"High" r squared (= .28), although no significant predictor (VIFs below 5)

Questions about the implementation and application of the PLS-SEM method, that are not related to the usage of the SmartPLS software.
Post Reply
Alex_
PLS Junior User
Posts: 2
Joined: Thu Oct 15, 2020 3:37 pm
Real name and title: Alexander Brodsky (doctoral candidate)

"High" r squared (= .28), although no significant predictor (VIFs below 5)

Post by Alex_ »

Dear all,

in my model, I want to predict different facets of competence development (e. g. development of methodological competencies; defined as four different latent variables) through different aspects of working conditions (autonomy, colleages, ...; in total 11 different constructs specified as latent variables).
I specified four different models (for the different facets of competence development) and in case of two models, the analysis shows at least one significant predictor. But in the other two cases, none of the independen variables conducts an significant influence on the dependent variable/latent construct.
That´s of course also a possible result, but I´m wondering about the relatively high R-squared in these models (.28 in each model; r squared adjusted: .21). I checked for multicollinearity, but the VIFs are all below the proposed threshold of 5 (although two independent variables are above 3.5). Correlation analysis of the different latent variables used in the PLS-analysis show significant correlations between almost all of the variables; some with a coefficient of .7.

Questions:
- What might be an explanation for the "high" r squared? (are the high correlations the reason, although the VIFs are below 5?)
- Are you aware of any sources dealing with that problem? (of course, I found a lot about multicollinearity, but as I said, these values seem to be okay)
- What are possible solutions? I might delete some constructs in these two models, but the variables included address different aspects of working conditions, therefore that would be critical from a content-wise perspective).

I´m happy about any hints.

Thank you!
Alex
jmbecker
SmartPLS Developer
Posts: 1282
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: "High" r squared (= .28), although no significant predictor (VIFs below 5)

Post by jmbecker »

I think the most important question is sample size. If you sample size is too small, you might not have enough power to detect small to medium effects. Together with some substantial (even though not severe multicollinearity) this might explain the quite ok R² values while having insignificant paths.
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