st: Constrained multinominal logit problem

 From "Mathiassen, Astrid" <[email protected]> To <[email protected]> Subject st: Constrained multinominal logit problem Date Wed, 30 Apr 2003 13:28:53 +0200

```Hey,
I am doing a multinominal discrete choice analysis (McFadden's conditional logit model). My dataset contains aggregated information. My dependent variables are the number of times each of the 50 alternatives (of cars) were chosen. The corresponding attributes are, Zjr, (where j refer to alternativ and r to attribute f.ex. price).
For this procedure I use mlogit with 48 constraints. The constraints imply that the coefficient for each attribute should be equal, hence bjr=b1r for all j.
I have redefined the attribute-vector so that Zj*=Zj-Z1. Hence I use alternative 1 as the base category (reference).
I am a first time user of STATA and are a bit insecure on how this work. In particular, I do not understand the output report. The puzzle is that in the output report ONE, and the last listed elements in the attribute vectors, is dropped for each alternative - except for the last alternative where the estimates are reported for all elements of the attribute vector. This also happens in the special where attribute vector contains only one element. Hence, it cannot be explained by interedependency between the elements of the attribute vectors.
Could you help me understand what's going on? Or could you indicate an alternative procedure?

Thank you very much in advance
best regards,
astrid

My model statement look like this (number refer to the number of times the alternative was chosen):
mlogit depvar attribute1 attribute2 etc. [aweight=number], noconstant constr(2-49) basecategory (1)

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```