Comparison of SAS NLMIXED and GLIMMIX for the One-Way Layout with Random Batch Effects: A Simulation Study
Ambika Smith
Master's Candidate, Department of Statistics
Colorado State University
Friday, December 15, 2006
11:00 a.m.-1 p.m.
006 Statistics


For binomial trials in which the binomial experiment is repeated a number of times per treatment, variability in the conditions of the experiment from replication to replication is to be expected. This variability indicates the need for a model that allows for random effects. Generalized linear mixed models enable the modeling of random effects with non-normal responses. The objective of this paper is to compare the performance of the penalized quasi-likelihood method, implemented in SAS PROC GLIMMIX, and adaptive Gaussian quadrature, implemented in SAS PROC NLMIXED. Type I error rates for the overall F-test, bias of the variance estimates, and coverage rates of confidence intervals are compared across the statistical methods through computer simulations under varying binomial sample sizes, number of replications, number of treatments, and magnitudes of variance. The results for bias are difficult to evaluate due to the highly skewed distributions for the error variance estimates. The results for type I error and coverage show that both methods perform well with a large number of replications. For a small number of replications, the penalized quasi-likelihood method (GLIMMIX) tends to have lower type I error rates and higher coverage rates than adaptive Gaussian quadrature (NLMIXED).



  home for studentspeoplecoursesdistance degreeresearchconsulting seminars resources
PRIMESSTARMAPemail us equal opportunitydisclaimerweb master • last modified: August 11 2003