For my undergraduate dissertation I'm analysing a 75-item self-report questionnaire filled out by 1140 autistic participants (randomly split into two groups. I have split the sample and am carrying out a PCA on one sample, and a CFA on the other. Previous Rasch analyses have indicated that this questionnaire is meant to be unidimensional, but my one-factor model is coming back with terrible fit indices. I've tried multiple different models by now, including a 2-factor, 3-factor, 4-factor model (always using oblimin rotation, always removing items with cross-loadings, but varying different cut-offs/ inclusion criteria- for example removing items which don't correlate highly with other items).

None of these models have returned with good fits. I'm okay with this, I know it's not good to keep running the analysis until I get something, and I know there are many issues with the exploratory nature of this analysis and the very simple tests I'm using.

But I'm wondering if I'm doing something wrong, or if there's something obviously wrong that will cost me marks on my dissertation. Is this an appropriate analysis for a 75-item questionnaire? Or should I be trying to do something with parcelling or correlated error in SEM?

The basic code I'm using is below.

Thank you

one.factor.model <-   'Factor1 =~ Q11 + Q29 + Q14 + Q65 + Q72 + Q52 + Q68 + Q55 + Q2 + Q5 + Q28 + Q62 + Q22 + Q75'
one.factor.fit <- cfa(one.factor.one, data= half2)
summary(one.factor.fit, fit.measures=TRUE)

1 Answer 1


You're not doing anything wrong.

Scales which have been developed using things like exploratory factor analysis never fit well in new samples, especially if they have 75 items.

75 items is far too many to get clean single dimensional fit. Why are so many items needed?

Also, don't do PCA, do EFA.

  • $\begingroup$ Thanks for your answer. All the 75 items aren't really needed. Ideally, I want to create a more unidimensional questionnaire. I think my next move will be to use EFA to reduce the items down to a smaller questionnaire, and then use CFA on a separate sample to confirm the new unidimensional model. I just have one question: if I find, for example, 45 questions that form a strong unidimensional first factor, and I want to use CFA to test it. Would my CFA sample only include those 45 questions? Or would it include all 75? $\endgroup$
    – T August
    Apr 14, 2020 at 16:11
  • $\begingroup$ 45. But that's still way too many (IMHO). $\endgroup$ Apr 14, 2020 at 21:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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