Redundancy calculation seems to be incorrect
Redundancy calculation seems to be incorrect
According to the paper of Tenenhaus et al. the redundancy for a generic endogeneous block j is computed as the product of the communality of block j with the Rsqr of block j. This does not match in my case. Example is below:
R Square Communality Redundancy
0.4044 0.5324 0.1975
0 1 0
0.4495 1 0.231
0.5897 1 0.574
0.5201 0.8331 0.4119
0.5354 1 0.3325
R Square Communality Redundancy
0.4044 0.5324 0.1975
0 1 0
0.4495 1 0.231
0.5897 1 0.574
0.5201 0.8331 0.4119
0.5354 1 0.3325
AP
- 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:
Hi,
the redundancy and communality computations run fine in SmartPLS.
However, cv-redundancy and cv-communality outcomes are different things. You use the blindfolding procedure to obtain these values and use them to compute Q² and q² (predictive relevance, Stone-Geisser-Test).
However, as reported elsewhere in this forum, the blindfolding procedure has a bug. It does correctly compute the cv-redundancy and cv-communality outcomes but you must follow specific rules:
cv-redundancy: check only a single latent variable in the blindfolding routine - you are then correctly analyzing the checked one.
cv-communality: check all latent variables exept a single one in the blindfolding procedure - you are then correctly analyzing the unchecked one.
Cheers
Christian
the redundancy and communality computations run fine in SmartPLS.
However, cv-redundancy and cv-communality outcomes are different things. You use the blindfolding procedure to obtain these values and use them to compute Q² and q² (predictive relevance, Stone-Geisser-Test).
However, as reported elsewhere in this forum, the blindfolding procedure has a bug. It does correctly compute the cv-redundancy and cv-communality outcomes but you must follow specific rules:
cv-redundancy: check only a single latent variable in the blindfolding routine - you are then correctly analyzing the checked one.
cv-communality: check all latent variables exept a single one in the blindfolding procedure - you are then correctly analyzing the unchecked one.
Cheers
Christian
Last edited by cringle on Tue Feb 10, 2009 8:45 pm, edited 1 time in total.
Prof. Dr. Christian M. Ringle, Hamburg University of Technology (TUHH), SmartPLS
- Literature on PLS-SEM: https://www.smartpls.com/documentation
- Google Scholar: https://scholar.google.de/citations?use ... AAAJ&hl=de
- Literature on PLS-SEM: https://www.smartpls.com/documentation
- Google Scholar: https://scholar.google.de/citations?use ... AAAJ&hl=de
Hi,
you are completely right. There is something odd indeed.
We left the communality and redundancy measure in the current release (Results Report ---> Overview) by mistake (just some experimental stuff we did) and did not really evaluate the computations. The communality is okay and equals the AVE in the standardized case. However, the redundancy computations are different. In the ECSI-example, the outcomes for the LV Complaints (single indicator construct) match your findings. The same (almost) for quality. But when more than two paths run into a LV, the numbers are much too low. Even though your quoted equation seems simple, we followed a different approach in SmartPLS: first compute the dependent LV scores by the LV scores of the predecessors and the inner path coefficients. Use these new LV scores to compute the outer correlations and hence to get the red.
Just compute these numbers via MS-Excel (Red. = R² * Com.).
Best
Christian
you are completely right. There is something odd indeed.
We left the communality and redundancy measure in the current release (Results Report ---> Overview) by mistake (just some experimental stuff we did) and did not really evaluate the computations. The communality is okay and equals the AVE in the standardized case. However, the redundancy computations are different. In the ECSI-example, the outcomes for the LV Complaints (single indicator construct) match your findings. The same (almost) for quality. But when more than two paths run into a LV, the numbers are much too low. Even though your quoted equation seems simple, we followed a different approach in SmartPLS: first compute the dependent LV scores by the LV scores of the predecessors and the inner path coefficients. Use these new LV scores to compute the outer correlations and hence to get the red.
Just compute these numbers via MS-Excel (Red. = R² * Com.).
