|
Hi!
This is perhaps a very simple problem (hope so), but ...anyway:
I am fitting a multinomial logit model. The response variable has
four values. The very idea is to compare different groups. The qualitative
variables included in the model and the group -variable are coded into
dummies.
The values of the response variable are labour market states and
the question in the study is to compare the effects of different
group-membership.
The basic syntax is (in UNIX -environment):
proc catmod data=data1;
weight count;
direct var1 var2 var3 ... group;
response logits/out=data2;
model x=var1 var2 var3 ... group;
/ml pred=freq;
run;
So I get the model as parameter estimates etc. I also get a data set
(data2), which contains the predicted probabilities (as frequencies)
for each combination of the regressor values.
If I want to present results which tell the model-based probabilities
among different groups _when the effects of other variables are being
held constant_, how should I simulate/estimate/calculate those "clean"
probabilities.
Thanks!
regards,
Jukka Halme
| 
