# Re: st: Computing interaction effects after svy:heckman

 From "Joao Ricardo F. Lima" To statalist@hsphsun2.harvard.edu Subject Re: st: Computing interaction effects after svy:heckman Date Tue, 12 Aug 2008 14:15:03 -0300

```Dear Stas,

thanks for your response. I really want to report the results
separately for rural and urban locations. Could you give me more one

This is my output after mfx

. mfx

Marginal effects after svy:heckman
y  = Fitted values (predict)
=  .18253958
------------------------------------------------------------------------------
variable |      dy/dx    Std. Err.     z    P>|z|  [    95% C.I.   ]      X
---------+--------------------------------------------------------------------
age |    .211089      .00614   34.35   0.000   .199046  .223132   37.2801
age2 |  -.0025012      .00008  -32.03   0.000  -.002654 -.002348   1524.69
years_~y |   .2013313      .01286   15.65   0.000   .176123   .22654   8.85091
yearss~e |  -.0511932      .01299   -3.94   0.000  -.076658 -.025728   7.97836
place_live *|   3.154186      .10662   29.58   0.000   2.94522
3.36316   .837693

Can I just calculate the effects of years_study doing:

Dummy=0 effect years_study =  .2013313
Dummy=1 effect years_study =  .2013313 + ( -.0511932 )= 0.15014

In this case, if the person lives in the rural, one more year of study
increase the lnincome in 0.2013313. If lives in the urban, one more
year of study increase the lnincome in 0.15014.

Thanks a lot,

Joao Lima

2008/8/12 Stas Kolenikov <skolenik@gmail.com>:
> -margeff- and -inteff- are user written additions, probably with a
> very specific objective in mind. The reason for interactions in
> regression models is to modify the main effects, so that effects of
> years of study depend on the place where you live. Probably you would
> want to report the results separately for rural and urban locations.
>
> Note that your model is only identified by the variables -race- and
> -region-. I presume those are categorical, and should enter through
> the set of dummies, e.g. as -xi : ... i.race i.region)-. Even with
> that, I would not say that's a terribly strong form of identification,
> especially if those are not terribly significant in the selection
> equation.
>
> On Tue, Aug 12, 2008 at 10:22 AM, Joao Ricardo F. Lima
> <jricardofl@gmail.com> wrote:
>> Dear Statalisters,
>>
>> I'm using the heckman procedure with survey data. I have an
>> interaction term (yearsstudy_placelive) that is a product of a
>> continuous variable and a dummy variable. My model is
>>
>> *****************
>> svy linearized: heckman lnincome age age2 years_study place_live
>> yearsstudy_placelive, select (ocup = age age age2 years_study
>> place_live yearsstudy_placelive race region) log
>> ****************
>>
>> where place_live is a dummy (0=rural; 1=urban)
>>
>> Iīm using margeff, because itīs easy to calculate the total effect of
>> age. However, I actually donīt known how to calculate the effect of
>> the interaction term in the regression model. I only can use inteff
>> after probit or logit.
>>
>> Then, I really appreciate an advice of how I can calculate this effect.
>>
>> Thanks a lot,
>>
>> Joao Lima
>> --
>
>
>
> --
> Stas Kolenikov, also found at http://stas.kolenikov.name
> Small print: I use this email account for mailing lists only.
>
> *
> *   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/
>

--
-------------------------------
Joao Ricardo Lima
Professor
UFPB-CCA-DCFS
+553138923914
-------------------------------

*
*   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/
```