Accelerated Failure-Time Mover-Stayer Regression Models for the Analysis of Last-Episode Data

This article introduces new models for the analysis of last-episode data—that is, data about the timing of the last occurrence of a repeatable event. The new models are elaborations and extensions of a particular model for backward recurrence times from a renewal process introduced by Allison (1985), and they are accelerated failure-time mover-stayer regression models that permit the separation of covariate effects on event timing from those on event occurrence. Using the data of last residential move from the 1990 Census of Population and Housing, an application of the models, which are found to fit the data adequately, shows distinct patterns in the effects of home ownership, gender, race/ethnicity, education, marital experience, and age group on residential mobility. The patterns differ regarding the relative importance and direction of influence between the effects on the duration of residence in the same house or apartment and the effects on the lifetime probability of not experiencing any residential move.