Age-Period-Cohort Analysis of Repeated Cross-Section Surveys: Fixed or Random Effects?

Yang Yang, Duke University
Kenneth C. Land, Duke University

Yang and Land (2003) developed a mixed (fixed-random) effects model for the age-period-cohort (APC) analysis of micro datasets in the form of a series of repeated cross-section sample surveys that are increasingly available to demographers. To estimate the mixed effects APC models, Yang and Land applied the statistical methodology of hierarchical regression models by treating cohort and period effects as random. A key assumption of such models is that cohort and period are uncorrelated with other regressors. An alternative approach to model estimation could be based on a purely fixed cohort and period effects regression model, which does not require the uncorrelated effects assumption and large numbers of cohorts and/or periods. We examine these assumptions and compare the fixed vs. random effects model specifications for APC analysis. We continue to use the data on verbal test scores from 15 cross-sections of the General Social Survey, 1974-2000, for substantive illustrations.