0
$\begingroup$

I am developing a new instrument consisting of items grouped into defined domains (constructs), such as personality traits. Each construct is measured through a set of 4-6 items (questions). To validate the instrument, I consider running an exploratory factor analysis (EFA): a) separately for each construct, to assess whether the items are unidimensional, and b) on all the constructs: not across all items, but by combining items of each construct via indexing into a single score, and then running the FA across the resulting approx. 15-20 constructs. This appears more effective then running an FA over all the (approx. 90-110) items, which would also require a much larger sample size?

Is this approach valid? Thoughts?

$\endgroup$
1
  • $\begingroup$ This site is much better suited to more specific questions than to questions of, to paraphrase, "here is my approach: please comment." I'll be surprised if a topic as complex and multifaceted as this will draw a satisfactory answer, and you may want to contact a consultant/coach/mentor. But at any rate suppose you tell your N; to what sort of party you will be reporting and with what expectations for scientific rigor; to what degree any of these scales have been previously validated; and whether you are planning to cross-validate. Cheers ~ $\endgroup$
    – rolando2
    Commented Apr 10, 2020 at 20:34

1 Answer 1

0
$\begingroup$

I would not consider this a valid approach. I would recommend reading Boateng et al. (selected because it is publicly available, not because it is necessarily a seminal work) and some introductory psychometrics and scale development books. If I were reviewing a new scale I would expect to see item level EFA, CFA, Cronbach's alpha, inter-item correlations, and comparison to an existing measure at a bare minimum. Looking at single factors/sub-scales and then only testing those is not a reasonable shortcut to take.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.