Data Analytics Corp. wrote:
>Thanks for all the helpful advice. I'll start to look for these
references. But I would still like to do something in Stata. It seems
that rologit would do the trick, but....>
I forgot about tricking a Cox PH program to fit the model you want, which someone else mentioned. -clogit- might be able to do it---I suspect -rologit- is just a -clogit- wrapper.
Still the effects of object similarity mean you probably need a mixed model, and therefore want to look at the Rabe-Hesketh et al stuff, which will use -gllamm-. This will be OK so long as the number of objects isn't larger than, oh, 4. Warning: -gllamm- is slow because it uses both adaptive quadrature and numerical derivatives.
I was messing around a bit with -xtmelogit- last year but couldn't figure out the identification constraint on the random effects covariance matrix necessary in Stata. (SAS has many covariance structures available and an easy way to impose arbitrary constraints but I wish it had better solution algorithms.) The easiest method to identify the model is to fix one variance to a constant (1) and all of its covariances to 0.
jverkuilen wrote:
> Walt,
>
> Let me consult some references tomorrow---my dissertation was on a related probem. Otherwise see "the" Google. From memory, though, see:
>
> -Fligner and/or Verducci, lots of articles.
>
> -John Marden, Chapman & Hall book on ranking models. His book has a summary of the literature to the mid 90s. (Marden was a committee member.)
>
> -Fellow Illinois alum Albert Maydeu-Olivares (now at Barcelona), who has an article on the unrestricted Thurstone ranking model. (I can't recall the journal but he has all his reprints online.) This would require -gllamm- or -xtmelogit- in Stata.
>
> -See also the article by Sophia Rabe-Hesketh (and co-authors?) on ranking models on (or at least cited on) http://gllamm.org which was in Psychometrika a few years back. She probably has code.
>
> -Can -rologit- accommodate ties? You might be able to trick it that way.
>
> -In general though you usually have to map this kind of response to a mixed binary regression with a decidedly peculiar set of identification constraints.
>
> JV
>
> -----Original Message-----
> From: "Data Analytics Corp." <[email protected]>
> To: "Stata Listserve" <[email protected]>
> Sent: 8/21/2008 10:03 PM
> Subject: st: use of rologit
>
> Hi,
>
> A potential client asked for a form of analysis sometimes done in market
> research called maximum-difference (maxdiff) analysis. Basically,
> consumers are presented with a series of product attributes (say, four
> at a time or a "quad") and then are asked to select the one they prefer
> the most and the one they prefer the least. For a quad, the other two
> non-selected attributes just fall in the middle. This is repeated
> several times with the attributes rotating following an experimental
> design. Since the attributes are rated, I was thinking of using rologit
> to estimate a model. But then I found a statement in the reference
> manual that says that "rologit does not allow for other forms of
> incompleteness, for instance, data in which respondents indicate which
> of four cars they like best, and which one they like least, but not how
> they rank the two intermediate cars." But this is exactly my problem.
> Does anyone know how to trick rologit, or know of an ado file to handle
> this problem?
>
> Thanks,
>
> Walt
>
>
>
>
--
________________________
Walter R. Paczkowski, Ph.D.
Data Analytics Corp.
44 Hamilton Lane
Plainsboro, NJ 08536
________________________
(V) 609-936-8999
(F) 609-936-3733
[email protected]
www.dataanalyticscorp.com
*
* 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/
*
* 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/
