I am trying to estimate a mixed logit model to understand taste variation among the people. In my sample, there are 5000 people as decision makers and each of them make repeated choices ranging from 1 to 300 times. So I end up having 160000 observations.
I tried to use PROC MDC first (code listed below), then this model takes 160000 observation at one time and I can't specify individual person as cluster.
****************************************************
proc mdc data=gdp.matlab_model_0610 type=mixedlogit;
model response =
ini_dly_beta_trun
dly_sav_min2
freq
distance
rev
hub
mj_index
prm_airlines
marketfare
seat
/
nchoice=2
mixed=(normalparm=ini_dly_beta_trun dly_sav_min2);
id id;
run;
**************************************
Then I tried PROC GLIMMIX, then it can't converge. Damn it....
******************************
proc glimmix INITGLM data=gdp.y7;
class cluster;
model response =
ini_dly_beta_trun
dly_sav_min2
freq
distance
rev
hub
mj_index
prm_airlines
marketfare
seat /
dist=binary link=logit
solution ;
random ini_dly_beta_trun
dly_sav_min2
/subject=cluster ;
;
run;
*******************************************