Hi all,
I am currently using SmartPLS and came across some issues with a consistent PLS algorithm. I would like to get feedback if I still should use SmartPLS or switch to other software.
I do get errors when using PLSc, as per the image below.
As far as I understand, the issue can be caused by violation of assumptions (e.g. consistent PLS algorithm is "PLSc is very strict about the common factor model assumptions." )
- https://www.smartpls.com/documentation/algorithms-and-techniques/consistent-pls-problems
Some people also report a similar problem.
- https://forum.smartpls.com/viewtopic.php?f=3&t=3766
My question is, though, what to do when encountering this issue. These URL discussed why and how the problem occur, but did not touch on what to do with these problems.
Basically, my model should use PLCs according to PLS-related papers as the model includes reflective latent construct. But I can't get a proper result with PLSc while I get a seemingly acceptable result from normal PLS.
In this case,
(it will be nice if your opinion is also backed up by papers)
Your suggestions and feedback are highly appreciated!!
Regards,
Ted
There are a good number of packages in R, including pls and lavaan. You can search the latest versions of these packages and use them.
Thanks. PLSc-SEM mimics a factor model. It uses the same PLS scores but ex-post changes the coefficients. This works without problem if you have high loadings and a high model fit. Otherwise, the change does not work. But why don't use just PLS: Article Estimation issues with PLS and CBSEM: Where the bias lies!
Dear Babak Jamshidi
Thank you for your reply! Yes, I will look into them when SmartPLS doesn't work for me.
Dear Prof. Christian M. Ringle ,
Thank you for your comment and explanation.
I have read some of your your papers and they are indeed really helpful for my current research (e.g. When to use and how to report the results of PLS-SEM is the basis of my analysis at the moment).
Besides, your suggested paper is very helpful too. Especially, the comparison of results between the study 1 and 2 is interesting.
Anyway, as you comment "if you have high loadings and a high model fit", I guess this may be one reason why PLSc doesn't work for me as the model fit is not very high.
The reason for using PLCs is because some papers recommend to use when reflective scales are in the model.
As far as I understand, the debate is still hot, and a safer approach is to consider both sides!
e.g.
"Composite-based models should opt for PLS, whereas factor-based models should use CB or PLSc." (Dash & Paul, 2021)
"If a researcher uses PLS for factor models (reflective), consistent PLS with measurement error correction should be used."(Evermann & Rönkkö, 2021)
Thanks from New Zealand.
T
Hey Ted,
I am not sure what exactly the problem is. However, you could estimate your model using the R package cSEM. This package provides a verify function which tells what your problem is. There are several reason why an estimation can be inadmissible, e.g., standardized factor loading estimates above one or reliability estimates above 1...
I agree with you that if you you want to estimate a reflective measurement model, you should use PLSc, see also Benitez et al. (2020).
Best regards,
Florian
Ted Totsuka your model is 2nd order then there is one option to use regular PLS since you only get a contracts score in regular PLS.
Further, after applying for the 2nd order you well know what is the problem.
Florian Schuberth
This is what I get from csem (thanks for this tool btw).
In nutshell, it looks like I have a NPD issue here.
I will keep looking into it.
Regards,
T
Verify admissibility:
inadmissible
Details:
Code Status Description
1 ok Convergence achieved
2 ok All absolute standardized loading estimates
I agree that is the problem. Thia can happen because of attenuation. If you want to stick to PLS, as a workaround, you could try to use Cronbachs alpha instead of rhoA... Otherwise, you can use SAM (structural after measurement). This can also be done in cSEM...
- Badges
- Science topic