuted. But when work is ignored without having been refuted, this is science working poorly. This is precisely the case with stereotype accuracy. Scores of studies demonstrate at least moderate accuracy, but it is, to this day, far more common to see major reviews and textbooks repeat the naïve assumption that stereotypes are inaccurate, as if those scores of studies were never conducted.

Paradigmatic of this is a recent review in an outlet of record for psychology that concluded gender stereotypes are mostly inaccurate.16 Reaching this position required ignoring and failing to cite or consider 11 published papers reporting 16 separate studies that found gender stereotypes to range from moderately to highly accurate. Future research may end up refuting these 11 papers on gender stereotypes, and criticism of these papers and their findings should always be on the table—no scientific work should be beyond scrutiny. But, at present, the work has not been refuted.  The relevant research was completely ignored, and a review claiming comprehensiveness and nuance should not be in the business of ignoring work that fails to conform to desired narratives and positions.

Testaments to the inaccuracy of stereotypes still dominate textbooks and reviews of stereotyping. To borrow an expression, it is a dead horse that is still up and trotting around. Many think of ‘science denial’ as something primarily characterizing right-wing disbelief in climate science and evolution. Apparently, however, it is alive and well in social scientists’ own resistance to the overwhelming evidence of accuracy and rationality in many people’s stereotypes.

We are going to end with a true story. A couple of years ago, one of us (Jussim) was having breakfast with some other very famous social psychologists after a conference. I can only report my lived experience. I had given a talk on the value of intellectual diversity in academia, and had briefly mentioned the work on stereotype accuracy. They did not like it, and, like many social psychologists before them, were criticizing it without refuting any of it. As it became clear that, unless they actually could refute my work I was not going to change my views about it, one of them said, and this is a close paraphrase, “But the Nazis relied on stereotypes.” Even though one of the most sure-fire ways to derail a nuanced and evidence-based conversation is to throw in “But Nazis!,” it’s not a completely ridiculous point; the Nazis did advance vicious propaganda that so dehumanized Jews that it paved the way for genocide. Indeed, as I and my collaborators reported in a 2020 study, crude images of Jews and slick antisemitic conspiracy theories can still be found in some of the more extreme networks that fester in nastier corners of social media.17

Recognizing how propaganda can promote and exploit pernicious stereotypes is important, and we are glad people are outraged by it. But what goes on in bleak corners of 4chan or Parler is, however disturbing, not representative of what the social science has discovered about what most people who have been studied think about most groups they have been asked about.

If people estimate the average daily high temperature in March in New Jersey, Nazis are not involved in figuring out whether they are right or wrong. Similarly, if people estimate the proportion of Black adults with college degrees, the standardized test scores of Jews, or the yearly income of male and female doctors, one can figure out how accurate they are by comparing their estimates to criteria.

Which gets us back to the evidence. If the overwhelming evidence does not change someone’s belief in the so-called ‘inaccuracy’ of stereotypes, what could? A foundational premise of the sciences is that they self-correct in the face of new evidence. Nothing in this essay should dissuade anyone from continuing efforts to combat discrimination, disinformation, and propaganda. We hope, however, that, with respect to the longstanding claim that ‘stereotypes are inaccurate,’ those who are willing to consider the actual evidence might be persuaded that a little scientific self-correction is long overdue.

References

• This study is the first of its kind to investigate authoritarianism and political discrimination in academia, relying on survey responses from both the perpetrators and targets of discrimination.

• Across three Anglophone countries, a significant portion of academics discriminate against conservatives in hiring, promotion, grants and publications. Over 4 in 10 US and Canadian academics would not hire a Trump supporter, and 1 in 3 British academics would not hire a Brexit supporter.

• Gender-critical feminist scholars appear to experience even more discrimination than conservatives. Only 28% of American and Canadian academics would feel comfortable having lunch with someone who opposes the idea of transwomen accessing women’s shelters.

• Most professors do not back cancel culture in its most authoritarian forms. Only 1 in 10 academics supports firing controversial professors. Nonetheless, while most do not back cancellation, many are not opposed to it, remaining non-committal.

• Right-leaning academics experience a high level of institutional authoritarianism and peer pressure. In the US, over a third of conservative academics and PhD students have been threatened with disciplinary action for their views while 70% of conservative academics report a hostile departmental climate for their beliefs.

• In the social sciences and humanities, over 9 in 10 Trump-supporting academics and 8 in 10 Brexit-supporting academics say they would not feel comfortable expressing their views to a colleague. More than half of North American and British conservative academics admit self-censoring in research and teaching.

• Younger academics and PhD students, especially in the United States, are significantly more willing than older academics to support dismissing controversial scholars from their posts, indicating that the problem of progressive authoritarianism is likely to get worse in the coming years.

• A hostile climate plays a part in deterring conservative graduate students from pursuing careers in academia. Conservative and liberal graduate students differ far more in their perceptions of whether their politics fit academia than they do on questions related to how well academia pays, the isolating nature of the work, and other aspects of the profession.

• One policy option is for government to proactively apply the law to universities, instituting sanctions for institutions that repeatedly breach individuals’ academic freedom while opening up a means for plaintiffs to appeal around their universities to a regulatory ombudsman. While this report makes no policy recommendations, this approach has been largely adopted by the British government.

For the executive summary, click here.

Click "download" for a PDF

4.15MB ∙ PDF file

Download

Download

Introduction

This report represents the most comprehensive survey-based investigation to date of academic and graduate student opinion on political discrimination, the punishment of academics for speech, and experiences of hostility and self-censorship for political beliefs. It examines evidence from the perspective of both victims, largely concentrated among the minority of 5-10% conservative or gender-critical academics; and perpetrators, involving a substantial minority or sometimes even a majority of scholars. The report is divided into four parts.

Part I concentrates on what I term hard authoritarianism, notably the experience of being disciplined or threatened for speech – or backing the firing or disciplining of controversial academics; as well as support for policies such as mandatory reading list quotas that abridge academic freedom. Part II examines soft authoritarianism, encompassing views on political discrimination, as well as experiences of hostility and self-censorship in scholarship, teaching, and other aspects of academic life. This part replicates the findings of many existing studies but also uses new methods to go beyond them. Part II is subdivided into three sections, the first on the political leanings of academics, the second on the extent of chilling effects and self-censorship, and the third on support for political discrimination.

Part III goes off campus to ask whether the patterns of hard and soft authoritarianism we find among academic staff and graduate students are also present among college graduates working in organizations off campus. We’ll see that there are differences, but also many similarities, between the two professional realms. Part III also discusses potential policy reforms to address the problems highlighted, comparing and contrasting interventionist approaches against those that are more hands-off.

The report is meant to be read as an integrated whole but can also be accessed in a modular way by clicking on a heading in the table of contents. The reader can return to the table of contents at any time by clicking on any heading in the text. The same is true for the table of contents of tables and figures.

The Surveys

This study relies largely on survey data of academics and PhD students at universities in the United States, Britain, and Canada. It also summarizes prior work on staff and students in the United States and Europe and contextualizes findings with surveys of the general population and the wider advanced degree-holding public. The core of the study concentrates on academic attitudes, where there has been less research than on student attitudes.

In order to mitigate the effects of social desirability bias – people giving socially acceptable responses – I use a list experiment.1 This is an innovation in work on political discrimination, and gives a more accurate indication of how prevalent the willingness to discriminate on political grounds is. In the analysis, I focus more on the social sciences and humanities because political considerations are a larger aspect of these fields’ conceptual foundation, and are thus assumed to exert greater influence over the culture and practices of these disciplines.

The UK data comprise the highest quality sector-wide academic survey to date, based on a 61-76% response rate from the approximately 1100 current and retired academics that happen to be on YouGov’s 500,000 strong national British panel. Since these are respondents who are in the YouGov system answering other surveys, there is no danger that they are self-selecting into the survey due to their interest in answering questions on academic issues. US and Canadian surveys represent a mix of online surveys using convenience samples with low response rates from a large target pool, and platform-based surveys with very high response rates from smaller target populations. I also fielded a UK online mailout survey to academics in the top 100 (out of 143) universities to compare with the relatively randomly-selected YouGov UK sample and the online North American surveys.

Though response rates to the North American and British mailout surveys are only around 2-4%, the data they provide tell a very similar story to the non-selective YouGov and Prolific surveys (with 61-87% response rates), and to previous questionnaires fielded by other scholars. The result is a triangulated body of replicated knowledge on political discrimination and chilling effects that has not been contradicted and thus should no longer be seriously disputed.

In addition to replicating previous studies of soft authoritarianism (political discrimination and chilling effects), I break new ground by examining support for hard authoritarianism – in this case dismissal campaigns for controversial academics, and penalties for failing to implement mandatory curriculum diversity quotas. I also use the surveys to create quantitative models that, in a departure from existing studies, examine the relationships between the ideological skew of staff, political discrimination, self-censorship, and support for cancelling controversial scholars or employees.

Table 1 shows the methods and population sampled in each survey discussed in this report. More details of the surveys may be found in the Appendix.

Part I: Hard Authoritarianism

As we shall see, there is a small minority of academics that supports ousting those with controversial views. Administrators are often willing to use the university’s disciplinary apparatus to enforce this sentiment, and it is not clear that academics are sufficiently motivated to resist this. While it is rare for an academic to be fired, especially on ideological grounds, a number of worrying cases have recently come to light in which life has been made so uncomfortable for a person that a dissident scholar has been forced to leave.

No-Platforming and Firing

This forms part of a growing climate of political intolerance. In the United States, the Foundation for Individual Rights in Education (FIRE) maintains a database of disinvitation incidents. These are presented below in Figure 1. Incidents rose steadily in the 2000s, driven mainly by left activists, and spiked in 2015 when Nicholas Christakis was mobbed by a group of Yale students and Jonathan Haidt and Greg Lukianoff published their seminal piece on the phenomenon in the Atlantic, “The Coddling of the American Mind.” Since then, incidents have remained at an elevated level compared to the pre-2010 period.

The National Association of Scholars (NAS) maintains a database of (mainly) American academics who have experienced campaigns calling for their dismissal. The database records 4 incidents apiece in 2015 and 2016, 9 in 2017, 13 in 2018, 12 in 2019, and a striking 65 in 2020. The database relies partly on reports from users, thus it is unclear whether the surge in 2020 reflects a disconcerting new rise in intolerance or the prevalence of real-time over historic reporting since the site was established in June 2020. Regardless, 65 cancellations in one year suggests the problem cannot be dismissed as purely anecdotal.2 Meanwhile, under the radar, in just one 3-week period in the summer of 2020, John McWhorter of Columbia University received 150 messages from academics sharing their anxieties about their work climate, and he continues to receive at least one per week.3

In Britain we see a similar, if slightly delayed, pattern for deplatforming and dismissals. The trend in combined free speech incidents, as catalogued in two separate lists, with a minority of overlapping cases, is presented in Figure 2 and shows a noticeable rise from 2013, and especially from 2018. Note that 2020 data are incomplete, but already show a higher level of incidents compared to 2017.

