Gebrenegus Ghilagaber: Analysis of Survival Data with Long-term Survivors

In the analysis of duration data with censored observations the main assumption is that censoring time is independent of event time. This assumption is violated in many situations. For instance in the analysis of data on family formation and marriage duration, individuals with a tendency to remain single or remain married over long periods may be overrepresented among the censored observations. In this work, we use a mixture model where parameters of a binary logistic model (for the conditional probability of long-term survivors given censoring) are jointly estimated with those of a continuous intensity model for marriage duration. The advantage of this model is that it allows simultaneous estimation of two sets of effects of covariates: one for the probability of the event and another for the timing of the event. We provide a comparative analysis between standard survival models and mixture models that account for long-term survivors. Our results illustrate that failure to account for long-term survivors may yield misleading results that could plague the purpose of the analysis.

The presentation will begin with introduction to the basic features of survival data.

Joint work with Paraskevi Peristera.