Statistically nonsignificant results with small effect sizes (f square) in large samples

Frequently asked questions about PLS path modeling.
Post Reply
FJAM
PLS Junior User
Posts: 2
Joined: Tue Jul 03, 2018 9:13 am
Real name and title: Francisco José Arenas Márquez

Statistically nonsignificant results with small effect sizes (f square) in large samples

Post by FJAM » Tue Jul 03, 2018 9:47 am

Hi, good morning

Using Cohen's f square, I am getting a small effect size (0.024) for a result that is statistically non-significant using a large sample (12000 individuals). Does this make sense?

I know that as the sample size grows, it is quite possible to find statistically significant results with no effects (f2<0.02), but I'm not sure if small effects (f2>0.02) in non-significant relationships are correct.

The confidence interval for this f square stimation is as follows (SmartPLS V 3.2.3):
Original Sample (O) - Sample Mean (M) - 2.5% - 97.5%
0.024 - 0.045 - 0.000 - 0.180

Thank you very much,
Francisco

jmbecker
SmartPLS Developer
Posts: 821
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Statistically nonsignificant results with small effect sizes (f square) in large samples

Post by jmbecker » Wed Jul 04, 2018 8:10 am

Sure it is possible. You might sample an exceptionally large effect where there is indeed no effect. Even a very large sample size might not prevent you from doing so. In addition, the f² effect sizes are always positive, but your confidence interval also indicates that you are touching the zero, hence the effect size might also not be significantly different from zero.
Dr. Jan-Michael Becker, University of Cologne, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Ja ... v=hdr_xprf
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de

FJAM
PLS Junior User
Posts: 2
Joined: Tue Jul 03, 2018 9:13 am
Real name and title: Francisco José Arenas Márquez

Re: Statistically nonsignificant results with small effect sizes (f square) in large samples

Post by FJAM » Wed Jul 04, 2018 9:20 am

Thank you very much for your answer!
F.Arenas
jmbecker wrote:
Wed Jul 04, 2018 8:10 am
Sure it is possible. You might sample an exceptionally large effect where there is indeed no effect. Even a very large sample size might not prevent you from doing so. In addition, the f² effect sizes are always positive, but your confidence interval also indicates that you are touching the zero, hence the effect size might also not be significantly different from zero.

Post Reply