A Review and Discussion of Residuals for Mixed Models
Talk for NCCC-170 meeting 2019
Residuals are a key tool used to diagnose models. As a statistical consultant for researchers in many areas, I often find myself reminding my clients to visualize residuals to assess model assumptions. Many of my clients are working with mixed models, and I recently realized that I often recommend the use of certain residual types without a full understanding of the implications of selecting one type over another. This led me to have an interest in better understanding the many residuals types for mixed model. In this talk, I will provide a review of the residual types available for linear mixed models (marginal, conditional, studentized, etc.). I will explain how the residuals are computed and how these computations differ between R and SAS. I will also discuss what I have learned from the literature about how to select a residual type when assessing a model. Lastly, I will briefly touch on residual types for generalized linear mixed models and list some unanswered questions. If time permits, I will pose these remaining questions to the attendees to discuss as a group.
