if we propose in a theoretical model that a particular relationship can have multiple mediators, do we have to gauge the effect of a mediator separately in the relationship of predictor and criterion variable in structural equation modeling or we can enter all the mediators in the model between predictor and criterion, will it be ok ???
Razia, I would include all mediators in the model and test the complete model. This is similar to multiple regression: The path coefficients are partialled effects controlling for all other variables in the model. If your hypothesis is that the indirect effect is mediated through one or more mediators, I don't think that testing the mediators separately is necessary, as the effects may change dramatically when a variable is left out.
You need to first test individually the suspected mediator. If found significant by the Sobel Test or Preacher Test, then report in the final SEM equation. Recall that mediation and moderation effects are indirect effect measurements. We will report them only if they are significant.
There are procedural steps outlined at this link: https://en.wikipedia.org/wiki/Mediation_(statistics)
Relevant articles on mediation measurement are also attached. In particular, the articles by Preacher and Hayes are quite informative. Cheers.
https://en.wikipedia.org/wiki/Mediation_(statistics)
Razia, I would include all mediators in the model and test the complete model. This is similar to multiple regression: The path coefficients are partialled effects controlling for all other variables in the model. If your hypothesis is that the indirect effect is mediated through one or more mediators, I don't think that testing the mediators separately is necessary, as the effects may change dramatically when a variable is left out.
HI,
The attached recent published articles with advanced statistical analyses my help.
good luck
Thanks Sir Paul, Ma'am Karin and Sir Muayyad for insightful feedback, in fact all the replies are so helpful, thanks again
You can model multiple mediators in a model. But when come to testing, you need to test each and every mediator separately. You can proceed to do bootstrapping to confirm the mediation results.
Sir Zain, Thanks for your valuable guidance, so it means that i need to do the CFA for every path in SEM or pooled CFA as suggested in your book chapter, and also test its structural model of every mediator separately, am I correct
Yes, better do separately if you interest on Baron&kenny approach. However, you can try hayes approach in which combination of ML-SEM with bootstrap to increase the accuracy of parameter estimate for mediator construct
Thanks Sir Wan Mohamad, actually I'm more interested in Hayes approach, but a little confused about how to calculate the contrasting effects of multiple mediators simultaneously.
I believe Preacher and Hayes (2008) recommend using a multiple mediator model approach to examine mediation effects of several mediators, where the total indirect effect, specific indirect effects (of each mediator beyond other mediators in the model), and pairwise contrasts (relative magnitude / strength of each mediator) are estimated.
Am not personally familiar with the above approach so I can't explain further, but hope the above is still useful for you.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling