Age-period-cohort models have a long and interesting history and area popular choice in a number of disciplines, includingepidemiology, demography, and the social sciences. However, the fitting and interpretation ofthese models requires great care because of the well-knownidentifiability problem that exists; given any two of age, period, andcohort, the third is determined.In this paper we review the identifiability problem and models thathave been proposed for analysis, from both frequentist and Bayesian standpoints. Anumber of recent analyses that use age-period-cohort models aredescribed and critiqued before data on cancer incidence in WashingtonState are analyzed with various models, including a new hybridBayesian approach based on an identifiable parameterization.
- age-period-cohort models
- random walk priors