MM Algorithms for Variance Component Estimation and Selection in Logistic Linear Mixed Model

摘要

Logistic linear mixed model is widely used in experimental designs andgenetic analysis with binary traits. Motivated by modern applications, weconsider the case with many groups of random effects and each group correspondsto a variance component. When the number of variance components is large,fitting the logistic linear mixed model is challenging. We develop twoefficient and stable minorization-maximization (MM) algorithms for theestimation of variance components based on the Laplace approximation of thelogistic model. One of them leads to a simple iterative soft-thresholdingalgorithm for variance component selection using maximum penalized approximatedlikelihood. We demonstrate the variance component estimation and selectionperformance of our algorithms by simulation studies and a real data analysis.