Modeling Long-Term Cohort Survival Using Repeated Cross-sectional Data

ABSTRACT : Studies since the late 1970s have shown how differential rates of mortality of members in a birth cohort affect the aggregate mortality rate. In short, as frailer members of a cohort are selected out, the aggregate mortality rate converges toward the rate of the more robust members remaining alive in the cohort. Thus, the aggregate mortality pattern may not look at all like the mortality pattern for any subpopulation within the larger population. Another consequence of this selection process is that the aggregate composition of characteristics that influence survival---i.e., indicators of frailty---change across the life course of the cohort as well. This process wreaks havoc on estimating parameters of life course processes and spawned large scale interest in dealing with "unobserved heterogeneity." Here I show that changing cohort composition can be useful for estimating the influence of fixed covariates on survival using cross-sectional data. I develop the method, test it via simulation, and then demonstrate it on sample data from the General Social Survey.