Survival Data

Abstract

A duration is the time until some event occurs. Thus, the response is a non-negative random variable. If the special case of a survival time is being observed, the event is considered to be absorbing, so that observation of that individual must stop when it occurs. We first consider this case, although most of the discussion applies directly to more general durations such as the times between repeated events, called event histories (Chapter 7). Usually, the distribution of durations will not be symmetric, but will have a form like that in Figure 6.1 (this happens to be a log normal distribution). This restricts the choice of possible distributions to be used. For example, a normal distribution would not be appropriate. Suitable distributions within the generalized linear model family include the log normal, gamma, and inverse Gaussian.