SEM (instead of GSEM) for Logistic Regression

[Yawo Kokuvi]

Hello,

Good Day.

I know that GSEM is the ideal model to use when fitting a CFA/Path Analytic models with binary DVs.

If however, a reviewer wants some modification indices, which GSEM is unable to produce, is it advisable to assume linearity and run the model with SEM, compare the estimates with those GSEM, before drawing conclusions from the modification indices?

Thanks - Yy

[Marcos Almeida]

I hope you get more insightful advice, but I believe a sound strategy would be simply explain what you know, but the reviewer ignores: estat mindices will not apply to gsem.

On top of that, perhaps expanding on a sensitivity analysis would be a nice way to corroborate the model.

[Yawo Kokuvi]

Thanks very much, Marcos. I agree with your suggestion. Could you please shed some more light on sensitivity analysis that could corroborate the model?

cheers, Yy

[Marcos Almeida]

Hello Yawo. That will depend on the model. For example, by using parcels, testing relationships between variables, comparing full gsem model with a "simple" model, etc.