<>
Although the justification for many things with regard to instrumental
variables estimators is only asymptotic, I don't see why the Durbin-Wu-
Hausman test implemented by Baum/Schaffer/Stillman's -ivreg2- option -
endog()- would be problematic in a sample of size 186. That does not
sound particularly small to me, but the ability of the test to
distinguish between OLS and IV may be fairly weak with that amount of
data.
I have simulated the performance of -endog()- for different sample
sizes, random subsets of the griliches76 dataset used in -help ivreg2-
in which I have added random noise to the variable considered
endogenous. Here is the average p-value for the Durbin-Wu-Hausman
endogeneity test, derived from 1000 replications at each sample size
(en):
en | mean
---------+----------
50 | .398134
100 | .3549545
150 | .3360662
200 | .3105404
250 | .2703655
300 | .2614716
350 | .2356522
400 | .1967296
450 | .1733359
500 | .1310702
550 | .1082535
600 | .0737662
650 | .0503589
700 | .0329359
750 | .0164609
---------+----------
It is apparent that a test that can comfortably reject the null that
OLS is appropriate at a sample size of 600+ has little power to do so
for sample sizes of 150-200. However, I am not aware of any
modification to the test that would enable it to be more powerful in
smaller samples.
Kit Baum | Boston College Economics & DIW Berlin | http://ideas.repec.org/e/pba1.html
An Introduction to Stata Programming
| http://www.stata-press.com/books/isp.html
An Introduction to Modern Econometrics Using Stata | http://www.stata-press.com/books/imeus.html
On Mar 28, 2009, at 02:33 , Stephen wrote:
I am trying to test for endogeneity but I think I the command that I
am using to test for endogeneity in STATA is not very powerful when
the sample size is small.
I have a panel of 31 countries and 6 four-year periods.
I am usinging: ivreg2 y x1 x2 (x3 = z1...zn), endog(x3)
where y is the dependent variable, the xi are the independent
variables so that x3 is endogenous, x1 and x2 are exogenous and the zi
are instruments.
Does anyone know of a test of endogeneity in STATA witha small sample
correction?
In order words is there a command in STATA that can test for
endogeneity when the sample size is small?
*
* 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/
