ABSTRACT
This paper presented a sweeping unification and extension of the previous work on survival data which allowed for the study of individual covariates and multivariate analysis. The finely stratified conditional analysis for individual failure times developed previously by Kaplan and Meier for a single survival curve and by Mantel and Haenszel for two or more curves (in terms of 2 2 contingency tables) was generalized to the conditional logistic model. The fundamental role of the proportional hazards model was emphasized and a comprehensive calculus was developed using the novel idea of partial likelihood. Highly flexible models allowing individual timevarying exposures, multiple events, and stratification of different sorts could be constructed and analysed in a coherent framework.
