Assumptions associated with the hierarchical generalized linear model include conditional independence, distributional assumptions for each level of the hierarchy, functional form of the covariates, exchangeability, and independence of the random effects with the person-level covariates. This chapter describes using the error components defined at each level of the hierarchy to check the validity of the assumptions made. Model selection, that is, which model from a set of models should be used, employs global measures to minimize overall model error subject to the number of model parameters. If adopting a Bayesian approach, a comparison of replicated data generated from the posterior distribution can be used to assess the appropriateness of model assumptions. Approaches to assessing the validity of causal inference assumptions are detailed.