Research Seminars

This paper investigates whether tax evasion can be beneficial for an optimal income tax schedule. Past theoretical discussions have presented mixed outcomes as to whether allowing taxpayers to opt into uncertainty could indeed enhance overall tax revenues. In this study, we conducted an original real effort experiment in an online labor market with almost 1,000 participants to test this hypothesis empirically. Our findings show significant positive labor supply responses to the opportunity to evade (increased labor supply by 37%). More importantly, the expected tax revenue significantly and substantially increased by up to more than 50%. Strikingly, this effect persists when comparing effective tax rates: Lowering effective tax rates through probabilistic enforcement (the opportunity to evade) is more efficient than simply lowering statutory tax rates. Our findings suggest that the opportunity for tax evasion can increase tax revenues beyond what a corresponding decrease in nominal rates would achieve.

Exploiting the 1848 revolution in the Austrian Empire, I study how a sudden institutional change impacts economic development. In contrast to much existing research, which focuses on the rule of law and the security of property rights, I argue that the main impact occurred through the expansion of the upper-tail useful human capital (Mokyr, 2005) following the revolution. While the revolution was defeated, important liberalizing reforms persisted in the areas of (i) the press, (ii) technical education and (iii) knowledge-promoting voluntary associations – the information space, for short. Pushing back against the existing consensus that the revolution had no discernible impact on the empire’s economic development, I combine newly collected data on the pre-1848 and post-1848 information space with data on the use and location of steam engines in 1841–1863 to establish the causal link between the post-1848 reforms and the diffusion of this important industrial technology.

While risky choice under time pressure is a topic that has attracted much attention in the literature, the effects of time pressure on risk attitudes seem to be poorly understood. In this paper we present evidence from an economic experiment where participants have to allocate an experimental income to two risky assets, with and without time pressure. We assume Rank Dependent Utility preferences, where risk attitudes are captured by both the utility curvature and the probability weighting function. Using the allocation data, we fit structural models which allow us to disentangle the effect of time pressure on the two components of risk attitudes. We then correlate the risk attitudes to various demographic and personality traits.

In this paper, we estimate the rise in smoking prevalence in East Germany caused by exposure to West German TV, with a focus on gender differences. We leverage survey data on smoking prevalence from 1989 (before the fall of the Berlin Wall) and 2002 (twelve years following reunification) and exploit exogenous variation in exposure to the West German TV signal to identify the causal effect. We document that exposure to West German TV increases smoking prevalence, but this rise is concentrated entirely among women, with a negligible impact on men. Our estimates imply a 10.5 percentage point increase in the probability of smoking among women and a 70.7% increase in the number of cigarettes smoked. These findings are robust across various tests and suggest that the convergence in smoking rates between men and women will likely lead to an increase in women's mortality and healthcare costs.

This paper provides a comprehensive assessment of the impact of education finance on students’ long-term outcomes and the mechanisms through which these effects operate. For identification, I exploit an intergovernmental transfer reform in Norway, which generated exogenous variation in school funding based on the school-aged demographic composition at the local level in the mid1980s. The main takeaway is that exposure to an additional $100 per pupil in education funding over nine years of primary and lower-secondary school leads to nearly $250 in higher annual earnings. This effect corresponds to an Internal Rate of Return (IRR) greater than 6% and a Marginal Value of Public Funds (MVPF) ranging from 1.5 to 2.5. The increase in earnings is mediated by higher educational attainment, higher likelihood of obtaining a degree in high-paying fields, and improved cognitive abilities. The effect is larger and more significant for students from low-educated parents and those at the lower end of the earnings distribution, while no significant impact is found for students with at least one college-educated parent. At the municipal level, the funding shock led to an increase in teacher hiring without affecting capital expenditures.

Cost of medicines is on an exponential rise and the pressure towards speedy access to medicines comes from all directions: pharmaceutical industry, patients, clinicians, politicians, and ultimately, the public. In Europe, standard reimbursement of a medicine typically requires efficacy approval from the European Medicines Agency (EMA), clinical and cost-effectiveness assessment of a Health Technology Assessment (HTA) body, and decision by a national regulator. With the intention to provide speedy access, efficacy evidence standards are often lowered in the name of hope that the given medicine may bring about clinically meaningful benefit nonetheless. This shift brings significant uncertainty into clinical and pharmaco-economic assessments, and ultimately into decision-making.
Using the example of a cost-utility analysis of an advance therapy medicinal product (ATMP) for spinal muscular atrophy, I will illustrate a prototypical case of modelling uncertainties. These uncertainties include the use of incomplete data sets, short-term data cut-offs used for long-term extrapolations, or challenges with discounting rates for benefits and costs of chronic conditions. Further challenges include modelling based on data from single arm trials, modelling with numerical values without statistical significance, label approved by EMA for a wider population than included in the clinical trials, or the question of selective approach to the inclusion of indirect costs.
This shift leads to a number of open normative questions such as: who is to pay for the cost of uncertainty? How should we set up willingness-to-pay-thresholds? What will be the role of the new European regulation on HTA (2021/2282)? How to prevent the public from paying thrice: for basic research of medicines, for the actual patented treatment, and for post-market evidence generation?

