st: Pairwise correlation and underidentification test

["Lim Boon Leong" <lboonl@streamyx.com>]

Dear Statalists,

I have three suspected endogenous regressors and eight instrument variables
(IVs).  According to the advice of Cameron and Trivedi (2009), one should
check for the relevant of the IVs by looking at the pairwise correlation as
a rough guide.  The pairwise correlation coefficients between these
endogenous regressors and IVs are significant at 5% or lower level, and the
values of correlation coefficients ranges from the lowest -0.0640 to the
highest -0.4081.  And quite a number of correlation coefficients fall
between -0.1195, -0.2138 to -0.3644.

Despite these significant correlation, the results for first-stage
regressions (I am using xtivreg2) show that I cannot reject the null
hypothesis of underidentification and weak id.

						Underid
(Weak id)
Variable  	F (9, 105)	p-val		AP Chi-sq (7) P-val	AP
F(7, 105)
X1		2.37		0.0177	0.43		0.9997	0.06
X2		2.80		0.0055	0.08		1.0000	0.01
X3		3.17		0.0020	0.15		1.0000	0.02
 
The AP Chi-sq test shows that it is underidentified or do not fulfill the
rank condition which also means that there is not enough correlation between
the IVs and the endogenous regressors.  But how about the F-statistic which
show that these IVs are significant at 5% to 1% levels for X1, X2, and X3?

In order to pass the underidentification test, as a rough guide what should
be the values for pairwise correlation coefficients between endogenous
regressors and IVs.

Any comments are most welcomed.

Thanks
 
Best regards,
Lim Boon Leong 



*
*   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/