1
$\begingroup$

I know I previously saw a thread on this topic, but I cannot locate it. So, please forgive my redundancy.

I have an 8-item questionnaire administered to a (small) sample of 59 patients. The data consist of 7-day means of ordinal responses on a 0-10 scale (e.g., 0="No pain", 10="Worst Possible Pain"). Reliability is decent: Cronbach's alpha is 0.87. A parallel analysis (using the fa function in R's psych package) suggests that one factor underlies the responses. However, confirmatory factor analysis (using the cfa function in R's lavaan package) suggests that a 1-factor model is inappropriate (CFI=0.687, RMSEA=0.297).

Aside from the small sample size, are there any explanations for this discrepancy? If you know of any journal articles that address this issue, that would be most helpful.

If you think this question is more appropriate on a different Stack Overflow site, please offer your suggestions.

Thanks in advance.

$\endgroup$
1
$\begingroup$

They both may suggest a single factor but the CFA stats you are reporting reflect fit ... Some of those 8 items may not be reflective of the underlying factors as you intended ... Check the loadings and perhaps drop the appropriate items to watch the CFI go up and rmsea go down. Return here to let everyone know how you got on @drpaulburke

$\endgroup$
1
  • $\begingroup$ Thanks for your suggestion. Unfortunately, this is an operational measure, and removing items affects content validity. Do you think model fit would be better if we had a larger sample? $\endgroup$ Mar 3 '20 at 14:14

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.