Previous evidence on the sunk cost fallacy primarily comes from laboratory or online experiments in which the sunk cost consists of a single upfront payment. In this paper, we study sunk cost effects in a highly relevant field context (exercising), where sunk costs may be dispersed over time and influence habit formation. In particular, we investigate how a (random) two-month discount on the membership fee affects the exercising behavior of new members of a large fitness chain and test whether individuals mentally account for the timing of this discount, i.e., the dispersion of sunk costs over time. The experimental design allows us to exclude any selection into discounted memberships and thereby isolate potential sunk cost effects. Despite a highly powered design, we do not find any evidence of sunk cost effects. These results cast doubt on the relevance of the sunk cost fallacy in the health domain. They also suggest that the common practice of offering discounts on membership fees does not undermine habit formation.

Despite the vast medical literature on the health effects of menopause, its economic implications remain understudied. During women's reproductive years, hormones like estrogen are protective of health. Menopause, associated with a gradual yet pronounced decrease in estrogen, induces a distinct change in women’s health trajectory that may have an impact on employment. This paper uses detailed data on the reported natural age of menopause for women in the Health and Retirement Study (HRS) to estimate the impact of menopause on health and employment. To address issues in identification that arise when analyzing menopause, such as confounding factors and measurement error, this paper uses the genetic predisposition for the timing of menopause, specifically the associated polygenic risk score (PGS), as an instrument for the reported age of natural menopause. There are three principal findings. First, consistent with the medical literature, crossing the menopause threshold there is an economically substantive and statistically significant acceleration in health conditions. The decline in health is more than triple that for pre-menopause years. Second, menopause is associated with a substantial reduction in the likelihood of working for pay by just under 2 percentage points every year after menopause, which accumulates to an 18 percentage point reduction in ten years. Combining these estimates, the associated IV estimate of the impact of health on employment indicates that the diagnosis of an additional medical condition reduces the likelihood of working for pay by between 49 to 77 percentage points depending on the exact specification. This is a substantial effect given that 78% of women work for pay prior to menopause. The key take-away is that essentially an additional diagnosis of a medical condition results in exit from employment for middle age to older women.

This study explores how, when income inequality is perceived to arise from both effort and luck, excuses (self-serving belief distortions) can influence acceptance of inequality. In a controlled laboratory setting involving a real-effort task, participants make redistribution decisions between themselves and a partner. The study varied the degree of uncertainty about the role of effort and luck in determining initial earnings endowments. Belief elicitations indicate that increased uncertainty caused participants to be more likely to attribute their partner’s success to luck. Furthermore, the use of excuses (attributing others’ outcomes to luck) was found to reduce willingness to redistribute earnings to their partners. These findings highlight how excuses about the role of luck versus effort may contribute to the persistence of inequality even if many individuals have meritocratic principles, and how variation in the degree of uncertainty about the causes of inequality, across individuals or societies, may contribute to different degrees of biased beliefs and inequality. The paper also shows evidence of excuses in another sense: According to a structural model of fairness views, individuals tend to adopt a fairness view – egalitarian, meritocratic, or libertarian – to justify an allocation that benefits themselves.

Does having children make you lose your mind? This study investigates the causal relationship between the number of children and the risk of dementia among parents, addressing longstanding questions about how parenthood shapes cognitive health in later life. Using comprehensive administrative data for all parents born in Sweden between 1920 and 1950, we derive completed fertility from multi-generational registers and identify dementia onset through in-patient registers and cause-of-death records. To estimate causal effects, we apply instrumental variable (IV), leveraging quasi-experimental variation from offspring sex composition and twin births, and within-family designs in survival models. Baseline associations reveal a U-shaped gradient between the number of children and dementia risk, with significantly higher risks for childless individuals and those with more than three children. However, IV and within-sibling estimates indicate no adverse effects of having multiple children on dementia risk, and suggest that parenthood generally reduces dementia risk. These findings challenge recent studies that report accelerated cognitive decline with higher parity, suggesting that observed risks are driven by confounding factors. Further analyses explore potential mediators, including educational attainment, labor market outcomes, and social/geographical proximity to offspring, highlighting how midlife motherhood penalties attenuate the protective effects of parenthood for women.

