A Fixed-Effects Proportional Hazards Model for Clustered Data, Designed to Yield Non-Parametric Estimates of Regional Fertility Patterns

Ali Cevat Tasiran, Gothenburg University
Insan Tunali, Koc University

We examine the evolution of fertility patterns in Turkey, using a flexible fixed-effects proportional hazards model of waiting times. The model is designed to exploit the clustering features of survey data. It permits correlation between observed (household level) and unobserved (regional) factors. The latter are absorbed in a cluster-specific baseline hazard, which is treated as an incidental function of time. We use stratified partial likelihood estimation on spell data for the first through the fourth births recorded in the 1993 and 1998 DHS, and find statistically significant cluster effects. We back out the integrated baseline hazard and use it to isolate the differences in the tempo of fertility across clusters. The patterns are consistent with the interpretation that the war like conditions that existed in the Eastern part of the country in the late 80s and early 90s depressed fertility. We also find evidence of replacement behavior and gender preference.