1. Download MG-PLS here.

2. Open the file in Microsoft Excel.

3. Enable Macros when it is opening.


The concept behind Henseler's test is simple. It is actually the probability that the bootstrapped estimates of Group X is different from that of Group Y. In other words, each bootstrapped estimate of Group X is compared with each bootstrapped sample of Group Y. A 5000-bootstrapped sample will then have 5000*5000 comparisons.
How to Open MG-PLS
The first step of any tests in MG-PLS

MG-PLS beta is a two-in-one excel based tool for computing multi-group analyses of partial least squared path analysis (or sometimes I call it variance-based structural equation modeling). There are two types of multi-group analyses that MG-PLS can handle. The computation of the first test is derived from the pairwise group difference test proposed by Henseler's (2007). The other test is based on Omnibus test of group difference documented in Sarstedt, Henseler, and Ringle (2011). Omnibus Test is so far the first test in the literature that is proposed to be able to reveal the statistical significance of group differences between three or more groups. In MG-PLS, it is able to test the group difference of parameter estimates between up to 5 groups simultaneously.

 

It is important to clarify that MG-PLS is unable to compute or estimate any parameter estimates from the partial least squared algorithm. Rather it relies on the boostrapped parameter estimates generated from other partial least squared modeling softwares such as SmartPLS, WarpPLS.


Also, the excel "program" only simulates the formulas proposed by the literatures, so there might be some degree of discrepancy between the result produced by MG-PLS and that generated by the scripts (run in R or other language platforms) of the authors. I did some initial tests and found the MG-PLS is able produce the result almost identical to Henseler's (2007) results. However, I am still not fully confident if MG-PLS is able to perform Omnibus test reliably because it requires a Monte-Carlo resampling procedure in the permutation process, which I am not sure excel is able to perform as other programs do. Also, I realise that the results of Omnibus test produced by MG-PLS tend to be over sensitive (very easy to reach statistical significance).


References:

Henseler, J. (2007). A new and simple approach to multi-group analysis in partial least squares path modeling. In: H. Martens & T. Næs (Eds.), Causalities explored by indirect observation: Proceedings of the 5th international symposium on PLS and related methods (PLS’07) (pp. 104–107). Oslo.


Sarstedt, M., Henseler, J., & Ringle, C. M. (2011). Multigroup analysis in partial least squares (Pls) path modeling: Alternative methods and empirical results. Measurement and Research Methods in International Marketing, 22, 195-218. Doi 10.1108/S1474-7979(2011)0000022012



Derwin Chan Research
​    Introduction​

The citation looks funny because it's not a software. However, I would be appreciated if you let me know you are using it.


Chan, D. K. C. (2014). MG-PLS [Computer software]. Available from www.derwinchan.com.

How to cite?
To compute the p-value of an Omnibus Test using MG-PLS, you have press the cloud-like button, and "Processing" will appear to indicate that the computer is processing the analysis. When it's done, the word "Completed" will be shown. The process may take quite long, please be patient. The more permutations it does, the longer it takes for completing the analysis. By default, the number of permutations is 1000. Sarstedt, Henseler, and Ringle (2011) suggested 5000 permutations is desirable, but I strongly recommend trying 1000 permutations first and see how long it takes before trying 5000. For old computers, it might take more than an hour to process.
The F-value and its associated SS-between and SS-within values are computed once you have copied the bootstrapped parameter estimates in the sheet named "Bootstrapped Samples".
MG-PLS
Omnibus Test
The concept behind Omnibus test is Analysis of Variance (i.e., ANOVA), so it concerns about the ratio between between-subject variance and the within-subject variance. However, it adopts a Monte Carlo resampling method to perform N permutations, such that the p-value of Omnibus test is estimated by the probability that the F-value of the group difference based on the bootstrapped values is higher than that of the random samples.

Before running any test, please copy the boostrapped parameter estimates of each group into the columns. It is recommended that 5000 bootstrapped re-samples are adequate (Sarstedt, Henseler, & Ringle, 2011). Therefore, MG-PLS, by default, has prepared 5000 empty cells for each group.

When all the bootstrapped estimates have been entered, please click the bottom of workbook to select the sheet named "Results".
Henseler's Group Difference Test
Please copy the parameter estimates of the original sample below. Otherwise the result would not be accurate.
The result of all pairwise comparison will then be shown in the rows below. Henseler (2007)'s conditional probability H(x) illustrates the probability of group difference, so I use "1- H(x)" to reveal the p-value. "G1-G2" means the test of difference between group 1 and group 2.