Hello Camila,

The regression method is pretty straightforward and works regardless of structure, and is well suited to studies in which you might wish to evaluate higher-order factor structure.

Bartlett's method has some virtues, but there shouldn't be substantive cross-loadings (e.g., the solution should be congruent with the principle of Thurstone's idealized "simple structure")

Anderson-Rubin method should be used only with orthogonal structures (no oblique rotations).

Good luck with your work.

David Morse thank you! I still struggle to understand which one I should use though for a regression. I use orthogonal rotation. There are no cross-loadings, maybe I should go for Bartlett then? My impression is that Bartlett is the best of the options.

What are the data and what is the exact relationship you are trying to prove?

Koen Van de Moortel I am making a model with PROCESSmacro. There are solely mediators. The data is scale from a survey,

Camila Sneli Mørch That's not really an answer to my question. If you tell me exactly what you are testing, I might have ideas...

Koen Van de Moortel I am testing the order of mediators between X and Y to compare it to my prediction. Additionally, I am testing for what the coefficients are between the mediators to evaluate them. Any advice on how I can be more precise?

Camila Sneli Mørch If you could just tell me concretely what your x and y are: length, time, mass, gender, cost,..., and most importantly: what do you expect the model function to be? Linear, exponential, periodic, whatever... This is most important!