Hi!
I want to estimate a Difference in Difference with GMM estimations using the command xtabond2.
I have two questions: Is there any good reason why GMM estimations are problematic for a DID context?
Second: Is the following code from below correct for such an estimation? The omitted constant as well as the Hansen and Sargan lead me to believe that there must be some error.
N=849, n=283, T=3years
dependent variables: con (Con. Vote Share)
Independent variable: i.treat##i.post (DID estimator)
Control variables: lagvote (lagged con. vote share), population, const (constituency fixed effects)
In the results I omitted the constituency estimations for a better overview.
I want to estimate a Difference in Difference with GMM estimations using the command xtabond2.
I have two questions: Is there any good reason why GMM estimations are problematic for a DID context?
Second: Is the following code from below correct for such an estimation? The omitted constant as well as the Hansen and Sargan lead me to believe that there must be some error.
N=849, n=283, T=3years
dependent variables: con (Con. Vote Share)
Independent variable: i.treat##i.post (DID estimator)
Control variables: lagvote (lagged con. vote share), population, const (constituency fixed effects)
Code:
. xtsum treat post const lagvote population
Variable | Mean Std. dev. Min Max | Observations
-----------------+--------------------------------------------+----------------
treat overall | .1107185 .3139682 0 1 | N = 849
between | .1572736 0 .3333333 | n = 283
within | .2718444 -.2226148 .7773852 | T = 3
| |
post overall | .3333333 .4716824 0 1 | N = 849
between | 0 .3333333 .3333333 | n = 283
within | .4716824 0 1 | T = 3
| |
const overall | 273.7244 156.8698 2 531 | N = 849
between | 157.0551 2 531 | n = 283
within | 0 273.7244 273.7244 | T = 3
| |
lagvote overall | 42.08616 12.7877 6.63736 69.91752 | N = 849
between | 11.94309 8.288684 62.84266 | n = 283
within | 4.606984 28.43678 55.31837 | T = 3
| |
popula~n overall | 11.50537 .1096006 11.1961 11.87054 | N = 849
between | .1090479 11.19899 11.86021 | n = 283
within | .0122013 11.46214 11.5484 | T = 3
Code:
. xtabond2 con i.treat##i.post i.const lagvote population age_50 if conwin == 1, gmm(L.con) iv(i.const i.post la
> gvote i.treat) robust orthogonal small
Dynamic panel-data estimation, one-step system GMM
------------------------------------------------------------------------------
Group variable: const Number of obs = 510
Time variable : year Number of groups = 170
Number of instruments = 173 Obs per group: min = 3
F(174, 169) = 0.09 avg = 3.00
Prob > F = 1.000 max = 3
---------------------------------------------------------------------------------------------------------
| Robust
con | Coefficient std. err. t P>|t| [95% conf. interval]
----------------------------------------+----------------------------------------------------------------
1.treat | 2.443444 .8396104 2.91 0.004 .7859689 4.10092
1.post | .5576694 1.377677 0.40 0.686 -2.162004 3.277343
lagvote | .8216299 .1354844 6.06 0.000 .5541701 1.08909
population | 1.936205 .9694733 2.00 0.047 .0223673 3.850042
age_50 | -.7147711 .317661 -2.25 0.026 -1.341866 -.0876763
_cons | 0 (omitted)
---------------------------------------------------------------------------------------------------------
Arellano-Bond test for AR(1) in first differences: z = . Pr > z = .
Arellano-Bond test for AR(2) in first differences: z = . Pr > z = .
------------------------------------------------------------------------------
Sargan test of overid. restrictions: chi2(-2) = 0.00 Prob > chi2 = .
(Not robust, but not weakened by many instruments.)
Hansen test of overid. restrictions: chi2(-2) =37205.46 Prob > chi2 = .
(Robust, but weakened by many instruments.)
