st: Different results from GLLAMM after each run?

[Daniel Schneider <daniel.schneider@stanford.edu>]

I am somewhat puzzled by how much my results change when I use GLLAMM. 
Here is the code I have been using:

gen chosen=alt==dv

tab alt,gen(a)
eq a2: a2
eq a3: a3

gllamm alt iv1 iv2 iv1Xiv2 ,expand(respid chosen m) i(interviewer ) 
lin(mlogit) family(binom) nrf(2) nip(4) eqs(a2 a3) adapt

matrix a=e(b)

gllamm alt iv1 iv2 iv1Xiv2 ,expand(respid chosen m) i(interviewer ) 
lin(mlogit) family(binom) nrf(2) nip(8) eqs(a2 a3) adapt from(a)

matrix a=e(b)

gllamm alt iv1 iv2 iv1Xiv2 ,expand(respid chosen m) i(interviewer ) 
lin(mlogit) family(binom) nrf(2) nip(12) eqs(a2 a3) adapt from(a)



so I am using GLLAMM to run a random effects model for a multinomial 
logit regression and I am increasing the nip-value step by step using 
the prior coefficients as starting values. Note that iv1Xiv2 is the 
interaction of iv1 and iv2.

The results that I am getting when re-running these commands over and 
over again are quite similar, but they are not identical. In fact the 
results for iv1 look very similar but both iv2 and the interaction term 
do show some fluctuation. Nothing substantive but the p-value can range 
from .03 to .09 which is somewhat unfortunate.

Is there something that I am missing or doing wrong? Any ideas why these 
results might be so unstable? Any suggestions are appreciated!


Daniel
*
*   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/