Ahn,Taeyong posted on Thursday, September 14, 2017 - 12:05 am
additional questions.I found that the r3step of GMM(w/o covar.) gives me a different result from GMM (user guid ex 8.1). And du3step of GMM(w/o covar.) also gives me a different result from GMM(from user guid ex 8.6). I think that the GMM(user guid ex 8.1) and GMM (from user guid ex 8.6) result are more proper. I wonder what results I should report.
The models in 8.1 and 8.6 not only have c on x, but also i, s on x. The latter is not covered in the automatic 3-step methods. That is, there are no effects from covariates that don't go through c. In reality of course there might be such effects (influence of x within classes).