Re: st: mediation by subgroup

[Maarten buis <maartenbuis@yahoo.co.uk>]

--- On Mon, 7/3/11, Fabio Zona wrote:
> I run the same model  in two subgroups and wish to
> test whether the coefficients significantly differ in the
> two subgroups (i.e. chow test).
> 
> However, each model includes a mediation effect.
> Specifically, each model includes:
> --- three predictor variables
> --- a mediation effect (the effect of one predictor is
> mediated by a Mediator)
> --- control variables

I assume you are using linear regression on an untransformed
dependent/explained/y variable (if you do not then mediation 
becomes a lot more complex). In that case the solutions seems
to me so simple that it does not deserve its own name: Just
estimate one model, add interactions, and test whether these
interactions are all equal to zero. In the example below:
assume that race has an effect on income through education 
(grade) and union is a control variable and you want to compare
married and unmarried women (the dataset contains only women).

*----------------- begin example ---------------
sysuse nlsw88, clear
gen byte baseline = 1
gen c_grade = grade - 12

reg wage i.race##i.married    ///
         c.c_grade##i.married ///
         i.union##i.married   ///
         , coefl

test 2.race#1.married 3.race#1.married     ///
     1.married#c.c_grade 1.union#1.married ///
     1.married
 *-------------- end example --------------------
(For more on examples I sent to the Statalist see:
http://www.maartenbuis.nl/example_faq )

Hope this helps,
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/