Analysis of related durations using fixed effects: a unified proportional hazards framework and an application to a study of breastfeeding durations in Malaysia.

The authors contrast 2 estimators that arise in the Cox Proportional Hazards model when durations are allowed to be correlated through a unit specific fixed effect. The 1st estimator is a Stratified Partial Likelihood estimator that leaves the time-dependence function unspecified. The 2nd is a Full Likelihood estimator which requires parametric assumptions on the time-dependence function. Both estimators admit time-varying covariates but differ in other respects. The authors employ the 2 estimators to investigate the determinants of breastfeeding durations in Malaysia and examine the differences between them. (authors)