A report by Civitas in December 2020 discovered that over half (53%) of all 137 British universities experienced demands for censoring speech around alleged “transphobic” episodes during 2017-2020. Many UK universities were also targeted by activists seeking to suppress other forms of free speech. Over half (55%) of universities had at least one open-letter campaign, 37% reported social media drives, and nearly a quarter bore the brunt of a campus pressure group drive seeking to curtail academic freedom or free speech. These forms of political intolerance thereby encompassed a majority of British universities.4

Many other cases occur beneath the radar: in a one-month period from  June 13 to July 14, 2020, the UK’s Free Speech Union received 10 calls for assistance from academics or contractual staff asking for help to defend against online campaigns or university disciplinary charges brought against them for public statements. None are listed in the data compiled by Wanstall or Biggs. A call for evidence on Twitter in mid-2019 by Professor Kathleen Stock of the University of Sussex turned up 28 testimonies of restrictions, threats, and harassment from British gender-critical academics in just 7 days.5

While successful dismissals are rare in academia, there have been a number of notable cases, including those of Bo Winegard of Marietta State College and Noah Carl and Jordan Peterson at Cambridge University. In Britain, cases include Chris Hill, formerly at the University of Central Lancashire in July 2018, and Andrew Dunn, formerly a Social Policy lecturer at the University of Lincoln.6 Peterson’s case involved guilt by association with an Islamophobic fan whom Peterson did not realize to be anti-Muslim. The Hill case involves remarks some found offensive, a view upheld by the university. Hill claims that accusers were misrepresenting his views out of hostility to his conservatism. Likewise, the Dunn case involved little more than anti-conservative discrimination leading to the lecturer being forced out. 

There are also an unspecified number of disciplinary actions that have not been reported by the media, many of which are subject to privacy rules, but some of which are alluded to in comments I collected in my NAS survey, such as:

“One professor at East Georgia State College was fired for ‘sexual harassment’ for disagreeing with the school’s sexual harassment policy. I have heard many stories from credible sources of [sociology] professors, students and graduate students being subjected to secret title IX trials and summarily had their lives ruined by their respective schools. Kansas State University being the worst offender.” –Politically right-wing sociologist, US

The Iceberg Model of Discrimination

            We can think of the threats to academic freedom using the metaphor of an iceberg, with items that make the news – such as deplatformings and dismissals – as the visible symptoms of a much deeper problem. Figure 3 and Figure 4 present some of the main findings of our survey, from the perspectives of the victims and the perpetrators of discrimination.

As can be seen, deplatformings or dismissals affect only a tiny minority of academics. Yet when one polls academics, particularly centrists or conservatives, on their experiences, we find that there are some forms of political discrimination that are experienced by a majority of political minorities on the faculty. By and large, most right-leaning academics feel a hostile environment or self-censor in some form or another.

Surveys of the rest of academia show that conservative scholars are not imagining these things. Only a small minority of academics is willing to endorse dismissal campaigns. Nonetheless, depending on the question asked and the way that it is asked (i.e., whether a “list” method is used to get around social desirability bias) a substantial minority or an overwhelming majority engages in discrimination. Among PhD students, the results are even more extreme than they are among other academics, with over 80% of American PhDs willing to discriminate against right-leaning scholars on at least one dimension.

Disciplinary Action and Bullying

In July 2020, a high-profile statement from 150 leading liberal writers and academics, including J.K. Rowling, Noam Chomsky, and Salman Rushdie, warned of the rise of a “cancel culture” in academia, the arts, and the media. “It is now all too common to hear calls for swift and severe retribution,” they write, “in response to perceived transgressions of speech and thought.” Noting the “hasty” decisions of administrators to fire transgressors, they continue that “We are already paying the price in greater risk aversion among writers, artists, and journalists who fear for their livelihoods if they depart from the consensus, or even lack sufficient zeal in agreement.”7

However worrying the trend in no-platformings and dismissals, this is merely the most visible manifestation of a deeper and more pervasive problem in academia that rarely makes the headlines. The most authoritarian aspect of academia that flies under the radar is disciplinary action and ideological bullying, much of which stem from fellow academics acting in the role of departmental manager or peer. A study for the Universities and Colleges Union (UCU), the main union for academics in Britain, surveyed several thousand British and European academics about their experiences of being disciplined for speech.8 The authors asked, “Have you ever been subjected to informal or formal disciplinary action, or the threat of disciplinary action (up to, and including, dismissal) because of academic views expressed in [the following]?” Questions were asked about “academic views expressed” in teaching, research, within institutions, in public fora, and elsewhere. Results appear in Figure 5.

While I am unable to obtain the UCU study’s crosstabulations by scholars’ ideology, I can compare their UK and EU data to American and Canadian survey results I have collected on my online mailout survey using the same UCU question wordings. These suggest that right-leaning academics are more likely to mention that they have been the target of disciplinary action for their public statements, research, or teaching.

My US online survey results show that right-leaning academics experienced 16-100% more academic discipline than colleagues on the left on 5 of 6 measures. I did not put these questions to my UK academic samples, but according to my Prolific PhD samples (with 72% of PhDs eligible taking the survey), right-leaning US and Canadian PhD students report 3 to 6 times the level of disciplinary action as PhD students with centrist or leftist views. Thirty-eight percent of conservatives were disciplined, or threatened with discipline, for their views in at least one of the following five spheres: teaching, research, speech in private or public forums, or speech elsewhere.

The Prolific North American PhD survey and mailed out North American academic survey contain only small samples of right-leaning respondents (N= 40 PhDs and 43 academics). To rectify this, I ran a survey of the membership of the US-based National Association of Scholars (NAS). Two-thirds report a right-wing identity, and 13-28% said they experienced disciplinary action on any given dimension.9 Results show that among this group, there is a substantially elevated level of victimization, with 43% reporting being disciplined, or threatened with discipline, for their views in one of the five spheres listed above.

These results suggest that the disciplinary threat level is higher for right-leaning academics and especially doctoral students. It helps explain why I find that right-leaning UK academics are 3-4 times as likely as those of other political ideologies to report a “hostile climate” for their beliefs and over twice as likely to say that they self-censor. I estimate that right-leaning UK academics and PhD students are 50-100% more likely to experience disciplinary threats than leftist and centrist academics.

Across a broader range of disciplinary measures in Figure 6, we see the right-leaning membership of the NAS generally reporting 2-3 times as much authoritarianism as the overwhelmingly left-leaning respondents in the Karran and Mallinson EU/UCU survey.

Approximately 6-9% of NAS members experienced the most severe forms of treatment for their views, including being moved to another department or center (6%), demoted (8%), being physically attacked (8%), or having research or facilities withdrawn (9%). A quarter of NAS respondents reported being falsely charged or threatened with charges for their views, and nearly a quarter were given more, fewer, or different research, administrative, or teaching duties in response to things they said or wrote. More generally, around a quarter were disciplined or threatened for views expressed in teaching and public or university forums.

Finally, colleagues were the source of most of the problems reported by respondents in all datasets. Half of NAS respondents reported psychological pressure from colleagues for their views and 36% said they experienced bullying for expressing them. The connection between colleagues, departmental authority, and higher-level administration is important. Departmental colleagues wield considerable power, both through informal peer pressure and via their formal departmental managerial roles as well as through faculty and college-level committees that administer the university.

Hard Authoritarianism Testimonials

Some reported being targeted by the university’s disciplinary apparatus. The following are testimonials showing cases of hard authoritarianism.

“The university is right now considering disciplinary proceedings because of innocuous items I posted on my private blog. I think I’ll be alright, but conservative academics are somewhat persecuted for their views.” – Right, Geography, US

“Only recently has the attitude of the administration turned to punitive measures for people who are outspoken; thus most of us are being very careful about what we say and make sure we are covered by our collective agreement to fend off the administration.” – Centrist, Classics, Canada

“Yes, every year I am called into my boss’s office because of some controversy surrounding course curricula. At times, I have been misunderstood. Other times, I have to justify my work with students.” – Right, Education, US

“I have been dismissed as undergraduate programs chair in my department because I am a gender critical feminist. Students currently have a petition underway supporting this dismissal because my faculty association is supporting me in a grievance about it. I am also facing an interestingly timed investigation into my research ethics, supposedly ordinary procedure any time a faculty member is featured in the news but the first time I’ve been investigated.” – Very Left, Anthropology, Canada

“A professor who declined to refer to a student by a false pronoun (a male wanted to be called ‘she’) was visited by the diversity people and had all of his work and his lab scrutinized. He was placed under specific administrative guidelines for how he was to deal with this student.”  – Right, Communications, US

“Professor Dennis Gouws of the Humanities Department has been subjected to years of harassment, denial of sabbatical, placement on official probation, threats of dismissal, and finally administrative decree commanding him to stay silent or be fired (which he was forced to obey). The reason was his written and spoken criticism of gender feminism.” – Right, History, US

“A friend of mine at another university got a negative tenure vote despite a strong record for having unpopular opinions (and research findings). Fortunately the provost overruled. One of his collaborators at a third school was subject to intense legal harassment (FOIA requests etc) though this was *mostly* not internal to his university.” – Right, Sociology, US

“I had heard from a peer that one of their close friends had been in the program the year prior to ours, and halfway through his department head sat him down and told him he ‘wasn’t cut out to be a teacher,’ and that he was being cut from the program.  My peer believed it was because he expressed strong conservative views in class, perhaps not very diplomatically or respectfully.” – Right, Education, US

Several Leave-voting and conservative academics in my UK YouGov survey report instances of authoritarianism from administrators or colleagues, impact on their freedom to teach and research, and even threats to their jobs.

“Yes, indeed I have lost two senior jobs because I voted leave.” – Tory Leaver (Supporter of Brexit)

“I have been called in for a meeting with University marketing, my Head of Department, and an HR officer after I published an article in a peer-reviewed academic journal [redacted]. They asked why I had not explicitly condemned conservatism as immoral within this article. I explained that I did not believe it was appropriate for me to use my position as a researcher to subjectively pass judgement on modern political ideology. I was told that there are some subjects I shouldn’t remain neutral on, and that I have a moral duty to condemn those on the political right. I was told that, if I insisted on remaining impartial within my research, I was not to further research this subject and warned I may face disciplinary hearings if I did.” – Tory Leaver

“Given the derogatory views regularly expressed by my colleagues about Leave supporters, including the VC sending a University-wide email referring to us as ‘Little Englanders’, I have no doubt that if my views were known then it would negatively affect the attitude of my colleagues towards me significantly. It probably wouldn’t be career-ending, but it would reduce my influence, make it harder for me to deliver my teaching and leadership responsibilities, and quite likely force me into a position where I would have to move institutions.” – Tory Leaver

“I have to be careful about where I place my research because of two different areas of my research. Colleagues have attempted to stop me teaching.” – Tory Remainer

“I had a professor who was very left wing and who reduced the promotional chances of anyone that was centrist or slightly right of centre.” – Tory Remainer

“A previous line manager had a large photo of Jeremy Corbyn on her desk. When I failed to approve (I said nothing) she had me removed from the programme despite very positive feedback.” – Centrist Remainer

“Yes – I avoid making political statements – have also had head of department voice strong disapproval for the sort of research I do and to use the ethics approval system to prevent certain research topics being studied.” – Left Labour Leaver

These comments go beyond the chilling effect of social opprobrium from colleagues and identify the potential for abuse of power by those in the hierarchy within the university. At all levels, the institution can deploy sticks such as assigning unpopular administrative tasks or courses, limiting access to grant funding or promotion, or even forcing a staff member to leave.

