Hi

first, PLS does not test common factor models. Hence, if you have (theoretically) a common factor model in mind (i.e. a latent underlying factor causes the respective set of indicators) avoid PLS.

Second, the difference between EFA and CFA has to do with the degree of expectations/hypotheses that you have on your indicators: Do you have any idea which indicators reflect which factor - then stick to CFA. If you have no idea, conduct an EFA. However, i find the lack of any theory the most worse situation when collecting data. It is doubtful that EFA can rescue oneself in such a situation :)

There were some discusions about CFA vs. EFA on researchgate. You should search a bit.

Best,

Holger

As far as I know,  EFA or CFA aren't applicable in smartpls. 

Although using both of them, is somhow controversial with today's standards. 

Dear Dr. Holger, thank you for your explanation and guide for further information search.  Dr.  Saeed, thank you for the additional information.

Hi Burhan,

If your measures used are modified versions of the original scales or new items you developed , reviewers usually suggest to add the results of an exploratory factor analysis of the constructs. Of course, you should use SPSS for EFA, although you use PLS in your path analysis.

Best,

Derya

Hi Derya,

Thanks for your explanation.

Regards,

Burhan

Dear Burahan I  asked same question from Prof Hair. His answer is here and I think its good reference  for you and other researchers.  

Is it necessary to do EFA (Exploratory Factor Analysis) before applying PLS-SEM and CB-SEM?

       Answer:     " No, it is not necessary to apply EFA first. I often do it to examine and understand the structure of the data and to compare to theory. But the statistical objective of EFA is different from CFA and you often will get difference results between the two statistical methods. Thus, theory is the driver in proposing measurement structures to test for reliability and validity."

Hi, when you are in confusion about theory and construct, and if your construct is reflective, go for EFA. We always follow EFA for adapted and constructed items.. go for pilot.

Aktar

Thank you Derya Doğan, Holger Steinmetz I did some slight modifications to the measurements, and I used PLS to analyse my hypothesized relationship without doing an EFA. Now, the reviewer suggests me to do EFA in SPSS, I have conducted that, but there are cross-loadings(but it does not make sense theoretically), and the loaded items to each factor are a little bit different from the results suggested by PLS. In such circumstances, what's your suggestion?

Hi Kara,

as I noted: PLS is not SEM and not CFA. The variables are composites no latent variables. Factors are latent variables (a subset). Hence, if your *theoretical model* is a common factor model (i.e., you think that there is a latent attribute causing the item response) then better test it via CFA (as you have an expectation about the structure).

Cross-loadings do not validate the common factor model. They are only a problem when the factor structure is wrong and the cross-loadings *appear* as a fix (but the don't change the structural problem).

Best,

Holger