We investigate howchoosingone of two products influences beliefs about their quality. Ina laboratory experiment, we deal with the endogeneity in choices by carefully constructinginformation that we provide to participants. This information is both sufficiently clear to allowus to predict choices and sufficiently unclear to leave room for participants to distort their beliefsabout product qualities. Simultaneously, we vary the choice set and whether participants —after viewing both products — can choose a product or simply have one of the two productsassigned to them. We find that choosing to own a product — rather than passively owningit — increases the perceived quality gap between owned and non-owned products (i.e., thechoiceeffect). Thischoiceeffectis driven by non-chosen products. In particular, rejecting aproduct causes it to be perceived as worse than if that same product was simply not assignedto be owned. We show that having participants focus on product qualities before making theirchoice eliminates thechoiceeffect, suggesting that attention is an important driver. Thechoiceeffectexplains several empirical observations and provides support for active choice policies overopt-out defaults.

We investigate how having children impacts the quality of couples’ relationships, a proxy of the non-material gains from being in a relationship. Using a novel measure of relationship quality (RQ), we perform a dynamic difference-in-differences estimation around the birth of the first child. We find a sharp and lasting decrease in RQ immediately after birth. We attribute this effect to changes in household specialization. Traditional gender based specialization prevails after birth, regardless of the baseline distribution of tasks within the couple. Leveraging heterogeneous changes in household specialization after birth, we find that couples undergoing larger rearrangements also suffer larger RQ drops.

The proliferation of standardized testing has raised concerns about its distortionary effects on both school and student behavior. While previous literature has extensively documented high-powered misconduct by teachers and administrators, there is little systematic evidence on how exam pressure may lead students to commit academic dishonesty. This paper investigates interpersonal student cheating during high school exit exams in Vietnam, a setting where educational success is highly coveted. Using individual-level data from a large province, I leverage the quasi-random assignment of students to test rooms to estimate peer effects on test day. I find that students from low-ranked schools performed better when taking exams with students from elite schools. High-achieving elite peers are particularly valuable. However, the gains were concentrated in multiple-choice and quantitative tests but absent in essay exams. Moreover, the positive effects virtually disappeared after a testing overhaul increased the stakes of the exams. Backed by institutional details, these suspicious patterns provide credible evidence that discreet interpersonal cheating was once prevalent: non-elite students quickly formed networks with elite peers to cheat for their own benefit. It took a major reform to reshape student incentives and eliminate this malpractice

This study examines how parental religiosity influences children’s education. Using Turkish census data for primary-school completion outcomes of individuals born in 1924-1984 and withinyear variations in fasting duration across provinces as a proxy, I find that a 30-minute increase in daily fasting during the enrollment year reduces primary school completion by 0.37 to 0.80 percentage points, with stronger effects for females. These results are not driven by income or teachers’ beliefs. Cumulative exposure to Ramadan before school enrollment confirms these findings. Parental religiosity has consequences for gendered outcomes, including fertility and labor market participation.

How do ties with elite peers affect the social mobility of non-elites who have historicallyfaced social exclusion? Using novel data on high school and university graduates in five colo-nial Indian provinces between 1894 and 1921, I examine the effects of elite peers, defined asupper-caste students, on non-elites in high schools and colleges. Exploiting the plausibly ran-dom variation in the share of elite peers across all graduating cohorts within the same schools,I find that exposure to more elite peers reduces the probability that non-elite graduates, par-ticularly those from merchant castes, complete university or become lawyers. These effectsare driven by social rank rather than economic differences between elites and non-elites. Thenegative effect is strongest in private schools run by local Indian elites and among college stu-dents graduating with the highest grades in their high school examinations. Overall, the resultssuggest that exposure to elite peers in settings with significant social distance between elitesand non-elites may hinder rather than foster upward mobility among non-elites.

