Analysis of dependent survival data by conventional partial likelihood
methods produces unbiased estimates of the regression coefficients bu
t incorrectly estimates their variance. Here we compared the conventio
nal partial likelihood methods with two alternative methods for analyz
ing dependent survival data. The first alternative method estimated th
e regression coefficient by the partial likelihood approach but adjust
ed the variance to account for clustering. The second alternative meth
od used marginal likelihoods to estimate both the regression coefficie
nt and its variance. We evaluated the performance of the three methods
using simulated and actual data. Simulated data were used to examine
bias, efficiency, type I errors, and power. An Old Order Amish genealo
gy was analyzed under these models to illustrate their performance on
real data. The simulation study showed that all three methods provided
unbiased estimates of the regression coefficient, but the efficiency
of the estimated regression coefficient varied according to the simula
tion conditions. The standard partial likelihood method showed increas
ing type I error as the dependence increased within clusters. Both alt
ernative methods had acceptable levels of type I errors at all depende
nce levels. In the analysis of genealogic data, the regression coeffic
ient was similar in the three methods showing stable estimates of the
regression coefficients. The variance estimates from the alternative m
ethods were slightly different from the conventional method, suggestin
g a flow level of dependence. This study displays the effect of violat
ing the independence assumption and provides guidelines for using alte
rnative statistical methods. (C) 1996 Wiley-Liss, Inc.