Short version below. For the full CV, including teaching, refereeing, complete presentations, and the full list of awards, download the PDF. Click “Abstract” under a paper to read its abstract.

Download full CV (PDF)

Publication

• Passive Inequality and the Dilemma of Meritocracy (with Timo Freyer).
  Experimental Economics, 28(2), March 2025, 379–400.
  Abstract

  In meritocratic societies, inequality is considered just if it reflects factors within but not outside individuals’ control. However, individuals often benefit differentially from other people’s efforts. Such passive inequality is simultaneously just and unjust by meritocratic standards, confronting meritocrats with a dilemma. We conducted an experiment with a representative US sample to investigate how people deal with this dilemma. In the experiment, impartial spectators redistribute payments between pairs of individuals. We vary whether initial payments result from luck or effort and whether spectators redistribute between individuals who worked themselves or individuals who benefited from the work of real-life friends. We find that spectators treat inequality based on the efforts of individuals’ friends as if individuals had worked themselves, and very different from inequality resulting from differential luck. This indicates that most people accept inequality if it is merited at some stage, which may explain opposition against redistributive policies.

Working papers

• Political Representation Gaps and Populism
  Revise & Resubmit, American Political Science Review
  Abstract

  Populists claim to fill political representation gaps: differences between the positions of established parties and the positions of ordinary people. I ask whether such gaps exist and how they relate to populist support. To this end, I compare the positions of citizens and parliamentarians in 26 European countries on identical survey questions asked to both groups. A German time series from 2009 to 2025 lets me follow the gap as parties repositioned. On economic questions, gaps are small on average and differ in sign across countries. On cultural questions, immigration above all, the gap is large: in the average country, the median parliamentarian is about one standard deviation more culturally liberal than the median voter. This cultural gap is systematic in four ways. It appears on nearly every non-economic issue, in every country in the sample, and for most established parties relative to their national mean voter, and every major demographic group, immigrants included, tends to be more conservative than its parliamentarians. Right-wing populist parties fill this gap. Citizens who face larger cultural gaps are more likely to support right-wing populists, but not populists of other kinds, and the cultural gap predicts the vote share of right-wing populists where the economic gap predicts it much less.

• From Weimar to Today: Mapping Populism in German Parliaments (with Paul Behler)
  Abstract

  The rise of populist parties is often compared with the collapse of the Weimar Republic, but populism has rarely been measured over a long enough period to make the comparison. We analyse every speech in the Weimar Reichstag from 1919 to 1933 and in the Bundestag of reunified Germany from 1991 to 2018. To measure populism, we tailored and validated a machine-learning model that scores each speech on the two components of populism: people-centrism, the appeal to a virtuous ordinary people, and anti-elitism, the attack on a corrupt elite. In both republics, populism is similarly common, is similarly distributed across the ideological spectrum, and rises modestly over time. In both, left-wing parties were initially the most populist and were later overtaken by right-wing parties. The form of populism differs. In Weimar, the increase came from rising anti-elitism among right-wing parties; in the Federal Republic, it came from a general rise in people-centrism across parties.

• Information Interventions Can Increase Opposition to Immigration
  Abstract

  An influential literature finds that people hold beliefs about immigrants that are too negative, for instance by overestimating their number, and that correcting these misperceptions makes attitudes toward immigration more favourable. I ask whether this finding depends on how immigrants, and beliefs about them, are usually measured. In a pre-registered survey and experiment with 2,104 Germans, I focus on the immigrant group most present in public debate, asylum seekers. I elicit beliefs about their number the way the media report it, in absolute terms rather than as a share of the population. Attitudes toward asylum seekers predict voting for populist anti-immigration parties better than attitudes toward other immigrants, and while most respondents oppose the arrival of asylum seekers, they do not oppose immigration by other groups. Two established findings then reverse. Respondents overestimate the share of immigrants in the population, as in earlier studies, but they underestimate the number of asylum seekers. In an information experiment, correcting this underestimate does not reduce opposition to immigration but, if anything, increases it. The reversal follows from two measurement choices: which group of immigrants beliefs are elicited about, and whether the quantity is asked as a share or as a count.

• Do Mainstream Party Positions Cause Far-Right Support? (with Salvatore Nunnari)
  Abstract

  Right-wing populist parties have grown across Western democracies. One explanation holds that mainstream parties took positions more liberal than those of their voters and left restrictive voters unrepresented. Whether such a gap causes populist support, or merely accompanies it, is hard to tell from observational data, because party positions and voter support are determined jointly. We test the positional channel in a pre-registered information experiment with 5,040 Germans, fielded before the 2025 federal election. Respondents were randomly given information about the immigration position of the main centre-right party, the CDU. The information shifts what respondents believe about the party’s position, without changing their own immigration attitudes, their beliefs about other parties, their coalition expectations, or how important they consider the issue. Political responses differ by group. Among voters with restrictive immigration preferences, perceiving the CDU as more liberal raises support for the populist-right AfD and lowers support for the CDU in stated vote intentions; incentivised behavioural measures point the same way, but those estimates are imprecise. Liberal and moderate voters show no comparable response, and the feeling of being represented moves in opposite directions for liberal and conservative voters. The population average effect is therefore smaller than the effect among voters with restrictive preferences. Where a mainstream party positions itself matters for populist support through whom it leaves represented.

