RE: st: RE: loglink and normality, mixed model

["Feiveson, Alan H. (JSC-SK311)" <alan.h.feiveson@nasa.gov>]

Hi Ruud - I experimented with some simulated clustered normally distributed data (y) with positive subject-specific means, but large enough error term so that some values of y were negative. -meglm- worked fine. In my simulation, I purposely generated all the subject-specific means to be positive. I think the problem may be that with a log link, all the subject-specific means must be positive, so that if the estimated variance of the random intercept is too large at any stage of the estimation, some subject-specific means would become negative, thus conflicting with the model.


Al


-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Ruud van Lieshout
Sent: Tuesday, February 25, 2014 4:08 AM
To: statalist@hsphsun2.harvard.edu
Subject: RE: st: RE: loglink and normality, mixed model

hi Al, 

well in theory you are right, but my question also holds for negative values ..The model can generate negative values (eps ~N(0, sigma^2))  but meglm can't handles those !! Why not?

regards,
Ruud



*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/