[Limdep Nlogit List] Q: Weighting Discrete Choice Data

[Koetse, M.J.J. (Mark)]

Dear Matthew,

I too have had some problems with this. Solution in my case was simple
because I had defined my weighting variable in a wrong way.

Since every software package does it differently, check whether you have
defined the weighting variable correctly. This can be done by dividing
all variables, also the constant, by the weights and estimating the
model with the redefined variables by OLS. These are the correct results
and may differ from the results achieved by the WTS command. 

Hope it helps, good luck.

Kind regards,
Mark Koetse 

-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au
[mailto:limdep-bounces at limdep.itls.usyd.edu.au] On Behalf Of Matt Weber
Sent: 01 November 2006 16:49
To: limdep at limdep.itls.usyd.edu.au
Subject: [Limdep Nlogit List] Q: Weighting Discrete Choice Data

Dear LimDep Users,

This is a re-post as I am still hoping for advice on one issue.
I have discrete choice survey data I am analyzing with NLOGIT
techniques. 
In an effort to correct for non-response bias I am
trying to weight the data from certain zipcodes above that
received from other zipcodes.  I am attempting to use the "WTS"
command with a weighting variable I have defined to achieve this.

Has anybody any experience with weighting discrete choice data
for a regression?  My approach using WTS is giving odd results,
all the regressors are going from highly significant to highly insignif.
I have tried using a unit weighting variable, i.e. a constant,
and this yields odd results too, whereas I would have
expected results identical to the unweighted regression.  I have
read the LimDep manual on this and still am unclear whether the
WTS command is applicable for discrete choice.

A Sincere Thanks for any thoughts on this,

Matthew Weber
University of Arizona
_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au