A growing experimental literature studies the determinants of cooperation in infinitely repeated games, tests different predictions of the theory, and suggests an empirical solution to the problem of multiple equilibria. To provide a robust description of the literature's findings, we gather and analyze a metadata set of experiments on infinitely repeated prisoner's dilemma games. The experimental data show that cooperation is affected by infinite repetition and is more likely to arise when it can be supported in equilibrium. However, the fact that cooperation can be supported in equilibrium does not imply that most subjects will cooperate. High cooperation rates will emerge only when the parameters of the repeated game are such that cooperation is very robust to strategic uncertainty. We also review the results regarding the effect of imperfect monitoring, changing partners, and personal characteristics on cooperation and the strategies used to support it.

We use a novel experimental design to reliably elicit subjects’ strategies in an infinitely repeated prisoner’s dilemma experiment with perfect monitoring. We find that three simple strategies repre‑ sent the majority of the chosen strategies: Always Defect, Tit‑for‑Tat, and Grim. In addition, we identify how the strategies systematically vary with the parameters of the game. Finally, we use the elicited strategies to test the ability to recover strategies using statistical methods based on observed round‑by‑round cooperation choices and find that this can be done fairly well, but only under certain conditions.

We study the role of commitment in communication and its interactions with rules, which determine whether information is verifiable. Our framework nests models of cheap talk, information disclosure, and Bayesian persuasion. It predicts that commitment has opposite effects on information transmission under the two alternative rules. We leverage these contrasting forces to experimentally establish that subjects react to commitment in line with the main qualitative implications of the theory. Quantitatively, not all subjects behave as predicted. We show that a form of commitment blindness leads some senders to overcommunicate when information is verifiable and undercommunicate when it is not. This generates an unpredicted gap in information transmission across the two rules, suggesting a novel role for verifiable information in practice.

This paper presents a dynamic equilibrium model of a taxi market. The model is estimated using data from New York City yellow cabs. Two salient features by which most taxi markets deviate from the efficient market ideal are, first, matching frictions created by the need for both market sides to physically search for trading partners, and second, regulatory limitations to entry. To assess the importance of these features, we use the model to simulate the effect of changes in entry, alternative matching technologies, and different market density. We use the geographical features of the matching process to back out unobserved demand through a matching simulation. The matching function exhibits increasing returns to scale, which is important to understand the impact of changes in this market and has welfare implications. For instance, although alternative dispatch platforms can be more efficient than street-hailing, platform competition is harmful because it reduces effective density.

We conduct an experimental analysis of selective disclosure in communication. In the model, an informed sender aims to influence a receiver by disclosing verifiable evidence that is selected from a larger pool of available evidence. Our experimental design leverages this model’s rich comparative statics, allowing us to systematically quantify the effects of selection relative to concealment. Our findings confirm the key qualitative predictions of the theory, suggesting that selection, rather than concealment, is often the dominant distortion in communication. We also identify deviations from the theory: Some senders are “deception averse” and overcommunicate relative to predictions; receivers respond too optimistically to both concealed and selected evidence, with errors of similar magnitude. Yet selection generates greater overall distortion in receiver behavior because it is far more prevalent than concealment.