Interaction effect in MIANALYZE is not estiamted

[marina_unn]

Hey!

I am using a mutliple imputation and then run a GEE model on the 10 imputed datasets. The GEE model includes interaction between time and treatment (denoted as "randomisering"). I manage to pool all the model coefficients with MIANALYZE, but the interaction term is not estimated (appears as 0 in the MIANALYZE table). I have checked that interaction term in MIANALYZE is called exactly the same as in the gmparms table, but still get no estimate.

proc genmod data=fatigue_migee;
by _imputation_;
class record_id gender randomisering timeclass;
model fatigue_total = randomisering  time_weeks randomisering*time_weeks
   gender inclusion_age/ dist=normaL LINK=identity maxiter=200 covb;
repeated subject=record_id/ withinsubject=timeclass
 type=exch;
 ods output GEEEmpPEst=gmparms parminfo=gmpinfo CovB=gmcovb ;
run;


proc mianalyze parms(classvar= level)=gmparms;
class gender randomisering;
modeleffects Intercept randomisering time_weeks time_week*randomiser gender inclusion_age;
run;

Is there something I am missing? I see no reason that interaction term is not estimated, I have seen estiamtes in each of the 10 imputed datasets, so it should be possible to pool them.

Thank you in advance! 

[SAS_Rob]

Could you post the entire LOG that contains the GENMOD and MIANALYZE steps in it?  Seeing the output might also be helpful in diagnosing what is going on.

[marina_unn]

Hi @SAS_Rob ,

Sure, I can copy the log output here, I think it does not say so much. I will also inlcude the MIANALYZE output table, which shows interaction effect as zero.

MIANALYZE LOG

40 proc genmod data=fatigue_migee;
41 by _imputation_;
42 class record_id gender randomisering timeclass;
43 model fatigue_total = randomisering time_weeks randomisering*time_weeks
44 gender inclusion_age/ dist=normaL LINK=identity maxiter=200 covb;
45 repeated subject=record_id/ withinsubject=timeclass
46 type=exch;
47 ods output GEEEmpPEst=gmparms parminfo=gmpinfo CovB=gmcovb ;
48 run;

NOTE: The variables in the WITHIN= subject effect are sorted by internal (unformatted) values. To sort by formatted values specify
the ORDER=FORMATTED option following the variable in the CLASS statement.
NOTE: Algorithm converged.
NOTE: Algorithm converged.
NOTE: The above message was for the following BY group:
Imputation Number=1
NOTE: Algorithm converged.
2

NOTE: The data set WORK.GMCOVB has 350 observations and 9 variables.
NOTE: The data set WORK.GMPINFO has 450 observations and 5 variables.
NOTE: The data set WORK.GMPARMS has 450 observations and 9 variables.
NOTE: PROCEDURE GENMOD used (Total process time):
real time 6.57 seconds
cpu time 6.57 seconds.

NOTE: PROCEDURE MIANALYZE used (Total process time):
real time 0.07 seconds
cpu time 0.07 seconds

44
45 %LET _CLIENTTASKLABEL=;
46 %LET _CLIENTPROCESSFLOWNAME=;
47 %LET _CLIENTPROJECTPATH=;
48 %LET _CLIENTPROJECTPATHHOST=;
49 %LET _CLIENTPROJECTNAME=;
50 %LET _SASPROGRAMFILE=;
2 The SAS System 08:39 Monday, March 23, 2026

51 %LET _SASPROGRAMFILEHOST=;
52
53 ;*';*";*/;quit;run;
54 ODS _ALL_ CLOSE;
NOTE: Writing EXCEL(EGXLSSX) file: /saswork/prod/SAS_work3C1D0000F862_dvh-p-grid-01.hn.helsenord.no/#LN07206
55
56
57 QUIT; RUN;
58

 And here I upload the GMPARMS table which have estiamted interaction effect (time_weeks*randomisering), and the output of MIANALYZE, where this interaction effect is zero.

[SAS_Rob · Accepted solution]

Your MODELEFFECTS statement should match your MODEL statement as far as the variable names is concerned.  I think in your case there is some truncation that is going on from the GENMOD output.  All you need to do is add the NAMELEN= option to the GENMOD code.

proc genmod data=fatigue_migee namelen=100;
by _imputation_;
class record_id gender randomisering timeclass;
model fatigue_total = randomisering  time_weeks randomisering*time_weeks
   gender inclusion_age/ dist=normaL LINK=identity maxiter=200 covb;
repeated subject=record_id/ withinsubject=timeclass
 type=exch;
 ods output GEEEmpPEst=gmparms parminfo=gmpinfo CovB=gmcovb ;
run;


proc mianalyze parms(classvar= level)=gmparms;
class gender randomisering;
modeleffects Intercept randomisering  time_weeks randomisering*time_weeks
   gender inclusion_age;
run;

[marina_unn]

Thank you! That was exactly the issue that caused the problem!