In the earlier analysis (Hessler et al., 1990), logistic regression was used to estimate
the probability of survival/death. At that time, the exact dates of death were unknown,
thus the dependent variable was dichotomized into dead/alive. This estimate
of death/survival was made under three conditions: using 1966 interview data to predict
survival to the 1974 and 1986-1987 interviews, and using the 1974 data to predict
survival to the 1986-1987 interview. The goal of this first analysis was to test
a theoretically driven predictive model.