Edsel Peña

Department of Statistics

University of South Carolina


Estimation of an Event-Time Distribution

This talk pertains to the important statistical problem of estimating a distribution function, in particular, the distribution of the random variable pertaining to the occurrence of an event of interest such as the failure of a machine, time to death, time to an insurance claim, etc. First, given a simple random sample, we compare and contrast the resulting parametric and nonparametric estimators, and consider the impact of model misspecification. Second, we consider the situation where the event times are right-censored, as is typical in biological, medical, public health, engineering, and economic situations. For this setting, we present and discuss the Kaplan-Meier or the product-limit estimator. Third, we consider the situation where there are multiple event times per experimental unit such as in the situation where the event is recurrent, and discuss peculiarities and difficulties in this setting. Finally, we tackle the setting where there are covariates, predictor, or marker variables. This will entail a discussion of the popular Cox proportional hazards model. We then describe a general model for recurrent events which simultaneously takes into account several aspects in recurrent event modeling. For these latter models we consider the problem of estimating a baseline distribution function. The talk is meant to be an overview and will thus refrain from delving deeply into mathematical technicalities.


Back to Colloquium Series