<>
You don't have to go that far to find this issue arising:
sysuse auto
reg price mpg
reg price mpg, robust
Any program that calculates robust or HAC (Newey-West) estimates of
the VCE (regress, ivregress, newey, ivreg2 from SSC) will also
construct robust (or HAC) estimates of the F-statistic, that is,
'robustified' F stats. See Wooldridge, Introductory Econometrics, 4th
ed., Appendix E:
"A Wald statistic that is robust to heteroskedasticity of unknown form
is obtained by... and similarly for a test statistic robust to both
heteroskedasticity and serial correlation. The robust versions of the
test statistics CANNOT be computed via sums of squared residuals or R-
squareds from the restricted and unrestricted regressions." (p. 810)
So the intuition that because the sums of squares have not changed,
the degrees of freedom, the R^2 have not changed, you should get the
same F is faulty.
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 May 19, 2009, at 02:33 , Herman wrote:
I am running a series of regressions using the regress and newey
procedures. However, the F-statistic is different when I run regress y
x than when I run newey y x. I thought the newey procedure only
adjusted the VCV matrix leaving the residuals (used to compute the
F-stat) unchanged, but apparently that is not case.
Does anybody know the exact formulas that Stata uses to compute the
F-statistic under the newey procedure?
*
* 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/
