topic Re: Proc mixed in Statistical Procedures

Proc mixed

DJGS — Wed, 03 Nov 2021 21:00:17 GMT

Hello! I am using Proc Mixed. If the null likelihood ratio test is not statistically significant (I tried with 4 different covariance structures), is it nevertheless valid to report the type 3 tests of fixed effects? If it is not valid, what would you suggest I do? My model is pretty simple -- there are three repeated measures per individual, with two fixed effects, and one interaction. Thanks, as always, for your help!

Re: Proc mixed

SteveDenham — Fri, 05 Nov 2021 13:39:04 GMT

I assume all of the Type 3 tests are non-significant as well. That would lead to the non-significant null likelihood, and then you should report those tests.

SteveDenham

Re: Proc mixed

DJGS — Fri, 05 Nov 2021 15:23:20 GMT

Thanks Steve... weirdly, 2 of the fixed effects are non significant, but one is (p=0.0144). Not sure what to do.. I can report it all, but I am not sure how to interpret this. Thanks for any additional thoughts.

Re: Proc mixed

DJGS — Fri, 05 Nov 2021 15:24:09 GMT

Forgot to mention that I am dealing with a very small sample size...