Re: st: NBREG for ordinal scales

["David Barron" <mothsailor@googlemail.com>]

> If, in fact, the DV has a Poisson or negative binomial distribution,
> then that is your justification for using count models.  But how do
> you know that it does have such a distribution???  And why would you
> even think that it does, given the way the questions are worded?  Is
> there some "classic" piece these authors cite, or do they just do it
> without citation?  There must have been somebody who started this -
> if you could find out who it was that might help solve the puzzle.
But of course data of this form cannot possibly have a Poisson or
negative binomial distribution because, as somone has already pointed
out, these distributions are unbounded on the right, while the ordinal
scales being talked about only take a few values.   This is quite
apart from thinking about the random process that generates the
observed data in the first place.  I would imagine that the Poisson or
neg bin might be a reasonable approximation in some situations, but if
expected values are toward the upper end of the range of values then
they can't possibly be.

I don't really undestand why one would want to use these methods when
apparently superior alternatives (such as ordinal logit) exist.

--
=================================
David Barron
Said Business School
University of Oxford
Park End Street
Oxford OX1 1HP
*
*   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/