indirect effect Vs Mediation

[ela...@yahoo.com]

Hi,

I have a query that has caused serious issues . I have read several articles in which authors are claiming mediation but they never show up their statistical findings. Some authors are strongly of the view that there is no difference between mediation and indirect effect. furthermore if we use product of coefficient approach , then we have to find effect size as well and then only we will be able to decide weather it is a indirect effect or not (many calling it as mediation or not). Ringles's approach in his PLS SEM is based upon Barron and Kenny as per his 2013 article. So he is more or less asking us to have a direct relationship between X-Y to determine mediation or indirect effect . Which is opposite to what Wrap PLS does. 

I believe that in both cases using Wrap PLS or PLS SEM the algorithm uses X-Y relationship in calculations that is total effect. 

I am unable to understand the real difference between indirect effect and mediation in this new discussions particularly I have a straight question.

Q. If X-Y relationship is not important at all then why we must include it in our theoretical frameworks and even in calculations to measure complete or partial mediation. 

If any articles are there that discuss these issues please do attach them. ( I mean papers published by different authors in different fields of business)

Regards

[Ralph Foorthuis]

Hi,

 

You are right that it is indeed often not made clear what mediation is conceptually and how it differs from indirect effects (if at all). My own opinion is that if you have a model with what you could call “indirect effects” (A -> B -> C), you should always do the tests that mediation analysis prescribes. For example, even if your theoretical causal (inner) model does not presuppose a direct relationship between A and C, you should still verify this direct statistical relationship (not only simply between A and C, but also between A and C when controlling C for B). Furthermore, if you discover that the effect of B on C is not present anymore when controlling for A, then your model may have a problem. So, whatever label you want to attach to it, I think you should always do these analyses and interpret the meaning of the results.

 

An article I like is Mediation in Experimental and Nonexperimental Studies: New Procedures and Recommendations (Shrout & Bolger, 2002).

 

Regards,

Ralph Foorthuis

--
--
You received this message because you are subscribed to the Google
Groups "PLS-SEM" group.
To post to this group, send email to pls...@googlegroups.com
To unsubscribe from this group, send email to
pls-sem+u...@googlegroups.com
 
---
You received this message because you are subscribed to the Google Groups "PLS-SEM" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pls-sem+u...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

[Ned Kock]

WarpPLS calculates indirect and total effects separately, and calculates the effect sizes for each and every one of them as well. This is a unique feature of WarpPLS that is largely unexplored in the test of mediating effects:

 

http://bit.ly/12UrTeK

 

We recently had a paper under review for a major journal, which is now getting close to acceptance, where one of the reviewers wanted us to use Baron & Kenny and Preacher & Hayes.

 

We did, along with the outputs generated by WarpPLS, and showed that the results were essentially the same, with the difference that the WarpPLS results were more completed (they included effect sizes).

 

In the end we of course did what the reviewers wanted, as doing the opposite is often fatal in the review process. If you want to take a look at examples, both Moqbel’s and Gaskins’s dissertations provide extensive reports of indirect and total effects:

 

http://www.scriptwarp.com/warppls/pubs/Moqbel_2012_PhDDiss.pdf

 

http://www.scriptwarp.com/warppls/pubs/Gaskins_2013_PhDDiss.pdf

 

Sometimes reviewers want to know if an indirect effect refers to full or partial mediation. In these cases, the criteria used by Baron & Kenny can be used (see link below), but instead of building smaller models one can use the correlations among LVs reported by WarpPLS.

 

http://warppls.blogspot.com/2010/07/testing-significance-of-mediating.html

 

Using LV correlations, and respective P values, is equivalent to the process of building smaller (or simpler) models employed by  Baron & Kenny, as the latter aims at capturing the bivariate relationships to be compared with the multivariate ones in the full model.

 

 

-----------------------------------------------------------

Ned Kock

http://nedkock.com

 

 

[ela...@yahoo.com]