03-09-2018 10:24 AM
I have an outcome variable that has 4 categories (very low, low, moderate, and high) and a predictor variable with 2 categories (yes/no). When I run a chi-square test, the association is (very) significant. However, when I run it using proc glimmix, dist=multinomial, the association is no longer significant. I understand that I am treating the outcome variable differently when using proc glimmix, but why would it be so different? And is it best to do the analysis using proc glimmix and treat the outcome as multinomial?