Research

Available here

Abstract: In sequential interactions, both the agent’s intention and the outcome of his choice may influence the principal’s action. While outcomes are typically observable, intentions are more likely to be hidden, leaving potential wiggle room for the principal when deciding on a reciprocating action. We employ a controlled experiment to investigate how intentions and outcome affect the principal’s actions and whether principals use hidden information as an excuse to behave more selfishly. We find that principals react mainly to the intention of the agent. When intentions are not revealed by default, principals tend to select into information based on their inclination to behave more prosocially. While information avoidance is frequent and selfishness is higher with hidden information, we do not find evidence of a strategic exploitation of moral wiggle room.

Available: here

Abstract: How do men and women differ in their persistence after experiencing failure in a competitive environment? We tackle this question by combining a large online experiment (N=2,086) with machine learning. We find that when losing is unequivocally due to merit, both men and women exhibit a significant decrease in subsequent tournament entry. However, when the prior tournament is unfair, i.e., a loss is no longer necessarily based on merit, women are more discouraged than men. These results suggest that transparent meritocratic criteria may play a key role in preventing women from falling behind after experiencing a loss. 

Abstract:  Are gender gaps in economic behavior and labor market outcomes associated with biological sex, or with gender-typical traits? Using an online U.S. sample and machine learning, we develop and validate a gender typicality measure capturing masculinity and femininity as separate dimensions. We show that confidence, competitiveness, and risk-taking are associated with masculinity, whereas altruism and equality and efficiency concerns are associated with femininity. Gender typicality also predicts labor market outcomes: masculinity is associated with higher income, managerial positions, performance-based pay, and willingness to negotiate. By leveraging latent heterogeneity, our framework offers a new empirical lens for analyzing gender differences.

Abstract:  The Bem Sex Role Inventory (BSRI) is among the most widely used measures of gender identity in economic research, yet its original classification has not been tested in over two decades. We provide a contemporary validation of the BSRI's item classifications using Bem's original between-subjects elicitation procedure on an online U.S. adult sample. Of the 40 gendered items, 38 retain their original classification: all 20 masculine items are confirmed, and 18 of 20 feminine items remain feminine. No item reverses direction. Yet the magnitude of desirability gaps has gradually attenuated since the original study, with female raters driving much of this compression. Our findings indicate that while the directions of the BSRI remain intact, the intensity of gender differentiation has weakened over time. 

Abstract:  Generative AI raises individual productivity in experiments, yet many workers do not use it on the task in front of them. We test whether ambiguity about whether AI use is permitted is itself a barrier. In a preregistered 2 × 2 experiment, 3,026 US and 572 German workers completed an incentivized writing task under an ambiguous platform default that may discourage AI. Participants were randomized to receive a one-sentence statement explicitly permitting AI use, a one-paragraph summary of scientific evidence on AI’s productivity benefits, both, or neither. Explicit permission raised task-level AI use by 39 percentage points in both countries, cut completion time by 0.4 standard deviations in the US (0.6 in Germany), and improved writing clarity by 0.4 standard deviations in both countries. Gains in depth, creativity, and use of examples were smaller. Productivity information alone had more muted effects. Even when explicitly permitted, roughly half of workers did not use AI, citing a preference for producing their own work. Permission left socio-demographic adoption gaps unchanged but widened the gap by prior attitudes toward AI from 14 to 47 percentage points, a pattern replicated in Germany. Explicit task-level permission is a low-cost and scalable organizational policy for AI adoption, but its effects are strongly conditioned by workers’ prior attitudes