All,
I checked the Statalist archives and a rather good econometrics textbook,
but came up with nowt for this problem.
In preparing to run some complex panel regressions, I'm running OLS models
(incorporating a lagged dependent variable) to test for all sorts of
non-sphericity in the errors, including endogeneity as well as
autocorrelation and the rest.
Unfortunately, I'm running into some bizarre difficulties in testing for
the endogeneity of my explanatory variables, in particular my LDV. When
saving the residuals (E) after -reg- and then correlating E with the X's,
I keep getting zero correlations. Let me show you one example:
pwcorr e ledconpc ed2-ed13 edpollch lagconch laglabch lagldmch clmargin
cdmargin edenp class restnw rurwales indswles perwmids eastmids eanglia
corndevn wessex roseland if e(sample), sig
| e ledconpc ed2 ed3 ed4
------------+----------------------------------------------
e | 1.0000
|
|
ledconpc | -0.0000 1.0000
| 1.0000
|
ed2 | . . .
| 1.0000 1.0000
|
ed3 | -0.0000 0.0992 . 1.0000
| 1.0000 0.0000 1.0000
|
ed4 | 0.0000 -0.0020 . -0.0853 1.0000
| 1.0000 0.9300 1.0000 0.0002
This is a snapshot from the original model (-reg-): the last 9 independent
variables are regional dummy variables. All of the X's from the regression
show zero-correlations with E, including (puzzlingly) the LDV (LEDCONPC),
which ought to be correlated with E. But this was despite finding very
significant autocorrelation (as well as heteroscedasticity).
Some private correspondence some time ago indicated to me that I should
change the specification if this happens. I've done this too, by dropping
the regional dummies and other non-essential background X-vars: but I get
the same result as above.
Following Wooldridge (2003: 507), I've also run a 'reduced test' of
endogeneity: i.e., -regress- the LDV on the other variables, save the
residuals and then -pwcorr if e(sample)- as above. This also produces the
same results: every X-var shows a zero-correlation with E.
Something is up, but I have little idea what. I've completely run out of
diagnostic options here, so any suggestions as to where I'm going wrong
are most welcome.
CLIVE NICHOLAS |t: 0(044)191 222 5969
Politics |e: [email protected]
Newcastle University |http://www.ncl.ac.uk/geps
Wooldridge, JM (2003) INTRODUCTORY ECONOMETRICS (2nd ed.); Mason, OH:
South Western.
*
* 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/
