Dear Statalist,
I would like to ask about the result of first-stage F-statistics that I can obtain from ivreg2.
My regression model is:
y = INDP1 + INDP2 + INDP3 + control variables + country-fixed effect + year-fixed effect
As I suspect INDP1, INDP2, INDP3 are endogenous, I instrumented them with 4 exogenous variables.
(INDP1 is binary variable, INDP2 and 3 are continuous variables)
When I ran ivreg2 command, I got the following first-stage F statistics.
(Underid) (Weak id)
Variable | F( 4, 289) P-val | SW Chi-sq( 2) P-val | SW F( 2, 289)
INDP1 | 18.93 0.0000 | 3.54 0.1703 | 1.34
INDP2 | 26.18 0.0000 | 2.27 0.3221 | 0.86
INDP3 | 48.42 0.0000 | 2.71 0.2575 | 1.03
Weak identification test
Cragg-Donald Wald F statistic 0.42
Kleibergen-Paap Wald rk F statistic 0.17
Weak-instrument-robust inference
Anderson-Rubin Wald test F(4,75)= 0.63 P-val=0.6436
Anderson-Rubin Wald test Chi-sq(4)= 3.36 P-val=0.4996
Stock-Wright LM S statistic Chi-sq(4)= 11.83 P-val=0.0186
As shown in the first stage F-statistics, F-statistic of each endogenous regressor is larger than the conventional threshold of 10. However, the Kleibergen-Paap Wald rk F statistic (hereafter, K-P F statistic) displays a very low value of 0.17.
Further, the result of weak-instrument-robust inference displays a somewhat contradicting result.
For example, according to the Anderson-Rubin Wald test, I fail to reject the null hypothesis.
However, if I use the Stock-Wright LM S statistic, I can reject the null hypothesis at p<0.1.
So, my questions are ..
- Would anyone know the reason for the high F-statistic for each endogenous regressor but the very low K-P F statistic?
- Would anyone know the reason for the conflicting result of weak-instrument-robust inference?
I searched various research papers to study the original definition of first-stage F-statistics and find the reason, but yet to find something valid.
Sincerely,
I would like to ask about the result of first-stage F-statistics that I can obtain from ivreg2.
My regression model is:
y = INDP1 + INDP2 + INDP3 + control variables + country-fixed effect + year-fixed effect
As I suspect INDP1, INDP2, INDP3 are endogenous, I instrumented them with 4 exogenous variables.
(INDP1 is binary variable, INDP2 and 3 are continuous variables)
When I ran ivreg2 command, I got the following first-stage F statistics.
(Underid) (Weak id)
Variable | F( 4, 289) P-val | SW Chi-sq( 2) P-val | SW F( 2, 289)
INDP1 | 18.93 0.0000 | 3.54 0.1703 | 1.34
INDP2 | 26.18 0.0000 | 2.27 0.3221 | 0.86
INDP3 | 48.42 0.0000 | 2.71 0.2575 | 1.03
Weak identification test
Cragg-Donald Wald F statistic 0.42
Kleibergen-Paap Wald rk F statistic 0.17
Weak-instrument-robust inference
Anderson-Rubin Wald test F(4,75)= 0.63 P-val=0.6436
Anderson-Rubin Wald test Chi-sq(4)= 3.36 P-val=0.4996
Stock-Wright LM S statistic Chi-sq(4)= 11.83 P-val=0.0186
As shown in the first stage F-statistics, F-statistic of each endogenous regressor is larger than the conventional threshold of 10. However, the Kleibergen-Paap Wald rk F statistic (hereafter, K-P F statistic) displays a very low value of 0.17.
Further, the result of weak-instrument-robust inference displays a somewhat contradicting result.
For example, according to the Anderson-Rubin Wald test, I fail to reject the null hypothesis.
However, if I use the Stock-Wright LM S statistic, I can reject the null hypothesis at p<0.1.
So, my questions are ..
- Would anyone know the reason for the high F-statistic for each endogenous regressor but the very low K-P F statistic?
- Would anyone know the reason for the conflicting result of weak-instrument-robust inference?
I searched various research papers to study the original definition of first-stage F-statistics and find the reason, but yet to find something valid.
Sincerely,