The latter is especially damaging, not just for those who lose their jobs, but for the signal it sends across the sector. Academic jobs are incredibly competitive and specialized. A lecturer or professor on a permanent contract who is dismissed or forced to leave on a package is unlikely to find a permanent academic job in the same geographic area, especially if they are older, in a social science or humanities field, and have acquired a reputation as conservative. The skill sets and networks of academics may also make it difficult to make a lateral move to another profession mid-career. All of this instills a powerful sense of caution in many full-time academics, with a desire not to jeopardize a fulfilling and reasonably (though not highly) paid career. This leads them to self-censor, further constricting viewpoint diversity. This impairs the exchange of ideas that is vital for both the academic enterprise, and for beginning to intelligently negotiate between advocates on both sides of society’s major political divide.

Left-Wing Authoritarianism

Hard authoritarianism on campus is often spearheaded by a left-wing authoritarian minority. Recent work in social psychology identifies three distinct clusters within the umbrella concept of left-wing authoritarianism. The first revolves around economic orientations, with high agreement with phrases such as “The rich should be stripped of their belongings and status.” The second focuses on culture, notably “Deep down, just about all conservatives are racist, sexist, and homophobic.” A third is focused on authority, i.e., “I must line up behind strong leaders who have the will to stamp out prejudice and intolerance.” The last two are especially germane to my analysis. The underlying orientation is a preference for using power or violence to upend the current moral order and replace it with a new regime that compels people to conform to new principles. This orientation is characterized by moral absolutism, intolerance of dissent, and a preference for censorship, exhibiting considerable overlap on many psychological measures with right-wing authoritarianism.10

Testing Support for “Cancel Culture” Among Academics

How does left-wing authoritarianism manifest itself within the professoriate? Having surveyed the hardest forms of internal discipline in academia, I next move to examine the level of backing for “cancel”-style dismissal measures. To do so, I probe responses to controversial research findings adapted from actual cases, but worded in an abstract enough way that individuals are unlikely to make a connection to a particular high-profile incident. While most people do not endorse these research statements, and expect that most academics feel likewise, I sought to use them as a litmus test, to empirically measure the extent to which scholars may disagree with the substance of findings but agree that those undertaking unpopular research should not be driven from their posts. The questions I selected are as follows:

1. If a staff member in your institution did research showing that greater ethnic diversity leads to increased societal tension and poorer social outcomes, would you support or oppose efforts by students/the administration to let the staff member know that they should find work elsewhere? [Support, oppose, neither support nor oppose, don’t know]

2. If a staff member in your institution did research showing that the British empire did more good than harm, would you support or oppose efforts by students/the administration to let the staff member know that they should find work elsewhere? [Support, oppose, neither support nor oppose, don’t know]

3. If a staff member in your institution did research showing that children do better when brought up by two biological parents than by single or adoptive parents, would you support or oppose efforts by students/the administration to let the staff member know that they should find work elsewhere? [Support, oppose, neither support nor oppose, don’t know]

4. Please imagine a member of your organization has done work showing that having a higher share of women and ethnic minorities in organizations correlates with reduced organizational performance. Several thousand professionals, some from your organization, have signed an open letter calling for the staff member to be fired in order to protect disadvantaged groups from a hostile learning environment. A small group have started a counter-petition defending the staff member on grounds of academic freedom. Would you: [a) Sign the open letter, which called for the staff member to be fired, b) Support the views expressed in the open letter, but choose not to sign it, c) Not support nor sign either letter, d) Support the counter-petition, but choose not to sign it, e) Sign the counter-petition, f) Don’t know.]

Results are displayed in Figure 7. These show American faculty’s willingness to support a “cancel” campaign to dismiss an academic on the one hand, or, on the other hand, to oppose it publicly or privately. Findings reveal an important reservoir of support for academic freedom among staff at US universities, with just 7-18% of lecturers and professors, depending on the issue, willing to back campaigns to fire academics who dissent from norms on hot-button issues.

As a comparison, I include a question on whether academics would support or oppose firing a member of staff who wants immigration to be reduced. Eight percent of US faculty would favor the dismissal of a restrictionist academic, but 78% would oppose this, with 14% uncertain. Most of those surveyed would undoubtedly take a progressive position on these five questions, but most also recoil from campaigns to remove academics who adopt a dissenting view on them in their research.

The flip side of this largely positive portrait, of course, is that there is a 7-18% minority of American faculty who would support dismissal campaigns that directly violate academic freedom. In addition, the share of academics who oppose a given dismissal campaign is not higher than 52% for any issue other than the immigration question, and falls to just 31% in the example of a study finding that a higher share of women and minorities lowers organizational performance. Many are noncommittal, including about half of those on the question just cited – unwilling to cancel but also unwilling to oppose those who would seek to do so.

British YouGov data, in Figure 8, show very similar findings. Just 6-13% support dismissal across the four campaigns, but no more than 51% would oppose a dismissal campaign. Eighty-three percent of British academics would oppose firing a member of staff who favored reducing immigration, slightly higher than in the US data. For the British mailed survey, the numbers are similar: 4-18% favoring dismissal, but with a maximum of only 54% who would oppose these actions.

As in America, the softest opposition to dismissal was in the “women and minorities lower performance” case, with just 27% willing to oppose a dismissal campaign and 6 in 10 unsure. The British mailout online survey showed results nearly identical to those in the YouGov survey.

These findings replicate with PhD students: in the UK, just 25% would oppose dismissal in the “diversity lowers performance” case, but 65% would in the immigration restrictionist case and 31-34% would for the diversity, empire, and parenthood questions. For North American PhDs the analogous numbers are similar: a mere 19% opposing cancellation in the diversity-organizational performance case, 50% opposing dismissal for an academic who backs immigration restriction, and 29-50% against firing on the diversity, empire, and parenthood hypotheticals.

PhD students on both sides of the Atlantic are noticeably more likely to endorse dismissal than academics: in the UK, 9-10% of PhD candidates back dismissal for immigration restrictionists and those doing research on the parenthood, diversity, and empire questions, rising to 27% for the diversity-organizational question. In North America, fully 25% of PhD students would seek to force an immigration restrictionist academic from their job. On three other scenarios, between 11 and 17% back dismissal, and for the diversity-organizational case, fully 41% back cancellation. These figures are worrying inasmuch as they concern the generation entering academia. It seems that acquiring a post generates a new appreciation for job security, countering these impulses somewhat, but, as we shall see, younger academics are more politically intolerant than their elders, portending a rising illiberalism problem.

The general pattern, therefore, is that the majority of academics is foursquare against cancellation only in the case of an immigration restrictionist academic. For the hypothetical case of a scholar finding that race and gender diversity reduces organizational performance, only 19-31% of academics or PhDs across all five surveys said they would oppose, publicly or privately, an open-letter campaign to get the academic fired even as few actively endorse a cancel campaign.

Who Backs Dismissal?

Compiling responses to the first four binary questions (all except “restrict immigration”) in Figures 7 and 8 above into an index allows us to ask about the characteristics of people who support illiberalism when it advances the cause of perceived racial and gender justice, broadly construed. Combining the four questions, each with answers as 0 for anti-dismissal or 1 for pro-dismissal, into one measure (also scored as 0 or 1) allows me to develop an approximate “likelihood to expel” score for each person in the surveys.11

Using this measure, an average of 20% of current UK faculty, rising to 25% in the social sciences and humanities, back at least one dismissal campaign. On the other hand, just 5% of current British academics in the social sciences and humanities back all four illiberal measures.

Comparing across the US, Britain, and Canada in Figure 9, including academics and PhD students, shows that a majority of both staff and PhD students do not support dismissal campaigns on hot-button issues around race, gender, and sexuality. Indeed, only 2% of American academics and PhD students support all four dismissal campaigns.

However, fully half of North American social sciences and humanities (SSH) PhD students backed at least one illiberal measure, as did 36% of UK PhDs and 20-25% of academics. PhDs are more authoritarian than academic staff. Is this because they don’t have a job to worry about losing? Perhaps. But, as we shall see, another (less powerful) reason is because they are younger than professors. Age plays an important role among both professors and PhD students in predicting their willingness to cancel an academic for transgressing progressive values.

Figure 9 above shows that 24% of American SSH academics, 20% of Canadian SSH academics and 25% of current British SSH staff endorse at least one of four dismissal actions – even though only 1-2% back all of them. As a rough guide, therefore, around 1 in 10 SSH academics back any single dismissal campaign and close to a quarter support at least one of the four hypothetical cases offered. Against this, between a quarter and a half oppose a given dismissal campaign.

Yet in almost all cases there is a large undecided group. Around half are unsure in 3 of 5 cases – they neither support dismissal nor oppose it. By contrast, only 1 in 5 are unsure on the immigration restrictionist case while more than half are unsure on the diversity and organizational performance case. The unsure group is often the silent majority among staff and PhDs when it comes to the more controversial cases. I would argue that this lack of certainty reflects a cross-pressuring between a defense of academic freedom and job tenure on the one hand, and a desire to protect disadvantaged identity groups on the other. For instance, as we shall see, most British academics diverge substantially from the wider UK public by overwhelmingly (76%) supporting the notion that the virtues of political correctness in protecting minorities outweigh its threat to free speech.

Far Left Activists more likely to Back Dismissal

Far-left and activist scholars are a substantial minority of SSH academics in North America and Britain. Figure 10 shows that far-left scholars who agree, to a greater or lesser extent, with the statement “I would consider myself as an activist” make up between 8 and 18% of SSH staff across the US, Britain, and Canada, with far leftists at 16-28% and self-described activists at 26-38%. The British YouGov sample, which is arguably the most representative, shows a lower share of far-left activists than the UK mailout survey. On the other hand, the PhD surveys, which captured 62-85% of PhD students on Prolific Academic, found that 35% of 124 North American SSH PhDs identified as far left and 44% as activists. In Britain, the 77 SSH PhDs were less likely to self-identify in these ways, with 22% far left and 19% activist.

Ideology and Intolerance

Illiberalism is much higher among academics who identify as far left, especially if activist. Far-left activists are often in the forefront of “social justice” activism within the university, and many challenges to academic freedom stem from this group.

Thirty-seven percent of US academics, 40% of Canadians, and 40% of British faculty who identify as far left favor at least one firing campaign. Among far-left activists, 40% of US, 51% of Canadian, and 41% of British academics back at least one dismissal campaign.

Are Young Scholars Less Tolerant?

