Friday, March 17, 2006
Re: st: GLMs that fail - an effect (Hauck-Donner)
Allan was reporting on estimation problems when the success/failure was predicted perfectly, in Stata jargon. One solution out of this problem would be to switch to the Bayesian world and put some priors on your estimates (because otherwise you are estimating the probability to be 1, and the variance of the binomial with p=1 is p(1-p)=0, and so the information matrix is singular, and what not). That way, you can get some spread of the posterior. I am not sure if this is an option for clinical trials -- all I know is that it is a heavily regulated area... And I am not sure what do you mean by robust procedures here; you can specify -robust- option with -logit-, but it would not resolve the generic issue of lack of information (zero variance). Those are just my two cents here.
-- Stas Kolenikov http://stas.kolenikov.name
* * 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/
Links to this post: