Indirect Effect

Questions about the implementation and application of the PLS-SEM method, that are not related to the usage of the SmartPLS software.
Post Reply
Derick
PLS Junior User
Posts: 2
Joined: Thu Oct 04, 2018 4:06 pm
Real name and title: Mr. Teoh Kok Ban

Indirect Effect

Post by Derick »

Hi,

I have tested indirect effect in my study, where the t-value is significant at 0.05 level of significance but insignificant at 0.01 level of significance.

However, according to my 95% bootstrap confidence interval, it is (-0.083, 0.003) where it includes zero in between the confidence interval.

Hence, I would like to ask shall I consider this as insignificance due to the confidence interval or significance due to the p-value?

Thank you in advance for the prompt reply.
jmbecker
SmartPLS Developer
Posts: 1281
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Indirect Effect

Post by jmbecker »

That is a good question.

1) Certainly, your effect is on the edge of being significant.
The problem with the concept of significance is that it is a dichotomous decision on a continuous phenomenon. Why 5% Because it is a conventional standard? But where is the difference between 4.9% and 5.1%?? There is none and still people count the one as significant and the other as non-significant.
Bayesians would say that there is some but not overwhelming evidence for the effect. They usually try to avoid making the dichotomous decision, but try to quantify the level of support. You will find that also in many modern discussions about frequentist approaches related to recent calls to abandon the concept of significance at all (https://www.nature.com/articles/d41586- ... yfw-_ZKnTQ)

2) Another aspect is that different methods come to different conclusions.
You are likely using some bias-corrected percentile or bias-corrected and accelerated confidence intervals. That is something different as using a t-distribution based test, which underlies the p-value. If you would use studentized intervals you would get an interval that agrees with your p-value.
However, in the realm of indirect effects, researchers (mostly Preacher and Hayes) have found that the distribution of the indirect effect in mediation does not follow conventional distributions and therefore non-parametric bias-corrected percentile intervals should be used.

Thus, if you want to make a dichotomous decision you should prefer the confidence interval because that is the better method. However, you may also discuss that there are divergent findings and that your effect might be on the edge of being significant and that more research (additional studies) is needed to come to a final conclusion.
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