Blindfolding in M3

This forum is closed, and read-only.
Locked
plskarsten
PLS Junior User
Posts: 3
Joined: Wed Jul 05, 2006 6:00 pm
Real name and title:

Blindfolding in M3

Post by plskarsten »

I have three questions regarding the blindfolding procedure in M3.

Firstly, whatever constructs or combination of constructs I check in the pls-algorithm wizard, there will always be CV Comm. and CV Red. measures calculated for all LVs of my model. What is my mistake?

Secondly, is it possible to calculate all Qsquare-measures simultaneously (PLS graph allows for only one blindfolded LV at a time)?

Lastly, there are large differences (up to 0.5 incl. sign changes) when I compare Qsquares derived from M3 with those of PLS-Graph. How is this possible-esp. taking into account that Chin 98, p. 318 noted that even at low omission distances (5-10) results for Qsquares will be consistent?
ad: used omission distance was 31

Thank you for your help!
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:

Post by cringle »

Dear Karsten,

ad 1: you need this kind of information as a result of the blindfolding procedure. Please specify the problem or what you do not understand.

ad 2: yes this is possible in SmartPLS

ad 3: the source of different results is connected to the way of treating missing values. We indicate (see information on M3) that we use mean replacement, PLSgraph uses pairwise deletion. Our test did not show huge differences.

Best regards
Christian
plskarsten
PLS Junior User
Posts: 3
Joined: Wed Jul 05, 2006 6:00 pm
Real name and title:

Post by plskarsten »

Thank you for these answers.

Ad 1:
To specify my first question: What effect has checking constructs in the blindfolding wizard of M3? That means: How do checked items compared to unchecked ones affect calculation of CV communality / CV redundancy measures? I hope this might provide the answer to the next question as well.

Ad 3: I do not have any missing values. So this is not very likely the reason for the differences in the results between Smart-PLS and PLS-Graph. I experimented a little and found out that CV communality measures run very close to those of PLS Graph. However, CV redundancy measures tend to differ a great deal across various omission distances (SSO is o.k. but SSE is less than in PLS-Graph). Do you have any other ideas what might cause these differences in CV red.?
For calculation I checked all of the LVs of my model in the blindfolding wizard of M3.

Thanks a lot!
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:

Post by cringle »

Ad 1:
the option to check or uncheck boxes is not connected (this is something we must take car of in the next release) - you get always results for all latent variables.

Ad 3:
we ran into the same problems and carefully recalculated our implementation in Excel. we have not been able to come up with the results in plsgraph. especially the outer blindfolding relationships caused some headaches. for example, the outer estimates (full model) for a latent endogenous variable are at follows:

0,8343
0,8838
0,8338
0,8878

the plsgraph blindfolding (D=3, cv.red.) computations for these outer relationships are:

0,7765 0,7868 0,7608
0,9436 0,8717 0,9606
0,8391 0,0103 0,9292
0,8537 0,8412 0,7364

especially the extremely low value in the second round look suspicious. this kind of unexplainable outliers have not been observed in smartpls. however, the results in this example differ a lot - as you noted for your cases:

PLSgraph SSO SSE 1-SSE/SSO
Eta1 800,0000 727,1253 0,0910

SmartPLS SSO SSE 1-SSE/SSO
Eta1 800,0000 642,6035 0,1967

it would be fantastic, if you could run your own excel computations. it takes a long time (we put much time and effort into this problem) but may proovides additional clarification. we work on a pairwise deletion implementation - a direct comparison of smartpls and plsgraph results is then easier!

best regards
christian
plskarsten
PLS Junior User
Posts: 3
Joined: Wed Jul 05, 2006 6:00 pm
Real name and title:

Post by plskarsten »

Hi Christian,

Once again: thank you for your quick reply. I will do as you suggested and cross-check QSquares using good old Excel :-). Because of year-end this may take a little longer. Pls. let me know if you want to know what came out of it.

Best regards
K.
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:

Post by cringle »

Hi Karsten,

we would be very much interested in those results. The three of us experimented on this issue about three weeks to make sure that the SmartPLS implementation works without error. When we add the pairwise deletion option in the future, it would be good to know if plsgraph works fine in blindfolding and to have a comparison for our results.

Best regards
Christian
sunnyarch
PLS Junior User
Posts: 3
Joined: Wed Aug 29, 2018 12:42 pm
Real name and title: Mr.Suresh Acharya

Re: Blindfolding in M3

Post by sunnyarch »

While calculating, the difference for Q2, between both smartpls 2 and 3 were all less than 0.02. Further Q2 values that were found from Smart Pls2 were on the higher side compared to those that came from smartPls3. Though all Q2 values Cross Validated Redundancy range from (0.34 to 0.53) in my studies . The q2 effect size (manually calculated using SmartPls 2) for model re estimations has shown consistently negative on 2 of the predictors. Need to calculate the q2 effect size again using SmartPls 3 manually to see if there are difference.
Locked