The declining share of the working-age population (20-64) has led many countries to introduce social security reforms to extend the working lives of older employees. While reforms such as raising the early retirement age can effectively achieve this goal, they are often perceived as forced. Partial retirement schemes, which allow individuals to work part-time while receiving a significant portion of their previous wages, offer a more flexible alternative. However, the impact of these schemes on overall labor supply remains ambiguous. On one hand, partial retirement may increase labor supply by encouraging part-time work over early retirement; on the other, it may reduce labor supply if full-time workers choose to shift to part-time work. This paper investigates these effects by studying the introduction of a partial retirement scheme for central government employees aged 61-65 in Sweden. The findings show a 6.5%drop in average earnings, suggesting that generous partial retirement terms, which replace a substantial share of prior income, incentivize a shift from full-time to part-time work, thereby reducing the overall labor supply.

Guilt Aversion and Inequity Aversion are pivotal concepts in understanding human behavior in situations involving trust dynamics. Inequity Aversion explains trustworthiness through a preference for fairness, as trustworthiness leads to more equitable distributions. Conversely, Guilt Aversion posits that people act to avoid the guilt associated with betraying others. As both preferences provide justifications for trustworthiness, distinguishing between them solely through observed behavior poses a significant challenge. In this work, I aim to disentangle the effects of Guilt and Inequity Aversion to identify the main driver of pro-social behavior in a theory-driven experiment based on the trust game. I show theoretically that by increasing the stakes for the trustor, the two preferences have opposite predictions on trustworthiness. The experimental design, informed by this theory, features a doubling of the trustor’s payoffs. The results indicate that the preferences for equality are the main determinants of trustworthiness.

A growing body of financial research shows that machine learning–based algorithms consistently outperform traditional financial benchmark models, often yielding significant statistical and economic advantages. However, the prevailing hypothesis within this literature is that these algorithms can replace—rather than complement—fundamental financial analysis. In this research, we explore how machine learning methods can enhance standard quantitative financial models, including well-known capital asset pricing frameworks such as CAPM, FF3, and FF5. In particular, we apply Elastic Net and Random Forest methods to statistically calibrate these financial models, using the traditional OLS approach as a benchmark, for monthly data of S&P sectors from 2000 to 2024. Surprisingly, our findings indicate that out-of-sample predictive performance is largely similar across most statistical methods, effectively ruling out Random Forest as a compelling alternative. Notably, the traditional CAPM with linear regression performs quite well and, given its low computational demands, remains a cost-effective forecasting tool. In conclusion, given the random-walk nature of financial returns, allowing for nonlinearity and interactions does not notably enhance predictive performance.

Extant literature proposes and finds empirical support for both a positive and a negative relationship between market integration and pro-sociality. According to a first strand of literature, market interactions help develop generalized pro-sociality, which extends to complete strangers the sense of particularized pro-sociality that is normally reserved to one’s family, clan, or kins in traditional societies. A second strand of theoretical literature posits that market interactions erode altruism. We test the relationship between market integration and tolerance of inequality – one relevant aspect of pro-sociality – in 30 villages of Bougainville, Papua New Guinea. Since this society is transitioning from self-subsistence to market integration, it provides a suitable setting to test the above hypothesis. We measure tolerance of inequality through the “power-to-take” game, in which two players with different initial allocations make a proposal over how to divide the sum of initial earnings. We measure market integration through the share of calories coming from market purchases vis-à-vis self-subsistence.

Overall, we find no significant relationship between market integration and tolerance of inequality. Rather, behaviour seems highly variable in each village groups. We find propensity for either egalitarian or non-egalitarian divisions both in relatively highly integrated and little integrated villages. Nonetheless, greater market integration seems to affect social norms justifying inequality and other mechanisms supporting cooperation.

International trade flows show strong persistence over time. This is true for yearly data and even more so for higher-frequency data such as monthly data. Standard gravity theory cannot explain the persistence, i.e., why lagged trade flows should enter as an explanatory variable. While some dynamic gravity models have been explored, the dynamics in these models are either driven by country-specific factors (such as capital accumulation or technology) or ad hoc (like assuming bilaterally specific capital). We provide a structural dynamic gravity framework where the persistence stems from firms’ sluggish adjustment of destination-specific prices, akin to sticky pricing a la Calvo (1983). Our theoretical framework provides a micro-foundation for a gravity equation with lagged trade flows as an explanatory variable. Using OECD trade data at high and low frequencies, we document the persistence of trade flows and estimate the parameter governing the share of firms that sluggishly adjust prices. Consistent with the literature on nominal rigidities, we find a high degree of price stickiness at monthly frequency and a lower degree at annual frequency. Our results help to explain the propagation of trade cost shocks to trade, prices, and welfare over the short and long run.

