st: question concerning generalized hausman test for clustered std errs using SUEST, regress and ivregress

[Woolton Lee <finished07@gmail.com>]

Hello,

I am trying to compute a Hausman test between two models both adjusted
for clustering for patient zip codes.  One model is simple OLS, the
other is IV where one variable has been instrumented.  I have tried to
obtain this using the SUEST procedure, but I am finding a problem with
the patient zip code clustered standard errors for the IV estimates
obtained from the SUEST procedure (which are clearly wrong).  I
estimate the first and second stage regression both via OLS because
SUEST will not accept the variance covariance matrix from IVREGRESS (I
keep getting an error indicating the model ..."was estimated with a
non-standard vce (unadjusted)" - this is even when I set the VCE
option in IVREGRESS to unadjusted).

The instrumented variable (in the IV regression) is sl_change_end.
This variable has the following coefficient and standard error
estimates (in parentheses) from clustered IV regression:  -1.16
(1.607)

Using OLS with clustered standard errors I get the following
coefficient (std err):  -0.125 (0.169)

When I run SUEST with zip code clustered standard errors I get for the
IV model: -1.16 (6.54)

and for the OLS model I get -0.125 (0.169)

The problem is that the standard errors I get from SUEST for the IV
estimate is clearly wrong so the result I get from the generalized
HAUSMAN test is that I can't reject the null that the coefficients
between the two models are the same.  It seems strange to me that
SUEST will not accept the covariance matrix from IVREGRESS.  Is there
no way to implement this test or am I just missing something?

Many thanks for your help,

Woolton
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/