'Logistic Regression With Multiple Random Effects: A Simulation Study of Estimation Methods and Statistical Packages'

摘要

Kim, Choi, and Emery (2013) provided a timely discussion of various estimation methods for logistic regression with multiple random effects and their implementations in available statistical packages. The article updates a similar study recently reported by Zhang et al. (2011). We would like to bring to the authors' attention the method of data cloning (DC) as another recent alternative to those studied in their article. Introduced by Lele et al. (2010), DC is a computational algorithm for approximating maximum likelihood estimates (MLEs) and their standard errors (SEs) in complex models. The method is particularly suited for obtaining MLEs in generalized linear mixed models (GLMMs), of which logistic mixed models are a special case, as it avoids the high-dimensional integration involved in the evaluation of the marginal likelihood. It entails merely the calculation of simple averages and variances and, unlike the methods considered by Kim, Choi, and Emery (2013), does not require numerical maximization or differentiation of the likelihood function.