When political pressure trumps academic freedom

Academic freedom has been declining over the past decade across the world. This column analyses over 25,000 economics working papers to show that, beyond direct intervention by authorities, political pressure also breeds self-censorship. After US federal agencies began screening research for a list of flagged words in early 2025, economists at universities most dependent on federal funding reduced politically sensitive content – even in research without any direct federal funding.

When the Academic Freedom Index Update 2026 (Kinzelbach et al. 2026) was recently published, scholars worldwide were unsurprised but nonetheless saddened by its findings. In line with anecdotal evidence, the data confirm that between 2015 and 2025, academic freedom substantially declined in 50 countries, including heavyweights such as the US, India, Indonesia, Germany, Türkiye, and Russia, while increasing in only nine, mostly small, nations. One manifestation of this pressure on scholarly liberty is the growing denial of empirical evidence on key issues such as climate change and vaccination by political authorities. When political dogma trumps scientific evidence, one is reminded of medieval times, when Galileo Galilei’s discoveries were met with fierce repression by the Inquisition.

Beyond denying scientific findings, governments can reshape the research agenda itself, by signalling which topics are welcome and which are not. Following a series of executive orders in January 2025, US federal agencies started screening research grants, and it was soon reported that agencies such as the National Science Foundation (NSF) were relying on lists of flagged words, including “women,” “gender,” “racial,” “discrimination,” and “pollution” (Yourish et al. 2025). The media started referring to a list of “banned words.” There was never a strict ban: using these words did not trigger automatic cancellations. But the perception of a blacklist was widespread, and there were real direct interventions. By January 2026, nearly 2,000 NSF grants had been cancelled or suspended (Kozlov et al. 2026).

There is, however, a subtler and potentially even more consequential question: do researchers self-censor when they feel that studying certain topics may harm their funding and career prospects? Václav Havel, the former Czechoslovak dissident and later Czech president, illustrated this dilemma in his 1978 essay “The Power of the Powerless” with the story of a greengrocer who displays the slogan “Workers of the world, unite!” in his shop window for the sole purpose of avoiding trouble (Havel 1978). If enough scholars adopt the greengrocer’s attitude, political pressure reshapes not just what gets funded, but what gets studied in the first place.

Our new study (Rohner et al. 2026) answers this question with current data from economics. We analyse 26,676 working papers posted between January 2020 and mid-2026 by NBER, CEPR, and on arXiv, and use a large language model to classify every occurrence of a flagged term by its context: distinguishing, say, a “race to the bottom” from research on racial discrimination. Our measure of politically sensitive content is whether a paper uses flagged terms in a gender, race, or environment context. We then compare researchers at US universities that rely heavily on federal research funding (“High-Fed,” above the median share of federal funds in university R&D expenditure) with two groups that are less exposed to US federal pressure: US universities less dependent on federal money (“Low-Fed”) and UK universities. Because economists circulate working papers before peer review, we can observe researcher behaviour immediately, and separately from the publication and grant-review process.

After the January 2025 policy change, the probability that a High-Fed paper contains flagged words in a sensitive context dropped by 3.4 percentage points relative to Low-Fed papers, and the gap widens to eight percentage points in 2026. A pooled difference-in-differences estimate confirms the decline (3.9 percentage points). The same pattern appears when High-Fed papers are instead compared with papers from UK universities; Low-Fed papers, by contrast, show no change relative to the UK, consistent with the response being concentrated where exposure to federal funding is highest.

These comparisons are across papers; the picture holds when we instead follow individual researchers over time. Comparing the same scholar before and after the policy change, US-based economists reduced their use of flagged words in sensitive contexts relative to UK-based colleagues. The declines are strongest when exposure is measured by universities’ reliance on NSF funding specifically, the agency most directly implicated in the keyword screening: 3.9 percentage points within the US and 6.5 percentage points relative to the UK. Moreover, the response appears to attach to the research team: the same author writes less sensitive content when publishing with a team based mostly at High-Fed institutions.

Crucially, this is not a mechanical effect of grant rules. Only 6.5% of US papers in our sample acknowledge NSF funding (17% acknowledge any federal agency), and the decline is just as strong when all papers acknowledging federal funding are excluded. Since the government has no direct control over working paper content, the sudden response, including in research without any federal money, points to self-censorship: researchers anticipated how their current work might affect future funding and career prospects, and adjusted accordingly.

What exactly changed? The decline is concentrated in the content most closely associated with what the administration characterises as “woke”: it is strongest for gender- and race-related uses. When we simply count flagged words irrespective of context, there is no differential change: the effect is confined to politically sensitive uses. Nor is it mere wordsmithing. Annotating each sensitive paper by whether its core research question is about gender, race, or the environment, we find that about two-thirds of the decline comes from papers whose research question itself concerns these topics: papers that could hardly describe their subject without the flagged vocabulary. The response is therefore unlikely to be pure lexical substitution: at least in part, fewer papers on these topics are being written.

The effect is remarkably broad-based: male and female authors, and authors across predicted ethnic backgrounds, responded to statistically indistinguishable degrees. Nor did scholars at highly exposed institutions publish less or shorten their papers: there is no High-Fed-specific drop in research output. US research output as a whole did, however, decline relative to the UK after the policy change, with the larger declines in papers with gender-, race-, or environment-related content.

Our findings take place in a context where democratic institutions more broadly are under strain in many countries around the world. The stakes are high: Democracy has been found in recent studies to foster peace (Rohner 2024) and prosperity (Acemoglu et al. 2019), while in contrast populist leaders on average lead to a drop of GDP by ten percentage points (Funke et al.2023). Populists around the world do not only target the independence of universities, but also that of judges, law firms, and media outlets. Together, these pillars of society constitute a crucial bulwark against the slippery slope toward authoritarianism. We must invest in strengthening the independence and freedom of all these vital checks and balances that safeguard the survival of democracy.

Source : VOXeu