Depending on your hypothetical model, you may ask yourself what analytical methods are most appropriate for analysis. 

Such questions you ask:  

Do you have a theory ready for testing? 

Are you looking to build a theory or predictive model? 

Are you still in the exploratory phase? 

In structural equation models, there are models for testing theory and others to build a theory. Simply, we use Covariance-Based SEM and Partial least Squares SEM. 

CB-SEM is mainly used when you have an existing theory to test. At the same time, PLS-SEM is suitable for theory building and prediction and in the exploratory phase of theory development.

We use CB-SEM when we need to fit the proposed model, but when we need to maximize the R square, an easy solution is PLS-SEM.

CB-SEM is Precisely Confirmatory based on a sound theoretical base, While PLS-SEM is Exploratory and confirmatory based on an insufficient theoretical base.

So, when the objective of your research is exploratory and confirmatory but not based on an insufficient theoretical basis, you need to use PLS-SEM. 

PLS-SEM makes predictions by explaining the variability of crucial target constructs.  At the same time, CB-SEM conducts the theoretical test by explaining the maximum fit between the correlation matrix and parameter estimates.

One exciting thing about PLS-SEM is that when looking at the sample size, you need a small sample size, at least 20 respondents are enough, but you need a larger sample or at least 200 when doing a CB-SEM.

This means that assumptions about the normal distribution of the data are less required in PLS-SEM, quite the opposite in CB-SEM; the attributes of a normal distribution must be strictly satisfied.

There is flexibility in PLS-SEM when you need to configure the model; you can use both formative and reflective relationships in model specification. You cannot do that in CB-SEM; you can only build a model with reflective relationships. 

Finally, when extracting hierarchical latent variable scores in PLS-SEM, the determinant nature of the factors explicitly calculates the estimation of the unobserved variables. In CB-SEM, the indeterminate nature of the elements is considered.

Was this article helpful?