ABSTRACT
Survival analysis is properly thought of as a univariate rather than a multivariate technique because there is only a single response variable, failure time, even though there may be many explanatory variables. Reduction to a binary response is most useful when the survival of each individual is easily classified as either very short or very long. One reason for the choice of a timescale is direct meaningfulness for the individual concerned, justifying the use of real time in investigating survival in a medical context. In the absence of censoring, the dependence of failure time on the explanatory variables can be explored through multiple regressions. With survival dichotomized, and with quantitative explanatory variables, discriminant analysis has sometimes been used to identify variables that are related to survival, although such use of discriminant analysis is better regarded as an approach to binary logistic regression.