Even among far-left activists, it is noteworthy that a bare majority would not support forcing dissenting voices from the academy in any given case. The UK data in Figure 11 show how support for dismissal rises as we move from all current and retired staff surveyed, to current staff only, to current SSH staff, and then to far-left and far-left activist faculty.

Is age still important when we take a person’s ideology into account? Maybe young people are just more leftist, so the two factors are confounded? To answer this question, I construct a model, with willingness to dismiss (0 for no, 1 for yes) on any of the four questions as my outcome measure. The results for Britain are shown in Figure 12.

Ideology and age are the main predictors of a scholar’s support for dismissal and have separate and important effects. Left-right ideology is the most important predictor of viewpoint intolerance on the four aforementioned items in Britain, but only “very left” (when contrasted to centrist) is statistically significant in picking out the intolerant academics in the British YouGov sample, with those on the moderate left not more likely to favor hard authoritarianism than those in the center.12 The average likelihood of an academic endorsing a dismissal campaign rises 50 points as we move from a “very right” academic to someone on the far left.

But age also matters. Younger staff, regardless of whether they are far left or moderate, tend to be more illiberal than older staff. This is concerning, as it points to the possibility that the potential for authoritarian activism is likely to rise in the years ahead. Much turns on whether this reflects the fact that younger academics prioritize social justice over academic freedom more than senior staff; or whether this is a maturity issue, with younger staff likely to outgrow their relative illiberalism as they age.

Among the other characteristics that one might assume are associated with favoring speech restrictions in the name of sensitivity are being female and nonwhite. However, in the model in Figure 12 above, neither reached conventional statistical significance. Ethnic minorities were slightly more favorable to removing academics that perform dissenting research on racially-sensitive issue topics when I disaggregated the outcome to focus only on the race-related issues. Being female was not a statistically significant predictor of opinion on any campaign. SSH academics were not more illiberal once I controlled for ideology. This is because SSH disciplines are more far left than STEM, and when this is screened out, the effect of SSH falls away. British academics who share their opinions on Twitter are also somewhat more likely to back expulsion, though this is only significant in some models.

The situation in North America in Figure 13 shows a very similar pattern, albeit with some shifting of the predictors and a stronger overall model. Far leftism is again a powerful predictor of illiberalism, along with age. Younger staff are consistently more likely to favor dismissal – indeed this is the most powerful predictor in the North American model. Interestingly, for North America, SSH academics are again no more illiberal than their STEM colleagues when you control for the fact that SSH has more far-leftist members of staff.

Activist and moderate-left academics are also more supportive of dismissal in North America than centrist staff, whereas in Britain this was not significant. These findings reflect the fact that centrist academics in North America express sentiments (such as feeling censored) that are more similar to conservatives whereas in Britain centrists report low self-censorship and seem closer to moderate leftists in their appraisal of questions of academic freedom.

In addition, women and minorities are more illiberal than men and whites in North America, whereas these characteristics did not predict higher levels of support for dismissal in Britain. The US findings for minorities are accounted for by a somewhat higher effect among African-American respondents (N=36), although black academics were not more illiberal than others in Canada or Britain, granted with very small numbers being polled. Finally, there was no significant difference between American and Canadian academics in their support for dismissal campaigns.13

Concentrating on the main drivers of illiberalism in the British YouGov data results in the chart in Figure 14. This shows that being far left increases the chance of favoring a dismissal campaign by about 25 points. Age matters as well: between age 30 and 70, the chance of backing the ouster of an academic for politically incorrect research decreases by around 15 points. Together, they exert a powerful effect. Thus a 30-year-old far-left British academic has a 1 in 2 chance of endorsing one of the dismissal campaigns, while a 70-year-old who is not far left has barely a 1 in 10 chance of doing so.

Among both staff and PhD students, there is a significant age gradation, with younger staff more intolerant. Are PhD students more pro-cancellation than academics? Figure 15 compares American PhD students and academics in the 23-33 age range, which contains 8 in 10 PhD students in my sample. There is also an important number of academics in this age bracket in my academic survey, resulting in a combined dataset for this age group consisting of 60% PhD students and 40% academics.

Results of a model that controls for gender, race, and age (PhD students contain more women and minorities and are younger) show that PhD students are more intolerant than academics, even controlling for age. However, the difference is concentrated among STEM PhDs. Among scholars in the social sciences and humanities, there is no detectable difference between PhDs and academics (this is confirmed in a model that excludes STEM participants). In addition, combining American, Canadian, and British respondents weakens the difference between academics and PhD students in Figure 15, leaving it statistically insignificant.

To examine the raw effect of studying for a PhD as compared to holding an academic position, where a person may feel the need to protect against risk to job loss, I run a model in Figure 16 looking at all age groups using a combined sample of 2 North American and 3 British surveys, combining PhD students and academics. Age is the strongest predictor of support for dismissal campaigns, followed closely by left-right ideology. PhD status is also strongly significant, along with being female rather than male. Being North American rather than British, or in STEM rather than SSH, does not predict pro-cancel sentiment.

There is an important wrinkle, however, that makes the transatlantic dimension important. Namely that American, and to a lesser degree Canadian, PhD students are noticeably more pro-cancellation than either British PhD students or American and Canadian academics. This is evident in the model in Figure 17. This shows that American PhD students 35 and under (Millennials) have a 56% chance of backing at least one of the dismissal campaigns and Canadian PhDs the same age have a 44% chance of doing so.

This compares to a 41% chance of supporting at least one dismissal campaign among academics 35 and under in Canada, a 38% chance among young American academics, a 36% likelihood for British PhDs, and a 32% chance for British academics. While there is only have sufficient data points for PhDs 35 and under in North America, it is clear they are more intolerant than academics the same age, especially in the US where Millennial academics are 18 points (38% vs. 56%) more tolerant than PhDs of the same cohort.

Millennial academics are twice as intolerant as those 55-64 in Britain, four times as intolerant in Canada, and close to three times as intolerant in the US. While younger academics are more pro-cancellation, there also appears to be a discontinuity, with the Millennial cohort substantially more intolerant than those in the next 36-45 cohort in three of four cases where I have comparable data.

Age matters more than any other factor for political toleration. In addition, being a Millennial is also important, with those 35 and under especially likely to support dismissal campaigns even when accounting for their age (continuously measured in years). While PhD students are significantly less tolerant than academics, this mainly concerns STEM PhD students in North America being significantly more pro-dismissal than STEM academics.

Support for academic freedom may wane as older academics retire, and as a less tolerant cohort of Millennial PhD students and academics replaces the more free speech-oriented Boomer and Xer generations. It is encouraging that Millennial academics appear more freedom-oriented than younger graduate students, which could be due to workplace socialization, role change, or the desire to be protected from the risk of dismissal. Nevertheless, young academics remain only half as tolerant as older academics. Intervening to familiarize undergraduate, master’s, and PhD students with the history and importance of academic freedom would seem an urgent task. Later we shall see, however, that by the time a student enters a PhD program, their views on these questions may be difficult to shift.

The emerging problem of Millennial illiberalism may stem from a rising culture of intolerance among younger cohorts rooted in a feelings-based ethic that some have termed “therapeutic totalitarianism.”14 Among American undergraduates, for instance, support for free speech appears to be very limited, with a reflexive intolerance for most forms of controversial speech. The Foundation for Individual Rights in Education (FIRE)’s 2020 survey asked 20,000 undergraduate students enrolled in four-year programs at 55 leading US universities the following:

Would you support or oppose your school ALLOWING a speaker on campus who promotes the following idea:

• Abortion should be completely illegal? 72% opposed.

• Black Lives Matter is a hate group? 75% opposed.

• Censoring the news media is necessary? 64% opposed.

• Some racial groups are less intelligent than others? 85% opposed.

• The US should support Israeli military policy? 50% opposed.

• All white people are racist? 74% opposed.

• Transgender people have a mental disorder? 72% opposed.

• Christianity has a negative influence on society? 56% opposed.

Only the small minority of conservative students consistently backed free speech across all questions.15 Figure 18 shows that just 7-18% of US undergraduates show “strong support” for free speech, with strong opposition at 40% or higher across five questions.

These findings resemble those of other surveys. A 2018 Knight Foundation survey found that students, by a 53-46% margin, favored the aim of a “diverse and inclusive society” over protecting free speech. Thirty-seven percent said it was acceptable to shout down a speaker, and 44% agreed that “people who don’t respect others don’t deserve the right of free speech.”16 Academics are more tolerant than PhD students, who might be more tolerant than undergraduates. However, it is unclear whether the rising cohort of new academics will import the intolerant ethos of the current generation of students from elite universities into the professoriate.

Conformity or True Belief?

According to Timur Kuran, many people may appear to support an authoritarian regime while privately opposing it. Faced with harsh sanctions for speaking out, many come to believe that others support the regime when in fact most oppose it.17 Sunstein terms this “pluralistic ignorance,” arguing that this characterized mass publics during the Nazi and Soviet regimes, and does so in organizational life today.18 Is this the nature of contemporary academia with respect to support for cancelling academics on “social justice” grounds? That is, do most professors oppose cancel culture but remain silent? My fourth illiberalism question helps shed light on the problem. Recall the earlier question on how individuals would respond to a finding that more women and minorities in organizations correlates with worse performance.

Among social science and humanities academics currently in post in Britain, just 7% would sign the open letter denouncing the professor. Another 11% would support the views expressed in it. Sixty percent wouldn’t support either side or said they didn’t know. Nine percent would support but not sign the counter-petition while 13% would sign a counter-petition. These figures suggest that while only 18% of serving British SSH academics take an illiberal stance, just 13% would publicly stand for academic freedom against a “social justice” challenge. Another 9% fit Kuran’s preference falsification pattern of privately backing the counter-petition while staying silent.

Looking across the range of five surveys in North America and Britain in Figure 19 shows limited evidence for preference falsification: in almost all surveys, those who opposed dismissal were more likely to say they would publicly sign a counter-petition than remain silently opposed.

Of course, it may be that when the rubber hits the road those who say they would sign the counter-petition against dismissal get cold feet. Arif Ahmed, a Reader (Associate Professor) in Philosophy at the University of Cambridge reports that it was very difficult for him to acquire the 25 signatures needed to get his free speech motion on the ballot at Cambridge, but it passed with 80% faculty support.19

Even so, the notion that there is a silent majority motivated to oppose cancellation but afraid to do so does not receive backing from the data presented here. If this were the case, I would expect to see a clear majority opposed to dismissal across all questions, but with most saying they would only do so privately.

In fact, there is a large group that is normatively undecided. Even if there was no fear of speaking out, there might not be a majority opposing a dismissal campaign. As Figure 19 shows, the pro- and anti-free speech positions have broadly similar backing, but there is a large group of undecided or neutral respondents, comprising 30-60% of the faculty, depending on the question. I found a similar result with students in my previous report.20

Moreover, as Figure 19 – which is limited to those on the pro-freedom side – shows, most of those on the pro-freedom side are willing to nail their colors to the mast. Indeed, slightly more are likely to say they would sign publicly than are supporters of dismissal. The only exception was among North American (88% American, 12% Canadian) PhD students, where there is more evidence for a pattern of preference falsification in which 14 of 22 (about 70%) of people would support but not sign their name to an anti-dismissal counter-petition.

