I am aware that the general rule of thumb is >0.3. However, my questionnaire has an item which has loading of around 0.245, and I would like to retain this item as i find it quite essential for my questionnaire. To justify, i heard that some papers quote factor loadings >0.2 as their cut-off but I could not find any based on my own literature search. Therefore, I would like to ask for help from anyone here if they have a record of papers that cite factor loadings >0.2 as a cut-off in their EFA.
Hi Chan Chee Ken
personally, I have not seen a single article or handbook that finds factor loadings < .3 acceptable and I have read many. Factor loadings are bacically correlations of an item with its corresponding factor. There is, however, a difference between mathematical and practical significance. A low factor loading may still remain mathematically significant whereas its practical significance - variance shared with the factor - is close to zero. This is why higher loadings are always preferable.
In the end, it is up to you to retain an items baed on its theoretical value, even without supporting literatur. This is to be stated as a limitation. As always, how you run the EFA influences its outcome. Please provide further details on the method of factor extraction, retention, rotation, perhaps even an output if available.
Best
Marcel
Sorry it's not my specialty. I was hoping for help. my best wishes
Chan Chee Ken, I think it's important to remember that EFA is primarily / ultimately a data-driven, not a theory-driven, procedure. Therefore, if the people in your sample don't regard certain items as being very related to other items, that's just the way the cookie crumbles, and I think you really have to accept that. Maybe it is something in the wording of the item that you'd like to retain that jeopardised its chances of being related to other items.
Whatever the reason for the low factor loading, I think you would be hard pressed to convince good methodologists that loadings < .20 are acceptable. I'm pretty sure I have seen loadings as low as that in some articles in which EFAs were used, but there were a lot of indications in those articles that the researchers didn't have much of a clue about what EFA was really about.
Marcel Grieger has asked for some important details up above. I think it would be good to address those questions.
My suggestion: https://www.researchgate.net/publication/355207581_Principal_Component_Analysis_versus_Factor_Analysis
Zenon Gniazdowski
kindly stop spamming/promoting your article under every question that is vaguely related to factor analysis.
Thank you all for the kind answers. Eventually, I have decided to stick with factor loadings 0.3 as the cut-off. To answer your question, Mr Marcel Grieger my factor analysis was done via Promax Oblique rotation in a tetrachoric correlation matrix with Unweighted Least squares (ULS) as the extraction method. Parallel analysis with eigenvalues >95th percentile criteria was used for factor retention.
It was enlightening to hear some of your thoughts and answers on this matter.
Regards,
Chan.
- Badges
- Science topic
- Similar topics
- Papers