Combining datasets for larger groups in MGA

Questions about the implementation and application of the PLS-SEM method, that are not related to the usage of the SmartPLS software.
Post Reply
janhofer
PLS Junior User
Posts: 1
Joined: Mon Dec 03, 2018 3:42 pm
Real name and title: Jan Hofer

Combining datasets for larger groups in MGA

Post by janhofer »

Dear members of the smartpls community,

I hope you can enlighten me on the following matter: Suppose you have the results of a survey e.g. a questionnaire about the individual contributions of employees to a specific networking platform. This survey was conducted in two different companies so you have two datasets of different sizes.

I built a reflective model in smartpls where I measure via different constructs (each consisting of multiple variables) the reasons why people contribute to a specific networking platform. All 25 items are measured on a 1-7 Likert scale.

I grouped the observations in four groups by age. The goal is to conduct a Multi Group Analysis to find the differences between the four groups. Unfortunately the size of each group using one of the datasets is too small for significant results.

Therefore my questions are: (1) Is it possible to combine the two datasets (of different companies) to get a larger overall sample and in the end larger groups for the MGA? (2) What tests ensure the "compatibility" of my datasets and can I conduct these tests in smartpls?

I am pretty new to smartpls and the matter of MGA. I tried to educate myself and searched for the past few days and read a lot about:
- intraclass correlation and anova (is this the right way to go?)
- pearson correlation (only for metric variables...)
- spearmans correlation coefficient (only about the relationship between two variables...)
- multilevel analysis (but this is more about aggregating variables than combining datasets if i understood this correctly)
- Omnibus Test, Levene Test and several other statistical tests (The problem here is, that I dont know how to apply these tests to my datasets and my model consisting of reflective constructs)

I scanned through:
- Klein, K. J., and Kozlowski, S. W. J. 2000. "From Micro to Meso: Critical Steps in Conceptualizing and Conducting Multilevel Research"
- Fichman, R. G. (2001). “The role of aggregation in the measurement of IT-related organizational innovation"
- Hox, 1995 "Multilevel Analysis"
- Bliese, 2000 "Within Group Agreement, Non-Independence, and Reliability - Implications for Data Aggregation and Analysis"
and several others but did not get much out of it in terms of practical advice.

Any help or a hint in the right direction would be highly appreciated.

Kind regards
Jan Hofer
jmbecker
SmartPLS Developer
Posts: 1281
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Combining datasets for larger groups in MGA

Post by jmbecker »

It is somewhat of a multilevel problem, but there exist no multilevel methods for PLS.

What you can do is a multi-group analysis of your two companies to show that the company does not affect the results.
Therefore, you want to first test measurement invariance using MICOM in the PLS permutation algorithm to establish that your measurement is invariant across the groups (https://www.smartpls.com/documentation/ ... ques/micom). This step is most important.
Second, you ideally also want to find no significant differences in your path coefficients for the different companies.
If both criteria are fulfilled you can certainly combine both datasets and build groups on age. If the second is not fulfilled you may want to add a dummy for your company to the structural model where you find the differences to control for this.
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
Post Reply