Page 1 of 1

Longitudinal and multi group studies

Posted: Mon Aug 26, 2019 12:09 pm
by shamilaj
Hi,

Please refer me to longitudinal studies that also used multi groups. I am trying to do a study using 2 groups and collected data during 2 different time points. I am looking for example studies to make sure what I am doing is correct.

Any help will be much appreciated.

Thank you.

Re: Longitudinal and multi group studies

Posted: Tue Aug 27, 2019 6:31 pm
by jmbecker
I'm not aware of longitudinal studies that use PLS-MGA. The problem is that PLS-MGA assumes independence of the groups. However, in a longitudinal setting groups are not independent, just like in a paired t-test.

If you have strictly paired data, I would rather model both time points in the same model (maybe even including autoregressive effects).

Re: Longitudinal and multi group studies

Posted: Wed Aug 28, 2019 7:56 pm
by shamilaj
Thank you very much Dr. Becker.

What you said makes sense. I did as you said and put both time points within the same model. To test difference between the 2 groups do I just look at p-value (condition 1 Vs condition 2) in the PLS-MGA or parametric test?

-Shamila

Re: Longitudinal and multi group studies

Posted: Sun Sep 01, 2019 9:05 am
by jmbecker
If you have everything in one model you can test parameters in t1 vs. t2 with the following approach
Rodríguez-Entrena, M., Schuberth, F., & Gelhard, C. (2018). Assessing statistical differences between parameters estimates in Partial Least Squares path modeling. Quality & quantity, 52(1), 57-69.

Re: Longitudinal and multi group studies

Posted: Sat Sep 07, 2019 8:22 pm
by shamilaj
Thank you very much Dr. Becker.

In this paper they talk about using MS Excel "in order to manually calculate the parameter difference for
every bootstrap sample" (p.63). I used SmartPLS 3.0 hence I think this operation is completed by the software itself. Correct me if I am wrong.

-Shamila

Re: Longitudinal and multi group studies

Posted: Thu Sep 03, 2020 6:33 pm
by meldarien
What about running the same model independently for t1 and t2? Then use the latent variable scores (or the unstandardized ones from IPMA) for partial least squares regressions in the structural model. Is this ok? Is there a literature that provides support or context?

Melvin Prince