SmartPLS 2.0 v 3.0

This forum is closed, and read-only.
Locked
THEPEDRO62
PLS Junior User
Posts: 2
Joined: Fri Aug 15, 2014 3:26 pm
Real name and title: Peter Green, Marketing Science Manager, Millward Brown UK

SmartPLS 2.0 v 3.0

Post by THEPEDRO62 »

Hi Sven and Team,

I've been trialing the new v3.0 with a view to purchasing some licenses for our team of marketing scientists here in the UK and just thought I'd drop by some feedback and questions.

Overall I love the new engine and analysis design and I'm sure in time I'll grow to love the new features just as much too! There are a couple of things that I've noticed are omitted from the 3.0 model reports that appeared to be included in 2.0 and I wanted to see if I'm just missing them somewhere or if there is a better way of thing about these concepts now? Any feedback massively appreciated please.

-'Measurement Model' and 'LV Scores Unstandardised' - to be honest I never could figure out how these were calculated, but they were useful to us and we used them. Is there a way to calculate these again?

-'Communality' and 'Redundancy' removed from the fit overview. We used these to create a surrogate Goodness-of-fit statistic, one which is SQRT(Avg Communality*Avg RSQ). Is there a reason these two statistics have been removed from the reports?

-LV index values (Averages) removed from report.

Any feedback on these changes would be appreciated please!

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

Re: SmartPLS 2.0 v 3.0

Post by jmbecker »

Hi Pete,
The unstandardized result can be found in the IPMA procedure. There you will have the unstandardized estimates for the outer and inner model as well as the index values for the LVs and MVs.
There are many reasons why we removed these results from the standard PLS report and put them to the IPMA, among them performance and design reasons, but also consistency of evaluation approaches. The unstandardized estimates not native PLS results, but are special transformations of the normal PLS results that are most useful within an Importance-Performance-Matrix Analysis.

The communality results are still there, just under a different name. You will see that in the old SmartPLS 2 that Communality and AVE always had the same results (they are basically the same). We therefore only report AVE anymore.
However, you should know that this surrogate Goodness-of-fit statistic is actually not a good indicator of model fit and you might reconsider using it at all.
Henseler, J., & Sarstedt, M. (2013). Goodness-of-fit indices for partial least squares path modeling. Computational Statistics, 28(2), 565-580.

Best,
Jan-Michael
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
THEPEDRO62
PLS Junior User
Posts: 2
Joined: Fri Aug 15, 2014 3:26 pm
Real name and title: Peter Green, Marketing Science Manager, Millward Brown UK

Re: SmartPLS 2.0 v 3.0

Post by THEPEDRO62 »

Hi Jan-Michael,

Very many thanks for your prompt and very useful reply - massively appreciated.

I've now found the unstandardised LVs and MVs in the IPMA procedure - I like that they have been given this location, it will slim down the report sizes and allows the user to focus on Importance-Perfomance - this is a very key part of path analysis for many of our clients. I also like the Rescaling versions from 0 to 100 - these could be very useful. I am however interested in gaining full insight into the "Unstandarised LV" scores - is there somewhere I can find the actual transformation formulae for how these scores are calculated please? Or can you post to this thread? Very many thanks.

I'm not sure I follow your thinking around AVE matching Communality. In all the PLS path models I can recall running, my AVE scores are displayed as 0.0000 for all constructs in both 2.0 and 3.0, whereas the old Communality scores were always a result somewhere between 0 and 1 in v2.0. Is this something that can be checked/confirmed please? I'm happy to take the discussion offline if necessary and forward on any example models for reference.

Thank you too for your guidance around the GoF - we're fully aware that this is a non-validated approach, but at this time we believe it's the best surrogate for an overall model best fit and we use it alongside the dependent RSQ in our results. Unless that is there are any better surrogate GoFs that you are aware of as having been developed in the last couple of years?

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

Re: SmartPLS 2.0 v 3.0

Post by jmbecker »

Hi,
I thiknk there are several post on these topics if you search the forum.

The unstandardized weights are simple weighted composites based on the rescaled weights that you will find in the 'Outer Weights' section. These weights are the normal weights devided by the standard deviation of the corresponding MV and divided this by the sum of all weights of a block.

Example:
If you have 3 MVs in a block, with weights w1, w2, w3 and standard deviation sd1, sd2, sd3 then you get:
wsd1 = w1/sd1
wsd1 = w1/sd1
wsd1 = w1/sd1

and you have to divided this by the sum of all weights of a block. Hence, the sum of the rescaled weights will equal 1:
w1_rescaled = wsd1/(wsd1+wsd2+wsd3)
w2_rescaled = wsd2/(wsd1+wsd2+wsd3)
w3_rescaled = wsd3/(wsd1+wsd2+wsd3)

The unstandardized LV scores have the same correlation pattern as their standardized counterparts.
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
Locked