Product Partitioned Dirichlet Process Prior Models for Identifying Substantive Clusters and Fitted Subclusters in Social Science Data

We introduce a new model-based clustering design using product partitions. This Bayesian specification simultaneously incorporates substantive clustering and model-fit subclustering on random effects from a Dirichlet process prior. The estimation algorithm directly includes variable context within clusters into a general clustering model that detects latent clustering effects pervasive in social science datasets based on posterior probability. The analysis of terrorist groups shows how this tool reveals important features in a dataset that are otherwise undetectable. AMS 2000 subject classifications: Primary 62F99; secondary 62P25; secondary 62G99