I've seen somewhere that communalities should be greater than half (0.5) for observed variables in factor analysis. Is that true? Is there any cut off points for this?

  • 3
    $\begingroup$ No cut points, generally. If a variable is "good" for FA its communality should be moderate or above moderate. Small communality says that the variable is hardly driven by common factors. Who wants to keep such a variable? Too high communality may be considered as unrealistic. (If iterations inflate a communality over 1, this diiificulty is called Heywood case, and then it is recommended to reconsider and extract less factors). $\endgroup$ – ttnphns Dec 23 '13 at 13:00
  • 2
    $\begingroup$ +1 to ttnphns. @Ehsan Abd - Remember that exploratory factor analysis is not essentially an hypothesis test but an exercise to see what productive sets of relationships might be uncovered. In practice in the context of survey research, with obtained latent factors explaining 35% - 65% of the variance in the observed variables, I often find that those with commonalities >~.3 are the ones I want to keep in the analysis. For other contexts in which common factors explain much more or much less, this value will be different--.6, .1, who knows. $\endgroup$ – rolando2 Dec 24 '13 at 3:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.