Re: st: Why do i get a conformability error in Gllamm6?

[Michael Ingre <Michael.Ingre@ipm.ki.se>]

On 2004-07-14, at 12.47, Daniela New wrote:

> I'm sorry to ask such a basic question, but i thought i could cluster
> first on the firms and then on the country in order to run 2 different
> random effects (i.e. writing i(code country)
This would have worked if your firms had been active in one country 
only (like members of families) but your firms are in all countries. To 
my knowledge, you can't estimate a random effect for both at the same 
time. I guess it might be possible in theory but very complicated to 
estimate two random effects nested in each other.

You could (in theory) have one random effect for each country by adding 
dummys and specifying one equation for each dummy. These random effects 
would describe the firm specific country effect for each of the 
countries. However, don't even try to estimate a model with 10 random 
effects. You are way past the practical limit.

If you are considering using time as one predictor in your model, I 
would suggest that you model it as a random coefficient. This would 
allow for firm-specific slopes of time, which (I think) often makes 
perfect sense.

. gen const = 1
. eq intercept: const
. eq slope: year

. gllamm y x year , i(code) nrf(2) eqs(intercept slope) adapt .....

Michael


*
*   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/