Hello Sascha,
>My understanding is that nests in -nlogit- take care of correlation in
>the error terms of latent utilities across choices of the same individual.
>
>What I want is non i.i.d. observations, i.e. allow for correlations in
>the error terms across different individuals ON TOP OF the nested
>structure across choices.
>
>As I wrote in my earlier message: In the restaurant example in the Stata
>manual we have nests for different restaurant types (fast food, family
>restaurants and fancy restaurants) but we still assume that choices made
>by DIFFERENT families are uncorrelated (i.e. they are i.i.d.)
>In real life, restaurant choices of families (those are the units of
>observation) in the same neighborhood may be correlated because >families
>in the same neighborhood have similar (unobserved) tastes.
>That's the case I am looking for.
I'm afraid I don't have up-to-date manuals, but it sounds like you could use a multi-
level mixed multinomial logit model, which can be estimated in gllamm; individuals
nested within families (if required), families nested within neighbourhoods, and
restaurant as an alternative-specific characteristic. Correlation amongst error terms
across units (individuals, families, neighbourhoods) is handled by latent variables.
Have a look at Skrondal & Rabe-Hesketh (2003) "multilevel logistic regression for
polytomous data and rankings", Psychometrika, 68(2): 267-287, and
www.gllamm.org.
Cheers,
Nick
--------
[email protected]
+44 (0)161 275 4738
CCSR, Dover St. Building,
University of Manchester,
Manchester
M13 9PL
*
* 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/
