If the latent construct is derived from a survey, can I correlate error terms by checking the questions to see whether the two questions are related to the same topic? If they are, I correalte them.
What other method or confirmation should I do?
Any other method if I want to improve my model fit?
Thank you all kindly for giving comments in advance!
You certainly can do, and the modification index will point to changes that can create a significant improvement in fit.
The problem, however, is that you may end up "over fitting" your data by introducing small effects that are unique to your specific sample. One way to check for this, if you have a large enough data set, is to divide your overall sample in half and develop the best fit for the first half, and then compare to the other half.
- Badges
- Science topic
- Similar topics
- Indexes