re: Re: st: medeff command negative percentage of total effect mediated

["Ariel Linden, DrPH" <ariel.linden@gmail.com>]

Yes, I have experienced this myself, using a variety of different modeling
approaches. It all depends on the variables you have, and how they perform
in one or both models (treatment to mediator) (mediator to outcome).

You may want to read this paper that Kristian Karlson and I just wrote on
mediation.* We test a variety of approaches, using various combinations of
mediator and outcome variable types. I believe that we may have encountered
this behavior in some of our simulations (to the best of my recollection).

* Linden A, Karlson KB. Using mediation analysis to identify causal
mechanisms in disease management interventions. Health Services and Outcomes
Research Methodology. Epub ahead of print [March 2013]


Date: Tue, 4 Jun 2013 18:20:25 +0100
From: Maria Chatzivasileiadi <mariachatzivasileiadi@gmail.com>
Subject: Re: st: medeff command negative percentage of total effect mediated

Dear Ariel,

Thank you very much for your reply. I hadn't read this article and it
does provide some useful references that I will go through.
I had a priori hypothesized that the indirect and direct effects would
be at opposite directions, what I am not sure is whether the program
takes into account that it is dividing a positive by a negative sign
and therefore gives a negative percentage of total effect mediated.
Have you used it before and encountered something similar?

Many thanks,
Maria

2013/6/4 Ariel Linden, DrPH <ariel.linden@gmail.com>:
> This not an unusual finding, and it is discussed in almost every paper on
> mediation. I suggest you read the paper by Adam Glynn* as a starting
point.
> You can go from there to his reference list to learn more if that does not
> suffice
>
> *Glynn, A.N.: The product and difference fallacies for indirect effects.
Am.
> J. Political Sci. 56, 257-269 (2012)
>
> Ariel
>
> Date: Mon, 3 Jun 2013 11:16:20 +0100
> From: "worldpatient ." <worldpatient@gmail.com>
> Subject: st: medeff command negative percentage of total effect mediated
>
> Dear Statalisters,
>
> I am using the user-written command -medeff command in Stata/IC 11 for
> Windows to test for causal mediation effects.
> My outcome is binary and my mediator is continuous.
>
> My question has to do with the interpretation of the percentage and
> 95%CIs of total effect mediated. In my analysis, the direct and total
> effects although not significant are negative while the indirect
> effect is positive and very small.
> As a result the percentage of the total effect mediated is negative
> and the CIs range from -203 to 46..
>
> Is there a way that I could interpret the percentage of the total
> effect mediated and the CIs when this percentage is negative and the
> CIs are so wide?
>
> Many thanks,
> Maria
>
>


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/