[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: Testing the Equality of Coefficients Across Two xtabond Arellano-Bond Regressions

From   Ana Fernandez <>
Subject   st: Testing the Equality of Coefficients Across Two xtabond Arellano-Bond Regressions
Date   Thu, 18 Jun 2009 11:02:18 -0400

Dear Statalist,

I am trying to get stata to test the equality of coefficient estimates
following two xtabond arellano-bond regressions.  Normally I would run
suest and lincom following two regressions but this doesn't work after
xtabond because xtabond is is gmm estimation. Consider the following:

xtabond pctfemmanager if fed==1, lags(1)
xtabond pctfemmanger if fed==0, lags(1)

(This corresponds to specification "pctfemmenagaer = a +
b*L1.pctfemmanager" , and fed is a 1/0 dummy variable).

Now I want to test whether the coefficient on lagged pctfemmanger (the
main dependent variable) in regression1 is statistically significantly
different from the coefficient on lagged pctfemmanager in regression2.
Normally I would run the following test of equality of coefficients
(after ols or fixed effects this works for example), but this doesn't
work with xtabond because xtabond doesn't produce score terms:

xtabond pctfemmanager if fed==1, lags(1)
estimates store reg1
xtabond pctfemmanger if fed==0, lags(1)
estimates store reg2
suest reg1 reg2

At this point I get an error message that says "reg1 was estimated
with a non-standard vce (gmm)"
If I hadn't gotten this error I could have just tested with following code:
lincom [reg1_mean]L1.pctfemmanger-[reg2_mean]L1.pctfemmanager

Do you know any other way to do this test? How can I test the equality
of coefficients in two xtabond regressions? I am even willing to
assume that the two samples for regression 1 and 2 are independent so
that the covariances of the estimates between the two regressions are
zero--perhaps there is a way to do this using the "test" command
rather than "suest"? Or maybe "hausman"?

Ana Fernandez

I tried running a single regression

xtabond pctfemmanager fed pctfemmanager_lag1Xfed, lags(1)

where pctfemmanager_lage_1Xfed is the interaction of pctfemmanager and
fed in order to use the "test" command to execute the above task.  I
expected this regression with interaction to give equivalent estimates
as running two regressions on the samples for fed=1 and fed=0 but the
estimates were not identical (which must have to do with xtabond since
running it with ols/reg would have given equivalent estimates).

*   For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index