Hi France

have you thought about testing some other measurement models rather than adding correlated errors or cross loadings guided by MIs. If there are lots of MI's this suggests the model is not adequate to begin with, and adding these additional parameters to improve model fit does not result in a 'good model'. For example, if you have a one factor model with correlated errors it's not correct to say that you have a unidimensional model. What you have is a model with one latent variable and other latent variables disguised as correlated errors.

Constraining 2 factor loadings may improve the fit of the model, but without justification this is just another statistical fit.

Sorry I can't be more helpful

Mark

Run a reliability test after dropping the newly constrained item. If the alpha value improves, remove that item from the measurement model.

Best,

A

I don't think 2 paths constraining is the right thing to do. To get some perspective on why we constrain, you can refer to Byrne's book on SEM using AMOS. Idea is that one path per construct is enough.

I think you may need to review model specification. Use MIs to do that, rather than delete items or correlate them unnecessarily. Is sample size adequate?

Hi,

I would totally second Mark's advise. Using MI's is almost never a good idea as the usefulness of the MI require that the overall structure is correctly specified and the model misfit is only due to omitted paths. However, when a measurement model does not fit it is also likely that the structure itself is the problem and a more specific / diversified structure is the correct one.

Hence, take (again) a look onto the question wording and re-consider your intuitive measurement theory: Which of these items can really be assumed to be caused by an underlying latent and which of the items presumably measure something else. Just following "scale"-label of a set of items almost always leads to a misspecified model. Scales are often causally heterogeneous.

Best,

Holger

Dear Mark, thank you for your good explanation. You seem not to support the idea of constraining two factors though it is possible. But i didnt get you well when you said "it is just another statistical fit"

Dear Muhammad. Thank you for your advice. I will consult the Byrne' s book and will come back to you in case of any further enquiry.

Dear @Ali Mahmood, thank you for such a helpful advice. I will work it out and see whether it will help improve my model.

Rgds

France