[Mediation-information] question about 'mediation'

Kosuke Imai kimai at princeton.edu
Tue Nov 22 23:20:23 CET 2011


The total effect of the mediator is b_mediator.  The indirect effect represents the effect of the treatment (not the mediator) that goes through the mediator. 

Kosuke

Department of Politics
Princeton University
http://imai.princeton.edu

On Nov 22, 2011, at 4:00 PM, Daniel Lim wrote:

> Thanks so much for the response. To delve a little further, what I'd really like to do is say what the total effect of the mediator is. To further sketch out the scenario:
> 
> M = b1_treatment*T
> Y = b_mediator*M + b2_treatment*T
> Y = b_treatment*T
> b_treatment = ATE = direct eff + indirect eff
> 
> I get all of that.  What I'm unclear about is how I can interpret the indirect effect in conjunction with b_mediator.  Could I say something like "the effect of the mediator is (indirect effect + b_mediator) per unit mediator" or is the indirect effect already accounted for in b_mediator?  Or is the situation more like:
> 
> total effect of mediator = (1-p)indirect_effect + b_mediator
> b_mediator = p*indirect_effect + independent effect of mediator
> 
> In substantive terms, I'm not only interested in the effect of the treatment but that of the mediator.
> 
> On Tue, Nov 22, 2011 at 5:58 AM, Kosuke Imai <kimai at princeton.edu> wrote:
> Hi Daniel,
> 
> If you regress Y on the treatment (but not the mediator), then you will get the average total effect which equals the sum of direct and indirect effect.
> 
> Kosuke
> 
> Department of Politics
> Princeton University
> http://imai.princeton.edu
> 
> 
> On Nov 22, 2011, at 5:18 AM, Daniel Lim wrote:
> 
> > Dear Professor Imai, I'm a poli sci PhD student at UCLA and I'm currently working on a paper involving your mediational model.  I had a quick question regarding the interpretation of results:
> >
> > Say that we are looking at the following linear models:
> > M = b1_treatment*T
> > Y = b_mediator*M + b2_treatment*T
> > In a nonmediational setting, we would interpret the effect of the mediator as b_mediator per unit M.
> >
> > When we apply "mediate" to the above, we get a result like the following:
> > Mediation Effect: X
> > Direct Effect: Y
> > Total effect X+Y
> > How do we interpret this result in conjunction with the original outcome model?  Is X additive with b_mediator or does it supplant that result?  In other words, are we saying b2_treatment=X+Y or b2_treatment+b2_mediator=X+Y (or are both these interpretations completely off)?
> >
> > Thanks,
> > --
> > Daniel Lim
> > PhD Student
> > Dept of Political Science, UCLA
> > daniel.k.lim at gmail.com
> > 818-480-2679 (USA)
> >
> >
> 
> 
> 
> 
> 
> 
> -- 
> Daniel Lim
> PhD Student
> Dept of Political Science, UCLA
> daniel.k.lim at gmail.com
> 818-480-2679 (USA)
> 
> 




More information about the Mediation-information mailing list