• Would Europeans Accept Immigrants If They Knew Them?
  Abstract

  Between January 2015 and December 2016, the member states of the European Union received more than 2.5 million first-time asylum applications. Across the 15 European destination countries studied here, the share of applicants recognised at first instance ranged from 18 percent in Hungary to 88 percent in Switzerland. These decisions were taken by officials applying the law, not by voters. I ask whether citizens, faced with the actual applicants, would have admitted the same people. To answer this, I use an existing vignette experiment fielded in the 15 destinations, in which about 18,000 respondents accepted or rejected hypothetical asylum seekers with randomly varied characteristics. I combine it with Eurostat data on the people who applied in 2015 and 2016. Reweighting each respondent’s decisions to the real mix of applicants in his own country gives the share of the actual applicant pool that citizens would have accepted. In Germany, the largest recipient, citizens would on average have admitted 53 percent of one group of young Syrian men whom the administration recognised at 97 percent. Across the 15 destinations, average citizen acceptance was below administrative recognition in 11. Applying a majority rule to each applicant profile implies 18 percent fewer admissions than the administrative recognition rates produced. Recognition rates also vary about four times as much across destinations as citizen preferences do. Elected governments sustain asylum institutions whose decisions differ from what citizen majorities would decide.

Work in progress

• Mainstream Party Responses to Populist Pressure on Immigration (with Hasin Yousaf)
  Abstract

  Populist parties have grown across Europe, and mainstream parties have responded by accommodating populist demands on immigration. Yet populist parties keep rising, which suggests that accommodation does not work. We ask to what extent mainstream parties accommodated populist demands at each level of politics, and how voters respond to the different types of accommodation. We measure accommodation at four levels, first for Germany and then for other European countries: the rhetoric in the parties’ press releases, the positions in their election programmes across election cycles, the laws that were actually enacted, and the enforcement of those laws as described in the reports of asylum lawyers and practitioners. Language models score all four levels on a common scale built from a set of concrete populist demands, so that accommodation in words can be set directly against accommodation in positions, in law, and in practice. Accommodation has been largely rhetorical: party positions, enacted policies, and their enforcement have moved much less than the rhetoric. This can explain why voters continue to vote for populist parties despite the accommodation they hear. A survey experiment in Germany then tests how voters respond to the different types of accommodation: to restrictive statements, to proposed or enacted laws, and to enforcement outcomes such as deportations carried out.

• Trade-off Rhetoric in Political and Technocratic Speech (with Guido Tabellini and Frederik Schwerter)
  Abstract

  Every policy has costs. Whether the people who advocate a policy name them differs across speakers, and this project asks whether central bankers acknowledge the trade-offs of the policies they advocate more often than elected politicians, and if so, why. We collect speeches by both groups in Europe and the United States. The European comparison sets 547 speeches of the European Central Bank from 2018 to 2023 against the 73 responses of the party-group leaders in the European Parliament’s annual State of the European Union debate. The American comparison sets 600 speeches of the Federal Reserve against 400 State of the State addresses of governors, from 1995 to 2023. Language models extract every policy a speaker discusses and record the benefits and the costs the speaker names for it, so that a policy counts as presented with a trade-off only when both appear. In the automated coding, central bankers state trade-offs several times as often as elected politicians. Under controls for the length and the topic of the speech, the year, and the number of policies a speech covers, the gap narrows by about half and remains large. A hand-coded check of 20 speeches under a written coding manual reproduces the direction of the gap at much lower levels for both groups; reconciling the two measures is part of the ongoing work. The second part of the project asks why politicians leave the costs out. One hypothesis is that balanced speeches, which name both the benefits and the costs of a policy, are less persuasive to voters than one-sided speeches, so that speakers who face elections have a reason to omit the costs. A survey experiment measures how persuasive balanced and one-sided speeches are.

• Silicon Politicians: Digital Twins of Swiss Politicians and Voters (with Ray Duch)
  Abstract

  Can we build digital twins of politicians that behave like the real ones? This pre-registered study builds a digital twin of each member of the Swiss parliament from nothing but the politician’s social-media record, and a matching set of twins that represent the Swiss population, following the digital-interview method that Cerina and Duch developed for ordinary citizens. The test is behaviour. The twins vote on ten measures that go to a popular vote in Switzerland in 2026, and we compare their votes with the recorded roll-call votes of the parliamentarians and with the real ballot results. The predictions for the popular votes of 27 September and 29 November 2026 are registered before those votes are held, so they are forecasts of behaviour that has not yet happened rather than fits to known outcomes. Accuracy is scored against a baseline that always predicts the majority, and we test whether a richer social-media record produces a more accurate twin. If twins built this cheaply reproduce the behaviour of real politicians and voters, the positions of both, and the gap between them, can be tracked continuously instead of once per survey wave.

