--- On Thu, 23/7/09, Afridi wrote:
I read this question on the group homepage. It is few years old
question. But i could not find answer to this question from anyone
here. I will be thankful if someone can give me some direction for
such kind of case. i am working to a smilar case and facing same
problem as stated below.
Having had the same question in the past, (but have not received any reply from Statalist) it is now very clear to me that the answer Maarten Buis has posted must be the absolute fact for the estimation of nlogit: no within branch variation=no utility differences= No nlogit estimation.
Nlogit is at its best should you have choice-specific attributes.
You can still create choice-specific variables by the standard procedure i.e. multiplying branch alternative (dummy values) times the explanatory -individual specific- variables in order to "generate" this variability. That's what you might try, but be dilligent enough with the results you get. Are they really significantly different than mlogit(LR test)? Are IVs not 1?
Seqlogit is an excellent way to estimate sequential -non attribute specific- choices, should this be applicable to your case -i.e. there is a time element. It can model sequential multinomial choices, but it has its own drawbacks which may be effectively accounted for by excellent sensitivity analysis by its sd option.
Should this not be applicable -you have simultaneous outcomes- maybe mlogit is your best chance.
Dino K.
U o Patras
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
