Journal of Biometrics & Biostatistics

Creation of Interpretable Summary Measures in Displaying Results from Mixed-effects Logit Models.

Abstract

Xian Liu, Bradley E. Belsher and Daniel P. Evatt

The authors of this article developed new approaches to present analytic results from mixed-effects binary logit models in longitudinal data analysis. We first described basic specifications of mixed-effects logit models, the derivation of the fixed and the random effects, and nonlinear predictions of the response probability and the corresponding standard errors. Particular attention was paid to the interpretability of the conventional odds ratio in the longitudinal setting. The authors contended that without information on averaging of the random effects for two population subgroups of interest, the regression coefficient of an explanatory variable and its antilog in mixed-effects binary logit models are not interpretable. We recommended the computation of the conditional effect and the conditional odds ratio to aid in displaying a covariate’s effect on the longitudinal binary response. An empirical illustration was provided to demonstrate how to create interpretable summary measures for aiding in the interpretation of the results from mixed-effects logit models when analyzing binary longitudinal data.