categorical interaction term in ESTIMATE function (GEE model)

[katy-barry]

hello, I am trying to test for an interaction between a categorical variable and my main exposure variable (also categorical). I have tried so many combinations to try to get the odds ratios for the interaction term. can someone please help me:

Class Level Information
Class	Value	Design Variables
earlystart	0	0	0
	1	1	0
	2	0	1
repeat1	0	0
	1	1

Parameter Information
Parameter	Effect	earlystart	repeat1
Prm1	Intercept
Prm2	earlystart	1
Prm3	earlystart	2
Prm4	repeat1		1
Prm5	earlystart*repeat1	1	1
Prm6	earlystart*repeat1	2	1

proc genmod data= final_chomage_dernier;

class ntt chomage_dernier (param=ref ref= '0') 

earlystart(param=ref ref= '0') repeat1 (param=ref ref='0');

model chomage_dernier= earlystart repeat1 earlystart*repeat1/ dist=bin link=logit;

repeated subject=ntt/ type=unstr;

estimate "repeat 1" repeat1 1/exp;

estimate "earlystart*repeat" earlystart*repeat1 1 1 /exp;

estimate "earlystart*repeat" earlystart*repeat1 0 1 /exp;

run;

thank you!

[cminard]

You're not specifying enough parameters in the estimate statements. I did a simulation to demonstrate. I guess you have repeated measures, but I ignored that. I am assuming independent observations in my example. But the estimate statements should be similar.

I set this up so that the odds of an event in earlystart2 should be about 6x greater than the odds of an event in earlystart0 and earlystart1.

data one;
do earlystart=0 to 2;
do repeat1=0 to 1;
do n = 1 to 1000;
if earlystart in (0,1) then event=1*(ranuni(133)<.2);
if earlystart=2 then event=1*(ranuni(6113)<.6);
output;
end;
end;
end;
run;

data two;
set one (drop=n);
id+1;
run;

proc genmod data=two;
class event earlystart (ref='0') repeat1 (ref='0');
model event (event='1') = earlystart|repeat1 / dist=bin link=logit ;
estimate 'earlystart1 vs. earlystart0' int 0 earlystart 1 0 -1 repeat1 0 0 earlystart*repeat1 .5 .5 0 0 -.5 -.5 / exp;
estimate 'earlystart2 vs. earlystart0' int 0 earlystart 0 1 -1 repeat1 0 0 earlystart*repeat1 0 0 .5 .5 -.5 -.5 / exp;
estimate 'earlystart2 vs. earlystart1' int 0 earlystart -1 1 0 repeat1 0 0 earlystart*repeat1 -.5 -.5 .5 .5 0 0 / exp;
run;