If factors from an exploratory factor analysis are highly correlated with one another, does this indicate that the constituency items are measuring similar constructs?
Or do the items measure different but related constructs?
If they measure the same construct, why wouldn't a single factor be extracted instead of having two highly correlated factors?
It also depends on the rotation method, but you should use parallel analysis and that meet the other parameters of the AFE to perform the AFC and confirm the proposed model, then if you should take more into consideration the correlation between dimensions
Yes, a high correlation means the two constructs are measuring the same thing, at least in a statistical sense. One reason the extraction would find two factors is that it occurs before the rotation.
To get a better sense of how your items are related, I would recommend examining an ordered correlation matrix, where you begin by listing the variables that load most highly on the unrotated first factor, and then the variables that correspond to the unrotated second factor. This will allow you to examine the correlations within the first set of variables, the correlations within the second set, and the correlations between the items in the two sets. If there are indeed two factors, then the correlations in the two "within" sets should be notably higher than those in the "between" set.
Hello Jameel,
A lot of what comprises the "best answer" hinges on what sort of theory, if any, you are relying on as regards this set of variables that you are factoring. When you use oblique rotation it is quite possible that factors can and will correlate (sometimes highly, though you didn't indicate what you consider high in the context of your data set).
One possible answer is that two (or more) correlated factors are really indicants of a second-order factor on which they both/all load.
A second possible answer is that you really don't have distinct factors. One common instance of this is when people factor questionnaire/attitude/opinion/sentiment measures wherein some items or stimuli are positively phrased and some are negatively phrased. It is not unusual for two apparent factors to emerge in EFA, one capturing the positively worded stimuli, the other capturing negatively worded stimuli. And, yes, the factors correlate (albeit negatively). Others might have preferred a single, continuum factor as the basis for such an arrangement.
A third possible answer is that you've identified latent variables that aren't all that different. For example, measures of self-concept often correlate with specific measures of self-efficacy. This takes us back to the underlying theory point.
Do note that David Morgan's good suggestion is akin to looking for convergent/discriminant validity information in your data set (back to the theory, once more!). As well, Cristian's suggestion walks you back to the earlier step of deciding on the appropriate number of factors, which is always an important step.
Good luck with your work.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts