Page 1 of 1

total indirect effects or total effect in (multiple) mediation

Posted: Mon Feb 11, 2019 4:23 pm
by Ingrid Snijders
Dear all,

When doing a (multiple) mediation analysis, I got confused whether I should use the smartpls-output value of the total indirect effect or the total effect? In the model the values for the c-path (from IV to DV) differ; one is significant (total indirect effect) and the other one is not (total effect), which means that in the case of path a x b (IV-M-DV) is significant, it concerns either a partial mediation if path c (IV-DV) is significant, or full mediation if path c is not significant.

Very much appreciated if anyone would be willing to reply asap. Thank you very much!

Ingrid

Re: total indirect effects or total effect in (multiple) mediation

Posted: Wed Feb 13, 2019 12:36 pm
by jmbecker
It depends on what you want to assess, but likely you want to look at the indirect effect (specific or total). Establishing an indirect effect validates your mediation hypothesis, i.e., that there is an intervening variable M that explains part of whole of the effect of A on B.
The total effect is the effect of A on B without any mediator. The total indirect effect is the effect though all mediators simultaneously. The specific indirect effect is the effect through one specific mediator.

See also https://www.smartpls.com/documentation/ ... /mediation

Re: total indirect effects or total effect in (multiple) mediation

Posted: Fri Feb 15, 2019 7:28 am
by Ingrid Snijders
thank you very much!

Re: total indirect effects or total effect in (multiple) mediation

Posted: Tue Sep 20, 2022 7:28 am
by paulinalo
Hi everyone,

What is your opinion (conclusion) on following mediation analysis:

Direct Effect: X -> Y B:-0.182 (negative), p=0.287 (insignificant)
Total Indirect: X -> Y total indirect effect B:0.164, p=0.047 (significant)
Specific Indirect:
X -> M1 -> Y B:0.044, p=0.274 (insignificant)
X -> M2 -> Y B:0.065, p=0.484 (insignificant)
X -> M1 --> M2 -> Y B:0.055, p=0.202 (insignificant)

Since the total indirect effect is significant and the direct effect is insignificant the conclusion should be FULL MEDIATION, however there is no specific indirect path that significant. Opinion? Please not the B of the direct effect is negative

Thank you,
Paulina Lo

Re: total indirect effects or total effect in (multiple) mediation

Posted: Tue Sep 20, 2022 11:00 am
by jmbecker
It could be a problem of insufficient statistical power. What is your sample size?

The total indirect effect is the sum of the three individual indirect effects. Its effect size is larger so it is easier to detect as significant with the same variability (uncertainty) around the estimated effects and sample size.

Overall, it seems that there could be a mediation going on, but you have not enough certainty to conclude something specific here. With larger sample sizes for example, it could also be that the direct effect becomes significant; or that some of the effects vanish. Too many possibilities to make any specific conclusion.

Re: total indirect effects or total effect in (multiple) mediation

Posted: Tue Sep 20, 2022 1:37 pm
by paulinalo
Dear Dr. Becker,

Many thanks for your reply.
My current sample size is 150. I am still collecting more sample and will try again with bigger sample size.

Sincerely,
Paulina Lo

Re: total indirect effects or total effect in (multiple) mediation

Posted: Tue Sep 20, 2022 1:44 pm
by paulinalo
Dear Dr. Becker,

It just cross my mind when I am looking at the p value of the total indirect effect is 0.047, which is very close with 0.050.
So would be better if temporary conclusion is NO EFFECT so both indirect and direct effects are not significant?

Sincerely,
Paulina Lo

Re: total indirect effects or total effect in (multiple) mediation

Posted: Wed Sep 21, 2022 9:46 am
by jmbecker
Yes. Overall, the uncertainty (standard error) of your estimates is too high at the moment to make meaningful conclusions.
You could do a power analysis and try to estimate how many observations you would need to have sufficient power to detect effects as small as those that you observe at the moment. 150 is certainly not a lot for such small effects as you usually get when looking at specific mediation paths.