I was trying to do a test of reliability for my survey items. In addition to Cronbach's alpha I'm looking at communalities. My criteria is that survey items with communality below 0.4 will be dropped. But when I looked at my communality table, I saw that some items had .99 for communality. Is this problematic? What should I do with these?
With a sample size of 104, any factor analysis is going to be shaky at best. The best approach is probably to collect more data (not really that useful an answer, but its true). This page gives some useful advice.
Fabrigar et al (1999) indicate that large communalities can often lower the sample size required, but almost 1 is probably way too high. I would drop the offending items and re-run the analysis (using ML, principal axis and another method of your choice) and see what the results are. If they still produce heywood cases, then FA is probably not the right approach.