Page 1 of 1

Indirect effects of multiple mediators

Posted: Fri Jun 04, 2010 3:38 pm
by derwinchan
Dear PLS users,

I am trying to perform a mediation analysis which involves 1 IV, 1 DV, and 3 mediators. I am quite familiar with the 4 steps approach by Baron and Kenny (1986), but I found it quite difficult to calculate the indirect effects through each mediators using PLS.

Please advise.

Many thanks,
Derwin.

Posted: Tue Jun 15, 2010 11:55 pm
by Diogenes
Hi Derwin,

See one example at
NEUFELD, D. J.; DONG, L.; HIGGINS, C. (2007). Charismatic leadership and user acceptance of information technology. European Journal of Information Systems, 16, p.494–510. [see p.505].

I hope this help you.

Best regards,

Bido

Posted: Wed Jul 28, 2010 2:40 am
by derwinchan
Thanks very much indeed!

Could Bootstrapping work for 3-Path Mediated Effect?

Posted: Mon Jan 03, 2011 10:40 am
by nicefreak
Hello everyone,

first of all, one article which I found quite useful for the assessment of multiple mediators is

TAYLOR; MACKINNON; TEIN (2005). Tests of the three-path mediated effect. Organizational Research Methods, 11 (2), p. 241-269.

I will draw on their model (p243) to phrase my question more clearly:
exogenous variable: X
two mediators: M1, M2
endogenous variable: Y

X -> M1: (path) Beta1
M1 -> M2: Beta2
M2 -> Y: Beta3
X -> Y: Beta4
X -> M2: Beta5
M2 -> Y: Beta6

Thus the whole mediation consists of (Beta1*Beta2*Beta3) + (Beta1*Beta6) + (Beta5*Beta3). Unfortunately TAYLOR et al. only give hints how best to assess the first term.
If you follow the article Prof. Bibo has quoted, you could assess the other two terms by splitting the model, once leaving out X, once leaving out Y and do the Sobel test.

Later on in my model, I plan to split my sample and do a multi-group analysis. By this means I would like to show, that the moderators work differently each time. Thus it would be useful to have the same tool to compare all the effects.

The idea I came up with was to use the "Bootstrapping the indirect effect" - method (Preacher and Hayes 2008) to calculate the standard deviation and thus to calculate the t-values
For the 2nd and 3rd term, this would be quite a standard procedure, for the first term, you could calculate
(total effect (x -> Y) - indirect effect (calculated for 2nd term) - indirect effect (calculated for 3rd term) - direct effect (X -> Y))

Could this work? Has anybody tried it before?

Thank you for your suggestions

Hope to put it right

Posted: Mon Jan 03, 2011 10:36 pm
by nicefreak
It should be easier to calculate the first term by just multiplying each Bootstrap sample (Beta1*Beta2*Beta3) to get the standard deviation!

Posted: Thu Jan 06, 2011 6:15 pm
by jmbecker
In your model you have two times M2 --> Y. (Beta3 and Beta 6)

If I suppose that it should be M1 --> Y for Beta 6

the indirect effect of X on Y is (Beta1*Beta2*Beta3) + (Beta1*Beta6) + (Beta5*Beta3)

You do not need the total effect to calculate the indirect effect. Just simply multiply the betas.

Re: Indirect effects of multiple mediators

Posted: Sun Nov 30, 2014 11:47 am
by davetol
Please, I am a doctoral candidate. i am presently analysing my data. i have 1IV, 2mediators and 1DV and the are all second order constructs(reflective-reflective). What are those things that i need to report in my study starting from missing data? Thanks.