Dictators, Democracies and Discoveries: Political Institutions and the Creation of Knowledge

Sebastian Hager, Felix Radde, Carlo Schwarz and Fabian Waldinger
Presented by Fabian Waldinger at the nber Economics of Talent Conference, Cambridge ma, 2 October 2026 (video). No discussant.
Source note: this piece is written from the talk alone. No public paper was available (an ssrn version dated August 2026 exists but could not be obtained), and the stream never showed his slides, so it has no figures and every number below is one the speaker said aloud.

The planner’s case and the republic’s case

There are two old arguments about whether a dictator is good for science, and they are really the two sides of the socialist calculation debate, moved over from the economy to the laboratory. The planner’s side says frontier science is a coordination problem, and a state that can point resources at strategic targets without asking anyone will hit them faster; Waldinger’s examples are Chinese ai research and the Soviet space program. The other side is Hayek’s knowledge problem in a lab coat. Popper, Hayek and Merton made it after the Second World War, and Michael Polanyi gave it its best form in “The Republic of Science” (1962): nobody, least of all a ministry, knows where the next result will come from, so you get more science by letting researchers wander and pick fights with the reigning paradigm.

The evidence used to be anecdotes, and they point both ways. On one side are Galileo’s trial, the Nazi dismissals of 1933 and Erdoğan’s purge of Turkish universities after the failed coup of 2016. On the other is the fact he opens with: in the Nature Index of top publications the us share has sat at about 20% for ten to fifteen years and the eu’s at 12–13%, while China’s has risen from about 7 or 8% to about 22%, overtaking the United States. So “autocracies can’t do science” has a China problem, and the paper tries to settle the question with a century of data instead of a few vivid cases.

The data, which is most of the work

The unit is a country every five years, from 1900 to today, and the hard part is the denominator: nobody had a country-level count of universities for the whole twentieth century, so they built one. The early part is their World of Academia Database (now in, as he put it, “year 14 getting into year 15”), hand-typed from a pre-internet directory of faculty rosters (the captions garble its name; probably Minerva): six cohorts from 1900 to 1969, about half a million person observations. The directory’s compilers gave up in the late 1960s because the sector had grown too big to list, which is a decent growth statistic in itself. After that come Unesco’s World List of Universities and the World Higher Education Database, which give counts but no names.

Output comes from OpenAlex (all publications, and the top 10%, 5% and 1% by citations within discipline and year), from a Nobel measure dated and located by where and when the prize-winning work was done, and from a dataset of technological discoveries from Andrei Shleifer and co-authors, who used a large-language-model agent to pull first-of-their-kind discoveries out of English Wikipedia. Institutions are measured by V-Dem’s Liberal Democracy Index (ldi), which runs from about zero (Nazi Germany) to about 0.8 for the most liberal countries. Unlike Polity, it scores colonies, so India has its own score before independence.

Three rungs of identification, none claimed as the top

The first rung is two-way fixed effects with population and, optionally, gdp per capita, which he flags as “an endogenous control.” The outcome is the inverse hyperbolic sine of output, because of zeros; logs give the same answer. The headline: a 0.1 increase in the ldi goes with about 13% more publications, and with about 21% more Nobel-prize research, both within-country. For scale, he offered that “the Trump effect in the last 3 years is basically” 0.2 of the ldi (an aside, not an estimate). Top-10% and top-1% papers show “very large” effects, with no numbers given. The size of the academic sector moves too: more universities and, up to 1969, more academics. (He gives a 24% figure while discussing the universities table, but the captions have him saying “publications,” so which outcome it belongs to is unclear.)

The second rung is event studies around sharp institutional breaks, defined as the ldi moving by more than 0.2 between periods. There are 70 positive and 28 negative ones. His examples: Poland at the fall of the Iron Curtain (about 0.3 to 0.8), Tunisia in the Arab Spring (about 0.1 to 0.6), and Weimar Germany sliding from about 0.5 to roughly zero under the Nazis. The identifying claim is that most breaks are coups or revolutions from below that don’t go after the universities first, and the pre-trends are close to flat. After a positive shock, publication counts rise by something like 81% (the captions say “25 years up,” so it is probably the effect about 25 years out, though whether that is the peak is not stated). Top-5% papers rise too, and he said he was “really amazed at how good the event studies look,” with no matching or other treatment of the data. Negative shocks give a noisier mirror image with some delay, and the full set of heterogeneity-robust staggered difference-in-differences estimators leaves the shape “not at all affected.”

