ABSTRACT
This chapter discusses the assumptions underlying the multilevel regression model, and outlines ways to test these assumptions. When distributional assumptions are not met, there are several approaches to deal with this. To establish an accurate confidence interval for a variance component, we need a technique that results in a confidence interval that is not symmetric. The profile likelihood method is one such method based on maximum likelihood estimation. More general methods to deal with non-normal data are using robust standard errors, bootstrapping and Bayesian estimation, all of which are discussed in this chapter.
