EFA or CFA for validity

Question

I need to test validity of measurement scales in my survey.

1) Is it true that both can be used to test measurement validity?

2) Or there are some types of validity that can be tested with EFA and some that - with CFA?

3) What is the best solution - which one should be used or both?

P.S. I have survey consisting of n-items (n-questions). Survey is related to consumer attitudes and behavior. The survey was constructed to measure 6 different constructs, each construct consist of different number of items (questions). I am writing a theoretical review of statistical methods that would be used to analyse the survey data. This is my first survey, I never did construct validity before, just read about it.

Answer 1

In the end, in order to validate that your instrument does in fact measure 6 specific constructs, you will need to test this model with confirmatory factor analysis (CFA). Exploratory factor analysis (EFA) may be informative in the early stages of evaluating an instrument, e.g. if you are not sure if your instrument measures 4, 5 or 6 dimensions. However, CFA is more informative about sources of misfit and potential methods effects. This includes cases of items with very similar wording, or reverse items. Such pairs of items have inflated loadings and may change the factor solution. CFA will notify you of such cases, EFA will not. In any case, if you want to test validity, you need a hypothesis, in this case a model, which you put to the test and potentially confirm.

Answer 2

Factor analyzing the scale itself does not tell you anything about the construct validity of the scale. Even if EFA/CFA suggests that there are indeed 6 dimensions, this does not tell you whether those 6 dimensions indeed represent what you think they should represent.

To obtain support for the construct validity of a scale, you have to examine whether the scale is related to other constructs to which it should be related to based on theory, prior empirical evidence, or just plain common sense (convergent validity) and not related to other constructs to which it should not be related to (discriminant validity). The same goes for any subdimensions of the scale. So, this involves collecting data not just with the scale itself (which is meant to measure construct X and its subdimensions x1, ..., x6), but also with other scales that are known to measure constructs Y and Z and then examining the relationships among those constructs (which can be done with factor analyses).