Sandwich Variance Estimation for random effect misspecification in Generalized Linear Mixed Models

Abstract

The literature clearly demonstrated how the random effect miss-specification in Generalized Linear Mixed Models (GLMMs) affect the model performance with respect to the Type II Errors of the Type III F-test. The method of Sandwich Variance Estimation (SVE) is a very popular method for improving the functionality of miss-specified models. This study attempted on examining whether the use of SVE could improve the Type II Errors of miss-specified GLMMs. A comprehensive simulation study comprising data from a Binary Logistic Mixed Model was performed of which the results clearly demonstrated that Type II Errors are being affected by random effect miss- specification. The novel finding of the study was that the adoption of SVE failed to contribute significantly to improve the functionality of GLMMs when random effects of the GLMMs are not correctly specified.