Dear Statalist,
I am running ivregress for the regression with one endogeneous variable X and its interaction with an exogeneous dummy D (0 or 1):
Y = b0 + b1 X + b2 X*D, where I instrument X and X*D with Z and Z*D.
I find the estimated coefficients in the first stage logical, however my worry is in the standard errors in the first stage.
1) If I don't include vce(robust) option, I get some p-values equal 1 in both first stage regressions;
2) with vce(robust) see the output below;
3) and with vce(cluster var) both first stage regressions lack estimation of standard errors and show "Warning: variance matrix is nonsymmetric or highly singular".
I couldn't come up with explanation for this issue with standard errors, and I am not sure if I can use these 2SLS estimates in the end.
Would be very much thankful for any insight.
I am running ivregress for the regression with one endogeneous variable X and its interaction with an exogeneous dummy D (0 or 1):
Y = b0 + b1 X + b2 X*D, where I instrument X and X*D with Z and Z*D.
I find the estimated coefficients in the first stage logical, however my worry is in the standard errors in the first stage.
1) If I don't include vce(robust) option, I get some p-values equal 1 in both first stage regressions;
2) with vce(robust) see the output below;
3) and with vce(cluster var) both first stage regressions lack estimation of standard errors and show "Warning: variance matrix is nonsymmetric or highly singular".
Code:
. ivregress 2sls Y (X c.X#D = Z c.Z#D) D, vce(robust) first
First-stage regressions
-----------------------
Number of obs = 1,120
F( 3, 1116) = 104.82
Prob > F = 0.0000
R-squared = 0.7058
Adj R-squared = 0.7050
Root MSE = 0.6680
------------------------------------------------------------------------------
| Robust
X | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
D | -3.10e-15 .122954 -0.00 1.000 -.241247 .241247
Z | .8933804 .0712486 12.54 0.000 .753584 1.033177
|
D#c.Z |
1 | 2.10e-15 .1007608 0.00 1.000 -.1977019 .1977019
|
_cons | .1350366 .0869416 1.55 0.121 -.0355508 .305624
------------------------------------------------------------------------------
Warning: variance matrix is nonsymmetric or highly singular
Number of obs = 1,120
F( 0, 1116) = .
Prob > F = .
R-squared = 0.8272
Adj R-squared = 0.8268
Root MSE = 0.4723
------------------------------------------------------------------------------
| Robust
1.D#c.X | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
D | .1350366 . . . . .
Z | -2.02e-15 . . . . .
|
D#c.Z |
1 | .8933804 . . . . .
|
_cons | 3.00e-15 . . . . .
------------------------------------------------------------------------------
Instrumental variables (2SLS) regression Number of obs = 1,120
Wald chi2(3) = 26.17
Prob > chi2 = 0.0000
R-squared = 0.0212
Root MSE = .74611
------------------------------------------------------------------------------
| Robust
Y | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
X | -.0004001 .0336982 -0.01 0.991 -.0664473 .0656471
|
D#c.X |
1 | .0269442 .049598 0.54 0.587 -.070266 .1241544
|
D | .1875806 .0797811 2.35 0.019 .0312125 .3439487
_cons | -.0700652 .0508648 -1.38 0.168 -.1697582 .0296279
------------------------------------------------------------------------------
Instrumented: X 1.D#c.X
Instruments: D Z 1.D#c.Z
I couldn't come up with explanation for this issue with standard errors, and I am not sure if I can use these 2SLS estimates in the end.
Would be very much thankful for any insight.
Comment