A Bayesian Approach to the Problem of Anticipatory Covariates in the Analysis of Retrospective Demographic Data

Gebrenegus Ghilagaber, Stockholm University

Some of the most important explanatory variables in retrospective demographic data, such as educational level and social class, are measured at the interview time but are used in investigations of behavior in life segments that occur before the interview. An interesting question would then be to what extent the changes in the behavior under study can be attributed to changes (misclassification) in the distribution of individuals across the various levels of, say education; and to what extent they reflect real differences in the behavior of interest. The present study uses Bayesian approaches in order to utilize available information in defining appropriate priors and estimating posterior patterns of relative intensities of the event of interest across the levels of the anticipatory variable. The issues addressed are illustrated by estimating effects of education and social class on entry into and exit from family unions among Swedish men born in 1936-64.

  See extended abstract

Presented in Session 39: Advances in Methodology