Mediation analysis using Smart PLS

Questions about the implementation and application of the PLS-SEM method, that are not related to the usage of the SmartPLS software.
Post Reply
Chu
PLS Junior User
Posts: 2
Joined: Thu Nov 24, 2016 3:58 am
Real name and title: Mr. Chuchai Smithikrai

Mediation analysis using Smart PLS

Post by Chu »

Hi everyone,

I need some advice for my mediation analysis. In my research model, there are 3 IVs, 3DVs, and 1 mediator.
To conduct the analysis, I perform the following steps:
1. Build the inner model by putting all the variables into the diagram.
2. Build the outer model by putting all indicators of each variable.
3. Connect all variables by drawing the arrows from 3 IVs --> mediator, and from mediator --> 3DVs. No arrow from IVs --> DVs.
4. Run the path-modeling estimation by selecting “PLS Algorithm".
5. Check outer model loadings and significance, indicator reliability, internal consistency reliability, convergent validity, and discriminant validity. The results show that everything is OK.
6. Check structural path significance in bootstrapping. The results show that all paths are significant.
7. I conclude that the model is correct, the mediator mediates the effects of all IVs on DVs.

My questions are:
Q1: Did I correctly conduct the mediation analysis in Smart PLS?
Q2: Did I correctly conclude the results?
Q3: Is there a more appropriate way to conduct the mediation analysis in Smart PLS?

Your advices are greatly appreciated. Thank you very much.
jmbecker
SmartPLS Developer
Posts: 1281
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Mediation analysis using Smart PLS

Post by jmbecker »

You should also connect the IVs and the DVs in your structural model.
A mediation analysis is then conducted by looking at the indirect effects from each IV via the mediator to each DV. Your mediation hypothesis is confirmed if this indirect effect is significant. If the direct effect of IV on DV is also significant you have a partial mediation, if the direct effect is not significant you have a full mediation.
If an indirect effect is not significant you don't have mediation (for this particular relationship).
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
Chu
PLS Junior User
Posts: 2
Joined: Thu Nov 24, 2016 3:58 am
Real name and title: Mr. Chuchai Smithikrai

Re: Mediation analysis using Smart PLS

Post by Chu »

Thank you Dr. Becker for your advice. In my model,I do not hypothesize any relationship between IVs and DVs, but you suggested that I should put arrows from IVs to DVs in the model. Is it OK to do that?
jmbecker
SmartPLS Developer
Posts: 1281
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Mediation analysis using Smart PLS

Post by jmbecker »

If you want to test a mediation hypothesis: yes. You have to do it to test whether there is a direct influence or whether the influence is really going through the mediator.
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
Ingrid Snijders
PLS User
Posts: 14
Joined: Sat Jul 02, 2016 2:25 pm
Real name and title: drs Ingrid Snijders

Re: Mediation analysis using Smart PLS

Post by Ingrid Snijders »

jmbecker wrote: Thu Nov 24, 2016 8:52 am You should also connect the IVs and the DVs in your structural model.
A mediation analysis is then conducted by looking at the indirect effects from each IV via the mediator to each DV. Your mediation hypothesis is confirmed if this indirect effect is significant. If the direct effect of IV on DV is also significant you have a partial mediation, if the direct effect is not significant you have a full mediation.
If an indirect effect is not significant you don't have mediation (for this particular relationship).
Dear Dr. Becker, I've run a mediator analysis with 5 IV, 3 MV, and 1 DV. My hypotheses are mainly focused on the prediction of the IV's on (eventually) the DV ( perhaps through mediation, but it is not my main goal to point that out). When including the path's between IV's and DV immediately in the model, after I've run the PLS-algorithm, R square rises substantially compared to the initial model (without the paths from IV's to DV) My question is how to correctly report the quality criteria for the measurement model?
I would be grateful if you could help me out here?
jmbecker
SmartPLS Developer
Posts: 1281
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Mediation analysis using Smart PLS

Post by jmbecker »

I'm sorry but your question is somewhat confusing.
Is your questions " how to correctly report the quality criteria for the measurement model?" which would not be connected to the question of mediation or is it about the increase in R-Square when you add direct paths?

Just to answer the second: If you want to assess mediation, you usually add the direct path. You do not compare models with and without. You may compare models with and without mediator, but not with and without direct paths, because your IV seems to by hypothesized to be predicting the outcome.
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
Ingrid Snijders
PLS User
Posts: 14
Joined: Sat Jul 02, 2016 2:25 pm
Real name and title: drs Ingrid Snijders

Re: Mediation analysis using Smart PLS

Post by Ingrid Snijders »

Dear dr. Becker,

Thank you for your reply. I meant the latter, so thank you for your answer. However, it is still puzzling to me:

When I include the mediating variables and draw direct paths from IV's to DV, the SmartPLS output show significant paths for some of the IV's with DV when I look at the Total indirect effects, however, the same paths are not significant when I look at the SmartPLS output under Total effects (or vice versa; total indirect effects value was not significant, and total effects value is significant). Therefore, I am not sure how to report the results from my model.

Furthermore, in the figure on mediation when( p1 x p2) is not significant one must examine p3 i.e., total indirect effect. But how to report on directs paths from IV('s) to DV; do I use the model with mediating variables, and then use the total effects value or the total indirect value for the path from IV to DV, or do a run another model excluding the mediating variables and use the total effects value?

Sorry for all these questions, but I hope you can help me out. I'm in the proces of submitting a revised manuscript, and I just want things to be correct. Thank you in advance!

-Ingrid
jmbecker
SmartPLS Developer
Posts: 1281
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Mediation analysis using Smart PLS

Post by jmbecker »

As you have posted under a different topic you might be confusing some of the terminology. You may also read some additional literature on how to test mediating effects (in general and in PLS). Some helpful information and additional literature can be found here: https://www.smartpls.com/documentation/ ... /mediation
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
Post Reply