Department of Biostatistics Seminar/Workshop Series
Mixed-effects Models with Variance Heterogeneity
David Afshartous, PhD
Assistant Professor of Statistics, Division of Clinical Pharmacology, Department of Medicine, Miller School of Medicine, University of Miami
Wednesday, September 30, 1:30-2:30pm, MRBIII Conference Room 1220
Intended Audience: Persons interested in applied statistics, statistical theory, epidemiology, health services research, clinical trials methodology, statistical computing, statistical graphics, R users or potential users
We study the impact of a certain type of random effects heterogeneity on the mixed-effects model, namely that in which the variance of the random effects differs between two identifiable subgroups in the population. Such heterogeneity may arise in a clinical trial in which the variability of the growth curves for the treatment and placebo groups is not equal. We extend the basic mixed model to account for this type of heterogeneity, and then formally examine its effect on random effects estimates, subject level Empirical Bayes estimates, fixed effects estimates and their corresponding standard errors. In order to develop further insight into the practical implications, we consider several different models in the context of growth curve data from an actual clinical trial. We demonstrate that this type of heterogeneity may be accommodated in practice via more than one model specification, yielding the same marginal model but with different random effects specifications. While Verbeke & Lesaffre (1997) show that the effect of misspecification of random effects distributions on standard errors of fixed effects is very minor, our simulations indicate that this is not the case for the random effects misspecification considered in this paper. Presentation | Link