Note: This study has been conducted in collaboration with Mario Larch and Dennis Novy.

Individual preferences for a specific majority threshold can be influenced by voters’ attitudes toward uncertainty. It has been theoretically demonstrated and experimentally verified that a higher majority threshold is associated with risk aversion, serving as a means to protect against the tyranny of the majority (Attanasi, Corazzini & Passarelli 2017). In this paper, we posit that the absence of ex-ante information regarding the likelihood of the voting outcome introduces an additional layer of uncertainty - namely, ambiguity - which motivates decision-makers to seek increased protection.

We model the impact of both the level of ambiguity and ambiguity aversion on the desired majority threshold of a voting lottery in a KMM environment (Klibanoff, Marinacci and Mukerji 2005). We assume that as the number of voters increases, so does the level of complexity – and consequently, the ambiguity – of the voting lottery, which in turn activates ambiguity attitudes. We test our predictions through a series of 32 classroom experiments conducted between 2020 and 2024, involving approximately 1,200 undergraduate and graduate students in Italy and France, with voter group sizes ranging from 7 to 281.

Our findings confirm a positive correlation between risk aversion and the desired majority threshold. Additionally, we provide support for our two novel predictions: first, that the desirable threshold is positively correlated with ambiguity aversion, and second, that it increases with the number of voters through this channel. These results highlight the significance of ambiguity in strategic voting.

We estimate the long-run labor market and health effects of breast cancer among Austrian women. Compared to a random sample of same-aged non-affected women, those diagnosed with breast cancer face a 22.8 percent increase in health expenses, 6.2 percent lower employment, and a wage penalty of 15 percent five years after diagnosis. Although affected women sort into higher quality jobs post-diagnosis, this is offset by a reduction in working hours. We argue that the hours reduction is more likely driven by an increase in the time preference rate, meaning that patients increasingly value the present over the future, rather than by an incapacitation effect or employer discrimination.

Until 2013, Sweden required transgender people to undergo surgical sterilization before changing their legal gender. We analyze whether the removal of this requirement led to an increase in the number of legal gender changes. Additionally, we evaluate whether people who changed legal gender with versus without mandated sterilization had different mental health and labor market trajectories during their gender transition. Our analysis uses population-wide administrative data from 2006 to 2020, including information on legal gender changes, medical records, and socioeconomic characteristics. We find that, starting in the first quarter after the requirement was abolished, three to four times as many people changed legal gender, driven by younger people with worse labor market attachment. Approximately 32.6% of trans women and 55.2% of trans men chose not to have surgery when it was not mandated. Despite this, we find negligible differences in earnings, sick leave, or mental health trajectories between people who changed legal gender before and after the abolishment.

Uncertainty about the long-term clinical and economic impacts of amyloid-PET hinders discussion regarding its potential implementation. This study evaluates the cost-effectiveness of incorporating amyloid-PET into the standard diagnostic workup in a memory clinic setting from healthcare perspective.

Nine hundred participants from the Amsterdam Dementia Cohort who were offered amyloid-PET as part of their diagnostic workup were linked to data from Statistics Netherlands on institutionalization, mortality, and healthcare costs over a five-year period. Inverse probability weighting was applied to balance the PET (n = 440) and no-PET (n = 460) groups. Primary cost-effectiveness outcomes were time in the community (TIC) and time to death (TTD), measured as restricted mean survival times. Incremental Cost-Effectiveness Ratios (ICERs) were calculated by dividing the difference in cumulative costs between the PET and no-PET groups by the difference in TIC and TTD. Uncertainty around the results was estimated using bootstrapping with 5000 repetitions.

Part 1: We consider a specific parametric version of Chicken and two different correlation devices, public and private, with the same expected payoffs in equilibrium, which is also the best correlated equilibrium payoff for the game. Despite our choices of the payoffs in the game, we find in an experiment that "following recommendations from a correlation device" vary significantly within and between our two treatments with two devices.

Part 2: We consider three simple 2x2 games, Symmetric Battle of the Sexes, Modified Battle of the Sexes and Chicken that differ only in one outcome of the game, which we call the "cooperative" outcome. We use a public correlation device which is a convex combination of two pure Nash equilibria for sending recommendations to the players. We find that following recommendations and thereby coordination in the game vary significantly among these three games. We can explain these differences by analysing players' inherent behavioural types to achieve the cooperative outcome in the game.

