P-values for R-square, F-square etc

Questions about the implementation and application of the PLS-SEM method, that are not related to the usage of the SmartPLS software.
Post Reply
rohit315
PLS Junior User
Posts: 4
Joined: Mon Jun 26, 2017 3:19 am
Real name and title: Rohit

P-values for R-square, F-square etc

Post by rohit315 »

Hello all

I saw that in bootstrapping while viewing R-square, F-square results, I get p-values after bootstrapping (I do get that it is because t-tests are run with sub-samples). While generally it is asked to just see if R-square values are so and so, or F-square values are so and so (like 0.02, 0.15, 0.35 etc) to determine the impact. However, what to do about the significance in this case?

As I understand, whichever F-square effect or R-square effect which has no significance (P>sig. level) cannot be interpreted for the population and should be just considered a sampling error. Is my understanding right?

Basically, the question arises because Hair et al. (2013) do not make this distinction and state that absolute values without any mention of significance level must be considered.
jmbecker
SmartPLS Developer
Posts: 1282
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: P-values for R-square, F-square etc

Post by jmbecker »

Generally, there were two considerations to include R² and f² into bootstrapping.
For R² in regression exists the F-Test to test whether R² is significantly different from zero. The same could also be interesting for PLS models, if R² is low. If R² is medium to high it is usually also significant. Yet, I would not use the t-test (p-value), but the confidence interval that we provide. It is more in line with the bootstrap based test philosophy.
Similarly, for f² it could be interesting to show the confidence interval of a certain effect size an whether it varies a lot or not.
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
mirf
PLS Junior User
Posts: 2
Joined: Fri Nov 09, 2018 10:39 pm
Real name and title: Farzana Asad Mir- Dr

Re: P-values for R-square, F-square etc

Post by mirf »

I have a follow up question for this.

1. Is p value of f-square also important while reporting f-square?
2. for path coefficients, if my p value is greater than 0.05 but confidence interval does not cross Zero, should i consider this statsitically significant or not?
3. Same scenario but for f-square i.e. 2. for f-square, if my p value is greater than 0.05 but confidence interval does not cross Zero, should i consider this statsitically significant or not?
hanifmauludin
PLS Junior User
Posts: 6
Joined: Mon Jan 23, 2012 7:33 am
Real name and title: Dr. Hanif Mauludin
Location: Indonesia
Contact:

Re: P-values for R-square, F-square etc

Post by hanifmauludin »

Dear Researcher

I need explanation about P Value.

1. Please explain whether it is permissible to accept the hypothesis with a sig P of 10%?
2. Maybe there is a theory that explains the use of P value in the field of economics ?
jmbecker
SmartPLS Developer
Posts: 1282
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: P-values for R-square, F-square etc

Post by jmbecker »

You should find an explanation of p-values in every good introductory statistics book. It is nothing specific to PLS, but also used for regression, t-test, whatever.
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
mytnaprofile
PLS Junior User
Posts: 1
Joined: Mon Dec 25, 2023 12:43 pm
Real name and title: mytnaprofile

Re: P-values for R-square, F-square etc

Post by mytnaprofile »

Hi

I have similar issues with the quoted comments below.

In my case, I found out that there are some sub-constructs with a higher HTMT value. I try to solve the higher HTMT value by checking the items' cross-loading; any pair of item differences less than 0.1 are removed.

a. As a result, only three HTMT values were recorded above 0.9. The two pairs of items' differences are greater than 0.1, hence, I decided to maintain these items. Except for one mediator construct with 0.976, one of the items' pair differences is below 0.1

b. When I removed this mediator item and assessed the structural model, the p-value of the f-square of the mediator was 0.069 (insignificant).

c. But, when I add back the mediator construct of 0.976 HTMT, the p-value of f-square is reduced to 0.034 (significance). However, when I remove it, I will get an insignificant p-value for the mediator.


Should I focus on solving the significance issues of the structural model or remain with the item with HTMT 0.976, but also look at Fornel & Lacker and cross-loading to indicate that discriminant validity is established? Or just report the finding that it is statistically insignificant?


mirf wrote: Thu Feb 04, 2021 3:05 pm I have a follow up question for this.

1. Is p value of f-square also important while reporting f-square?
2. for path coefficients, if my p value is greater than 0.05 but confidence interval does not cross Zero, should i consider this statsitically significant or not?
3. Same scenario but for f-square i.e. 2. for f-square, if my p value is greater than 0.05 but confidence interval does not cross Zero, should i consider this statsitically significant or not?
Post Reply