[Limdep Nlogit List] discrete choice and weighting questions

[sstewart at hwr.arizona.edu]

I'm forwarding this on behalf of Matt Weber ( maweber at hwr.arizona.edu), who has
had difficulty posting messages to the LIMDEP list. You can reply to him
directly, or to the list. I'll see that he receives any replies.
Thanks,
Steve


Dear LimDep Users,  I am a graduate student at the University of Arizona
struggling with completion of a Conditional Logit Model.  I am running
data from a Choice Experiment with 3 options, the third being the Status
Quo.  I am wanting to make sure my code is correct for interaction of
sociodemographic terms with a single ASC, and am also trying to weight
the data to correct for sample bias.

--> nlogit; lhs=choice; choices=O1,O2,SQ;
   model:U(O1,O2)=<ASC,ASC>+Bcost
*cost+Benvi1*envi1+Benvi2*envi2
   +Bsocio1*socio1+Bsocio2*socio2
   /U(SQ)=Bcost*cost+Benvi1*envi1+Benvi2*envi2;
   wts=popweigh$

Although the model spec. part of the code seems to work, I am not sure
what specifically about this code lets LimDep know I wish to interact
socio1 and socio2 with the ASC.  For example, if I wished to interact
socio1 and socio2 with the envi variables instead, how would I code it?

I am also having trouble with weighting.  I calculated a weighting
variable for each choice observation but when I run this none of the
regressing variables are significant any longer.  I tried a fake
weighting variable that is a constant, thinking this should give me the
same results as an unweighted run, but it does not.  So I believe I am
using the WTS command incorrectly somehow.  Use of the "noscale" option
within the WTS command does not make a difference.

Thanks for any advice anyone may have,

Sincerely,

Matt Weber
Univ of Arizona

-- 

-- 
Steven Stewart
SAHRA/Hydrology and Water Resources
University of Arizona
P.O. Box 210158-B Marshall Building 530
Tucson, AZ 85721-0158
(520) 437-2091
(520) 626-4479 FAX
sstewart at hwr.arizona.edu

---