Event history analysis

The aim of event history analysis is to explain why certain individuals are at a higher risk than others of experiencing the event(s) of interest. This can be accomplished by using special types of methods which, depending on the field in which they are applied, are called failure-time models, life-time models, survival models, transition-rate models, response-time models, event history models, duration models, or hazard models. Examples of textbooks discussing this class of techniques are Allison (1984), Blossfeld & Rohwer (1995), Kalbfleisch & Prentice (1980), Lancaster (1990), Singer & Willett (2003), Tuma & Hannan (1984), Vermunt (1997), and Yamaguchi (1991). Here, I will use the terms event history, survival, and hazard models interchangeably. A hazard model is a regression model in which the “risk” of experiencing an event at a certain time point is predicted with a set of covariates. Two special features distinguish hazard models from other types of regression models. The first is that they make it possible to deal with censored observations, which are observations containing only partial information on the timing of the event of interest. Another special feature is that they can deal with covariates that change their values during the observation period, which makes it possible to perform a truly dynamic analysis. Below I will first explain what is actually analyzed in an event history analysis. Then, I introduce the basic statistical concepts for both continuousand discrete-time analysis. As far as analysis tools themselves is concerned, I will discuss the Kaplan-Meier estimator, which is a method for describing event history data, as well as regression models for continuousand discretetime event history data. I will show that after organizing the data in the appropriate manner, an event history analysis can be performed using standard tools for Poisson and logistic regression analysis. Moreover, I will discuss how multilevel and mixture modeling tools can be used to deal with unobserved heterogeneity.