ABSTRACT
In standard regression techniques, there is no time component. However, many outcomes in people analytics and in other fields can happen at different times over a period of study. Promotion or attrition can occur at different times for different individuals. Subscription sign-up or cancellation can occur at differing points for different customers. This chapter covers the common elements of survival analysis, which considers the time to event as well as the event itself in explaining the outcome of interest. Using an example of employee attrition, the chapter covers Kaplan-Meier estimates of survival rates and survival curves as a way of establishing prima facie relevance of a variable to the outcome. Cox proportional hazard models are studied as a way of performing multivariate regression on a time to event outcome. The underlying assumptions behind Cox models are reviewed and methods for checking these assumptions are presented. Frailty models are then briefly studied as a method for conducting regression on a population where there are differing background risks of an event occurring.
