ESTIMATING INCREMENT-DECREMENT LIFE-TABLES WITH MULTIPLE COVARIATES FROM PANEL-DATA - THE CASE OF ACTIVE LIFE EXPECTANCY

A fundamental limitation of current multistate life table methodology- evident in recent estimates of active life expectancy for the elderly- is the inability to estimate tables from data on small longitudinal pa nels in the presence of multiple covariates (such as sex, race, and so cioeconomic status). This paper presents an approach to such an estima tion based on an isomorphism between the structure of the stochastic m odel underlying a conventional specification of the increment-decremen t life table and that of Markov panel regression models for simple sta te spaces. We argue that Markov panel regression procedures can be use d to provide smoothed or graduated group-specific estimates of transit ion probabilities that are more stable across short age intervals than those computed directly from sample data. We then join these estimate s with increment-decrement life table methods to compute group-specifi c total, active, and dependent life expectancy estimates. To illustrat e the methods, we describe an empirical application to the estimation of such life expectancies specific to sex, race, and education (years of school completed) for a longitudinal panel of elderly persons. We f ind that education extends both total life expectancy and active life expectancy. Education thus may serve as a powerful social protective m echanism delaying the onset of health problems at older ages.