Cross-Pressured Between Value Commitments

Earlier, we saw that more than 6 in 10 British academics and 5 in 10 American academics are noncommittal on whether to defend the hypothetical academic who found that diversity worsens organizational performance. Many in the academy are torn between the claims of self-esteem egalitarianism and those of freedom and reason. The upshot of this is that the problem of academic freedom in academia is predominantly a battle for hearts and minds, rather than just a collective action problem like Kuran’s “private truths, public lies” that we see among those who conceal their private views in authoritarian regimes. Among North American and British academics, there are few active authoritarians, but the share who care enough about academic freedom to even privately support (i.e., without saying anything) an academic against a campaign to fire them is generally not higher than 50%.

Even among right-leaning professors and lecturers, just half in Britain and 60% in North America express private or public support for the academic freedom counter-signatories in the diversity-organizational example. This indicates that many academics have not made up their mind on questions where academic freedom and “social justice” collide. This reticence likely also informs their belief that opposing such measures may out them as a heartless conservative. It does not support the idea that a clear “silent” majority favors academic freedom but is too scared to speak up: the “emperor’s new clothes” scenario in which it takes just one person to prick the balloon.

The noncommittal stance of most staff may also have to do with the fact that free speech activism on behalf of politically incorrect researchers may be implicitly associated with the right. In addition, many on the left may implicitly accept the group affirmation that extremists provide because, as Clark notes in a review of the psychology literature on political tribalism, “Extreme, rigid, dogmatic defenders of our political ingroups demonstrate tribal loyalty that can be appealing to those of us who care about our political coalition’s success, even if we are more moderate or have more nuanced beliefs and policy preferences.”21

Focusing only on centrist and conservative academics who do not endorse dismissal – options c) through e) above, and controlling for ideology, I find that women and younger lecturers are especially likely to opt to remain neutral, with significantly greater support for actively promoting academic freedom coming from those on the right compared to those in the political center.

Gender seems to match left-right ideology in explanatory power. The model in Figure 20 shows, controlling for ideology, that there is a 54% chance of a 70-year-old non-leftist British male academic publicly or privately backing a counter-petition, but this falls to 17% among a 30-year-old female non-leftist academic. Even among conservatives and centrists, women and the young appear to be more cross-pressured between progressive and academic freedom commitments.

Combining British, American, and Canadian respondents, we see a similar pattern. What is vital for the report, however, is to appreciate the connection between viewpoint diversity and resistance to intolerance. 35% of 700 centrist and 214 right-wing academics and PhD students say they would publicly oppose dismissal compared to 21% of 1,719 respondents on the left. Among those under 40, who represent the future of academia, public opposition is even more concentrated among conservative academics (34%) compared to centrists (20%) or leftists (11%). Thus the problem of hard authoritarianism does not simply arise because the larger post-1990s pool of far-left activists in the professoriate produces more agitation for disciplinary action and, as a result, a chilling effect. It also involves a weakening of resistance to authoritarianism due to the paucity of centrists and, especially, conservatives. This limits the ranks of those who would actively resist authoritarian measures, clearing the path for institutionalized illiberalism.

A Lopsided Trade-Off: Free Speech and Political Correctness in Academia 

The cross-pressuring among academics on the question of dismissal does not mean that most faculty resemble the median voter. In reality, the typical SSH academic trades free speech and emotional safety off considerably further in the direction of the latter.

Noting the way that processes of social closure operate, some argue that academia is becoming a moral community, with a set of norms that constrain the questions scholars may ask.  What is termed “social closure” within particular occupations like the arts or academia involves a boundary being drawn around a community, so enabling a group-based identity to be constructed.22 At the extreme, some consider these norms to be sacred values, with the campus a “safe space” that must be kept pure, free of heretics who would profane values of cultural equity and diversity. For instance, sociologist Christian Smith speaks of a teleological “liberal progress narrative” that has come to dominate the field of sociology in America. He summarizes it as follows:

Once upon a time, the vast majority of human persons suffered in societies and social institutions that were unjust, unhealthy, repressive, and oppressive. These traditional societies were reprehensible because of their deep-rooted inequality, exploitation, and irrational traditionalism ...there is much work to be done to dismantle the powerful vestiges of inequality, exploitation, and repression.23  

On Smith’s account of the liberal progress narrative, a set of “social justice”-based normative commitments are prioritized above even the search for truth.

The question of political correctness is a useful barometer for assessing how far outside the mainstream academia lies. More in Common’s Hidden Tribes report segments US and British society into values “tribes” based on their survey responses to ideological questions. The “Progressive Activist” segment of society makes up 8% of the American adult population and 13% in Britain. It is largely white, urban, and professional, leaning consistently left across all issues, and bulks larger in academia than elsewhere. In Britain, this group is 6 times more likely than average to post political content on social media, and in America, 3 times more likely. When it comes to political correctness, Progressive Activists stand apart from all other groups in support. For instance, 80% of Americans say “political correctness has gone too far” but just 30% of Progressive Activists agree. In Britain, the comparable numbers are 72% and 28%.24

Younger Americans are more pro-PC. Within the general population, the 2016 American National Election Study (ANES) pilot survey asked:

There’s been a lot of talk lately about “political correctness.” Some people think that the way people talk needs to change with the times to be more sensitive to people from different backgrounds. Others think that this has already gone too far and many people are just too easily offended. Which is closer to your opinion?

Among whites, and controlling for education level, Figure 21 shows that the youngest respondents (born in 2000) are, on average, 25 points more likely to favor the politically correct position than those born in 1940. Though ideology and party identity are far stronger predictors, age is still associated with support for political correctness among American whites when other variables are held at their mean values. That is, even when screening out ideology and partisanship, a 16-year-old remains 15 points more politically correct than a 76-year-old.

Most Britons and Americans oppose PC. However, most also oppose hate speech, which is often used to justify PC. When the question is phrased as a trade-off between speech restrictions and discrimination, a clearer picture of people’s priorities emerges. I use UK data due to the enormous sample size of YouGov’s UK panel, from which my academic survey is drawn. Consider a question asked across YouGov’s British panel:

Thinking about political correctness, are you generally in favor of it (it protects against discrimination), or against it (it stifles freedom of speech)?

Figure 22 shows that when anti-discrimination and free speech are set against each other in this way, 47% of 164,000 members of the British public oppose political correctness and 37% support it. Against this, the balance among the 603 academics in my YouGov sample (includes both STEM and SSH, active and retired) who also answered the PC question as part of separate polling that is included in the Profiles database, is 64% in favor and 31% against. Among the 176 currently employed academics in the social sciences and humanities, however, I find 76% backing PC and a mere 20% against. For comparison, even among the sample of 56,000 university graduates in the sample, the pro-PC tilt is just 48-41, much narrower than in academia. In a related vein, Samuel Abrams finds that in the US, 78% of Democratic, but only 39% of Republican, academics support safe spaces.25

Within the sample of 603 British academics who answered the political correctness question in Figure 22, the main predictors of their political correctness are ideology, politics, and activism, as shown in Figure 23. This suggests that the political skew of academia renders it less friendly to freedom of speech than other occupations when the competing value of emotional harm-avoidance comes into play.

In academia, two other characteristics are associated with increased support for political correctness: being a professor rather than a lecturer, and being female. Given the influence of professors in running universities, the former finding is especially concerning for those who fear for academic liberty. No significant difference was found by age, or by whether an academic is retired or active, however, once other factors had been taken into account.

Notice that age does not significantly predict attitudes to PC among British academics in Figure 23 above. This is in marked contrast to the picture among the general public in the US data in Figure 21. Figure 24 presents a sample of roughly 1,200 mainly non-academic PhD holders drawn from the UK YouGov panel. Here there is a big age gap: PhD holders under age 40 are over 30 points more inclined to support PC than PhD holders over 60.

But this age gradation does not exist among academics, especially SSH academics. According to Figure 25, there is an important divergence between British PhD holders who remain in academia and those who leave for other fields. The difference is especially noteworthy among academics currently teaching in the social sciences and humanities, where being older does not appear to correlate with diminishing enthusiasm for political correctness.

Therefore, while PhD holders under 40 inside and outside academia are fairly similar in their high support for PC – a finding that we’ll see also holds (on related measures) for current PhD students of all ages – there is a wide divergence between academic and non-academic PhDs over age 60. This points to possible academic socialization, especially within SSH fields, as a stimulant to retaining support for political correctness.

But there is another possibility, with wider implications. Namely, that the change is generational, and PhDs outside academia will henceforth remain PC as they age, altering the ideological composition of non-academic professions. On this reading, the high support for PC among older academics, especially in SSH fields, is because they are a left-modernist ideological vanguard in society that is leading where others will follow. If the latter interpretation is correct, this means that non-academic professional spheres are likely to become more PC in the years to come.

However, the age profile of left-wing frequent social media users is flatter than it is for PhD holders, indicating that ideological fervor counteracts the conservatism that is associated with being older (whether for life cycle or generational reasons). Among British non-academics who are frequent social media users, identify with the Labour Party, and are far leftist, there is no difference between young and old. Indeed, the over-60s in this group are as or more PC than their academic counterparts. This evidence suggests, in line with the model in Figure 23, that leftist consciousness, social media use, and gender are central, with age per se a smaller factor. Education level on its own appears to have only a modest impact, if any, on support for PC.26

Support for political correctness is grounded in empathy for historically disadvantaged groups, especially those based on race, gender, and sexuality. Yet, as Paul Bloom warns, empathy can be a flawed guide to morality.27 An effect of empathy for a particular person or group can be that others who deserve sympathy disappear from view. Worse, those who are viewed as harming the group with which one empathizes may cause one to become angrier than would be justified by a more wide-angle, evidence-based view.

This is illustrated by the fact that empathy for some identified victims can be used to rouse people to commit atrocities. As Bloom remarks, empathy towards white women was used to fan anti-black violence in the Jim Crow South, and feelings for the plight of the poor helped to mobilize anti-bourgeois genocide in communist regimes. In today’s context, empathy for minorities can lead to hostility towards whites or conservatives. Rather than see the whole and trust in empirical regularities to guide one’s anger and calibrate one’s response, powerful attachments and narratives take over, leading to flawed moral reasoning.  Moreover, when empathy for a particular subset of disadvantaged groups and hostility to their purported oppressors becomes a leading priority, values other than cultural egalitarianism – notably freedom and reason – which universities have long proclaimed, may become endangered.

Decolonizing the Curriculum?

Returning to my academic surveys, I sought to probe faculty views on a less dramatic test of academic freedom than support for firing heretics:

Please imagine there was a new initiative in the Humanities, Arts and Social Sciences at your institution, stipulating that on each reading list, at least 30% of readings must come from women and 20% from authors of color.

