Testing the Validity and Consistency of Hierarchical Age-Period-Cohort Cross-Classified Models

Ryan K. Masters, University of Colorado at Boulder
Daniel A. Powers, University of Texas at Austin

Hierarchical Age-Period-Cohort (HAPC) Cross-Classified Fixed Effects Models (CCFEM) and Random Effects Models (CCREM) have become widely used by social scientists to investigate temporal variation in outcomes across ages, time periods, and birth cohorts. We discuss the scope and application of HAPC models, and test the validity and consistency of model estimates using simulated data. Findings show that fitted HPAC-CCREMs and HAPC-CCFEMs estimate the "true" degree of age-, period-, and cohort-based variation in simulated data when applied to (1) APC data structures in which cohort membership is not a linear function of one's age and time period, or (2) data in which the functional forms of age, period, and cohort effects on the the outcome are not all assumed to be linear. Further, we highlight the advantage of fitting HAPCs with Markov Chain Monte Carlo simulations using Gibbs sampling.