A SAS Macro for Computing Pooled Likelihood Ratio Tests with Multiply Imputed Data

For multilevel analyses (e.g., linear mixed models), researchers are often interested in pooling, interpreting, and testing both fixed effects and random effects. PROC MIANALYZE has two shortcomings in this regard. First, it cannot easily pool variance estimates. Second, the significance tests of these estimates are Wald-type tests that are inappropriate for testing variance estimates. Likelihood ratio testing is a more flexible approach, as it can be used to compare models that differ in both fixed and random effects. The likelihood ratio test statistic requires a complex calculation that is not included in PROC MIANALYZE. This paper describes a SAS macro, MMI_ANALYZE, that fits two user-specified models in PROC MIXED, pools the estimates from those models (including variance components), and implements a pooled likelihood ratio test.