I have a general question and look for some suggestions regarding cross-loading's in EFA. Which number can be used to suppress cross loading and make easier interpretation of the results? There are some suggestions to use 0.3 or 0.4 in the literature. But can I use 0.45 or 0.5 if I see some cross loadings in the results of the analysis? Some people suggested to use 0.5 depending on the case however, can anyone suggest any literature where 0.5 is used for suppressing cross loading ?
In my analysis, if I use 0.5 it gives me 3 nice components, while with 0.4 I have few cross loadings where difference is 0.2
I would much appreciate your suggestions/comments
Best regards,
Davit
Dear Davit Gogilashvili,
Obviously, you know how to handle EFA and your using 0.5 for handling cross-loadings is well-justified. However, for further information, I refer you to the following links, which can hopefully take care of the targeted issue.
Best regards,
R. Biria
https://pareonline.net/pdf/v10n7.pdf
https://www.researchgate.net/.../290042054_An_Easy_Approach_to_Exploratory_Facto...
https://kttm.hoasen.edu.vn/sites/default/.../exploratory_factor_analysis_khang_updated.pdf
Dear Reza,
Thank you for your feedback. However, I cant access pdf files you have attached. I would much appreciate if you can send me on e-mail: [email protected]
Best regards,
Davit
Either way, you should not forget that you are actually not suppressing the cross-loadings, you are just choosing to ignore them more strongly. Even with your .5 rule, the cross-loadings you see with .4 are still there.
Just a thought to keep in mind :)
Best of luck,
Georg
Georg thanks for your comment.
Do you have any concrete suggestion?
Best regards,
Davit
Dear Davit,
to be honest, my concrete suggestion would be to take these cross loadings very serious. Even if you can justify ignoring
Dear Alejandro,
Thank you for your comment
I agree with you if the loadings in one factor are "higher" than other factors, cross-loadings are ok. Many researchers have the same argument.
Davit
Dear Georg,
Thanks for your suggestion
Firstly, I have checked multicollinearity and initially I had determinant < than 0.00001. I have tried to omit those items with high correlations and my determinant exceeded the above mentioned threshold. Logically, I was expecting not to have any cross-loading issue however, still I have a bit. One thing is clear each item assigned for each factor loads quite high 0.8 and cross loading does not exceed 0.5. This situation allows me to confidently assign those items with respective factors.
Best regards,
Davit
Dear Davit,
then it seems I misunderstood your initial statement
"...while with 0.4 I have few cross loadings where difference is 0.2".
It seems you have already decided how to deal with the cross loadings. But please consider that the cross loadings can give you insights into you items, your test, and your test takers' response behaviour. If on the other hand you want to choose your cut-offs to get the 3 nice components, obviously you will find citations to support these cut-offs.
Best wishes and best of luck with the data/test/analyses,
Georg
I suggest 0.5 could be the better option.You will get review for the same also to justify in the paper
- Badges
- Science topic
- Similar topics
- Computer Science
- Research Topics