Response categories were the same as in the diversity-organizational performance dismissal scenario, permitting us to gauge not only people’s views, but the extent to which academics censor their private beliefs:

a)     Publicly express opposition (e.g. at a department meeting)

b)    Privately oppose the initiative, but not say anything publicly

c)     Neither support or oppose the initiative

d)    Privately support the initiative, but not say anything publicly

e)     Publicly express support (e.g. at a department meeting)

f)     Don’t know

This kind of reading list campaign has occurred in a number of jurisdictions, with one Swedish academic, Erik Ringmar, complaining that he had to cancel an entire course on fascism due to his inability to find sufficient women writers on the subject to satisfy a quota of 40% females per reading list mandated by the University of Lund’s political science department.28 

Figure 26 summarizes the results across all surveys, focusing only on SSH subjects. It highlights that this question led more people to jump off the fence, with the share of neutral and don’t know answers, 16-26%, much lower than in the dismissal scenario. This is 2 to 3 times less uncertainty than for the question of whether to dismiss an academic who argues that diversity impairs organizational performance.

The second major aspect to note with this question is that a near majority of SSH academics support the decolonization agenda across all jurisdictions: between 44 and 48% of academics in the first four surveys back the idea of racial and gender reading list quotas. In all surveys, this comfortably exceeds the share who oppose quotas though there is significant opposition: between 29 and 34% of academics, depending on the survey.

Social science and humanities PhD students in both Britain and North America are especially enthusiastic about quotas, with 61% of British SSH PhDs and 70% of their North American counterparts in support. This tallies with the views of the 1,216 PhD holders under 35 in the YouGov non-academic sample, who leaned 67-20% in favor of political correctness. While these figures might be expected to dip somewhat when PhD students enter academia, have to make up their syllabi, and find their intellectual wings clipped, it is a worrying sign of how far the balance between social justice and academic freedom has tipped in the direction of the former among the youngest cohorts entering the academy.

Respondents who backed or opposed the reading list quotas tended again to say they would do so publicly rather than keep their beliefs to themselves. Naturally, a portion of those who claim they would stand publicly may fail to do so in the breach, but Figure 26 does not support a Kuran-Sunstein picture of mass conformity to social justice claims concealing majority support for academic freedom.

I also have data from my survey of National Association of Scholars members, where 91% of the 227 respondents opposed the decolonization initiative and just 3% backed it. Among the opponents, 61% – two-thirds – said they would publicly sign a counter-petition while 30% said they would keep their views private.

However, this doesn’t mean there is no concealment. Compared to the dismissal scenario, where there is less of a clear pattern of self-censorship among opponents, here there is more evidence of a chilling effect on those opposing the decolonization agenda. Figure 27 indicates that supporters of reading list quotas are far more likely than their academic freedom opponents to publicly state their views rather than keep them private. Among academics, supporters are between 4 (315%) and 8 (693%) times more likely to voice their opinion publicly than privately. By contrast, opponents of decolonization are much less willing to go public, being just 26-174% more likely to speak up than stay silent.

This tips over into a preponderance of silence when we shift from academics to PhD students, where opponents are between 33 and 58% more likely to stay silent than speak up – reflecting the heavy dominance of pro-decolonization sentiment among graduate students.

In particular, it is noticeable that among the 14-20% minority of PhD students who oppose decolonization quotas, there is more reticence to come out publicly than is the case among professors who oppose quotas. Among PhDs, a majority of opponents in Figure 28 say they would choose to keep their beliefs to themselves rather than voice opposition publicly. This could be related to the greater support for hard authoritarianism among PhDs compared to academics. This comports with a 2020 FIRE survey that finds that conservative students tend to self-censor more at relatively liberal universities.29

The final point to note is that in this question, across almost all surveys, opinion among academics under 40 generally contained 5-10 points fewer neutral responses than among the over-40s, possibly indicating that there are more settled views of this question among younger staff, in the direction of both diversity quotas and, to a lesser degree, academic freedom. This age-pattern was not apparent in the previous question on organizational performance, however, so it may not hold more generally.

To probe the malleability of this group of undecided scholars, I asked respondents:

If you had to choose, which of these do you think is more important:

a) that course content should be inclusive, representing the racial and gender makeup of the students.

b) that course content should feature the most intellectually foundational books and articles in the field.

c) Don’t know.

Results are plotted in Figure 29. These show that academics back content over quotas in both the UK and North America. North American faculty and, especially, graduate students are relatively more favorable to quotas over foundational content compared to British faculty, who clearly privilege content over quotas. Both surveys of UK social science and humanities academics show that they prioritize intellectual merit over quotas by 18-25 points. British PhD students break 50-50 between the two options. On the other hand, North American PhD students incline 15 points more toward quotas over foundational texts.

Note the difference between these numbers and those in the similar question in Figure 26 above. This time, I introduce a more explicit trade-off between quotas and intellectual merit (“foundational texts”) than was presented in the previous question asking people to merely support or oppose an initiative “stipulating that…at least 30% of readings must come from women and 20% from authors of color.” Among North American academics, support for quotas has slipped 6 points in the second version (45% supported quotas in the first question, falling to 39% support in the second).

In Britain, mention of intellectual rigor and the key value of education seems to have swayed academics even more than in North America, lowering support from 44% (YouGov) and 51% (mailout) in the case of support for the diversity initiative to 32% (YouGov) and 34% (mailout) on the question of quotas versus foundational texts. So too among PhD students: UK PhDs reduce support for the quota option from 61% in Figure 26 to 46% in Figure 29 while North American PhD students lower their support from 70% to 55%.

Diversity Statement Experiment

In order to examine the firmness of these beliefs, I conducted a survey experiment in each of the five surveys. Before answering the question, half the sample read an equality and diversity statement modeled on that of the University of California:30

Please imagine an organization is committed to the full realization of its historic promise to recognise and nurture merit, talent, and achievement by supporting diversity and equal opportunity in its education, services, and administration, as well as research and creative activity. This organization particularly acknowledges the acute need to remove barriers to the recruitment, retention, and advancement of talented students, faculty, and staff from historically excluded populations who are currently underrepresented. Many personally identify with underrepresented groups in the curricula (that is, the stories, histories, findings and research of women and BME members, among others) and the organization is determined to serve them through an education aligned with social justice and inclusion.

In the British YouGov sample, among those who didn’t read a statement, just 25% supported the view that course content should represent the racial and gender makeup of students instead of the intellectually foundational texts in a field. This jumped to 39% among those who read the UC diversity statement. The effect seemed to register across most academics, but not among Brexit supporters, as Figure 30 shows. Clearly many British academics have values that render them open to a Social Justice appeal.

The typical academic who voted to remain in the European Union in 2016 (the vast majority of academics) shifted 16 points after reading the diversity statement, from 27% to 43% support. By contrast, the minority who voted to Leave the EU barely budged, from 13% among those reading nothing to just 15% among those reading the diversity statement, showing no statistically significant difference.

Another important factor affecting a person’s view on course content is support for political correctness. PC supporters were, as one would expect, far more supportive of quotas than opponents of PC. In addition, as Figure 31 shows, the greater an academic’s support for PC, the more susceptible they are to being convinced of quotas after reading the diversity statement. Here it seems that institutional initiatives around equity and diversity, like diversity statements, strike fertile soil among the pro-PC majority of academics: recall that three-quarters of current social sciences and humanities academics favor the concept to help combat discrimination, even if it stifles free speech.

Looking across the wider array of results in Figure 32 unearths two patterns. First, culturally conservative groups, notably Brexiteers (known as “Leavers”) and those favoring immigration restriction, have low support for quotas and are not affected by reading the diversity statement. Second, groups that already have high support for quotas – young, women, the far left – shift less. It seems that a broad receptive group in the middle – center-left, Remain, men, pro-PC – move most toward backing quotas.

The first point to note is that 40-45% of younger academics (under 40) oppose racial and gender quotas for reading lists. While younger academics are more supportive of the “decolonization” agenda, a near majority are not, and their lack of movement after reading the diversity statement suggests there is an important group of young faculty who are aware of diversity claims, but reject them when they conflict with academic freedom.

That said, the survey experiment shows that there is a flexible component of academic opinion in Britain, just as there is with the student opinion I analyzed in a previous co-authored report, which I reprise in Figure 34. Yet these results only show that academic opinion can be swayed by a social justice/diversity appeal, which is less institutionalized in Britain than North America. Might an appeal to free speech shift this group away from the pro-decolonization position?

Academic Freedom Statement Experiment

The diversity statement had a profound effect on British academic opinion, suggesting that opinion in the academic center is malleable on this question. Might it be shifted in the other direction, away from quotas? The answer, it appears, is no.

In order to test whether academic freedom arguments might work, I introduced a new statement group, in addition to the diversity statement group and the control group who read nothing:

Throughout the ages, those who have expressed unorthodox opinions – on religion, on the government, on homosexuality, socialism or capitalism – have been subject to persecution or censorship. Britain [America] has an especially important history of resisting threats to freedom of expression from the authorities. Sometimes these threats take the form of being killed or jailed, while at other times they involve being turned into a social pariah, losing one’s livelihood or being subject to reputational shaming. Many have died for our right to freely speak our minds and exchange ideas, our precious inheritance. Liberalism is not easy because it is about tolerating ideas we don’t like, which is why it is rare in history and in much of the world. The role of the university is to permit a wide variety of opinions to be expressed and debated, even if they challenge social convention or offend people’s sensibilities. The quest to teach and research the highest truth must prevail over other priorities. Tolerating only views we agree with is contrary to the spirit of free speech.

Figure 33 summarizes the impact of the two treatment statements on my mailout sample of UK SSH academics. The results reveal that opposition to decolonization is similar among both those who read nothing (57%) and those who read the free speech passage above (60%). This shows that reading the free speech vignette has no impact on academics’ value priorities when academic freedom and diversity are in tension. However, those who read the diversity statement were nearly 20 points more likely to favor quotas – replicating findings in Figure 32 from the YouGov data. The “don’t know” responses are similar, indicating that many changed their view from the “foundational texts” position to the quotas position after reading the diversity statement.

For the North American academic and PhD surveys, and for British PhDs, neither the diversity nor free speech experimental passages made any significant difference to their responses to the quotas-versus-merit question. This suggests that diversity-based quota logic is already well-established among North American scholars of all ages and younger cohorts in British academia, with views fixed in both directions. Thus new messaging on academic freedom or free speech to PhD students and faculty may have little impact on their trade-off function between social justice and free speech.

The lack of responsiveness to the free speech message across all academic and PhD surveys stands in contrast to that of British undergraduates, who are more biddable to the free speech side. In previous work, I found that the free speech passage above had a significant impact on undergraduate student opinion on the question:

When in doubt, which policy should your university support?

a) prioritize free speech, even if this makes people upset;

b) prioritize emotional safety, even if this limits free speech;

c) don’t know.

Figure 34 reproduces that work, showing that 63% of undergraduate students who read the free speech passage backed the free speech option but only 49% of those who read nothing did so, and just 41% backed free speech among those who read a paragraph about the need for emotional safety for disadvantaged groups. Thus the free speech treatment moved opinion 14 points (49 to 63%). The emotional safety passage had a similar impact the other way, shifting views in the direction of emotional safety by 14 points (34 to 48%). Female undergraduates shifted more than male undergraduates in response to the emotional safety passage while both genders were equally moved by the free speech appeal.

