In the modification suggestions the largest one is e20 and e21 (90.00)
When I do this all of my problems are solved. Some researcher say I can and some doesnt agree. I couldnt find any references.
Prof. Mike crowson does the same in youtube but he didnt share any references. I am sharing the figures of what I am planning to do.
Yes, it can make perfect sense for factor residuals pertaining to endogenous factors to be correlated. That would be the case if F3 and F4 shared variance with one another after partialling out the exogeneous factors F1 and F2. In other words, F3 and F4 may still be linearly related after their common causes F1 and F2 are taken into account (e.g., there may be additional but omitted common causes of F3 and F4 other than F1 and F2).
However, I would not add a residual association only because it is suggested by modification indices. There should also be a theoretical/substantive rationale for adding the residual correlation. A modification index of 90 is large though and does suggest that the covariance between F3 and F4 is currently not fully accounted for in the model.
How large is the error correlation between the two factors in your revised model?
Christian Geiser Thank you very much for the reply. Unfortunately, this scale in the middle (F3+F4) is developed by myself. So, on this occasion, there is no theoretical background. The reason for this is that the concept of Achilles syndrome hasn't been studied since 1995 and that only study was conceptual. So I named the factors according to that book.
On the other hand, I can explain why the factors are correlated but I am having problems with the reference for correlating error terms. Do you know any study which made modifications between error terms or suggests it?
David L Morgan Thank you for the reply. I am not sure but is this what you were asking for?
This is a possible reference for correlating error terms: Hermida, R. (2015). The problem of allowing correlated errors in structural equation modeling: concerns and considerations. Computational Methods in Social Sciences, 3(1), 5-17.
Here is a very good textbook: Brown, T. A. (2015). Confirmatory factor analysis for applied research. Guilford publications.
Have you considered testing a model where your current two factors treated as one single factors?
Is it possible to use my scale as a single factor? Because in EFA while I was developing the scale I found 2 dimensions. So, I thought I must use it as two factors as found.
But on the other hand, these two factors are not like positive or negative features of the concept. They are actually quite related.
David L Morgan
Michael Fitzgerald Unfortunately Hermida explains why not to correlate error terms
- Badges
- Science method
- Similar topics
- Methodology
- Research Methodology
- Qualitative Research