Difference in difference with Genmod

[Molina]

Hi,
I am trying to figure out how to conduct a difference in difference analysis with a proc Genmod.
My outcome is a continuous variable (Y), and I would like to know if being a case (Yes/no) influences this outcome.

I have repeteated measures ( time = 0 to 12), and several not time-dependent variables (Adj_var).

My  code :

PROC GENMOD DATA=GEE_PREPARATION2;
CLASS id time(ref="0") Adj_var (ref="0") case(ref="0") / param=ref;
MODEL Y= case time Adj_var  case*time;
REPEATED SUBJECT=id / TYPE=UN CORRW; RUN;

And... I don't understand my results at all. For time*Case (1 1) it gives me a coefficient "22.8271". I thought it was the result of my Diff in Diff because it's the same results when I calculate it by hand DiD=[(Y11 - Y10)-(Y01-Y00)].  But he didn't take into account my adjustment variables (Adj_var).

If I try with this code :

PROC GENMOD DATA=GEE_PREPARATION2; 
CLASS id time(ref="0")  case(ref="0") / param=ref; 
MODEL ville= case time case*time; REPEATED SUBJECT=id / TYPE=UN CORRW; RUN;

I have the same results.

I must have missed something, but I don't see what...

[StatDave]

Well, that seems consistent with this note. Parameter estimates for CASE and TIME should be adjusted for the effects of the other variables. So, their estimates should change with vs. without those additional variables unless, of course, they are all uncorrelated. 

[Molina]

Thank you StatDave_sas, 

I'm totally agree with you, if I adjust with VAR_ADJ, the parameter estimates for CASE and TIME are changed. What I find curious is that my interaction parameter TIME*CASE does not change with or without VAR_ADJ. 

The purpose of my model is precisely to adjust the difference in difference estimate (TIME*CASE). That's why I'm so embarrassed with this result. 😄