Observed choices in Social Dilemma Games usually take the form of bounded integers. We propose a doubly-truncated count data framework to process such data. We compare this framework to past approaches based on ordered outcomes and truncated continuous densities using Bayesian estimation and model selection techniques. We find that all three frameworks (i) support the presence of unobserved heterogeneity in individual decision-making, and (ii) agree on the ranking of regulatory treatment effects. The count data framework exhibits superior efficiency and produces more informative predictive distributions for outcomes of interest. The continuous framework fails to allocate adequate probability mass to boundary outcomes, which are often of pivotal importance in these games.
