I do my PhD in geriatric care and use factor analysis in two scales with 28 items and 48 items. I have more than 1000 participants (20 participants: 1 item).
My question: what is the best option to remove the items: the communalities or the factor loading. In my data i have some items with communalities less them 0,4 but have factor loading high than 0,3 (for Hair et al 2000) is acceptable for sample with more than 300 participants. If I remove some items I increase the % total of variance (just a few) and reduce the nº of factors (10 to 9 or 8)
I attach in file with same of results What is the best choice?
if i were to weigh on this issue, unless you have a VERY sound theory of why you should keep a 10-dimensional scale, i would completely ignore SPSS’ software default of eigenvalues-greater-than-one and use a much more robust method for factor extraction like Horn’s Parallel Analysis…
just out of pure intuition, with a sample size greater than 1000 i think i a lot of those eigenvalues are more of a sampling artifact than real dimensions within your scale…unless you are really expecting to have 10 dimensions
uhm… so you’re in the process of validating a translated scale, huh?
in that case, i think your method of choice may not be the most appropriate one to answer your research question. this scale as already been developed and it’s already known that 6 dimensions exist within it. what you should be trying here is a confirmatory factor analysis and not an exploratory one, because what you’re trying to do is testing for invariance across cultures/language translation of whatever latent variable that test is measuring.
you have quite a big sample size so you should have no problems running a confirmatory factor analysis or some sort of structural equation model of your choice. but definitely not an exploratory factor analysis (unless you’re using it as some sort of descriptive guidance)