st: Re: ivprobit marginal effects

[Christopher Baum <baum@bc.edu>]

That is because, as help ivprobit postestimation indicates, the default action of predict is to compute xb, the latent variable, rather than the probability of  a positive outcome, option pr.

Using the example from help ivprobit:

. webuse laborsup, clear

. ivprobit fem_work fem_educ kids (other_inc = male_educ), nolog

Probit model with endogenous regressors           Number of obs   =        500
                                                  Wald chi2(3)    =     163.88
Log likelihood = -2368.2062                       Prob > chi2     =     0.0000

------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
   other_inc |  -.0542756   .0060854    -8.92   0.000    -.0662027   -.0423485
    fem_educ |    .211111   .0268648     7.86   0.000     .1584569    .2637651
        kids |  -.1820929   .0478267    -3.81   0.000    -.2758316   -.0883543
       _cons |   .3672083   .4480724     0.82   0.412    -.5109975    1.245414
-------------+----------------------------------------------------------------
     /athrho |   .3907858   .1509443     2.59   0.010     .0949403    .6866313
    /lnsigma |   2.813383   .0316228    88.97   0.000     2.751404    2.875363
-------------+----------------------------------------------------------------
         rho |   .3720374   .1300519                      .0946561    .5958135
       sigma |   16.66621   .5270318                      15.66461    17.73186
------------------------------------------------------------------------------
Instrumented:  other_inc
Instruments:   fem_educ kids male_educ
------------------------------------------------------------------------------
Wald test of exogeneity (/athrho = 0): chi2(1) =     6.70 Prob > chi2 = 0.0096

. margins

Predictive margins                                Number of obs   =        500
Model VCE    : OIM

Expression   : Fitted values, predict()

------------------------------------------------------------------------------
             |            Delta-method
             |     Margin   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       _cons |  -.1417621   .0624493    -2.27   0.023    -.2641605   -.0193638
------------------------------------------------------------------------------

. margins, pred(pr)

Predictive margins                                Number of obs   =        500
Model VCE    : OIM

Expression   : Probability of positive outcome, predict(pr)

------------------------------------------------------------------------------
             |            Delta-method
             |     Margin   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       _cons |   .4627798   .0161664    28.63   0.000     .4310943    .4944654
------------------------------------------------------------------------------


. margins, dydx(_all) pred(pr)

Average marginal effects                          Number of obs   =        500
Model VCE    : OIM

Expression   : Probability of positive outcome, predict(pr)
dy/dx w.r.t. : other_inc fem_educ kids male_educ

------------------------------------------------------------------------------
             |            Delta-method
             |      dy/dx   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
   other_inc |   -.014015   .0009836   -14.25   0.000    -.0159428   -.0120872
    fem_educ |   .0545129   .0066007     8.26   0.000     .0415758      .06745
        kids |  -.0470199   .0123397    -3.81   0.000    -.0712052   -.0228346
   male_educ |  (omitted)
------------------------------------------------------------------------------


. mfx compute

Marginal effects after ivprobit
      y  = Fitted values (predict)
         = -.14176214
------------------------------------------------------------------------------
variable |      dy/dx    Std. Err.     z    P>|z|  [    95% C.I.   ]      X
---------+--------------------------------------------------------------------
other_~c |  -.0542756      .00609   -8.92   0.000  -.066203 -.042348   49.6023
fem_educ |    .211111      .02686    7.86   0.000   .158457  .263765    12.046
    kids |  -.1820929      .04783   -3.81   0.000  -.275832 -.088354     1.976
------------------------------------------------------------------------------

. mfx compute, pred(pr)

Marginal effects after ivprobit
      y  = Probability of positive outcome (predict, pr)
         =  .44363395
------------------------------------------------------------------------------
variable |      dy/dx    Std. Err.     z    P>|z|  [    95% C.I.   ]      X
---------+--------------------------------------------------------------------
other_~c |  -.0214364      .00242   -8.87   0.000  -.026176 -.016697   49.6023
fem_educ |   .0833791      .01057    7.89   0.000   .062664  .104094    12.046
    kids |  -.0719183      .01888   -3.81   0.000  -.108927  -.03491     1.976
------------------------------------------------------------------------------

. 
end of do-file

. 



Note that -margins- in Stata 11 gives you the same fitted value of -0.1418 as does mfx_compute. Likewise, margins, pred(pr) gives you a probability. That value agrees with mfx compute, pred(pr).  The marginal effects from the two commands do not agree because mfx compute evaluates the derivatives at the point of means, whereas margins computes average marginal effects.

help mfx  shows you that the default setting is 'discrete', that is, evaluate the marginal effect of a dummy going from 0->1. 'nodiscrete' turns that off. For the margins command, it is the continuous option, which is not the default. Dummies by default are treated as dummies by both mfx and margins.


Kit Baum   |   Boston College Economics and DIW Berlin   |   http://ideas.repec.org/e/pba1.html
An Introduction to Stata Programming   |   http://www.stata-press.com/books/isp.html
An Introduction to Modern Econometrics Using Stata   |   http://www.stata-press.com/books/imeus.html

On Mar 2, 2010, at 11:56 AM, Torres, Margarita Liliana Vides Morales de wrote:

> Dear Kit:
> I am working on a model using ivprobit over 8 different sets of data, when
> looking at the result from mfx command (Stata10) all of them are negative
> and its absolute value greater than 1.
> As an example one of the result says:
> Marginal effects after ivprobit
> y = Fitted valued (predict)
>  = -2.02  How should I interpret this number?
> At the same time the first derivative dy/dx are the same as the regression
> coefficients.
> One last thing where can I find documentation of treatment of a dummy in
> the marginal effected calculated by stata 10? the help of mfx just said
> that is calculated with their mean, does it apply for the dummies variable
> as well?
> 
> When working with probit mfx result is positive lower than 1. and the
> coefficients are different from the regression, all is perfect.
> 
> Thanks in advance.
> 
> Liliana Vides
> 


*
*   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/