Hello Sarita,

Covariance-based and variance based (partial least squares) structural equation modeling methods aren't better or worse; they're different, with different aims, assumptions, and philosophical perspectives about model-building and evaluation.

It's fair to say that the PLS-SEM advocates often seem to take on an evangelical world-view of the method as being uniformly superior; you have to take that stance with a grain of salt.

Here's a laundry list of common distinctions of the methods (as offered by PLS-SEM promoters):

https://www.smartpls.com/documentation/choosing-pls-sem/choosing-between-pls-sem-and-cb-sem

Good luck with your work.

Dear Sarita Sood,

I think it is not about which is better or worse. I think our main role is to recognize each approach strengths and weaknesses.

Starting from CB-SEM, although some estimation techniques used within CB-SEM to handle non-normal distribution data set, however, the measurement model will be poorly loaded and often is not going to pass through the reliability and validity tests. However, this strick demand for a specific data distribution has a positive side as well which is that you have a set of model fit indices to help you to decide whether it is a good model or not.

In the other hand, due to no prior data distribution requirements for VB-SEM, the model fit indices are still in the development process and not yet sure if they will be widely accepted, even though that some software as SmartPLS is presenting a set of fit indices, however, the designers of it asks the researchers not to use them. Furthermore, although that VB-SEM is so often referred to as prediction approach of SEM, however, it is still in the early ages.

In addition, the following are the arguments for choosing PLS as the statistical means for testing structural equation models (Urbach & Ahleman, 2010) by the researchers, which may also give you more insights about the differences.

• PLS makes fewer demands regarding sample size than other methods.
• PLS can be applied to complex structural equation models with a large number of constructs.
• PLS does not require normal-distributed input data.
• PLS is able to handle both reflective and formative constructs.
• PLS is better suited for theory development than for theory testing.
• PLS is especially useful for prediction.
• PLS can handle first and second-order together.

Best Regards,

Belal Edries

I recommended

Article Estimation issues with PLS and CBSEM: Where the bias lies!

Article PLS-SEM or CB-SEM: updated guidelines on which method to use