SURVIVAL TREES - AN ALTERNATIVE NONPARAMETRIC MULTIVARIATE TECHNIQUE FOR LIFE-HISTORY ANALYSIS

In this paper an extension of tree-structured methodology to cover cen sored survival analysis is discussed. Tree-based methods (also called recursive partitioning) provide a useful alternative to the classical survival data analysis techniques, such as the semi-parametric model o f Cox, whenever the main purpose is defining groups of individuals, ei ther with complete or censored life history, having different survival probability, based on the values of selected covariates. The essentia l feature of recursive partitioning is the construction of a decision rule in the form of a binary tree. Trees generally require fewer assum ptions than classical methods and handle non standard and non linear d ata structures efficiently. Tree-growing methods make the processes of covariate selection and grouping of categories in event history model s explicit. An example concerning the analysis of time to marriage of Italian women is presented.