st: GLM / blogit /glogit

["Allan Reese (Cefas)" <allan.reese@cefas.co.uk>]

Hi friends, can I get an opinion on when to use the various commands for
fitting observed counts?  I've looked round for discussion of the models
but find most references come back to Stata - so I'm chasing my tail! 

Consider some grouped data: r successes from n trials with covariate x
measured for each group.
The command options are

glm r x, fam(bin n)
blogit r n x
glogit r n x

The first two give the same parameter estimates but different df and
goodness of fit.  My interpretation is that if x is determined once for
each group (eg a shared treatment) then the number of groups is
appropriate, but if each observation of x is made independently (ie is a
covariate for the individual) then the expanded count may be valid.

glogit however uses LS not ML estimators and gives different parameter
estimates.  But the LS estimators are biased, and I can't find any
indication when or why you might prefer the LS model.  Isn't it just
something that was computationally easier before we had good glm
software?

glm , irls gives different estimates yet again, presumably because of
reweighting at each iteration.

Comments please.

Regards
Allan  



***********************************************************************************
This email and any attachments are intended for the named recipient only.  Its unauthorised use, distribution, disclosure, storage or copying is not permitted.  If you have received it in error, please destroy all copies and notify the sender.  In messages of a non-business nature, the views and opinions expressed are the author's own and do not necessarily reflect those of the organisation from which it is sent.  All emails may be subject to monitoring.
***********************************************************************************


*
*   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/