Is there a command to test for significant differences between the
coefficients of running a model with different samples?
____________________________________________
J. Benson Durham
Economist
This question is a FAQ on Statalist. You can't do that unless you can
get the separate estimates of the model into the same estimation
problem (and into the same estimated covariance matrix). For instance,
if you have two samples of size N, you might use SURE (sureg), which
would freely estimate two coefficient vectors with different \sigma^2
and allow you to test constraints of equality. (That will work for any
larger number of samples, for that matter). If the samples are of quite
different sizes, you could pool them and "dummify" the sample
indicator, interacting it with each regressor; then you have to worry
about the imposition of a common error variance. But if you bring me
two printouts and ask for a test of "significant differences", the Mark
1.0 eyeball will be the only tool available.
