I would approach that as a fixed effect model at level 3, i.e., run 10
logits, m.b. xtlogits, and treat the samples as independent for
testing the equality of coefficients purposes (i.e., var[beta hat DE -
beta hat UK] = var[beta hat DE] + var[beta hat UK]). The LR test for
overall equality of coefficients is obtained by adding up the
likelihoods from each country and relating them to the likelihood of
the pooled model.
Note there are issues with scaling in panel random effect binary
models: the normalization of logit, xtlogit, re and gllamm is to set
the variance of, essentially, e_it to a known constant. The identified
parameter function, however, is beta/sqrt(total variance of the
error), so don't be too surprised when estimates from logit and
xtlogit, re differ all by exactly the same factor like 1.3: it means
that variance of u_i is 0.7 (= 1.3^2-1) of that of e_it, and only the
latter was accounted for in logit.
On 4/19/06, Martina Brandt <[email protected]> wrote:
> Can anyone recommend a textbook/paper/working example using multilevel
> modeling for international comparisons, perhaps also comparing different
> methods?
> In our special case we deal with: binary outcome; person-level,
> household-level & country-level; very small number of obs. at level 3
> (10), very high numbers of obs. at levels 1 & 2
>
> Any help would be appreciated!
>
> Best,
>
> Martina
> *
> * 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/
>
--
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/