Hi
Is there a way to conduct multinomial regression for a weighted subsample datasets? For e.g. I used mi estimate: mlogit DV IV1 IV2 IV2[iweight=ngwt] for the full sample set N = 5777. However, out of the full sample set, I want to conduct multinomial regression on households that have received at least one welfare assistance (e.g. TANF, SNAP, WIC, etc). So I created a new variable called assist and replaced it with a binary (y/n) coded as 1 if assist (households with any of the welfare assistance), new sample set N = 676.. Then I ran a mi estimate:mlogit. However, the number of observations were the same as the full sample set 5777. Why is that?

Secondly, if I want to change my condition to households receiving at least ONE welfare assistance rather than ANY, e.g. so that I can compare households who did NOT receive welfare assistance vs households who received at least ONE, how would I code it? Someone suggested creating a by id and sort with egen?
Can I still use mi estimate:mlogit function with weights?
Please advise.
Thank you
Lena