PLS-MGA - Comparing More Than Two Groups

PLS is broadly applied in modern business research. This forum is the right place for discussions on the use of PLS in the fields of Marketing, Strategic Management, Information Technology etc.
Post Reply
PLS Junior User
Posts: 3
Joined: Tue Dec 03, 2019 7:34 am
Real name and title: Siqi (HDR Student)

PLS-MGA - Comparing More Than Two Groups

Post by gsq551 » Fri May 29, 2020 10:05 am

Hi forum members,

I have two questions about comparing more than two groups in PLS-MGA.

1. How can I calculate the significance of group difference?

I understand that for comparing more than two groups, I should go with OTG or pairwise comparison given that the embeded MGA algorithm only allows to compare two groups. I tried the OTG R Code, but I have no idead how to apply that code on my own data. As Prof. Ringle answered in another post (viewtopic.php?f=12&t=16122&p=27484&hili ... sis#p27484), I also can try pairwise comparison with a control of familywise error rate, and do 6 comparison in case of 4 groups.

However, may I ask the detail process to do pairwise comparison in SmartPLS 3? Which test should I run to get the path coefficient and significance? By running permutation test of each two groups or MGA algorithm or just running the normal PLS alogrithm for each group and calculate the difference manually?

2. What if a sample size of a group is lower than the minimum required sample size of the model?

I have four observed groups in my dataset, but only two of them match the minimum required sample size of the model. I know I should remove these segments, but I was wondering Is there any way to keep them as they are important for the research objective?

I really appreciated for your help.

SmartPLS Developer
Posts: 1129
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: PLS-MGA - Comparing More Than Two Groups

Post by jmbecker » Tue Aug 18, 2020 4:55 pm

1. I would not use the OTG. It is a strange procedure that is not really evaluated in scientific studies. I would stick with pairwise comparisons and correct for the family-wise error using an appropriate technique. You do this by running PLS-MGA or Permutation for each combination of two groups. For each comparison you get the p-value for the difference and correct this p-value accordingly.

2. That is very unfortunate. If you want to conduct a multi-group analysis each of your groups needs to meet the required minimum sample size. The only solution is to collect more data.
Dr. Jan-Michael Becker, University of Cologne, SmartPLS Developer
GoogleScholar: ... AAAJ&hl=de

Post Reply