st: RE: Question about Gllamm (mlogit link)

["Verkuilen, Jay" <JVerkuilen@gc.cuny.edu>]

Karine B�gin wrote:

>>>I am using gllamm to compute multinomial logistic regressions with random effects.  After looking at my results, I noticed that depending on which value of my dependent variable I chose to specify as the base outcome, I get different results (different LL, coefficients, z-score).  However, when performing multinomial regressions on a non-nested sample in Stata (using the mlogit command), the results I get are identical (same LL, coefficients are of the opposite sign but of the same value) notwithstanding the reference category specified for the dependent variable. I was wondering if anyone has an idea as to why gllamm yields different results depending on the base outcome specified or if there would be a way to obtain the results (coefficients, p-value, ...) for all oppositions at once.<<<

Sounds to me like the models aren't actually statistically equivalent. The mlogit estimates are statistically equivalent after change of baseline choice, but the mixed models may not be if the covariance structure of the random effects may differ after application of identification constraints. Getting that right is tricky. You may want to build the model up a few terms at a time to see where things go wrong. 

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/