I like the idea of reporting lots of these, especially if it underlines
that the various pseudos are measuring
various different things and will not agree numerically (although
perhaps they will often rank different models
more or less consistently).
Still, that won't stop
People asking, "But which is best?"
People wondering if there is a way of averaging or combining the
pseudos.
People cherry-picking what seems "best" for their favourite story.
Naturally, I do the latter all the time when I choose graphs, but I put
that down to experience and judgment....
Nick
[email protected]
Mike Lacy
>
>I have fitted a partial proportional odds model in Stata 9.2 using
>- -gologit2-. In their book, Regression Models for Categorical
Dependent
>Data using Stata, Long and Freese suggest that McKelvey & Zavoina's R2
>most closely approximates the R2 from linear regression models
This is not necessarily true. That is a good measure, but
see: (self-promotion mode on)
Lacy, M. 2006. "An Explained Variation Measure for Ordinal Response
Models with Comparisons to Other Ordinal R2 Measures." Sociological
Methods and Research 34:469-520.
For software (not sufficiently documented to merit submission to
SSC): http://lamar.colostate.edu/~socmgl/r2o.ado
I did find, however, in parallel to what Richard Williams suggests in
another post, that while various ordinal pseudo-R^2 fit measures
might differ in their ability to approximate an underlying linear
model R^2, all the various measures have pretty comparable
performance as aids to model selection.
*
* 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/
