3
$\begingroup$

I was looking at the following question, which appears to suggest that multiple indirect effects can be added together to calculate a total indirect effect. Multiple mediation analysis in R

But reading the abstract of this paper leads me to believe that there is a way to calculate the total indirect effects all together, and that calculating them separately and adding them together may be incorrect.

What would be the proper way to calculate a total indirect effect?

In this example, could you add the indirect effects from ACY, ADY, BCY and BDY? Would it be more appropriate to keep A and B separate? Or is there a different way to summarize the indirect effects? Model with 2 main effects and 2 mediators

If I want to know how much of CY is not explained by and indirect effect of A or B, could I subtract ACY and BCY from CY?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
4
$\begingroup$

It only makes sense to calculate the sum of indirect effects from the same predictor to the same outcome. OLS regressions decompose variance into explained and unexplained components. SEMs decompose covariances, too. Your mediation model decomposes the total effect of (e.g.) A on Y (given B) into a direct effect (AY) and multiple indirect effects (ACY and ADY). If you add AY+ACY+ADY, that total effect should match an OLS regression slope regressing Y on A (controlling for B), assuming your model is saturated by allowing A to covary with B and C to covary with D.

Improve this answer
$\endgroup$
4
  • $\begingroup$ Thanks very much, that makes sense. I just added another question at the end, do you have any suggestions about this last question? $\endgroup$
    – canderson156
    Jun 15, 2022 at 13:03
  • $\begingroup$ It sounds like you just want 1 minus R-squared for C. That has nothing to do with Y. $\endgroup$
    – Terrence
    Jun 16, 2022 at 14:03
  • $\begingroup$ I see I wasn't clear, any thoughts on my updated question? $\endgroup$
    – canderson156
    Jun 16, 2022 at 17:08
  • $\begingroup$ It still does not make sense. CY is not being decomposed. AY (and BY) are being decomposed. If all the effects have the same sign, you could calculate the proportion of the total AY (or BY) effect that is accounted for by the indirect path via C and/or D. $\endgroup$
    – Terrence
    Jun 17, 2022 at 9:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.