Thank You all for your comments. Yes I am trying to do something very
similar to Angrist 1998, but for South Africa in this case. So if I
understand correctly, having high standard errors is a common problem
when one uses Ivs, however since the z-values are low and p-values high
would it still be safe to assume there is causality however with low
prediction power?
Going back to Asutin's point about clustering, I have only one year of
data so I was not sure if I should still go ahead and cluster? I am also
not clear about what variable to use for clustering in this case. Any
suggestions?
Thanks once again for all the help.
Shruti
Shruti Kapoor, Adjunct Instructor
Economics Department, Occidental College
Office: Fowler 214 (enter through Fowler 211)
Phone: (323)- 259-1322
e-mail: [email protected]
Fax: (323)- 259-2704
-----Original Message-----
From: Shruti Kapoor
Sent: Thursday, November 19, 2009 9:29 PM
To: '[email protected].'
Subject: IV reg est with high standard errors
Hi All,
I am doing IV regressions using ivreg2, while my OLS estimates are all
significant with the right signs etc, the 2SLS results have unusually
high standard errors.
Here is a sample of the results, could you please provide me some
insight on how I can correct for the standard errors.
The command I am using is ivreg2 wrk4pay age agetry black asian color
yrschl urban (morethan2children= twoboy twogirls)
IV (2SLS) estimation Estimates efficient for homoskedasticity only
Statistics robust to heteroskedasticity
Number of obs = 90511
F( 9, 90501) = 1640.98
Prob > F = 0.0000
Total (centered) SS = 18904.19625 Centered R2 = 0.1552
Total (uncentered) SS = 26897.32544 Uncentered R2 = 0.4062
Residual SS = 15970.85841 Root MSE = .4201
Robust
wrk4pay Coef. Std. Err. z P>|z| [95% Conf. Interval]
morethan2c~n -.0191096 .1308071 -0.15 0.884 -.2754868 .2372676
age .0166344 .0066042 2.52 0.012 .0036904 .0295785
agetry -.0015381 .006236 -0.25 0.805 -.0137605 .0106843
black -.2965384 .0119606 -24.79 0.000 -.3199807 -.2730961
asian -.2025844 .0093118 -21.76 0.000 -.2208352 -.1843335
color -.0619381 .010215 -6.06 0.000 -.0819592 -.0419171
yrschl .0160991 .0017973 8.96 0.000 .0125764 .0196218
urban .0629951 .0150462 4.19 0.000 .0335052 .0924851
boyfirst -.0084626 .0030072 -2.81 0.005 -.0143565 -.0025686
_cons -.1698414 .0221104 -7.68 0.000 -.2131771 -.1265058
Underidentification test (Kleibergen-Paap rk LM statistic): 57.237
Chi-sq(2) P-val = 0.0000
Weak identification test (Kleibergen-Paap rk Wald F statistic): 28.638
Stock-Yogo weak ID test critical values: 10% maximal IV size 19.93
15% maximal IV size 11.59
20% maximal IV size 8.75
25% maximal IV size 7.25
Source: Stock-Yogo (2005). Reproduced by permission.
NB: Critical values are for Cragg-Donald F statistic and i.i.d. errors.
Hansen J statistic (overidentification test of all instruments): 0.175
Chi-sq(1) P-val = 0.6761
Instrumented: morethan2children
Included instruments: age agetry black asian color yrschl urban boyfirst
Excluded instruments: twoboys twogirls
Dropped collinear: white
Shruti Kapoor, Adjunct Instructor
Economics Department, Occidental College
Office: Fowler 214 (enter through Fowler 211)
Phone: (323)- 259-1322
e-mail: [email protected]
Fax: (323)- 259-2704
*
* 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/
