I am preparing a model of personality. EFA revealed 18 factors. In second factor, the first item shows a loading 1.08. According to my understanding, these loading are correlations. So does one magnitude represents, highest value of correlation? Is such a value common in large sample data? Shall I accept or reject that item? Does the loading of items can range more than .9?
Dear Gitanjali,
it seems Rimantas is right, Heywood case. However, only in the case that you actually speak of the communality of that item instead of its' factor loading (has to be clarified), otherwise it seems odd (impossible). As Chalamalla points out some definitions, communality (to which the Heywood cases refer) means: how much of the variance of an item can be explained by all factors taken together. 1 would 100%, and yes, values higher than 1 mean that you can explain more than 100%, which is a problem for interpretation and a Heywood case. The question is rather not whether this occurs in large samples, but: when does it not when there are correlated factors (and are allowed...) which additionally load very high on the same items and all factors claim to add to the explanation while being the "only source" of the information they contribute, which when factors are correlated, they are not.
You will find how to deal with it searching for the mentioned "Heywood"
Best wishes,
René
It might be useful to inform the reader the specific type of estimation method you used: Principal Axis Factoring (PAF)? ML-Robust? IPAF? etc.. Regardless, I would urge you to abandon the traditional EFA estimation procedure and go straight to using ESEM to help you address this item-factor loading issue. You will also observe meaningful associations among the derived factors. Good luck with your project...
Thank you Rene, Rimantas and Augustine.
I shall check Heywood case and then think of proceeding with an appropriate step.
- Badges
- Science topic
- Similar topics
- Psychology
- Psychometrics