Hello @Rick_SAS,
Thank you for your reply. Much appreciated. So, I get why the intercept would change and I'd get an estimate for the new reference group I've set, but I'm not sure why I wouldn't get different estimates (the conditional effects?) for each group (comparing male vs female outputs) for all of the other estimates except the 2-way interactions and the 3-way interaction that include the group variable? Sorry if that's confusing. Still learning (grad student). When I change the reference groups in a non-multiple imputation model, I get different estimates for the intercept, the group variable, and the 2-way interactions that do not include the group variable. Here is the code I'm using just in case you can spot anything obvious I might be doing wrong.
/*FORMAT*/
proc format;
value gender1f2m
1 = 'female'
2 = 'male';
run;
proc format;
value genderref
1 = '2female'
2 = '1male';
run;
/*EPII X GENDER X DHEAcort*/
ods select none;
Proc mixed data = pandemic.auc_mi_imputed_dheacort method=reml covtest ;
class gender1f2m;
model covidanx_z = epii_mc gender1f2m DHEA_cortisol_winsorized_mc
epii_gender epii_DHEAcort gender_DHEAcort epii_gender_DHEAcort
anx_z
diff_s4_wake BlackvsOtherRace1Black0Other incometoneedsw pubc_mean/ solution;
format gender1f2m genderref.;
by _Imputation_ ;
ods output SolutionF=mixparms ;
run;
/*POOLING 2 IMPUTATION ONLY*/
ods select all;
proc print data=mixparms (obs=27);
var _Imputation_ Effect Estimate StdErr ;
title 'REG Model Coefficients and Covaraince Matrices' '(first 2 imputations)';
run;
/*POOLING*/
proc mianalyze parms=mixparms edf=234;
class gender1f2m;
modeleffects intercept epii_mc gender1f2m DHEA_cortisol_winsor
epii_gender epii_DHEAcort gender_DHEAcort epii_gender_DHEAcort
anx_z
diff_s4_wake BlackvsOtherRace1Bla incometoneedsw pubc_mean;
format gender1f2m genderref.;
run;
Thank you very much!
Best,
Joe