The third rung instruments own-country change with regional waves of democratization among neighbors, a design he credits to a name the captions garble as “Achiladal” (probably Acemoglu and co-authors). The first stage is strong and the results “very very similar,” he says, but he did not show them and conceded the exclusion restriction is a worry.

The surprising part: headcount explains the volume, not the stars

Democracies have more universities and more academics, so maybe they publish more simply because more people are writing. To check, the authors control for the number of authors in the publication data. For all publications this “takes away a lot of that coefficient,” so raw volume is “predominantly driven by size.” For highly cited papers the coefficient shrinks much less. And for Nobel-prize research there is “basically almost no change.”

So the 21% Nobel effect is not about buying more lottery tickets. On his reading, democracies make the best scientists more productive or attract them from elsewhere, and the talk cannot yet tell which. Volume you can get by hiring; the top of the distribution responds to something else.

The breadth result is the direct test of Polanyi (1962). Take OpenAlex’s 256 fields and compute the share of a country’s output in its 25 largest, about 10% of fields (the 25 can differ by country and year). For all research, once country fixed effects are in, institutional change is unrelated to concentration. For top-5% and top-1% research, more democratic countries are less concentrated. That is a within-country result on changes; his gloss in levels (the us excellent across many fields, China in a narrower set) is about top-1% work only. Autocracies can be very good at a few things; what they lack, at the frontier, is the wide search.

The last part answers the objection that autocracies do their best work where journals can’t see it. More democratic countries also produce more of the Wikipedia discoveries. By producer, university discoveries (crispr-Cas9, the first heart transplant) look like the publication results, government discoveries (Nasa’s moon landing) are positive, and industry is noisy with a coefficient similar to universities’. Military discoveries are the one category where democracies produce fewer, which is the sign you would expect, though he stressed it is not significant. Independent tinkerers like the Wright brothers don’t respond to institutions at all. (The authors added a de-duplication pass of their own, because the data still held “three [Volkswagen] Beetles and four V2 rockets.”)

What the room pushed on

Questions came throughout. Does every country’s bomb count? No, only the first atomic bomb; where “new” begins for planes and cars is left to the dataset’s builders. Does China drive the unweighted regressions? Dropping one continent at a time leaves the results “very very similar.” On income, a questioner suggested that China’s story is growth driving both science and institutions. Waldinger allowed that “here it has a bit of a bigger effect, it depends on the specification” (with no slide on screen, it is not clear what “here” referred to) but said that overall the result “doesn’t seem to be driven by these kind of changes in income,” and he would not lean hard on an endogenous control in either direction. He then joked that his PhD adviser (the captions’ “Steve Fishisk,” presumably Steve Pischke) would disown him for running cross-country regressions at all.

The sharpest question was about reverse causality inside the regressor: when Orbán goes after the universities, Hungary’s ldi falls because he went after the universities. Waldinger accepted part of it. Regimes usually “first do other stuff” before turning on academia, and most of the sample’s shocks are coups not aimed at academia, but “there is a little bit of reverse causality concern that we can’t even with these event studies address.” On citations, a questioner worried that the world simply starts citing a country more once it liberalizes. He shares the worry and points to the Nobel measure as a partial, indirect answer. (Fair enough, though a Nobel is also somebody’s judgment, just a much more expensive one.) They considered counting only home-country citations but dropped the idea, because insular regimes would then just cite themselves. On whether today’s Hungary and Turkey use science differently from the Soviet Union or China: “honestly I don’t know.” Whether the effects come from researchers trained under a regime or from senior people already in place, which is also how you would separate productivity from sorting, awaits an individual-level follow-up on migration, held up for now by OpenAlex author identifiers that break down when people move countries. Asked whether small research sectors are concentrated simply because they are small, he reframed it as a “size interaction” and answered, “Interesting… well, no, possibly.”

He closed with Popper (1945), from The Open Society and Its Enemies: science grows by mutual criticism, so “the only possible way of planning its growth is to develop those institutions that safeguard the freedom of this criticism.” It suits a talk whose cleanest finding is that the Nobel effect does not run through headcount, which is the one input a planner can most easily buy. And by his estimates, the only inventors no regime seems to touch are the ones working in their garages.

References