• Court Rulings and Asylum Flows
  Abstract

  Court rulings on asylum are natural experiments in migration policy. A ruling arrives on a single known date, is not chosen by the legislature, and changes what future applicants can expect: the German Federal Constitutional Court’s decision of 18 July 2012 raised the benefits asylum seekers receive, and French rulings in 2010 and 2012 removed countries from the list of safe countries of origin. This project uses such rulings to estimate how far a change in expected treatment alone moves the number of asylum applications and the countries they come from. The design compares the monthly application series in the months after a ruling with the months before it and with the counterfactual path that a forecast model trained on the preceding years predicts. The same event window at placebo dates and synthetic-control comparisons with countries the ruling did not affect serve as checks, and a restrictive German reform of 2014 provides a case in which applications should move in the opposite direction. A registry of 27 asylum-related court decisions from 14 jurisdictions provides the set of candidate cases.

• Digital Twins of the United States Supreme Court (with Ray Duch)
  Abstract

  Can we build digital twins of judges that predict their actual decisions, and do they beat the existing forecasts? How the United States Supreme Court will decide a pending case is already forecast by panels of legal experts, prediction markets, a public forecasting crowd, and machine-learning models trained on case characteristics. The crowd has averaged about 82 percent correct over five terms, the best machine-learning model 74 percent. We build a digital twin of each sitting justice from the justice’s own record and let the twins vote on cases decided after the language model’s training cutoff, so the test is out of sample. Their accuracy is compared with the existing benchmarks at the level of each justice’s vote and of the case outcome. A second part of the project turns the twins into a tool. For cases whose briefs are filed but not yet argued, we render systematic variations of the arguments in the briefs and let the twins rate each variant, and we register before oral argument which arguments the twins predict will move which justice. The longer-run aim is to ask how the reasoning of a court relates to the preferences of the public it decides for.

• The Political Direction of Published Social Science: Evidence, Questions, or Framing?
  Abstract

  What leading social-science journals publish has become more political. This project asks in which direction it has moved, to the left or to the right, and why. Three explanations are possible: the evidence itself has changed, researchers ask different questions, or the same results are framed differently. I code 3,830 primary research questions from abstracts published in 13 leading journals in economics, political science, sociology, and psychology. For each abstract, language models extract the research question, judge whether it is politically relevant, and if so place the question, the reported result, and the framing of the result on a left-to-right scale, with independent raters and double coding as checks. The change in the political position of published findings over three decades is then decomposed into the three channels: the part that comes from the results themselves, the part that comes from which questions get asked, and the part that comes from how results are framed. The design separates a literature that has followed the evidence from one whose questions or framing have shifted.

• Selecting Misperceptions: Which Immigration Statistics Do People Care About? (with Frieder Neunhoeffer)
  Abstract

  A large literature corrects people’s misperceptions about immigration and measures how their attitudes change. The statistics it corrects are chosen by researchers. We review 54 publications that elicit quantitative beliefs about immigration, covering 322 study-country-statistic observations. Of those, 80 percent concern a single question, how many immigrants there are, and 91 percent concern immigrants as a whole rather than any subgroup. Crime, welfare use, and cultural integration together account for under 20 percent. We run three studies in the United States and Germany, each pre-registered before fielding. In the first, an AI interviewer asks ordinary citizens in open-ended interviews, with follow-up questions to each answer, which facts about immigration matter to them, and we compare their answers with the statistics used in the literature. In the second, a survey measures beliefs about 100 statistics, those the literature uses, those citizens named, and items that bridge the two, along with how important respondents consider each one. In the third, an information experiment corrects the statistics citizens named as important and compares the attitude effects with corrections of the statistics the literature has used.

Positions

• Aug 2026 — Visiting Scholar, Oxford University (Nuffield College)
• Feb 2026 — Visiting Scholar, Stanford University (SIEPR)
• 2025–present — Research Fellow, IAST–Toulouse School of Economics
• 2023–present — Fellow, Institute for European Policymaking, Bocconi
• 2023–2025 — Postdoctoral Researcher, IGIER–Bocconi University

Education

• 2023 — Ph.D. in Economics, Bonn Graduate School of Economics
• 2019 — M.Sc. Economics, University of Bonn
• 2017 — B.Sc. Economics, University of Bonn

Selected awards

• 2025 — Top 40 under 40, Capital magazine
• 2021–2022 — Econtribute research grant