The factor tries to reduce the minimum of factors in the analysis mostly the two-factor model seems to fit better. You must corroborate with your theoretical framework and perform the confirmatory factor analysis or even SEM analysis.

Hi Attila,

although there is a common objection of doing EFA and then CFA using the same sample, there is a slight variant that at least provides a basis for further future research: Test a one vs. a two factor model in a CFA or SEM framework but include also external / other variables that function as validation criteria (i.e. variables that should be related with the one or two theoretical variables if they truly are what you think they are. These new variables add new implications for the data given the model structure. A model, even applied to you current data, will test those new theory-based predictions, while "repeating" those that you already explored with the EFA (concerning the factor structure). Hence it is not simply "repeating the exercise".

In this thread, I wrote a posting that explains this in more detail:

https://www.researchgate.net/post/Can_I_use_EFA_before_SEM_analysis

Be aware, however, that others may find this unconvincing and your result is only a further piece in a longer research chain. But at least it is a solid basis for deciding how to proceed (as in any exploratory research).

Best,

Holger

Dear Holger, thanks for the suggestions.

I would love to know what exactly you did and how it worked out--in some sense as an example for others here. Perhaps you want to present this. That is, what were your items, interpretation of both, the one-factor and the two factors, your idea about potential validation variables, and the results of the SEM.

I would offer to contribute by guiding you, for instance through R/lavaan how to set up the models.

Think about that. Would be a very nice working example.

Best,

Holger

Dear Holger, sorry for not reacting before. Thanks for offering your help. I have been studying literature on different dimension testing methods, like higher order FA, bifactor model, mokken scaling. Literature suggests the first two are only advised if there are at least three tentative domains. Mokken scaling seems to suggest one dimensionality, but I found an article that doubts the aplicability of mokken scaling for dimension testing. The CFA carried out in Rcmdr suggests that the 2 factor model provides a better fit. As for SEM (which I am not really knowledgeable about) a statistician collague of mine said that my scale is two simple for doing SEM on it. I tried other factor retention methods since my post offered by various R packages and they suggested 2 factors. I will probably stick with the two factor solution, yet, all questions are meant to test, the same construct but from two different aspects.