Page 1 of 2

Quality statistics - Release date for next SmartPLS release?

Posted: Wed Dec 07, 2005 10:44 am
by stefanbehrens
Given some of your feature announcements (quality-statistics, cross-loading, etc.), I am anxiously awaiting your next SmartPLS release. Hence, I was wondering, if you have already set/ communicated a release date. Could you specify a likely timeframe?

Btw - great work on the software so far!

Regards,
Stefan Behrens

Posted: Sat Dec 10, 2005 10:11 am
by cringle
Hi,

due to limited resources (time and funds) this might take a while – implementing and extensively testing takes a lot of time... and we are involved in a lot of other projects at the moment.

However, the most important features are included in SmartPLS 2.0 M2 – all additional things can be easily calculated in MS-EXCEL.

Maybe in February... but not for sure…

Best regards
Christian

Excel?

Posted: Sat Dec 10, 2005 11:32 am
by stefanbehrens
Christian,

thank you for your swift reply. Not sure how long this would take, but would you mind pointing me into the right direction on how to calculate
- Stone-Geisser Q-square and (more importantly)
- Crossloadings
using current SmartPLS 2.0M2 Output and Excel? I'd be happy to do it (and post a corresponding sheet for the community's benefit) if I knew how to do it.

For the crossloadings I am currently working around the problem by duplicating the model in SmartPLS 1.0 (which provided those stats).

Also, when I registered my copy of SmartPLS, I was informed that the copy expires some time in January, 2006. Will there be new activation keys before the next version is released?

Kind regards and have a nice weekend,
Stefan

Posted: Sat Dec 10, 2005 7:16 pm
by cringle
Hi Stefan,
I'd be happy to do it (and post a corresponding sheet for the community's benefit) if I knew how to do it.
that sounds great - I definitely wanted to spend more time on these things, but other things stop me... I send you one of my Excel-files next week (they are on my PC at the University...).

Well, we take care of the license issue – you are right ;) – it would be great to give new licenses with a new release so that people need to update SmartPLS… Soon, there will be an announcement in the forum.

Best regards
Christian

[/quote]

Calculate cross-loadings in SmartPLS 2.0

Posted: Wed Dec 14, 2005 9:18 am
by stefanbehrens
Christian,

after a little reading I was able to figure out a way to calculate cross-loadings based on SmartPLS 2.0M2 output. To make things easier, I've built a little Excel tool to do the trick.

Here's the link for those who are interested (quick & dirty, but works): http://www.is-outsourcing-survey.org/PL ... adings.xls

Enjoy!

Stefan

Posted: Wed Dec 14, 2005 5:13 pm
by cringle
Hi Stefan,

good work! We had this in SmartPLS before, forgot to include it in the current release and will definitely include this feature in the next release, again!

Well, I created two spreadsheets in Excel (cross-validated communality and cross-validated redundancy) for calculating the Q² predictive relevance.

http://www.smartpls.de/Sat_Com.xls
http://www.smartpls.de/Sat_Red.xls

This is an example for the latent variable Satisfaction in the ECSI model. However, for lager models, it might take some time to run all calculations. We plan to include the blindfolding procedure in the next release!

I also intend to demonstrate with these files how suggestions for additional extension in SmartPLS should be posted. Such easily to comprehend “cooking recipes” provide for a quit implementation into SmartPLS.

Best regards
Christian

Posted: Fri Jan 06, 2006 3:47 pm
by panhans
Hi Christian,

many thanks for your two upfront xls spreadsheets, I just need some help on their exact interpretation. Is the blindfolded cross-validated communality the same as the Stone Geisser Test Criterion? And if so, what is then the meaning of the blindfolded cross-validated redundancy test?

Many thanks!
Dirk

Posted: Fri Jan 13, 2006 7:49 pm
by stefanbehrens
Dirk,

basically, to test the predictive validity of structural PLS-models, you need to look at cross-validated redundancy (F-sq) of your main dependent variable(s).

F-sq < 0 indicates that your structural model has issues.

Cheers,
Stefan

Excel-Tool for Stone-Geisser Q-square (redundancy)

Posted: Sat Jan 14, 2006 4:43 am
by stefanbehrens
All,

after some more in-depth reading and some fooling around with Christian's Excel-sheets I was able to figure out a way to implement the procedure in a more flexible way.

The sheets I created take MV- and LV-score output from PLS and calculate the cross-validated redundancy measure for a given variable. By rejigging some of the calculations, I was able to keep the number of predictor LVs, number of target LV indicators and the omission distance flexible (though within certain limits). Thus, the present version works with up to 10 LVs (including the target LV), up to 5 indicators for the target LV and omission distances up to 10.

I used the data in Christian's sheet to validate the new model and, in fact, it produces the exact same value for F-square (0.461)

Unfortunately, the new sheet has become a little large (4meg) and is thus zipped for download: http://www.is-outsourcing-survey.org/PL ... eisser.zip

Enjoy!

Interestingly, the omission distance seems to have absolutely no effect on the F-square. I'd appreciate if somebody could explain why...

Stefan

Calculate Stone-Geisser's Q2

Posted: Mon Mar 27, 2006 12:33 pm
by beuningen
Hi all,
I have tried to calculate the Stone-Geisser's Q2 using the Excel sheets presented above. But I didn't succeed. Maybe someone can explain how to do this exactly? A general explanation of how to calculate the cv-redundancy index is also fine. I think I should report this statistic for one of my constructs which is a 1st-order reflective, 2nd-order formative construct (modeled in PLS using a partially aggregated measurement model).

I'm using my PLSgraph output since I've used it for all analyses. I found out that the path coefficients in smartPLS are not exactly the same so was also wondering what could be the problem there (since I have no missing data in my dataset).

Thanks in advance!
Jacqueline

Posted: Tue Mar 28, 2006 2:55 pm
by beuningen
Sorry, never mind the previous posting. I finally figured it out from the excel sheets posted before.
Greetings, Jacqueline

Posted: Sat Oct 07, 2006 7:18 pm
by tks
Hi all,

just to get it right: Can Q² be calculated for LV with reflective as well as with formative indicators or does it only make sense for reflective operationalisations?


Cheers,
Tom

Posted: Thu Oct 12, 2006 9:44 am
by Nicole Baumüller
Hi Tom,

did you already find an answer to your question concerning Stone Geisser with formative Indicators? I have the same problem and would be intersted in an answer as well!

Cheers
Nicole

Posted: Thu Oct 12, 2006 11:01 am
by cringle
Hi,

it makes sense in reflective measurement models. Please check Tenenhaus et al. 2005 (viewtopic.php?t=21). They provide excellent explanations and a very good formal description that should clarify everything.

Best regards
Christian

Posted: Tue Oct 17, 2006 3:25 pm
by tks
Hi,

thanks for your reply. Another useful reference is

Herrmann et al. (2006): Varianz- und kovarianzbasierte Strukturgleichungsmodelle. In: zfbf 58: 59, 61.

Cheers,
Tom