As a result of consecutive eight modification, fit indexes were acceptable. Modifications were made in variables under the same factor. Is there a limit for modification in CFA? Thanks.
Hi Ali,
once you observed a misfit to the data (optimally by taking the chi-square test seriously), you leaving the "confirmation road" and move towards exploration. If you did not blindly follow the modification indices (which you should not do) but used a combination of theoretical reasoning and inspection of local misfit (d-separation-failures, standardized residuals) the chances are higher to get some solid results. Nonetheless, you are running the chance of a) capitilization on chance (i.e., finding some data-specific, non-replicable result) and b) fitting a still wrong or even worse model to the data. The first problem can be and should be ruled out by a replication but the first (and this is more problematic) can only be solved by enriching the model. Don't move into the trap of thinking that mere replication will "validate" the modified model. Wrong models (beyond capitilization on chance) will replicate well despite being wrong.
With regard to your question: Of course, some slight changes will matter less than tinkering until the cow comes home :)
For more info about this "enrichment" issue see this thread your (in which you will find many of this cross-validation-recommendations which I don't think are of much use with regard to point "b" above.
https://www.researchgate.net/post/Can_I_use_EFA_before_SEM_analysis
Best,
Holger
Too many modifications will affect the integrity of the work.
- Badges
- Science topic
- Similar topics
- Indexes