probit - marginal effects

[Marry Lee]

Dear all!

When I use the single #, with county and year fixed effects, to run my DID regression,

 probit Outcome i.Treat#i.post_Treat  `provinceXyearFE' i.coun, cluster(coun)

I get this result:

 Treat#post_Treat |
               0 1  |  -1.011938    .361093    -2.80   0.005    -1.719667   -.3042089
               1 0  |          0  (omitted)
               1 1  |          0  (omitted)

then I run the marginal effects:

margins Treat, dydx(post_Treat) noestimcheck pwcompare (effects) post

I get the following results:

Pairwise comparisons of average marginal effects
Model VCE    : Robust

Expression   : Pr(Outcome), predict()
dy/dx w.r.t. : 1.post_Treat

------------------------------------------------------------------------------
             |   Contrast Delta-method    Unadjusted           Unadjusted
             |      dy/dx   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
1.post_Treat   |
         Treat|
     1 vs 0  |    .254434   .0719085     3.54   0.000      .113496    .3953721
------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.

Is it possible for Stata to compute right marginal effects for coefficients it omited in the probit regression?

Thank you!

[Maxence Morlet]

I recommend you take a look at the paper by Ai and Norton (2003) on why margins will yield erroneous results in nonlinear models for interactions terms.

The following commands are very helpful:

inteff by Norton, Wang and Ai

inteff3 by Cornelißen and Sonderhof

[Marry Lee]

Thank you Maxence Morlet ! I will take a look at that paper!

[Marry Lee]

Here, I want to answer to my self!

Aside from the fact that we cannot talk about 'marginal effects' with interaction terms, what I did is totally wrong!

Running the probit regression with one #, will only give the average marginal effect for 'post_Treat' with only one value of Treat:

Average marginal effects                        Number of obs     =        939
Model VCE    : Robust

Expression   : Pr(Outcome), predict()
dy/dx w.r.t. : 1.post_TCZ

------------------------------------------------------------------------------
             |            Delta-method
             |      dy/dx   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
1.post_Treat|
         Treat|
          0  |   -.254434   .0719085    -3.54   0.000    -.3953721    -.113496
          1  |          0  (omitted)
------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.

so the result with

 
 pwcompare (effects)

is exactly the same! so this is wrong.