01-28-2015 12:22 PM
Is it possible to produce likelihood ratio test (LRT) statistics for the pairwise comparisons with the LSMeans statement rather than Wald statistics?
My model has two variables, one has 3 levels and the other only 2. I noticed that analysis of the variable with only 2 levels resulted in differing conclusions. Using the Type 3 LRT, the p-value is 0.0355. Using the LSMeans Wald (or the Type 3 Wald), I only get 0.1250. Clearly, the LRT is a bit more powerful in this circumstance, and I can just use that p-value. That makes me worry about my other variable though, the one with 3 levels. To do my pairwise comparisons, it looks like GENMOD can only do a Wald test. Is it possible to get a LRT? I'd like the extra power.