Best
Christian
Administration Team
Hello Dr. Ringle,
concerning the blindfolding bug mentioned above:
In your posting from March 3rd I read:
"For example, you have two latent exogenous variables (LV1 and LV2) and one latent endogenous variable (LV3). The SmartPLS blindfolding results are identical to the LVPLS software,
1. for the cv-redundancy of LV3, if you check only the specific endogenous LV under analysis (LV3);
2. for the cv-communality of LV3, if you check all other LVs in the model (LV1 and LV2 and not LV 3) to obtain the results for LV3.
However in your posting from Jan 25 you state:
"cv-communality: check only a single latent variable in the blindfolding routine - you are then correctly analyzing the checked one.
cv-redundancy: check all latent variables exept a single one in the blindfolding procedure - you are then correctly analyzing the unchecked one."
I am a bit confused. Aren't these explanations contradictory? Which one is correct? Thank you very much for the information in advance and thanks for the very useful software!
Kind regards, Tilmann
concerning the blindfolding bug mentioned above:
In your posting from March 3rd I read:
"For example, you have two latent exogenous variables (LV1 and LV2) and one latent endogenous variable (LV3). The SmartPLS blindfolding results are identical to the LVPLS software,
1. for the cv-redundancy of LV3, if you check only the specific endogenous LV under analysis (LV3);
2. for the cv-communality of LV3, if you check all other LVs in the model (LV1 and LV2 and not LV 3) to obtain the results for LV3.
However in your posting from Jan 25 you state:
"cv-communality: check only a single latent variable in the blindfolding routine - you are then correctly analyzing the checked one.
cv-redundancy: check all latent variables exept a single one in the blindfolding procedure - you are then correctly analyzing the unchecked one."
I am a bit confused. Aren't these explanations contradictory? Which one is correct? Thank you very much for the information in advance and thanks for the very useful software!
Kind regards, Tilmann
- 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:
Hi,
it is:
1. for the cv-redundancy of LV3, if you check only the specific endogenous LV under analysis (LV3);
2. for the cv-communality of LV3, if you check all other LVs in the model (LV1 and LV2 and not LV 3) to obtain the results for LV3.
Could you provide the link this perious post (I could not find it right away).
Best
Christian
it is:
1. for the cv-redundancy of LV3, if you check only the specific endogenous LV under analysis (LV3);
2. for the cv-communality of LV3, if you check all other LVs in the model (LV1 and LV2 and not LV 3) to obtain the results for LV3.
Could you provide the link this perious post (I could not find it right away).
Best
Christian
Prof. Dr. Christian M. Ringle, Hamburg University of Technology (TUHH), SmartPLS
- Literature on PLS-SEM: https://www.smartpls.com/documentation
- Google Scholar: https://scholar.google.de/citations?use ... AAAJ&hl=de
- Literature on PLS-SEM: https://www.smartpls.com/documentation
- Google Scholar: https://scholar.google.de/citations?use ... AAAJ&hl=de
Thank you for the quick response!
The posting from Mar 03, 2008 can be found here: viewtopic.php?t=634&highlight=blindfolding
The more current posting I reffered to was within this topic.
Just to make sure:
=> for cv-redundancy, I check only a single latent variable in the blindfolding routine, I am then correctly analyzing the checked one.
=> for cv-communality, I check all latent variables except a single one in the blindfolding procedure, I am then correctly analyzing the unchecked one
Do I have it right now?
Kind regards,
Tilmann
The posting from Mar 03, 2008 can be found here: viewtopic.php?t=634&highlight=blindfolding
The more current posting I reffered to was within this topic.
Just to make sure:
=> for cv-redundancy, I check only a single latent variable in the blindfolding routine, I am then correctly analyzing the checked one.
=> for cv-communality, I check all latent variables except a single one in the blindfolding procedure, I am then correctly analyzing the unchecked one
Do I have it right now?
Kind regards,
Tilmann
- 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:
Yes!
The blindfolding procedure correctly runs in SmartPLS but we have a "wiring" problem with GUI in the current release.
This issue will be corrected in the next release.
Best
Christian
The blindfolding procedure correctly runs in SmartPLS but we have a "wiring" problem with GUI in the current release.
This issue will be corrected in the next release.
