Handle local minima problem in FA/PC rotations

[bwiernik]

All of the GPARotations rotations have pretty severe problems with local minima. This can lead to erroneous rotation solutions. This is a problem in psych implementation, because it only rotates the factor solution once and returns that result. By contrast, the fungible::faMain() function rotates from random starting values many times (100 by default) and then returns the "global" minimum across that set. We should adopt a similar approach.

[DominiqueMakowski]

That's very interesting thanks, I wasn't aware of that issue (and I'm quite a big user of EFAs 🙈)

Do you think we should replace the call to psych here to fungible?

parameters/R/factor_analysis.R

Line 60 in 900f234

psych::fa(

[bwiernik]

Not sure. What are all of the statistics you extract from the psych fa object?

[DominiqueMakowski]

here's the method of the parameters extraction. I think indeed the first step would be to provide a model_parameters() for fungible::faMain() objects

[bwiernik]

If it’s just the rotated loadings you use from psych, then indeed fungible provides a good alternative. If you are also extracting fit information, etc., I’m not sure if fungible does that.