[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: Dummy Variables vs. Subgroup Models in Logistic Regression
yet another option for multiple group analysis is to estimate a random
coefficients model with -gllamm-. If there are differences in
intercepts or slopes between groups, the corresponding variances
should be positive and significantly greater than zero. Of course
estimating a variance with just 8 observations is a tough task, and it
is far beyond the asymptotic domain of the maximum likelihood
estimates. With more groups, you can have somewhat better confidence
in your results, but you would have to interpret your results as if
each group gives a unique contribution to the random effects, and you
wanted to keep all interactions of race/gender/whatever in your model.
The non-zero variance would then mean that at least some of the
combinations are different from others.
That all is still a bit shaky.
* For searches and help try: