Performing MGA - sample size?

[agalvez]

Hi everyone

There are two groups in my sample: group 1 (50 obs.) and group 2 (150 obs.).

Q1

- Is there any problem when group 2 is three times group 1? Should both samples be equal?

Case Study on book "A primer on partial..." use 219 for G1 and 125 for G2", so that sample size difference is two times. Following this example, should I care if my samples size are 50 and 150? I know the problem now could be "50" if this is not enought in terms of power size.

Q2

When performing MGA, first I calculate PLS Algorithm (selecting group box) and then PLS-MGA to observe differences in Paths.

However, in the first step, PLS Algorithm, I have seen that item loadings, CR and AVE change when analyzing Group 1 and Group 2. For example, for the total sample the model is OK (AVE>0.50 and CR>0.70), but when I see the Group 1/2 specific report, these values change not being possible to guarantee the convergent validity (in some subscales, AVE<0.50 for Group 1, and OK for Group 2, but other constructs are Ok for Group 2 and bad for Group 1).

This makes me think I can not conduct PLS-MGA, probably because my sample its too small and that is the reason why values are ok (total sample) and bad (analyzing by groups, reducing sample).

Thank you.

[Hengkov]

Hi,

Q1: That could be a problem in measurement invariance. Please check.

Q2: No. use the full sample to the evaluation of the measurement model.

Best regards,

[jmbecker]

To be a little bit more specific:

1) You should check measurement invariance. If you have measurement invariance across the groups, you can analyze the measurement properties on the pooled data. If not, you might not be able to compare groups at all and can only analyze each group separately.
2) Uneven group sizes are a problem for the permutation test not for the other MGA procedures. As measurement invariance is tested by using the permutation approach, this could be a problem in your particular case.
3) I generally believe, that sample sizes of less than 100 are not really trustworthy and reliable. This also applies to each group's sample size. Your sample size of 50 is quite small and thus not really well suited for an MGA in my opinion. In addition, you probably have insufficient power to detect even medium group differences.

[agalvez]

Thank you very much Hengkov and Jmbecker.

I will think about your explanation.

Happy new year.