Hi Massoud,

  Take a look at :

Preacher, K. J., & Kelley, K. (2011). Effect size measures for mediation models: quantitative strategies for communicating indirect effects. Psychological methods, 16(2), 93.

  They provide some useful information about the interpretation, in terms of effect size, of mediation models/indirect effects.

- joe

Article Effect Size Measures for Mediation Models: Quantitative Stra...

Hi Massoud,

  Take a look at :

Preacher, K. J., & Kelley, K. (2011). Effect size measures for mediation models: quantitative strategies for communicating indirect effects. Psychological methods, 16(2), 93.

  They provide some useful information about the interpretation, in terms of effect size, of mediation models/indirect effects.

- joe

Article Effect Size Measures for Mediation Models: Quantitative Stra...

Hi,

the Sobel test is applied to calculate the statistical significance of the indirect effect not its size. The paper the other colleagues have posted should answer the latter question.

In addition, full medication means that there is no direct effect and that all causal flows are transmitted by the mediator.

Best,

Holger 

Hi Massoud

I think the general consensus nowadays is to favour the use of bootstrapping and effect sizes as opposed to causal steps and a sobel test. 

The limitations with the sobel test is that it relies on significance testing and normal distribution.  It could be argued that categorising something as full or partial mediation has its weaknesses and is misleading ....for example, if you're claiming partial mediation you're actually 'celebrating' model misspecification.  Also, you can actually have have multiple full mediators within the same model.

As such, I would be inclined to read the paper below regarding your approach and use the other paper Joseph provided to determine an appropriate effect size measure (in most situations kappa squared has its merits).

Regards

Tom

Massoud, you're getting good advice. I want to add that I agree with Tom about full mediation. Full mediation is, to my mind, a wild goose chase. To claim in, you're endorsing the null--that the direct path is exactly zero in the population.

In order to add something to this picture I wish to propose two more issues. Namely:

1. While analyzing mediation effects it could be of interest to additionally mention suppression effects. I do recommend the following paper: (which by the way discourages the use of Full and Partial mediation analysis and instead proposes the effect size focus)

Rucker, D.D., et al., Mediation Analysis in Social Psychology: Current Practicies and New Recommendations, 2011.

2. The idea of implicit mediation analyses has been in the literature for a while. In my opinion it is an interesting concept worth mentioning. For an example:

Gerber, A.S., et al., Social Pressure and Voter Turnout: Evidence from a Large-Scale Field Experiment, 2008.

Last but not least, I find the following book very useful "Introduction to Mediation, Moderation, and Conditional Process Analysis" by A.Hayes, 2013.

If you have a single moderator model I'd say that PRODCLIN is the way to go. 

Nadra and Massoud,

FYI, Tofhigi and MacKinnon have updated and improved MacKinnon's PRODCLIN with an R package, RMediation. The full text of the article describing RMediation is at the link.

Pat

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3233842/

Dear Massoud,

Nowadays, Sobel test is not the favorite test for determining the significance of mediation. Probably, you should more focus on the effect size. Please visit the following page for more information:

http://davidakenny.net/cm/mediate.htm

Best wishes,

Dr. Sanaz Aazami

Thanks everyone very much! I have been (and I am still ) reading all references suggested. Thank you for that. However, the same question, then will remain, when it comes to the effect size.

Is there an agreed effect size as to strong, medium and weak? 

Below please find what I have gathered more concretely from the link that Sanaz suggested. http://davidakenny.net/cm/mediate.htm

Two questions:

1) Is the below statement a generally agreed statement?

2) How about the ratio of indirect to total effect size? Could that be a better indication? 

[

There are two different strategies for determining small, medium, and large effect sizes. (Any designation of small, medium, or larger is fundamentally arbitrary and depends on the particular application.) First, following Shrout and Bolger (2002), the usual Cohen (1988) standards of .1 for small, .3 for medium, and .5 for large could be used. Alternatively and I think more appropriately because an indirect effect is a product of two effects, these values should be squared or rr. Thus, a small effect size would be .01, medium would .09, and large would be .25.

]

Thank you very much,

Massoud

Hi Massoud

Kappa squared is a meaningful effect size that you could use (there are calculators online)...although I think the mathematical criteria behind is quite advanced (for me anyway!) my understanding is that it represents the size of ab as compared to the maximum possible value that ab could be based on your parameters.

The limitation with using ratio measures such as the indirect/total effect size is that it assumes that the effect size to which you are comparing ab (in this case c) is larger than ab when that needn't be the case...for example you may have a suppressor effect which is limiting the size of c.

Cheers

Tom

Kindly find in detail the answer for your question in recently published paper...

Article Making Sense of Mediating Analysis: A Marketing Perspective

Thank you very much for all the good advice!

Halloo everyone..

I am conducting a study on Retinopathy and Depression. I carried out two separate mediation analysis using the Sobel test. Can anyone please tell me in simple words when to say that a variable is mediating the relationship between two other variables using the Sobel test. These are the results of the Sobel test internet calculator for both of my mediation models (z= -0.16 ; p= 0.87, (z = 0.31; p= 0.76). Greatly appreciate all your help...

Regards..Ghada

As you can see none of the mediation is significant. Your z value is very low and your p value is very high!

Thank you