We investigate subjective longevity beliefs in a large sample of the Swiss adult population, using the Click-and-Drag interface, a tool to empower subjects to intuitively submit full belief distributions. We collect data on longevity beliefs on archetypes, on a series of different health scenarios, and for themselves. We implement both a CDF elicitation task – in line with most of the literature – and a PDF version for the same task. Our results show that participants’ beliefs elicited with a PDF interface were more accurate than those using the CDF interface. We show additionally that providing participants with a visual support in the form of an average longevity distribution, substantially helps to improve accuracy and debias estimates. Moreover, providing visual aid right away clearly outperformed helping after a first unguided estimation. Our findings show the promise of eliciting full distributions, reveal a surprising outperforming of the PDF over the CDF visualization, and show that providing visual guidance is a powerful tool for improving longevity predictions.

Menu-dependence is a feature of a decision process according to which people behave differently not only because of the differences in the payoffs associated with the outcomes of their choices but also because of the payoffs associated with the choices they decided not to make. This phenomenon is involved in many explanations of other-regarding behavior based on the idea of reciprocity: a preference for rewarding the other player for acting on kind intentions and punishing her for acting on unkind intentions. Intentions, in fact, are derived by comparing what a player does and what he could have done and did not. We investigate the role of this mechanism in the context of a simple trust game, by looking at whether, as reciprocity theories predict, the likelihood of a trustworthy response is affected by the kindness of the trustor’s intentions. We find, first, no menu-dependence, in the sense that trustees do not vary their trustworthiness along with the perceived kindness; second, that trustors correctly anticipate trustors’ insensitivity to changes in kindness; third, that decision to repay trust is more intuitive and faster than the self-interested ones and, fourth, the decision to trust comes quicker than the outside option, when there is room for mutual gain. We think that these results may shed new light on the intentions-based explanations of reciprocal and trustworthy behavior.

Recent papers use models of job search to interpret quasi-experiments with heterogeneity in order to understand the effects of unemployment insurance and potential benefit duration (PBD), but disagree about key findings. We argue that amending this approach with analyses that let the data speak without restrictions yields insights into policy relevant parameters and the mechanisms behind effects that do not depend on the interpretative lens and assumptions of a model. The data also casts doubt on key assumptions of common models. We first investigate what we can (not) learn from (quasiexperiments with two interdependent outcomes using a model-free framework that allows for unrestricted heterogeneity. Our discussion shows that one cannot separate direct effects of PBD on wages from indirect effects through duration, but methods to examine their presence are key to understand the channels behind wage effects. We then re-examine the effects of longer PBD in Schmieder, von Wachter and Bender (2016). We first analyze the effects of PBD that quasi-randomization identifies. Duration effects of PBD almost exclusively prolong a few long spells, which helps to explain differences between studies. Dynamic selection into reemployment timing is non-monotonic, but does not change with PBD at short durations so that dynamic treatment effects are identified at these short durations. For wage effects of PBD, we find the conditions under which LATEs of PBD on wages are informative about consequences of PBD extensions and hence useful for policy to hold. We then examine what the data can say about channels and mechanisms behind wage effects. Using dynamic treatment effects and mediation analyses, we find PBD to affect wages directly. In consequence, the effect of duration on wages is not identified and we find at most limited evidence of its relevance. That wage loss operates through the firm fixed effect and not through duration speaks against individual-based causes such as skill depreciation or bargaining. The negative direct effect we find contradicts key assumptions of common models of job search unless there are positive effects on non-wage outcomes for which we find only limited evidence.

In this paper, we test the idea that religious leaders play a central role in shaping pro-sociality and religious (in)tolerance within their churches. Using controlled allocation tasks, we directly elicit in-group-out-group biases among pastors (N=200) and members of their churches (N=800) in Kenya. We first document remarkable heterogeneity in preferences across religious leaders, with one type of leaders being tolerant and the second type severely discriminating against Muslims and non-religious individuals. Next, we show that preferences of pastors are robustly positively related to the preferences of church members, which gives rise to two prototypical types of church communities, tolerant and parochial ones. In line with recent cultural transmission models, several findings support the interpretation that religious leaders directly influence pro-sociality of their followers: (i) both tolerant and parochial leaders aim to instill their preferences in church members, (ii) church members follow behavior in an experiment that exogenously provides information about leaders’ behavior, and (iii) the preference link is stronger for members with greater exposure to their religious leader. Together, our findings suggest that differences in preferences of religious leaders spillover and create distinct social groups with contrasting moral views how to treat out-group members.

Keywords: Religious leaders, Tolerance, Parochialism, Discrimination, Social preferences, Cultural transmission