It appears that British academics and PhD students have relatively fixed views. British academics are amenable only to a diversity-based appeal for restrictions on academic freedom, not to free speech appeals. North American academics and PhDs have settled views on the matter so are resistant to either message. Advocating for free speech is therefore more likely to bear fruit if directed at undergraduates. By the time students have been socialized into the diversity and emotional safety-oriented academic culture at the PhD level, it may be too late.

Age, Gender and Support for Academic Freedom

We have seen that views on the decolonization versus academic freedom question have crystallized in North America and among British PhDs, and are relatively impervious to competing perspectives. This suggests stronger divisions on this question could prevail going forward. The question this then begs is whether opinion is shifting away from merit toward quotas due to generational turnover. In order to explore what may be coming, I parse the views of faculty and doctoral candidates by age.

In general, younger faculty lean more toward quotas than older faculty. Figure 35 shows how academics under 40 are, depending on the survey sample, 4-23 points more likely to back the reading list quota initiative than those over 40, while PhDs under 30 are 10-29 points more likely to do so than PhDs over 30. The age differences are somewhat larger among North American than British academics and doctoral students.

When the question explicitly juxtaposes “intellectually foundational texts” against reading list quotas (see Figure 29), support for quotas falls across the board, but as Figure 36 demonstrates, the age gap remains – though only in the North American samples. This may point to the greater penetration of curriculum quota activism in North America, which has roots in the 1960s and with the multicultural education movement of the late 80s and early 90s. Only recently has there been a major push on this front in Britain.31

Gender is also a consistent predictor of attitudes to diversity/safety versus freedom questions. Figure 37 charts a wide gender divide that ranges from 12 to 38 points but is greater than 23 points in all but one survey. This echoes findings in student surveys, which also reveal consistent gender disparities on free speech versus emotional safety questions.32 Having said this, an important group of between 24% and 34% of female academics and graduate students favor intellectually-foundational texts as the basis for course reading lists rather than diversity quotas.

As new generations of PhD students enter the academy and, if the share of women rises, we should expect the balance of internal opinion to move in the direction of emotional safety over academic freedom. In addition to limiting the freedom of academics to set the course texts they believe to be most pertinent for teaching, further moves in the direction of emotional safety are likely to mean that research on controversial topics around race, gender, and sexuality may become increasingly off limits. This circumscribes and distorts knowledge produced by the university, reducing its value.33

Attitudes toward Nonconformity

If a new initiative mandates that people meet diversity quotas for reading lists, what is the punishment for those who refuse to yield their academic freedom to these rules? While one may see diversity as a positive goal, are those in favor willing to own the authoritarian implications of such a policy for those who refuse to comply? I put the question to my sample of over a thousand American and Canadian academics. Table 2 lists, among those in favor of quotas, their preferred sanctions, from least to most punitive, for academics who refuse to comply with the quota requirement:

The appetite to punish severely is not as strong as the desire for quotas. Only 3% recommended firing dissenters, though 19% supported formal disincentives from dismissal to being allocated less favorable teaching, research, and administrative positions. Another 27% favored social pressure while a further 28% backed mandatory implicit bias awareness training. These punishments are the mirror image of the complaints outlined in Figure 6, which found that 38% of right-wing PhD students in North America and 43% of the mainly right-leaning National Association of Scholars (NAS) members experienced formal or informal sanctions for expressing their views in research, public fora, and teaching. These showed that the most prominent form of punishment experienced by nonconformists was social bullying, followed by adverse teaching or administrative roles or reduced research funding.

The strongest proponents of severe punishment were academics under 40, with 11% of them advocating forcing dissenters to cancel their course and teach a quota-compliant module. By contrast only 2% of academics over 50 selected this option. Those who identified as activists and “very left” were similar in their inclination to punish, and being female had an effect only slightly weaker than age, activism, and ideology.

Yet much of this is an artifact of these groups’ stronger support for quotas over intellectual merit. When limiting the analysis only to those in favor of quotas, the relative authoritarianism of the under-30s, and lack of it among the over-65s, is what jumps out of the models. Ideology and being female no longer predict authoritarianism, though self-described activists are somewhat more inclined to punish than others (this effect was just outside the boundary of conventional statistical significance).

What this question does not ask is whether people would be willing to publicly object if the university decided to dismiss those who resist, or to punish them severely. We saw that for controversial research, about half of academics opposed dismissal campaigns, and for less controversial views – such as favoring reduced immigration – around 8 in 10 did. A little over half the opponents said they would be willing to publicly express their opposition. Assuming that a recalcitrant academic who does not wish to comply with mandatory reading list quotas attracts responses intermediate between these positions, this leads me to estimate that close to half of academics would oppose a dissenter being terminated, but that little more than a quarter would say so publicly. This would possibly clear a path for administrators to adopt a hard authoritarian position against those who refused to comply with a reading list quota directive.

Academic Freedom or Social Justice?

A related but distinct question is whether the tendency to favor quotas over intellectually foundational texts, or to back new diversity quota initiatives, reflects a general preference for social justice over the freedom to seek the truth, wherever it lies.  Here the question is:

When it comes to tension between the freedom for academics to publish research on their interests and concerns over social justice for disadvantaged groups, which comes closest to your view?

a)     I prioritise social justice, and have strong beliefs in this area

b)    I prioritise social justice, but don’t have strong beliefs in this area

c)     I have thought about it, but don’t have a strong view either way

d)    I haven’t thought about it much, and don't have a strong view either way

e)     I prioritise academic freedom, but don’t have strong beliefs in this area

f)     I prioritise academic freedom, and have strong beliefs in this area

When phrased in these more intellectualized and abstract terms, the balance of answers shifts away from the social justice position towards academic freedom. Whereas American and Canadian academics favored foundational texts over quotas 47-39, Figure 38 reveals that they prioritize academic freedom over social justice by a whopping 56-27 margin (58-26 in the US). Even among North American PhDs, a 15-point lead for the quota position has eroded to just a 6-point lead for the social justice view in this formulation. UK PhDs move less: from an even position between foundational texts and quotas to +4 for academic freedom. The difference can likely be attributed to both the intellectual appeal of the term “academic freedom” and the more abstract nature of the question. Elsewhere I find that more concrete trade-offs involving actual people, policies, or events tend to be resolved more in favor of the social justice position.34

It is also noteworthy that North American PhD students favor social justice over academic freedom by 6 points (9 points among SSH PhDs), and that social justice is nearly on par with academic freedom among British PhDs. This indicates that support for academic freedom may be less robust among the emerging generation of faculty.

The correlates of these positions are similar to those in the quota versus merit models, with younger scholars, those on the far left, activists, and women more inclined toward the social justice position. Being a racial minority is not statistically significant, bearing out a pattern whereby race is a less important correlate of views on academic freedom questions than age, gender, or ideology.

Figure 39 shows how gender and ideology interact to condition attitudes to academic freedom or social justice, with women increasingly diverging from men toward social justice as we move left on the spectrum, but with more limited gender disparities on the right. Those of “other” gender (29 in dataset) also tend to favor the social justice option over academic freedom. Crucially, men – even far-left men – tend to favor academic freedom over social justice.

In terms of statistical effect sizes, identifying as an activist had the strongest effect on prioritizing social justice, followed closely by left-right ideology, gender, and age. American academics were marginally more likely to favor academic freedom than Canadian scholars.

Not Against Academic Freedom, But Not For It, Either

The upshot of our analysis of the academic freedom vs. social justice questions is that most academics are cross-pressured between their commitments to cultural progressivism and their attachment to academic liberty and reason. On a positive note, just 20-25% of SSH professors and lecturers, rising to 36-45% of SSH PhD students in North America and Britain, would support at least one of four authoritarian measures against a dissenting conservative academic. This still leaves three-quarters of SSH academics as consistent opponents of drives to remove professors from their posts for politically incorrect research. Far leftists are considerably more likely to back illiberalism, but even here, the far-left faculty divides fairly evenly between supporters and opponents of dismissal even when using my expansive measure of support for firing (backing any one of four hypothetical campaigns). In any given campaign, the majority of far-left academics will oppose dismissal.

Fear plays a part for an important minority of pro-free speech academics, inclining them to remain silent rather than advocate publicly for freedom, but sincere belief and not fear is the main obstacle to tackling today’s illiberal campus climate. Activists and administrators are the instigators of authoritarian policies, but in the absence of the climate of opinion prevailing within the faculty, where half are hesitant about protecting academic freedom when it collides with progressive aims, their efforts would meet more resistance.

The fact that many academics do not publicly nail their colors to the academic freedom mast due to their beliefs is not a good sign. This is especially so if younger academics do not become more inclined to defend academic freedom as they ascend the academic ladder. The rising share of women in academia may also affect active support for academic freedom due to their somewhat higher predilection for diversity arguments – even though women are not more likely to endorse authoritarian punishments than men who share the same belief system.

Most academics and PhD students in the United States, Canada, and Britain don’t actively oppose academic freedom, but they don’t actively support it either. Most are unlikely to speak up for colleagues who dissent from social justice-inspired restrictions on their academic freedom. Thus progressive authoritarianism succeeds through a kind of sin of omission. This indicates that intervention would need to occur at the institutional level, to ensure activist-driven proposals cannot become university policy. At present, college authorities are generally silent on academic freedom or subordinate it to the countervailing imperatives of equity, protection from harassment, diversity, and college reputation.

Currently, the momentum inside the university lies largely with those who would abridge academic freedom and free speech in response to equity and diversity claims on behalf of disadvantaged groups. This is well-established in the American and Canadian cases.35 But there is also a rising volume of diversity initiatives at British universities, spurred on by the Race Equality Charter (REC) of Advance HE and ATHENA SWAN, influential charities that get universities to sign up for their monitoring and compliance pledges, building these into their research assessment scores.36 Both Universities UK (UUK), the industry body for universities, and the major sector union, the UCU, are strongly promoting this agenda. This gives internal diversity proponents within the professoriate a metrics-based rationale to move in the direction of mandatory targets or even quotas. In North America, such initiatives are even more entrenched as a combination of staff and student activists and better-funded (compared to Britain) equity and diversity administrators enforce compliance. These have truly become what Jonathan Haidt and John Ellis term “social justice” rather than “truth” universities.37

While Cambridge University’s governing body famously rejected, by an 80-20 margin, a proposed policy shift that would have mandated “respect” for identities over toleration of difference, the vote involved just 1 in 8 academics at the university, and included retired academics and some non-academic staff. While my survey results intimate that the vote would still have prevailed had all staff been polled, they also suggest that the final vote would have been closer.38

These findings also indicate that rebalancing internal messaging from equity-diversity toward academic freedom would have a significant impact on attitudes among undergraduate students, and – to the extent that messaging around emotional safety for protected groups is reduced – also among UK professors and lecturers. Where views have largely crystallized, as among North American academics and PhDs, and among UK PhDs, messaging is less likely to persuade.

