st: Multilevel (Panel) Multinomial Logit with GLLAMM

[Merle K <merle.k@hotmail.de>]

Dear all,

I have a rather long list of clarification questions regarding the estimation of a panel mlogit with the gllamm programme (Stata 13):
I have a panel of a number of political groups in 12 countries over 20 years, each year deciding which of three strategies they employ to reach their goals.
I definitely have two levels: the occasions or years and the respective organisations. However, I would also assume that within the same country, the organisations’ decisions will not be independent from each other. Would it be sufficient to use the cluster- option or should I include the countries as a third level?
          At the moment, I use a multinomial logit with random intercepts as described in chapter 9.3 of the manual. When adding the third level, would the random intercepts be the same for both levels? Following that example, would I simply add nrf(2, 2) eqs(a2 a3 a2 a3)?
          Also, intuitively, I would have included organisation or country fixed effects, would that be advisable? However, I would not know how to implement them as in nonlinear models I cannot just include organisation and country dummies without running into the incidental parameter problem.
          I do not use random coefficients because there is very little variation in most of the variables over time. Is this the correct
              reasoning?
This is my code ignoring the country level (mode being the variable that describes the three possible strategies):

tab mode, gen(mode)
eq mode1: mode1
eq mode2: mode2
eq mode3: mode3

gllamm mode explanatory variables, expand(group chosen m) i(orgid) lin(mlogit) family(binom) basecategory(1) eqs(mode1 mode3) nrf(2) nip(4) adapt trace

I would be extremely grateful for any feedback and thoughts!
Best regards,
Merle 		 	   		  
*
*   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/