st: Pls Help, fail overid test for some subsamples, not for fullsample - xtiverg2

[mparrat@bu.edu]

Hi,

I am estimating a model with panel data in which one of the covariates 
is endogenous.  I have chosen a set of instruments and I am using 
xtivreg2, fe r.   My instruments are significant (jointly and 
separately) and I fail to reject Hansen?s J statistic test, i.e., the 
model is overidentified.  The sample size is over 9000 observations (9 
years, 1000 units approx.)

My problem is that the model is no longer overidentified (reject 
Hansen?s J) when I run the estimation for some subsets of my sample, 
say for a group of units given by one specific characteristic (about 
6000 obs).  This, however, only happens for some subsets not all and 
these are not necessary the smallest.  Moreover, the estimate of my 
endogenous variable is not significant anymore for these subsamples 
(but continues to be for other subsamples).

So, what are the implications on these findings?  For example, does the 
rejection of the test for a subset of the sample invalidate the results 
on the full sample?  Also, I assume I could not say anything about the 
estimates of the non-overidentified models, that is, I cannot compare 
them with the full sample or other subsamples, which was the point of 
splitting the sample.  Am I right?

I appreciate any comments on this,

Best regards,

Monica



*
*   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/