Best
Christian
Prof. Dr. Christian M. Ringle, Hamburg University of Technology (TUHH), SmartPLS
- Literature on PLS-SEM: https://www.smartpls.com/documentation
- Google Scholar: https://scholar.google.de/citations?use ... AAAJ&hl=de
- Literature on PLS-SEM: https://www.smartpls.com/documentation
- Google Scholar: https://scholar.google.de/citations?use ... AAAJ&hl=de
-
- PLS Expert User
- Posts: 31
- Joined: Mon Jun 14, 2010 7:35 pm
- Real name and title:
Hey all,
I have a question about the Q2 (redundancy)!
I did the Blindfolding procedure and I want to analyse 1 construct (here KUZ, it is reflektive) and marked it .
Then I see the output under "Construct cross-validated Redundancy" under Total, but where do I have to look exactly, tp find the Q2 (Stone Geisser-Criteria).
Total SSO SSE 1-SSE/SSO
BA 302,0000 282,4235 0,0648
KL 1208,0000 943,8200 0,2187
KUZ 906,0000 442,9977 0,5110
PS 906,0000 661,0435 0,2704
Do I have to look at at KUZ in the last column (1-SSE/SSO)? Here I see that it is greater 0, that´s good isn´t it?
The redundancies in the PLS Algorithm are completely wrong, am I right?
Thank you very much for your help :)
Best regards,
Larissa
I have a question about the Q2 (redundancy)!
I did the Blindfolding procedure and I want to analyse 1 construct (here KUZ, it is reflektive) and marked it .
Then I see the output under "Construct cross-validated Redundancy" under Total, but where do I have to look exactly, tp find the Q2 (Stone Geisser-Criteria).
Total SSO SSE 1-SSE/SSO
BA 302,0000 282,4235 0,0648
KL 1208,0000 943,8200 0,2187
KUZ 906,0000 442,9977 0,5110
PS 906,0000 661,0435 0,2704
Do I have to look at at KUZ in the last column (1-SSE/SSO)? Here I see that it is greater 0, that´s good isn´t it?
The redundancies in the PLS Algorithm are completely wrong, am I right?
Thank you very much for your help :)
Best regards,
Larissa
-
- PLS Expert User
- Posts: 31
- Joined: Mon Jun 14, 2010 7:35 pm
- Real name and title:
YesHey all,
I have a question about the Q2 (redundancy)!
I did the Blindfolding procedure and I want to analyse 1 construct (here KUZ, it is reflektive) and marked it .
Then I see the output under "Construct cross-validated Redundancy" under Total, but where do I have to look exactly, tp find the Q2 (Stone Geisser-Criteria).
Total SSO SSE 1-SSE/SSO
BA 302,0000 282,4235 0,0648
KL 1208,0000 943,8200 0,2187
KUZ 906,0000 442,9977 0,5110
PS 906,0000 661,0435 0,2704
Do I have to look at at KUZ in the last column (1-SSE/SSO)? Here I see that it is greater 0, that´s good isn´t it?
No. There is a difference between redundancy and cross validated redundancy (using blindfolding). Please read Henseler/Ringle/Sinkovics (2009) --> Literature in the announcementsThe redundancies in the PLS Algorithm are completely wrong, am I right?
Best
CR
Administration Team
-
- PLS User
- Posts: 13
- Joined: Thu Apr 04, 2013 12:42 pm
- Real name and title:
Redundancy computations still different from "R² * Com&
Dear all, especially Prof. Ringle and the admin,
i am following up on Andreas Pufall's question:
the redundancies given from the PLS-Algorithm (not Blindfolding) in SmartPLS 2.0 M3 are still not computed as "R² * Com", correct?
Therefore, should i still compute them "by hand"?
also, i am still unsure how to interpret the redundancy-values. any advice?
thanks in advance
Frederic
i am following up on Andreas Pufall's question:
the redundancies given from the PLS-Algorithm (not Blindfolding) in SmartPLS 2.0 M3 are still not computed as "R² * Com", correct?
Therefore, should i still compute them "by hand"?
also, i am still unsure how to interpret the redundancy-values. any advice?
thanks in advance
Frederic