Ulric B. and Evelyn L. Bray Social Sciences Seminar

Abstract: Signaling games are central to the study of politics but often have multiple equilibria, leading to no definitive prediction. We demonstrate that these indeterminacies create substantial computational problems for currently used maximum likelihood routines: they fail to uncover the parameters of a crisis-signaling game, regardless of sample size, even if the equilibrium in the game generating the data is unique. To overcome these problems, we adapt estimators from the IO literature, demonstrating that they outperform current best practices and are well suited to problems in international relations. We fit a signaling model to data on economic sanctions and find that traditional methods return estimates that are highly sensitive to software choices and produce payoffs indicating that bargaining concessions are not valuable. Our solutions remedy the problems and uncover a novel U-shaped relationship between audience costs and the propensity for leaders to threaten sanctions.