PLS-MGA with SmartPLS 3

[beautifulgarbage]

Dear PLS community,

I am currently "struggling" with a multigroup analysis in the new (3rd) version of SmartPLS and have some questions. I hope you might be able to help me:

1. First, in SmartPLS 2 it was apparently not possible to conduct an MGA straightaway with the software, but you had to calculate the model separately for each group first and then use the output to perform a subsequent t-test in Excel or follow Henseler's nonparametric procedure afterwards (see Sarstedt et al., 2011). As I can see now, in the meanwhile MGA has been implemented in SmartPLS 3. But how can I interpret the results? From the p-values in the section "Bootstrapping Results" I can conclude to what extent the path coefficients in each group are significant, right? So the PLS-MGA module in Smart PLS 3 itself basically calculates the model for each group? And from the results in sections "Henseler'S MGA", "Parametric Tests" and "Welch-Satterthwait Test" I know whether there's a siginficant difference in the path coefficients between the two groups, so that I do not need to use any .xls etc. for this, is that right? And last, but not least, which test is when relevant? "Welch" is usually used when the variances in the two groups are unequal- does SmartPLS test whether this is the case as well?

2. Furthermore, I have a special case where I have to compare more than 2 groups (to be more precise 8). I thought this is not possible with SmartPLS and one has to perform an omnibus test of group differences with excel or another software following Sarstedt et al. (2011). However, by chance I found out the the MGA-module in SmartPLS 3 indeed can compare various groups to each other. If I tick "Select All" both in Block A and both Block B, then the software works with all 16 groups (in my case) and the ouput contains results on the 28 differences all in all (=7+6+5+4+3+2+1). Do you know whether it is just doing pairwise group comparisons and thereby does NOT account for the mutiple testing problem? Or what is behind the MGA when selecting more than one group in each block?
I would be very grateful if you could help me with this and/or recommend me some documentation on PLS-MGA with SmartPLS 3.

Thank you very much in advance and best regards
BG

[jmbecker]

Hi BG,

1) There is no build in test for unequal variances, as there is no established test for a PLS path model. You have to find out for yourself, which test is the right one in your situation, given the advice in the literature.
Generally, I would also look if there is consistency between the three tests. If all three tests point in the same direct, your confidence about the result should much higher as if there are divergent results.

2) SmartPLS does only compute multiple pairwise tests and no omnibus test for all your 16 groups. You have to account for the multiple testing problem on your own.

[MAN]

Hello,

I'm trying to do a multigroup analysis for my data (between males and females).

I've about 700 males and 192 females. The data is not normally distributed. Which test should I use in SmartPLS 3.0 please, the "Henseler'S MGA", or the "Parametric Tests" or the "Welch-Satterthwait Test" to compare between the two groups.

Many thanks,
MAN

[beautifulgarbage]

Dear JM Becker,

thank you very much (belatedly) for the reply, you have helped me a lot!

Best regards

BG

[MMVS]

Hi BG,

1) There is no build in test for unequal variances, as there is no established test for a PLS path model. You have to find out for yourself, which test is the right one in your situation, given the advice in the literature.
Generally, I would also look if there is consistency between the three tests. If all three tests point in the same direct, your confidence about the result should much higher as if there are divergent results.

2) SmartPLS does only compute multiple pairwise tests and no omnibus test for all your 16 groups. You have to account for the multiple testing problem on your own.

Hi,

I am currently in the same situation as the OP. Can anyone tell me if this is still the case for the latest version released (3.2.1)?
Just for clarity, how exactly does one conduct a multigroup analysis (especially to compare more than 2 groups) in the new (3rd) version? (e.g. Age, gender, education, occupation)

Regards,
Mark

[yingsha84]

In SmartPlS 3, I tried to conduct Multi-Group Analysis. However, I couldn't select any group variable to run this analysis. I can not find any variable in the block Group A or Group B. When I clicked "select all", nothing was selected. The two blocks are in red and empty. The button "Start Calculation" is not available for clicking. How can we make this function work? Can anybody help me? Thank you very much!

[Choukri]

Hi,
To make multigroup analysis (ex, gender ), double click on your date --> Generate date group --> group column 0 (select gender) then OK.

After identifying the grouping variable, now go to your model, calculate function --> multi group analysis (MGA) --> Group A (select gender_1) and Group B (select gender_2) then at the same funcion go to bootstrapping and select complete bootstrapping --> finely click start calculation.

Greetings

[yingsha84]

Thank you so so much! It is a great help!!! Your answer has solved my problem completely!

[cheok_my]

To yingsha84,

I have problem in running MGA in pls 3. There are two boxes in red. How to run the analysis? What should I do? Pls advise. Thanks a lot.

[Choukri]

Hi,

For example you would like to make PLS-MEG using Gender (Male 1 and Female 2)

Select Gender 1 in the first boxe, and select again Gender 2 in the second boxe, than run the MGA.

Kind regards

[Choukri]

Hi,
To make multigroup analysis (ex, gender ), double click on your date --> Generate date group --> group column 0 (select gender) then OK.

After identifying the grouping variable, now go to your model, calculate function --> multi group analysis (MGA) --> Group A (select gender_1) and Group B (select gender_2) then at the same funcion go to bootstrapping and select complete bootstrapping --> finely click start calculation.

Greetings

[Choukri]

Hi,

Take a look on this paper, the answer is there

http://www.emeraldinsight.com/doi/abs/1 ... 0000022012

Greetings

[lisuhuang]

I am conducting PLS-MGA, too. I divide the sample into male and female groups.
Which of "Henseler'S MGA", "Parametric Tests" and "Welch-Satterthwait Test" should I use to decide whether the gender has a significant moderating effect in the model?

[Choukri]

Hi,

Take a look on this paper, the answer is there

http://www.emeraldinsight.com/doi/abs/1 ... 0000022012

Greetings

[jmbecker]

You will find answers to your question in the paper mentioned by Choukri. Yet, there is no definite answer to your question. All methods have pros and cons.

In addition, the following tutorial might be interesting for you
https://www.youtube.com/watch?v=l0TfK4z12D8
https://www.youtube.com/watch?v=qssDqR_-Q7Y