Simulation evaluation of three models for correlated binary data with covariates specific to each binary observation
Titel:
Simulation evaluation of three models for correlated binary data with covariates specific to each binary observation
Auteur:
Hertzberg, Vicki Stover
Verschenen in:
Communications in statistics
Paginering:
Jaargang 26 (1997) nr. 1 pagina's 375-396
Jaar:
1997
Inhoud:
Binary data from human reproductive studies occur naturally in clusters and often exhibit strong intracluster correlation which results in overdispersion. Although several methods for analyzing clustered data, particularly those with covariates specific to each binary outcome, have been proposed recently, the properties of many of these models have not been rigorously evaluated in simulation studies, nor have they been compared amongst each other. In this paper, three such models which are applicable to the analysis of spontaneous abortions in humans are evaluated. In all three cases, the maximum likelihood estimates of effect are biased. In addition, naive logistic regression to the whole data set of simulated pregnancy outcomes, ignoring intracluster correlation, is shown to be biased toward the null in the presence of an effect.