Richard Williams wrote:
>>One caveat though, and this goes well beyond gologit2: researchers
often justify using simpler and easier methods because effects have
the same sign and statistical significance as in the more appropriate
and more complicated methods. <snip><<
This is right. One thing I always point out to the "but it doesn't make
a difference" crowd is that actually it often does. Skew and nearness to
the boundaries of the sample space seem to be the big culprits. It is
not difficult to find real examples where spurious interactions appear
due to the presence of skew in the dependent variable or a ceiling/floor
effect. As the various non-Gaussian models allow for skew and "know"
about the boundaries, this tends to be their real benefit. There are, of
course, many theoretical reasons to use other models, too, and far too
often statistical convenience trumps theoretical relevance.
Jay
*
* 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/
