ABSTRACT
This chapter focuses on multilevel generalized linear models (GLMs). It explores the application of exploratory data analysis techniques to multilevel data with non-normal responses. The chapter explains the crossed random effects that are appropriate and how they differ from nested random effects. It also focuses on a multilevel generalized linear statistical model, including assumptions about variance components. The chapter describes the model parameters from a multilevel GLM. It generates and interprets random effect estimates. The chapter presents a case study on college basketball referees. In the College Basketball Referees case study, the two primary level two covariates are home team and visiting team. The estimates of variance components provide evidence of the relative variability in the different Level Two random effects. The chapter illustrates the null distribution of the likelihood ratio test statistic derived by the parametric bootstrap procedure with 100 samples as compared to a chi-square distribution.
