Quality & Quantity 32: 297–324, 1998.
© 1998 Kluwer Academic Publishers. Printed in the Netherlands.
Analysis of Survival Data with Multiple
Causes of Failure
A Comparison of Hazard- and Logistic-Regression Models with Application
Uppsala University, Department of Statistics, P. O. Box 513, S-751 20 Uppsala – Sweden
Abstract. The purpose of the paper is to compare results of estimation and inference concerning
covariate effects as obtained from two approaches to the analysis of survival data with multiple
causes of failure. The ﬁrst approach involves a dynamic model for the cause-speciﬁc hazard rate.
The second is based on a static logistic regression model for the conditional probability of having
had an event of interest. The inﬂuence of sociodemographic characteristics on the rate of family
initiation and, more importantly, on the choice between marriage and cohabitation as a ﬁrst union, is
examined. We found that results, generally, are similar across the methods considered. Some issues
in relation to censoring mechanisms and independence among causes of failure are discussed.
In many areas of the natural, medical, and social sciences, there has been much
interest in the analysis of data representing the time to occurrence of certain events.
Engineers, for instance, study failure of speciﬁc machine components. Medical
scientists are concerned with hospitalizations, visits to a physician, and death or
relapse of patients in a clinical trial. Criminologists study crimes and arrests. In
the study of work and careers, attention is given to job changes, promotions and
unemployment. Demographers focus on births, deaths, marriages, divorces, and
In each of these examples, an event consists of some qualitative change that
occurs at a speciﬁc point in time. A central concept in the analysis of data repre-
senting times to occurrence of some speciﬁed event, is the hazard function. Such
The ﬁrst version of the paper has been presented at one of the Statistics Seminar Series organized
jointly by the Department of Statistics, University of New South Wales, and the Australian Graduate
School of Management. Dr. Chris Aisbett, Dr. Matt Wand, and Dr. A.Y.C. Kuk made some helpful