Hi Jayden,

you are almost correct: The degree of support for a model depends on the degree of restrictions in the model. While I find the CFI not very informative in the first place (and rather would recommend really testing the restrictions of the model with the chi-square test), you are correct that almost-saturated models have less support than highly restrictive models. On the other hand, this is not reason to enforce restrictions that are simply wrong. Hence, analyse which restrictions are implied by the model and simply consider this when evaluating the model.

Having said that, how does the structure actually looks like? What are the df and what does the chi-square test say?

And finally, it is strange that fitting models are judged with more suspicion than the vast majority of the models which show a clear misfit :)

Best,

Holger

You shouldn't have DF=0.

Practically speaking, you shouldn't ve aiming for "just-identified" model (desire is overidentified).

Hope this helps

Matt

The attached image has the model, which has df = 1. Prior to adding the residual covariance between A and B, model fit was quite poor (and df = 2). Modification indices then suggested adding either a covariance between A and B or a covariance between C and D.

After some rumination, I suspect that adding one covariance over the other fundamentally changes the interpretation of F. In the current image, F accounts for much of the variances in C and D. If the other covariance was instead added, F would account for much of the variances in A and B.

I think in a case like this, the model should be scrapped. What do others think?