Hi everyone,
I have read the discussion on how to conduct multigroup comparisons, and also searched the literature. Recent literature seem to emphasize the permutation approach, but this approach does not look too straight forward until the procedure is implemented in SmartPLS.
I understand that the t-test suggested by Chin (http://disc-nt.cba.uh.edu/chin/plsfaq.htm) is another useful approach. I have estimated a model based on the EPSI framework for customer satisfaction, and would like to test whether there are significant differences in the path coefficients between 3 subsamples (customers that are users/buyers/users and buyers, respectively). In that respect I have some questions I hope someone could help me out with:
1) One assumption for the test is that the variances across the samples are equal. The sample sizes are very different (256, 419 and 1679). What would be the most suitable test for testing the hypothesis of equal variances? I have tried Levene's test statistic (in SPSS), but this rejects the hypothesis, and the F-test for equal variances should probably not be used as the sample sizes are very different.
And should I test whether the variances are equal across the three samples for each of the path coefficients seperately?
2) I have a total of 10 path coefficients in the model. Should I assess whether there are differences one by one across the three subsamples?
3) Can I use the t-test and do a Bonferroni-correction of the significance level (if I want to test whether there are differences between one specific path coefficient across the three groups, that would give me a corrected significance level of 0.05/((3(2)/(3-1)) = 0.0167). And am I right when I think that as the hypothesis are whether there are differences in path coefficient, this would lead to a two-tailed test, I should look for the 0.0167/2 critical value in the t-distribution?
4) What I want to investigate is whether there are different drivers for the customer satisfaction in the 3 subsamples. The EPSI model has 4 exogenous drivers. Would it be ok to use a t-test to test whether there are significant differences in the total effects from each of the exogenous drivers to satisfaction? Or should I just analyze the path coefficients?
Any comments on these matters would be very much appreciated!
Thank you very much in advance,
Ingrid Lærdal
Multigroup comparison with more than two subsamples
Multigroup comparison with more than two subsamples
Ingrid Lærdal