Effect size f-square and q-square

Questions about the implementation and application of the PLS-SEM method, that are not related to the usage of the SmartPLS software.
Post Reply
ferry.jaolis
PLS Junior User
Posts: 2
Joined: Mon Jul 25, 2016 10:45 am
Real name and title: Ferry Jaolis, M.Res, Ph.D (Cand.)

Effect size f-square and q-square

Post by ferry.jaolis »

Dear PLS users/experts/developers,

Could someone please help with the computation of f-square and q-squared. As far as Prof Hair's book can tell, both of these effect sizes should be computed manually by excluding predecessors for target latent variables and check any changes in R squared and Q squared.

1). In smartpls 3.2.4 the f square values were included in the bootstrap report (are these the f square that once should be computed manually if using smartpls 2 ?)

2). While smartpls 3.2.4 provided the f square values in bootstrapping report, it doesn't provide the q square values in blindfolding report. So correct me if I am wrong, for smartpls 3.2.4 the f square values can directly be generated from bootstrapping report, while the q square values still need to be computed manually by Prof Hair's formula?

3). For the q square values manual computation, do we delete the path arrow of the predecessor LV to target LV or delete the entire predecessor LV in order to compute the changes in Q square (i.e. for q square computation) and/or R square (for f square computation) ?

Lots of thanks forum members,
Ferry
Holsteinerin22
PLS Junior User
Posts: 3
Joined: Thu Jul 28, 2016 3:32 pm
Real name and title: Levi

Re: Effect size f-square and q-square

Post by Holsteinerin22 »

Bumping up this question. Would love to hear an answer, too.
Thank you
jmbecker
SmartPLS Developer
Posts: 1282
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Effect size f-square and q-square

Post by jmbecker »

1) Yes, you do not have to calculate them manually in SmartPLS 3.

2) Yes, while we provide the f-square, you still need to calculate the q-square manually.

3) This is a good question. It is not yet decided on how to do it correctly for Blindfolding. That is why we have not implemented it in SmartPLS 3 so far. You should make your decisions explicit (write them into your manuscript) when you report the q-square, so that reviewers can evaluate them.
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
zameer
PLS Junior User
Posts: 1
Joined: Mon Oct 10, 2016 9:30 am
Real name and title: Zameer

Re: Effect size f-square and q-square

Post by zameer »

Dear PLS experts,

In SmartPLS 3.2.4,

1. Q² (=1-SSE/SSO) from Construct Crossvalidated Redundancy from blindfolding. Is this not the predictive relevance q square values? If not, then how do we manually calculate q square?

2. For f square, from bootstrapping results there is Original Sample (O),Sample Mean (M),Standard Deviation (STDEV),T Statistics (|O/STDEV|),P Values, which one is actually the f square values?

Looking forward for your expert advises.

Regards
Zytozid
PLS Junior User
Posts: 3
Joined: Thu Oct 20, 2016 9:08 am
Real name and title: Ulrich Heintze

Re: Effect size f-square and q-square

Post by Zytozid »

zameer,
the f square values can be found after calculating the PLS algorithm and looking at the results of it, then clicking on the "f square" link on the bottom right of your program. (for PLS 3.2.4)

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

Re: Effect size f-square and q-square

Post by jmbecker »

The large Q² is the predictive relevance of the structural model for predicting the indicators of an endogenous constructs.
The small q² is a quasi-effect size measure of the difference in Q² after including and excluding a certain predictor construct from the model. However, we do not automatically report the q² effect size as there are some unsolved conceptual questions that need expert judgment when calculating the results.
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
avinash
PLS Junior User
Posts: 3
Joined: Sat Jan 07, 2017 11:06 am
Real name and title: avinash

Re: Effect size f-square and q-square

Post by avinash »

Dear Experts
I have gone through Cohen 1992 power table, and also the sample size calculation based on the table in book on Premiers in PLS
Cohen 1992 look at effect sizes (small, large, medium). The book on Premier in PLS looks at min R square

I am working on a model (all reflective constructs)

A->B
A->C
A->D

B->E
C->E
D->E

E is the final endogenous construct
I calculated the effect size using the steps mentioned in the PLS SEM facebook page "f² Effect Size Computation - How to Do it in SmartPLS"
f² effect size = (R²(incl.) – R²(excl.)) / (1 - R²(incl.));

The effect sizes are
for B->E is 0.64
for C->E is 0.034
for D->E is 0.02

Kindly help me with these questions
What does minimum R squared mean? ( referred from Pg 21 table on how to select sample from book - Premier on PLS)
for my model
R squared included = 0.6306
R squared B excluded =0.393
R squared C excluded = 0.618
Rsquared D excluded = 0.623

I am confused. How to calculate the sample size from all this information
On what basis should I select the min R squared (0.1, 0.25, 0.5, 0.75)
I really need your help to progress.Kindly do advise as cohen paper has no direct reference to R squared and talks about small, medium and large effect sizes only.

Kindly do help.
Regards
Avinash
ahmad.m.k
PLS Junior User
Posts: 1
Joined: Wed Apr 05, 2017 5:15 pm
Real name and title: ahmad mofaddi
Location: Jordan
Contact:

Re: Effect size f-square and q-square

Post by ahmad.m.k »

please help us about computing q-squared by PLS 3 manually.
I mean the Q2 formula by pls 3 .
thank you advanced
User avatar
cringle
SmartPLS Developer
Posts: 818
Joined: Tue Sep 20, 2005 9:13 am
Real name and title: Prof. Dr. Christian M. Ringle
Location: Hamburg (Germany)
Contact:

Re: Effect size f-square and q-square

Post by cringle »

q² = (Q²_included - Q²_excluded) / (1 - Q²_included)
compuig
PLS Junior User
Posts: 1
Joined: Fri Dec 08, 2017 4:40 am
Real name and title: Lilian C., PhD student

q² effect size

Post by compuig »

Hello!

So, my structural model has one exogenous (IQ) and two endogenous variables (XV and TS). Hair et al. mention that in order to determine the q-square, we have to delete "a specific predecessor of that endogenous latent variable." Therefore, by deleting IQ I got IQ on TS, and by deleting XV I got EV on TS, however, is there something I should do to get IQ on XV? Am I missing something? Thank you!
User avatar
cringle
SmartPLS Developer
Posts: 818
Joined: Tue Sep 20, 2005 9:13 am
Real name and title: Prof. Dr. Christian M. Ringle
Location: Hamburg (Germany)
Contact:

Re: Effect size f-square and q-square

Post by cringle »

I don't fully understand the description of your model. Anyways, if you only have one explanatory for an endogenous latent variable, the q² computation does not work - or you could say that Q²_excluded is 0 and, then, plug in the value of Q²_included and 0 for Q² excluded into the equation: q² = (Q²_included - Q²_excluded) / (1 - Q²_included)


Best
CR
Post Reply