I have a repeated measures design for a Likert scale with 10 items. This scale is assumed to be unidimensional and this is one of the research questions along with the scales measurement invariance (MI) over time.

I conduct the analysis as follows: First assessment is randomly split into two pieces, I apply PCA in the one half, CFA in the second half and then I wish to make MI comparisons between the second half of the first assessment and the second assessment.

PCA is supporting a one factor construct but not very strongly (λ = 2.7, Prop: 73%) while CFA is problematic to provide one factor structure (c2/df > 3, RMSEA = 0.12, GFI = 0.87).

Under that circumstances does it make sense to run and provide also the necessary MI comparisons between the two assessments?

  • $\begingroup$ Regarding the CFA, might your problem be solved by dropping a problematic item? $\endgroup$ Jul 11, 2013 at 14:36
  • $\begingroup$ It is an idea, however, I do not feel very comfortable to do so since my purpose is to validate the original 10 - item instrument in Greek. $\endgroup$ Jul 12, 2013 at 5:43
  • $\begingroup$ Sure. That makes sense. Did you find a single factor at time two? $\endgroup$ Jul 15, 2013 at 14:51
  • $\begingroup$ MI - measurement invariance? (It could also be modification index, which cna be used in this context. $\endgroup$ Sep 27, 2013 at 14:26
  • $\begingroup$ No it doesn't. Measurement invariance tests assume you have a fitting model to start wth, and you don't. $\endgroup$ Sep 27, 2013 at 14:27


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.