Hugh Colaco <[email protected]>:
Selection on X (one of the RHS vars) does not introduce bias per se;
just estimate for those who work, and plan to generalize results to
that population. The main problem you face is one of endogeneity, due
most obviously to simultaneity (those with high wealth may have
different reservation wages which affects their observed wage) and
less obviously to measurement error (a big concern with measured wages
on the RHS). You should find excluded instruments for wage (findit
ivreg2 and read the help file and references). Another problem is the
huge skew in wealth, noting also that wealth may be zero or negative.
You might consider a cube-root transformation of wealth.
On Wed, Dec 2, 2009 at 12:58 PM, Hugh Colaco <[email protected]> wrote:
> Hi,
>
> I hope I am clearer this time in my example.
>
> This is the -heckman- example from the -help- menu using the -twostep-
> procedure:-
>
> webuse womenwk
> heckman wage educ age, select(married children educ age) twostep
>
> Suppose I want to take it one step further and examine how wage
> influences the personal
> wealth of women.
>
> Basically, I need to run the following model:-
>
> wealth = f(wage, control variables)
>
> I have data on the variable "wealth" for women who work AND women who
> don't work.
> I have data on some control variables only for women who work. I also
> want to include in the regression interaction variables with wage
> using some of the control variables that have data only for women who
> work.
>
> How do I take care of the sample selection bias in the wealth equation?
*
* 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/
