Page 1 of 1

f2 and q2

Posted: Sat Aug 31, 2019 3:10 pm
by zd18
Hello everyone,

I ran the PLS SEM for effect size f2 and q2 and found it a little odd. Is it possible to have an inverse outcome between the explanatory power (f2) and predictive power (q2). For e.g:

Factor A f2 = 0.36, q2 = 0.12
Factor B f2 = 0.25, q2 = 0.15
Factor C f2 = 0.08, q2 = 0.21
Factor D f2 = 0.03, q2 = 0.26

How is it possible to get a low explanatory power, but high predictive power and vice versa? If it is not possible, where did I went wrong with the analysis? Feedbacks are greatly appreciated. Thanks.

Re: f2 and q2

Posted: Sun Sep 01, 2019 9:09 am
by jmbecker
First, I would try to replicate the finding with PLSPredict. It has a much better foundation than the Blindfolding which is a method developed in the 1980s.
Second, it is generally possible if the factors with high explanatory power lead to overfitting. However, such consistency of inverse results would also be suspicious to me.

Re: f2 and q2

Posted: Mon May 11, 2020 2:31 am
by NomanMahmood
Dear jmbecker,

Is there any other way to calculate effect size (f2) for formative second order constructs. R2 is equal to 0.99999 (1.00).
Computed F2 in SmartPLS are too big because of it.