difference between Lavaan and SEMinR

[Benny Chen]

Hi all,

I know it may sound silly, but I am very new to R programming. 

I recently submitted a paper and got a comment that said I should use PLS-SEM (what I used is Lavaan package).

I did some reading and found some said Lavaan is factor-based SEM, whereas SEMinR is composite-based. I wonder what is the difference?  Is it really that PLS-SEM is better? And it would be helpful if anyone can provide references for the comparison of these two. 

thank you.

[laura trinchera]

Hi Benny,

whether using covariance-based SEM (estimated through the lavaan package) or component-based SEM (among which SEM-PLS, indeed PLS Path Modeling PLS-PM) strictly depends on the nature of the constructs you have in the model (and on the way you measured them). If all your constructs are "reflective" (i.e. are supposed to be latent variables that exist independently from their measures) there is no reason to prefer SEM-PLS to covariance-based SEM, i.e. you can keep using lavaan to estimate your model. If your model includes some "formative" constructs (i.e. unobserved variables that are defined as composites whose definition strictly depends on the variables you used as observed measures) then there will be some advantages in using SEM-PLS. You can refer to the recent book by J. Henseler "Composite-Based Structural Equation Modeling: Analyzing Latent and Emergent Variables" for a more comprehensive view of the difference between covariance-based and component-based SEM approaches.

To conclude, do not mix the software you used (i.e. lavaan) with the statistical model/technique you applied (i.e. covariance-based SEM). SEMinR is a software that allow to apply both covariance-based and PLS-PM.

Hope this will help.

Regards,

Laura

Laura Trinchera, PhD in Statistics
Associate Professor - NEOMA Business School
Department of Information Systems, Supply Chain Management and Decision-making
GoogleScholar: https://scholar.google.it/citations?user=VYMtEPMAAAAJ&hl=it

[Christian Arnold]

Hi Laura,

I agree with you that using PLS for formative constructs has an advantage: it is easier to implement. Whether you actually get "better" results with it is a different question: https://aaltodoc.aalto.fi/handle/123456789/20261

@Benny: It would be helpful if you mentioned the reasons the reviewer gave.

Best

Christian 

---

Von: lav...@googlegroups.com <lav...@googlegroups.com> im Auftrag von laura trinchera <laut...@gmail.com>
Gesendet: Dienstag, 4. Oktober 2022, 11:41
An: lavaan <lav...@googlegroups.com>
Betreff: Re: difference between Lavaan and SEMinR

--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/lavaan/625a6f98-06ec-4c77-9f4d-bceca2f2c2bcn%40googlegroups.com.

[Benny Chen]

Hi Laura

Thank you very much.

[Benny Chen]

Hi Christian

I provided the results of CFA and used CFI, chi-square divided by the degrees of freedom ratio, RESEA, and SRMR to evaluate my model fit.

The reviewer said "convergent and discriminant validity is very necessary, in addition to many important indicators for testing the model of this study provided by PLS-SEM." 

Kind regards

Benny

[Christian Arnold]

Hi Benny,

Exciting, whether this speaks for or against the reviewer ....  You can determine all of these coef with lavaan and semTools and you can expect the results to be more meaningful than the results obtained with PLS. On the topic of discriminant validity, there is a recent paper by Mikko and co-authors that you should definitely read: https://journals.sagepub.com/doi/full/10.1177/1094428120968614

HTH

Christian 

---

Von: lav...@googlegroups.com <lav...@googlegroups.com> im Auftrag von Benny Chen <chenshu...@gmail.com>
Gesendet: Dienstag, 4. Oktober 2022, 12:16

To view this discussion on the web visit https://groups.google.com/d/msgid/lavaan/7ab0c10c-0958-4c81-ad56-237dd8d1a1f2n%40googlegroups.com.