I understand that, in general, statistical models must be developed with a sample and validated with another sample. CFA is a kind of validation way for models that have been created using EFA. However, it is not the same as validating prognostic models, for instance. So, my question is, could both methods be used with the same set of data? I am reviewing an article dealing with scale validation that uses both.
Hello Rosa,
Both methods could reasonably be applied to the same "set of data" under these two circumstances:
1. The sample is split, an initial model is generated through EFA, then the held-out portion of the sample is used to run a CFA on the generated model.
2. The researcher elects to run a CFA, finds that it fails to fit in some non-trivial way. The same data set is then used to run an EFA, which yields some proposed model for later confirmation (with new/additional data), which is then offered as a tentative solution.
However, running an EFA on a given data set, then using the very same cases to run a CFA on the "winning" EFA model adds little, if any value. The reason is that doing so greatly increases the chance of overfitting the candidate model. Further, a number of model-data fit indices may be derived and applied to the EFA results, even though many factoring software subprograms do not routinely furnish these indicators (unlike sem/cfa software).
Good luck with your work.
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling