Dear all,
Based on an earlier thread about calculating the marginal effects after the teffects ipw command when using the probit model, I would like to find out if it is possible to do the same with the teffects ipwra model.
The earlier thread is this one.
http://www.statalist.org/forums/foru...binary-outcome
Specifically, I attempted the following. Note that abovepov40_1 is a binary outcome 0/1 if an observation falls above or below the poverty line.
teffects ipwra (abovepov40_1 $matchvars, probit) (treat $matchvars, probit), ate /* ATE */
Iteration 0: EE criterion = 1.142e-13
Iteration 1: EE criterion = 7.925e-25
Treatment-effects estimation Number of obs = 897
Estimator : IPW regression adjustment
Outcome model : probit
Treatment model: probit
------------------------------------------------------------------------------
| Robust
abovepov40_1 | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
ATE |
treatment |
(1 vs 0) | **.2125789** .0312211 6.81 0.000 .1513867 .2737711
-------------+----------------------------------------------------------------
POmean |
treatment |
0 | .6083379 .0274081 22.20 0.000 .554619 .6620567
------------------------------------------------------------------------------
.
. /* force dropping pscores */
. cap drop pscore
. /* predict pscores */
. predict pscore, ps
. /* predict pscores at different levels */
. predict p0 p1, ps
.
. /*generate inverse probability weights */
. g w = 1/p0 if treat==0&oindex3_poly!=.
(598 missing values generated)
. replace w = 1/p1 if treat==1&oindex3_poly!=.
(598 real changes made)
.
. /* probit adjusted regressions with inverse probability weights */
. qui probit abovepov40_1 $matchvars [pw=w] if treat==0
. predict xb0
(option pr assumed; Pr(abovepov40_1))
. qui probit abovepov40_1 $matchvars [pw=w] if treat==1
. predict xb1
(option pr assumed; Pr(abovepov40_1))
.
. /*calculate differences in potential outcomes */
. g te=xb1-xb0
. su te
Variable | Obs Mean Std. Dev. Min Max
-------------+---------------------------------------------------------
te | 897 **.2125789** .14611 -.4187791 .7202263
The question is that how can one derive the marginal effects from this teffect ipwra command. Moreover, would it be possible to calculate the marginal effects with the atet option instead of ate?
Thank you very much.
Tisorn
Based on an earlier thread about calculating the marginal effects after the teffects ipw command when using the probit model, I would like to find out if it is possible to do the same with the teffects ipwra model.
The earlier thread is this one.
http://www.statalist.org/forums/foru...binary-outcome
Specifically, I attempted the following. Note that abovepov40_1 is a binary outcome 0/1 if an observation falls above or below the poverty line.
teffects ipwra (abovepov40_1 $matchvars, probit) (treat $matchvars, probit), ate /* ATE */
Iteration 0: EE criterion = 1.142e-13
Iteration 1: EE criterion = 7.925e-25
Treatment-effects estimation Number of obs = 897
Estimator : IPW regression adjustment
Outcome model : probit
Treatment model: probit
------------------------------------------------------------------------------
| Robust
abovepov40_1 | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
ATE |
treatment |
(1 vs 0) | **.2125789** .0312211 6.81 0.000 .1513867 .2737711
-------------+----------------------------------------------------------------
POmean |
treatment |
0 | .6083379 .0274081 22.20 0.000 .554619 .6620567
------------------------------------------------------------------------------
.
. /* force dropping pscores */
. cap drop pscore
. /* predict pscores */
. predict pscore, ps
. /* predict pscores at different levels */
. predict p0 p1, ps
.
. /*generate inverse probability weights */
. g w = 1/p0 if treat==0&oindex3_poly!=.
(598 missing values generated)
. replace w = 1/p1 if treat==1&oindex3_poly!=.
(598 real changes made)
.
. /* probit adjusted regressions with inverse probability weights */
. qui probit abovepov40_1 $matchvars [pw=w] if treat==0
. predict xb0
(option pr assumed; Pr(abovepov40_1))
. qui probit abovepov40_1 $matchvars [pw=w] if treat==1
. predict xb1
(option pr assumed; Pr(abovepov40_1))
.
. /*calculate differences in potential outcomes */
. g te=xb1-xb0
. su te
Variable | Obs Mean Std. Dev. Min Max
-------------+---------------------------------------------------------
te | 897 **.2125789** .14611 -.4187791 .7202263
The question is that how can one derive the marginal effects from this teffect ipwra command. Moreover, would it be possible to calculate the marginal effects with the atet option instead of ate?
Thank you very much.
Tisorn
