Quantcast

Re: Marginal effects for Heckman two-step selection model

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Marginal effects for Heckman two-step selection model

Maarten buis
--- On Wed, 17/11/10, Lenka Vavroušková wrote me privately:

> I would like to ask you for a help. I have read all your
> responses in relation to this topic on Stata website and it
> seems that you are the expert in this field.
>
> Unfortunatelly I am not sure how I can calculate the
> marginal effects after the heckman two-step estimation and
> correctly interpret them. I used the mfx command (mfx,
> predict(xb)) to calculate unconditional marginal effects and
> I got the same values as the coefficients in outcome
> equation. I read your explation for it and it makes sence
> that the first derivation of linear function will be the
> coefficient.
>
> Is there a command in STATA 10 which will allow me to get
> the marginal effect which will represent the effect of the
> unit change of the explanatory variable on the dependent
> variable holding other variables at means? I would like to
> get values that I can interpret like in this example:
>
> "Considering all the explanatory variables at their mean
> value, a one year increase in education will cause a 11.6%
> increase in women’s earnings, for those working, and a
> 5.2% increase in the proportion of women working,
> corresponding to a 17.4% increase in total women’s
> earnings."
>
> Followed the Stata manual I used: mfx, predict(yexpected)
> and I got the marginals effects that are much different from
> the coefficients in outcome equation. But I am not so sure
> how I can interpret them. The manual says that they are
> marginal effects of all independent variables for the
> expected value of the dependent variable. Would be the value
> 17.4 % in example?
>
> Then used mfx, predict(psel) to obtain marginal effects for
> the probability of being observed (would it be 5.2 % in
> example?)
>
> and then use mfx, predict(ycond) to get the change in
> dependend variable for "working woman" - like the number
> 11.6 %?
>
> I have feeling that is not right :-(
>
> I have no logaritm or other mathematical transformations in
> my variables. But I have dummy variables - for age, salary
> and occupation. But there were not created in Stata, they
> were already created as dummy variables in excel. My feeling
> is that STATA will calculate the marginal effect to hold the
> other dummy variables from the same group at mean instead of
> 0 (I mean i.e. calculating marginal effect for the age group
> 20-29, it would make sense to hold the other age groups at
> 0) Is there is a way how deal with this?

The Statalist FAQ explains why the way to ask for help is to
contact the statalist and not individual members directly:
<http://www.stata.com/support/faqs/res/statalist.html#private>

The interpretation in terms of percentages comes from the log
transformation of income that is common in economics. Since
you do not do the log transformation the default output
from -mfx- is in absolute terms (i.e. euros/kg/m/...) when it
comes to the main equation.

-- Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------




     

*
*   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/
Loading...