Abstract: We propose novel methods to assess assumptions on random effects in linear mixed models. A key ingredient to the proposed methods involves creating missing data strategically from the observed data in order to detect multiple sources of misspecification on random effects. Random-effects assumptions traditionally tested separately by tests for variance components and tests for normality can now be assessed simultaneously using the proposed methods. The rationale underlying the new methods is applicable to other types of mixed effects models.
Key words and phrases: Ignorable missingness, missingness mechanism, model misspecification, nonignorable missingness, random effects.