Survival Analysis Using SAS®: A Practical Guide

Generalized R²

People who do a lot of linear regression tend to become attached to R² as a measure of how good their models are. When they switch to PROC LIFEREG or PROC PHREG, they often experience severe withdrawal symptoms because no similar statistic is reported. In this section, I show how a generalized R² can be easily calculated from statistics that are reported. Before doing that, I want to caution readers that R² is not all it’s cracked up to be, regardless of whether it’s calculated for a linear model or a proportional hazards model. In particular, R² does not tell you anything about how appropriate the model is for the data. You can obtain an R² of only .05 for a model whose assumptions are perfectly satisfied by the data and whose ...