I translated a scale from Engish to Arabic using the back-translation technique. After collecting data, the scale has a low reliability (α < 0.4), but when one item is removed alpha exceeds 0.7. I suspect this particular item has a translation issue, because the original English scale demonstrated high reliability across many published studies.
So is it right to use EFA or PCA to justify why I removed the troubling item in my research? This particular item does not load on factor 1 but on factor 2 (the orignial scale loads on 1 factor).
Thank you!
Sultan Sabr Your approach seems very reasonable to me. If your factor analysis clearly shows that the item in question does not load on the intended factor, this may be a reason to exclude (or modify or replace in future studies) that item, particularly if there are also plausible theoretical-substantive reasons (e.g., translation issues) as you write.
Note that exploratory factor analysis (EFA) and PCA are not the same thing (although they often lead to similar conclusions). For details, see, for example
Article Repairing Tom Swift's Electric Factor Analysis Machine
Hello Sultan,
Yes, if exclusion of one item makes that big a difference in your estimated internal consistency reliability values, it (the item) probably needs to be dropped or revised for your translated instrument. I agree with Christian Geiser on this point.
As a note, this sort of instance could most likely also have been discovered by having run ordinary (corrected) item-total correlations for each question/item.
Finally, common factor analysis makes very different assumptions about reality than does principal component analysis. You'll find that, in general, unless the number of variables is large, PCA tends to bias loading estimates upward in magnitude relative to what common factor analysis would yield. That, coupled with the assumption in PCA that there is no such thing as unique or error variance in manifest variables renders it not well suited to measurements of human belief, perception, sentiment, attitude, or opinion (or, for that matter, knowledge/skill attainment).
Good luck with your work.
Hello Sultan,
Because for cultural differences, there is a possibility that items may not necessarily behaves in the same way everywhere.
Item reliability is not always measure by using Cronbach alpha. However, Cronbach's alpha of below 0.4 is not acceptable. there is also possibility of data entry error. Make sure your data entry has no error. In addition, there may be possibility of response bias.
EFA and PCA are not same thing. I think for the items like yours, EFA would work best. EFA is based on item correlation Whears PCA is based on items variances.
Thank you all for the detailed answers, and thanks Christian Geiser for the article, it is very informative!
Hi Sultan,
I recommend reviewing the literature to check how the scale of your interest has been validated. If the scale is originally validated in the factor analysis framework under the assumption that there is an underlying cause of its items, you should use EFA/CFA consistently and decide to exclude the troubling items based on their results.
Best,
Gyeongcheol
- Badges
- Science topic
- Similar topics
- Biological Science
- Physiology
- Physiological Processes
- Stress