runmixregls: A Program to Run the MIXREGLS Mixed-Effects Location Scale Software from within Stata

Abstract

Hedeker and Nordgren (2013) present the stand-alone MIXREGLS program for fitting the mixed-effects location scale model to continuous longitudinal and other clustered data. This model can be used when interest lies in joint modeling the mean and dispersion of subjects' responses over time. The model extends the standard two-level randomintercept mixed model by allowing both the within- and between-subject variances to be influenced by the covariates and for the within-subject variance to additionally depend on a subject random-scale effect. In this article we present the runmixregls command to run MIXREGLS seamlessly from within Stata. We illustrate the notable advantages of using runmixregls by replicating and extending the two example analyses presented in Hedeker and Nordgren (2013). We then use runmixregls to demonstrate a new and important research finding. Namely, that ignoring the random-scale effect in the withinsubject variance function will lead to the regression coefficients in this function to be estimated with spurious precision, especially the regression coefficients of subject-level covariates.