Many thanks for your help.
I was not aware of the "boundary of the parameter space" issue.
In addition to Stata's help j_chibar, I also found Bosker & Snijders
(1999:90) very helpful.
Pierre
On 05/09/06, Joseph Coveney <jcoveney@bigplanet.com> wrote:
Pierre Walthery wrote:
I am fitting a 2 levels random intercept model. The model is a
multinomial logit. In order to fit it, I am using gllamm, with no
random effects specified. I then get the normal gllamm output with an
estimate of level 2 variance.
My question is: how can I test whether this level 2 variance is
significant - say by comparison with a one level mlogit model? I know
about the possibility to use a likelihood ratio test to compare two
nested models, but this would not allow me to test specifically for
level 2 variance, or is there anything I am missing here?
--------------------------------------------------------------------------------
I'm not very familiar with multinomial logistic modeling in particular, but
in general I thought that a likelihood ratio test is indeed how you would
test the hypothesis that the level-2 variance component is greater than zero
with -gllamm-.
The general form is
gllamm . . . , i(id)
estimates store RandomEffectsIncluded
mlogit . . . // use identical fixed effects as with -gllamm-
lrtest RandomEffectsIncluded ., force
There is some considerations involved in the assignment of the probability
associated with the likelihood ratio chi-square test statistic in these
cases. This is in order to take into account that the null hypothesis is on
the boundary of the parameter space. It is explained in -help j_chibar-.
Joseph Coveney
*
* 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/
*
* 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/