We develop a formal framework for analyzing how decision-making is informed by the extraction of statistical relationships from data. We define a dataset as a joint distribution over explanatory variables and an outcome. Across datasets, we assume a common and stable data-generating process, so that the conditional distribution of the outcome given a set of relevant predictors is the same in every dataset. A decision maker observes a dataset and seeks to extract decision rules that map subsets of explanatory variables into actions, trading off the rules' value against their complexity cost. We derive testable predictions that separate a rule's value from its cost and characterize the probability of extracting the value-maximizing rule that conditions on the relevant variables, substitution toward competing rules, and heterogeneity in rule extraction. We test these predictions in our own experiment and provide supporting evidence using data from Kendall and Oprea (2024). We also demonstrate how our framework can be applied to study the role of complexity costs.

We study, empirically and theoretically, how restitution helps restore cooperation after a breach or an exploratory defection. Restitution strategies propose a return to cooperation by cooperating against defection, and condition actions on the balance between cooperation given and received. We reanalyze experimental data from repeated Prisoner's Dilemma games and find empirical support for restitution strategies in general, and for a strategy we name Payback, in particular.  Besides explaining how subjects deal with conflicts, accounting for restitution strategies helps reconcile discrepancies between theory and experiments—such as the widespread use of non-equilibrium strategies like Tit-for-Tat and the limited predictive power of risk dominance under imperfect monitoring.

It has been amply shown that choice behavior is context-dependent. Evidence from the cognitive sciences suggests that this dependency may be driven by implicit associations. In this paper, we propose and study a choice-based model of contextual associations. We start by formalizing contexts by the set of concepts it contains. We then introduce associations by way of the implicit relationship between alternatives and concepts, which directly impacts the utility evaluation of alternatives. We study the empirical content of the model, establish conditions for identification, characterize its comparative statics, and propose several extensions. We provide an application to probabilistic voting, showing that our model rationalizes how political competition contributes to the polarization of parties’ values, as emphasized in the empirical literature.

I explore the trade-off between efficiency and strategic risk in the context of a repeated prisoner’s dilemma with imperfect public monitoring. Since with imperfect monitoring deviations can happen on the equilibrium path, to keep the value of cooperation high, players have to adopt more "lenient” and "forgiving” strategies, being more exposed to opportunistic behavior. For a broad class of one-dimensional public signals, I show that the maximal payoff achievable under a (symmetric) cooperative risk-dominant equilibrium is strictly lower than the maximal symmetric equilibrium payoff. This formalizes the fact that under imperfect monitoring, strategy adoption responds continuously to beliefs about the opponent’s cooperation, whereas under perfect monitoring, beliefs affect only the extensive margin—whether to cooperate at all. I leverage experimental evidence from Aoyagi and Fréchette (2009) to document suggestive patterns consistent with agents selecting strategies to maximize value rather than controlling for risk.