Actor-level variations in the amounts of uncertainty have been widely ignored in the growing literature on statistical models of strategic interaction in international relations. In this article, I provide a tool for testing theories about the level of uncertainty in strategic interactions. I show that ignoring potential variations in levels of uncertainty across different cases can be a source of bias for empirical analyses. I propose a method to incorporate this form of heteroskedasticity into existing estimators and show that this method can improve inferences. With a series of Monte Carlo experiments, I evaluate the magnitude and the severity of the bias and inconsistency in estimators that ignore heteroskedasticity. More importantly, the tools developed in this article have many interesting substantive application areas. Examples considered include measuring speculators’ suboptimal behavior tendencies in international currency crises, and capturing varying levels of signaling and Bayesian updating behavior in the recent strategic models of signaling.