Government policy can play an important role in conveying the will of the democratic majority and the letter of the law (at least among institutions in receipt of public funds), while issuing guidance to ensure that university administrators (though not academic staff) act in a politically-neutral manner. This intervention could also help set the tone at universities based on what society and the law – as well as most academics – expect. As Cass Sunstein notes, law often signals the public mood and can dispel misperceptions about the approved values of an organization.39

Part II: Soft Authoritarianism

Most academics reject hard authoritarianism. But the verdict is less clear for soft authoritarianism. As noted in Part I, faculty matter because they play a cardinal role in setting the culture and environment of departments on a day-to-day basis. Whether permanent, or on time-limited contracts as researchers and/or teachers, they do so by shaping how research and teaching are conducted, and by being responsible for the variety of decisions in which, for instance, papers are accepted for publication; grants are awarded or not; job offers and promotions are received; and students are admitted. More generally, faculty set the tone.

Taken individually, very few actions by individual faculty members are of decisive significance. What is of interest and significance are larger patterns, which help shape the culture of the university and ultimately influence the public contribution it makes. To what extent are academics free to pursue research or public engagement that may have a political interpretation, without paying a social or professional penalty?

This part of the report examines political discrimination against conservative and gender-critical academics and its corollary, the chilling effects produced within these target groups. In line with the “iceberg” models presented earlier, it pursues the question through a focus on both the perpetrators and victims of discrimination. The former involves half or more of staff and students at the university while the latter is narrowly concentrated on conservative and gender-critical academics and students.

Targeted scholars either experience direct victimization or feel chilling effects by anticipating that if their views were to become known to colleagues or the wider scholarly community, it would damage their careers. This is especially pointed in view of the scarcity of academic positions, with hundreds of applications for permanent posts. Hyper-specialization results in limited mobility – especially if a researcher wants to live in a given geographic area or work at a high-status university. Meanwhile, reputations travel quickly through the “invisible colleges” that affect hiring choices, making it very difficult for academics to move jobs. This produces even greater risk. As a result, dissenters prudently adapt to this manifestation of John Stuart Mill’s “despotism of custom” through self-censorship, limiting their academic freedom, and constraining the truth-seeking mission of the university.

Prior Work

In recent years, a growing body of work has begun to examine the possibility of discrimination on ideological or political lines. “Political” or “ideological” discrimination involves one’s professional judgment about another person or their work being affected by the extent to which they agree with the political or ideological orientation of the person.

This becomes even more urgent in light of the growing political polarization of the 2000s, initially in the US, but increasingly in Canada and the UK as well.40 Indeed, political polarization is connected with political discrimination. For instance, in 2014, Shanto Iyengar and Sean Westwood found that affective polarization – hostile feelings towards opposing political partisans – was far more pronounced than negative feelings based on race. Further, affective polarization impacts discriminatory behavior to a significantly greater degree than racial prejudice.41

While political discrimination could occur in any sector of society, it is of particular interest within academia, where political viewpoint may directly shape or affect the content of someone’s work. Accordingly, some studies focus on political discrimination within academia specifically. Key contributions to this literature include studies by George Yancey,42 Yoel Inbar and Joris Lammers,43 Nathan Honeycutt and Laura Freberg,44 and Uwe Peters and co-authors.45

Yancey found that 30% of his sample would discriminate against a Republican while 50-60% of anthropologists and English professors said they would be less likely to hire an evangelical Christian. The latter three contributions adopted a similar question wording so can be readily compared. Each used an online mailout survey to gauge the attitudes and opinions of a given sample. Inbar and Lammers sought to evaluate the attitudes of psychologists in the US, and focused on the attitudes of and towards political conservatives (n = 508). Honeycutt and Freberg extended that study, using a sample of academics in California across disciplines, repeating these questions for political liberals in addition to conservatives (n = 618). Peters et al. used an international sample of philosophers, and asked about a more variegated set of political orientations and viewpoints, rather than a simple left-right division (n = 794). These studies confirmed, for their respective samples, the same central finding as that of Sam Abrams (see Figure 40 below), namely that the American professoriate has an overwhelmingly left-liberal orientation and a paucity of political conservatives relative to the general population. For instance, in the Peters study, 75% of philosophers leaned left, 11% were moderate, and 14% leaned right.

On the willingness to discriminate politically, the findings were dramatic. In Inbar and Lammers’ study, over 1 in 3 of those who identified as liberal would discriminate against conservatives in hiring decisions, while 1 in 4 would discriminate against them in reviewing their grant applications.46 Honeycutt and Freberg replicated this result nearly exactly, but showed that the willingness to discriminate was symmetrical on both political sides. In their sample, 33% of liberals were willing to discriminate against conservatives in hiring decisions (answers varying from “somewhat” to “very much”), which was equally matched by conservatives’ willingness to discriminate against liberals, at 32%.47 This study also found that the more ideologically committed a respondent was, the more likely they were to be willing to discriminate.48 From the Peters study, the highest levels of willingness to discriminate were again reported for hiring decisions. For a right-leaning hire, over 55% of left-leaning philosophers reported being willing to discriminate “occasionally” to “all the time.” For a left-leaning hire, the equivalent proportion of right-leaning philosophers willing to discriminate was over 45%.49

A larger sample was collected in 2017 by Terrence Karran and Lucy Mallinson.50 These researchers surveyed 2340 academic members of the University and College Union (UCU), the main union for academic staff in Britain and an even larger complement of European scholars. The UCU survey focused on threats to academic freedom, germane to my study here, and principally on internal threats to academic freedom arising from managerial favoritism, unspecified forms of discrimination such as methodological tribalism, and imperatives flowing from research and teaching assessment exercises.

Qualitative comments gave insight into a series of “invisible” threats to academic freedom that are rarely identified. The most widespread was psychological bullying from colleagues, reported by 27% of respondents. There were also reports of disciplinary actions from department heads or managers. Such individuals, often senior staff in rotating appointed positions at department or School/Faculty level, hold considerable power over individual careers. Survey respondents mentioned cases of threatened or actual dismissal from employment, increasing an academic’s teaching or administrative load, removing them from a cherished teaching or administrative position, berating them during performance reviews, and withdrawing grant funding. Other instances I am aware of include departmental or higher-level decisions compelling conservative speakers to be balanced by progressives (a requirement not imposed on left-only panels), or repeatedly delaying an event without cancelling it.

These studies have been vital for highlighting dynamics within universities in terms of both hard and soft authoritarianism. Of course, each of these studies depended on participants opting in to take the survey. This creates a potential risk of self-selection bias. Among the surveys presented in this report, that risk also applies to the North American academic survey, the NAS academic survey, and the UK mailout academic survey, but not to the UK YouGov academic survey or the Prolific surveys of American, Canadian, and British PhD students. By comparing these samples, we are able to show that the online mailout samples do not differ systematically from the accidental, non-selective surveys. For more information on survey data and methods, see the Appendix.

In addition, previous surveys rely on participants’ willingness to report their attitudes in a way that is susceptible to social desirability bias. This report is based on a series of studies that address this through the use of a concealed list technique that can reveal the true willingness of academics to engage in political discrimination.

Do academics feel comfortable expressing their views on politically-salient or controversial issues, and have they decided not to express their views in research or teaching because of the risk of penalties? Reported chilling effects are not, in themselves, direct evidence of a problem. Someone’s fear of penalties may be irrational, and should thus be handled through personal resilience rather than external protection.

To address this, I consider a suite of attitudes that affect the social and professional context in which academics work. A key aspect of this is the degree to which the perceived political import of their work may affect their career. To assess such political discrimination, I focus on attitudes towards grant applications, papers submitted for publication, and applications for promotion. Another is the softer but arguably no less consequential issue of workplace social climate. Are academics comfortable interacting in an informal context with colleagues who hold politically-opposing or controversial views?

From Hard to Soft Authoritarianism

Hard authoritarianism, being fired or threatened for one’s views, is a more serious violation of academic freedom than soft authoritarianism, not being hired, promoted, awarded a grant, or published in a journal. But both matter for academic freedom. Active social bullying is more punishing than social ostracism, which is in turn worse than socially avoiding someone or not including them in one’s social circle. Even so, there is a sliding scale from hard forms of authoritarianism to softer “everyday” exclusions that can affect a large number of staff. All forms of political sanction matter, however, and ideally would be kept to a minimum given the importance of collegiality to the academic enterprise.

Part I examined both perpetrators’ willingness to endorse hard authoritarianism and victims’ experiences at the sharp end of such measures. Part II considers soft authoritarianism: political discrimination and its impact, in the form of chilling effects, on nonconforming academics. Here Mill’s “despotism of custom” rather than activist bullying or university disciplinary power is what ensures compliance.

This doesn’t mean people should not be allowed to discriminate in their social interactions. There is an important distinction to be drawn between the negative duty not to exclude, i.e., “I won’t associate with conservatives,” which should be unacceptable, and the positive desire to socialize more often with those who share common beliefs or identity, i.e., “I like to associate with neo-Marxists.” Limiting discrimination is not coterminous with positively mandating patterns of socializing, which is contrary to the spirit of free association that is central to liberalism. Something similar holds for inviting speakers to campus, where staff or student interests may organically result in invitations skewed to the left. This is quite different from actively rejecting conservative speakers by stretching the definition of racism to encompass their views and thereby justify excluding them from campus.

In addressing the question of academic freedom, I begin from the standpoint of the victims who experience discrimination or anticipate that they will be penalized for their views. Academics’ decisions to restrict what they teach or research, on the basis of concern about colleagues’ potential discriminatory reactions, is what is often termed a “chilling effect.” This is at the heart of soft authoritarianism, as opposed to hard authoritarianism, which relies on threats of disciplinary action or dismissal.

Is there a rational basis for scholars – notably conservatives or gender-critical feminists – to feel a chilling effect? To answer this question, there is a need to focus on perpetrator attitudes, namely those of the leftist and (to a lesser extent) centrist academic majority toward the conservative minority. If these views are indeed penalizing, then perceptions of a chilling effect are based in reality, and self-censorship becomes a reasonable response to the threat of genuine antipathy and unequal treatment.

The degree of faculty support for dismissing academics who take an unpopular position on a sensitive issue is one barometer of whether those who perceive a chilling effect are correct to do so. This connects hard authoritarianism to the soft authoritarianism of public opinion. The perpetrator test here is whether academics back the idea that someone should lose their job or post, or be subject to an online campaign, for research that is deemed to have contravened a moral norm.

Political Discrimination 

Soft authoritarianism arises from political discrimination. That is, the willingness of perpetrators to penalize someone for their perceived politics. Short of dismissal, this may involve discrimination in hiring, promotion, and refereeing; or social ostracism in collegial spaces. The surveys below test for these forms of discrimination. Though short of dismissal, these less dramatic forms of discrimination may nevertheless prove extremely important in shaping the perceptions of those who might otherwise have chosen to pursue their genuine research interests. In this manner, speech is subject to soft authoritarianism.