Page **1** of **1**

### Isolating indirect effect from Models with two mediators

Posted: **Thu Aug 13, 2015 6:38 am**

by **jta**

My mediation model consists of the following: Only one DV = Y; Only one IV = X ; Two mediators = M1 and M2.

Specifically, the mediation model is as follows:

Direct effect of X on Y = Path C';

Indirect effect of X on Y through only M1 = a1*b1;

Indirect Effect of X on Y through only M2 = a2*b2;

Effect from M1 to M2 = a3; and

Indirect Effect of X on Y through M1 and M2 in sequence =a1*a3*b2

I noticed that SMARTPLS does not produce outputs for each of the following indirect effects:

Indirect effect of X on Y through only M1 = a1*b1; and

Indirect Effect of X on Y through only M2 = a2*b2.

Please correct me if I have just overlooked the SmartPLS outputs.

### Re: Isolating indirect effect from Models with two mediators

Posted: **Thu Aug 13, 2015 12:26 pm**

by **jmbecker**

You are correct. SmartPLS only gives the results for the normal indirect effect that is the total effect minus the direct effect or in other words all possible indirect paths together. As complex path models like the ECSI can have a multitude of different specific indirect effects, it would be computationally to demanding to provide the specific indirect effect for each path. In addition, it would cloak the results with many additional information of which only a few are relevant for some researchers.

If you want to test specific indirect paths, you need to do that on your own by multiplying the respective direct effects as you have shown yourself.

To get the bootstrapped significance, you need to take the data from the path coefficients "Samples" output and calculate the specific indirect effects for each sample and then calculate the standard deviation over all these specific indirect effects. The t-values is then the original sample estimate of the specific indirect effect divided by the bootstrapped standard deviation. Percentile bootstrap confidence intervals are just the 2.5% and 97.5% percentiles of all these bootstrapping results.

### Re: Isolating indirect effect from Models with two mediators

Posted: **Fri Aug 14, 2015 8:24 am**

by **jta**

Jmbecker, I used bootstrapping with 5000 subsamples. To follow your suggestion, i need the excel file of the SmartPLS outputs of the direct path coefficients. This excel file should contain the direct path coefficients for each of the 5000 bootstrap subsambles. From the smartpls outputs, where can I find the direct path coefficients for each of the 5000 subsamples? I am asking this question because the Excel at the bottom-left of the calculation Results cant be opened. I am using a trial version of Smartpls.

### Re: Isolating indirect effect from Models with two mediators

Posted: **Fri Aug 14, 2015 8:50 am**

by **jmbecker**

There should be a results tab called "Samples". It is the farthermost right results tab (sometimes on smaller screens you have to click on an arrow to get to the tab). There you will find all the results, which you can copy to excel or any other program of your choice. Yet, this will not work in the student version. You have to have a professional license to copy results or to get excel reports (

https://www.smartpls.com/#pricing).

### Re: Isolating indirect effect from Models with two mediators

Posted: **Mon Aug 17, 2015 1:09 am**

by **jta**

I've tried to click the Excel. An Internal Error appeared on my screen as follows:

X Internal Error

Reason:

java.lang.OutOfMermoryError: Java heap space

How to solve such error?

### Re: Isolating indirect effect from Models with two mediators

Posted: **Mon Aug 17, 2015 7:17 am**

by **jmbecker**

### Re: Isolating indirect effect from Models with two mediators

Posted: **Mon Aug 17, 2015 8:36 am**

by **jta**

The problem about "Java heap space" already solved. Also, isolation of indirect effects already solved. Thank you.

One more question about this thread. Where can I find the outputs for correlations among constructs, loadings and cross-loadings, and factor scores?

Thank you.

### Re: Isolating indirect effect from Models with two mediators

Posted: **Mon Aug 17, 2015 11:07 am**

by **jmbecker**

You will find all these results in the normal PLS output:

correlations among constructs and factor-scores are under "Latent Variables"

loadings are under "Outer Loadings"

cross-loadings are under "Discriminant Validity"

### Re: Isolating indirect effect from Models with two mediators

Posted: **Tue Aug 18, 2015 1:16 am**

by **jta**

At the bottom, outputs are grouped as "Final Results", "Quality Criteria", "Histograms", and "Base Data". Under such group, there are no outputs labeled "Latent Variables" and "Discriminant Validity".

I am using the 30-day trial (not the student version). Does it mean that the outputs of the trial version are not complete? That is, are the outputs for the trial version not similar to the commercial version? I am asking this because we are planning to endorse your software to our department for procurement.

Thank you.

### Re: Isolating indirect effect from Models with two mediators

Posted: **Wed Aug 19, 2015 12:53 am**

by **Hengkov**

Hi,

After running the PLS algortihm you will find the output. There is no difference in output between the trial and full versions.

Kind regards,

### Re: Isolating indirect effect from Models with two mediators

Posted: **Wed Aug 19, 2015 3:31 am**

by **jta**

Thank you.

Boostrapping does not produce the "Latent Variables" and "Discriminant validity" outputs, but the PLS Algorithm does.

### test mediators effectS

Posted: **Sun May 08, 2016 6:35 am**

by **Amira**

Hello,

Please I would like de Know how can I test the mediators effect in smart pls 3 ?? How to test the mediating effect through the PLS software. I can not read the result sets of tables .

### Re: test mediators effectS

Posted: **Mon May 09, 2016 11:34 am**

by **Amira**

Amira wrote:Hello,

Please I would like de Know how can I test the mediators effect in smart pls 3 ?? How to test the mediating effect through the PLS software. I can not read the result sets of tables .