i want to check factorial and predictive validity of a 24 item questionnaire. For the factorial validity i calculated Measurement Models. Subsequently i wanted to calculate SEMs for predicitve validity.
The answers are on a 5-point Likert skala and pretty skewed. So i calculated with MLR. Sample size is 250. There were a few factorsolutions with 3 to 5 latent factors proposed by earlier EFAs. Testing those in Measueremt Models with my sample results in an unacceptable fit. The best solution has a RMSEA of .076 and a CFI of .817.
Deleting some items due to low factor loadings or setting factor loading to another latent factor after modification indices can better fit upto Rmsea=.69 / CFI=.90. In this case i still made only changes which still seam senceful because of the content or sample characteristics. Unfortunately this solution leads to unidentified models when going over to SEMs.
There is no senceful solution with a good fit. So SEMs are prohibited. Is there still a possibility to calculate some kind of predictive valdity? Maybe after an EFA or with path models?
Thanks for your recommendations!