Intervention studies often rely on microcoded data of social interactions to provide evidence of change due to development or treatment. Traditionally these data have been collapsed into small contingency tables. Such an approach can introduce spurious findings. Instead of treating each unit's contingency table independently, or collapsing the tables into single aggregate table, it is more efficient to analyze associations in all units simultaneously using hierarchical models. This article presents Bayesian hierarchical models to analyze several two-way categorical data with random effects that allow different levels of variation across several events. To illustrate this approach, the authors present an analysis of couples' interaction data from a recent study investigating how couples cope when one partner has become unemployed. (Contains 8 tables.)