There are 4 items in one variable and in EFA , items loaded in 2 different columns of other variable
This is quite common in social research (questionnaires with rating-scales).
In this context, you want to discover what variables combine to express the same underlying construct.
The default in most statistics packages is orthogonal rotation (Varimax), producing columns of factor loadings so that all factors are completely uncorrelated. (or, if creating summative scales, then the factors are minimally correlated.)
This is where theory comes into play. Must your factors be orthogonal? Maybe some correlation among the factors is acceptable. If so, then try a non-orthogonal rotation. (Oblimin, Promax)
see "https://hosted.jalt.org/test/PDF/Brown31.pdf" for a useful discussion.
A common mistake is to include all the rating-scale questions from the questionnaire in the same EFA. Think about your theory. Do you expect to find that some constructs are related to other constructs? Or that perhaps some construct "causes" some other construct? If so, then you want to see some factors correlated with other factors.
In such cases, do not include items that are intended to be "dependent" variables in your EFA. Remove "mediating" variables too, for the same reason.
On the other hand, do you really need to do an Exploratory Factor Analysis? If you are using an established set of scale questions, then you already know what items should go together - your theory in this context. Now that you have gathered your data, then you would want to run a Confirmatory Factor Analysis (CFA, not EFA). That is, you want to confirm that the items that should go together do go together.
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers