Of Discussions, Beliefs, and Algorithms: Essays in Experimental Economics

Abstract

This dissertation contains three empirical essays. The first essay examines the effects of a debating process prior to a collective decision. We conduct a lab-in-the-field experiment with school minors in Germany. We randomly assign some of them to discuss via chat how much they want to donate to a charity supporting incoming refugee minors. In our study, the pre-vote debate leads to higher donations. It does not directly affect trust among discussants, but subjects who are perceived as refugee-friendly after the discussion benefit from increased trust from their chat partners.

The second essay addresses belief formation and asks how breaking a social norm affects what people think others would do in the same situation. To study this question, I employ an online experiment with a representative sample of the UK’s general population exposing subjects to either a high or a low temptation to lie. I find that subjects in the tempting environment lie more and, importantly, hold more pessimistic beliefs about what others would do in the same situation. I also include additional treatments in order to argue that the observed effects are not driven by rational expectations. The study provides causal evidence that breaking a social norm leads to strategic belief distortion about other people’s behavior in order to maintain a positive self-image.

The third essay studies a situation in which a human decision-maker is assisted by an algorithm. We conduct an online experiment with US participants who repeatedly perform an estimation task while receiving (largely biased) recommendations from an algorithm. We analyze two interventions and ask whether they can help humans to assess the quality of algorithmic advice. First, we find that explaining the functioning of the algorithm in abstract terms reduces adherence to algorithmic advice, but it does not improve decisionmaking performance. Second, disclosing the correct answer after each round reduces adherence to algorithmic advice and improves human decision-making performance. While existing literature suggests that people abandon algorithms after seeing them err, this is not confirmed in our setting. This is likely because in our setting people can comprehend the reasons why some of the algorithmic predictions are inaccurate. Jointly, the three essays provide insights into (behavioral) economic concepts of trust, generosity, belief formation, and advice taking through the lens of experimental methods.