Abstract: We experimentally show that complexity shapes strategy selection in infinitely repeated prisoner's dilemmas in several fundamental ways.  Our subjects choose from a menu of pre-set strategies and we vary the presence of three potential drivers of complexity aversion by varying whether subjects have to  (i) implement their chosen strategies themselves, (ii) compute the payoff consequences of those strategies and whether they (iii) can use the simplicity of strategies as a coordination device.  We find that subjects display a strong aversion to algorithmically complex strategies when these drivers of complexity aversion are present, but not when they are removed.  In our data, complexity is at least as important in shaping strategic choice as strategic uncertainty, the most important determinant of strategy selection identified by the literature so far.  We show that these results are robust to learning and several measures of cognitive sophistication.

Abstract: In many strategic environments with asymmetric information, systematic decision mistakes can arise from two sources: failures in probabilistic reasoning when updating beliefs and incorrect strategic inference about opponents' private information or choices. Using a sequential belief elicitation design in persuasion games, we experimentally disentangle and quantify these two reasoning errors. We find that probabilistic and strategic reasoning errors are comparable in magnitude, largely orthogonal, and their patterns and overall degree are stable across various settings. These results suggest that an improved behavioral theory should account for both strategic naivete and non-Bayesian updating, without significant concern for their interaction.

Abstract: We experimentally investigate how characteristics of information structures beyond their instrumental value influence information demand. Using novel methods that minimize Bayesian reasoning errors, we find that while participants strongly favor structures with higher instrumental value, they systematically deviate from optimal choices in predictable ways. We identify four key characteristics of information structures that drive these deviations and show that they are as predictive of behavior as individual characteristics, with stronger effects among participants who are otherwise more likely to act optimally. Our findings suggest that these distortions stem from the cognitive complexity of evaluating information structures, with implications for models of information demand and the design of effective disclosure strategies.

(Previously circulated as "Choosing Between Information Bundles".)

Abstract: This paper experimentally studies how people choose sets of information sources (information bundles). The findings reveal that people frequently fail to choose the more instrumentally valuable bundle in binary choices, largely due to the challenge of integrating sources within a bundle to identify their joint information content. These mistakes cannot be attributed to inability to use information bundles, but rather to a choice-stage heuristic I call "common source cancellation (CSC)". This heuristic leads people to reduce the comparison of two bundles to the comparison of their non-shared sources in isolation, ignoring each bundle's joint information content. CSC robustly explains information bundle choices for all subjects, including those who make perfect use of each bundle.

Abstract: We explore theoretically and experimentally whether information design can be used by trustees as a signaling device to boost trusting acts. In our main setting, a trustee partially or fully decides a binary payoff allocation and designs an information structure; then a trustor decides whether to invest. In the control setting, information design is not available. In line with the standard equilibrium analysis, we find that introducing information design increases trustworthiness and trusting acts, and some trustees choose full trustworthiness with the most informative structure. We also find systematic behavioral deviations, including some trustees' choosing zero trustworthiness with the least informative structure and trustors' overtrusting in low informative structures. We finally provide a model of heterogeneity in prosociality and strategic sophistication, which rationalizes the experimental findings.

Abstract: The same underlying data can be communicated or perceived in different formats. A natural question is whether people prefer the formats that serve them best. We study five different summary reports of a series of binary signals in the canonical belief updating setting: Majority (which signal type appeared more often), Percentage (the percentage of each signal type), Difference (how many more of the dominant type), Count (the number of each type), and Sequence (the raw sequence of the signals). These reports differ in two key dimensions: instrumental value (IV) and information richness (IR). In an online experiment (N=200), we elicit subjects’ preference for the reports and measure their belief-updating performance across the reports. Our central finding is a systematic misalignment between preference and performance. Preference strengthens with both IV and IR, although the standard economic theory predicts only IV should matter. Performance, in contrast, improves with IV but deteriorates with IR conditional on IV, that is, participants perform worse with the very reports they prefer more. There is heterogeneity in preference, but the misalignment between preference and performance is robust across subgroups. Together, these results indicate that demand for information is shaped by features that go beyond instrumental value, and that people have imperfect meta-cognition about their own belief-updating behavior.