topic Re: Proc mixed interaction term producing strange output in Statistical Procedures

Proc mixed interaction term producing strange output

Dimi — Fri, 03 Aug 2018 10:19:25 GMT

Hi Everyone,

Thanks for taking the time to read this.

I'm having a bit of trouble with PROC MIXED -namely when modeling an interaction term between a continuous and binary variable.

The model part of my code reads like this:

model result = arm_studyall*visit_m arm_studyall visit_m /solution;

For clarity result is a continuous variable. arm_studyall is a binary variable and visit_m is a continuous variable.

When I execute my code the fixed part of the model produces an interaction term for both levels of the binary variable

I'm not quite sure why this is happening -as I would expect only one interaction term to be estimated.

Any insight would be helpful.

Thanks again.

Re: Proc mixed interaction term producing strange output

PaigeMiller — Fri, 03 Aug 2018 10:46:37 GMT

You get an effect (slope) of visit_m when arm_studyall is zero.

You also get an effect (slope) of visit_m when arm_studyall is one.

This is the exact meaning of an interaction between continuous and binary variable ... there are really two different slopes.

Re: Proc mixed interaction term producing strange output

FreelanceReinh — Fri, 03 Aug 2018 11:09:14 GMT

Hi @Dimi,

If you want only one of the interaction terms to have a nonzero estimate, just reorder the effects in your model statement, so that main effects are prioritized:

model result = arm_studyall visit_m arm_studyall*visit_m /solution;