Analysis of Grouped Survival Data: A Synthesis of Various Traditions and Application to Modeling Childhood Mortality in Eritrea

Abstract

This paper merges together some statistical methods used in the analysis of data involving rates of occurrence of an event. These methods are (1) indirect standardization with the multiplicative model, (2) loglinear regression for count data, and (3) proportional hazards regression for survival data. In many applications these approaches have been portrayed as belonging to distinct fields or as competing methodologies. In this paper it is demonstrated that (1) and (2) actually represent one special case of (3) in two different, but equivalent, parameterizations. One advantage of such synthesis is that computer algorithms developed for one setting can be exploited in another. Accordingly, we demonstrate how the General Loglinear Analysis Procedure in SPSS, and the GENMOD Procedure in SAS may be used to compute estimates of baseline and relative hazards (parameters common in survial analysis) and how these estimates may be interpreted in relation to standardization. The issues addressed are illustrated by empirical analysis of a data set on mortality experiences among 7,055 Eritrean children based on data from the 1995 Eritrean Demographic and Health Survey.