Bug: n_factors_f is not being used in FactorAnalyzer model fitting.

[pmdscully]

Bug:

n_factors_f is not forwarded to the FactorAnalyzer modelling process for group factors; and therefore the code always fits for 3 factors only. - Here's an example of code and outputs in colab.

Fix: add 1 line (an extra argument) into the call for group factors, around lines 183-185, as below:

        self.fa_f = FactorAnalyzer(rotation=self.rotation_fa_f,
                                   method=self.method_fa_f,
                                   n_factors=self.n_factors_f, # <--- This line.
                                   is_corr_matrix=True)
        self.fa_f.fit(self.correlations_matrix)
        self.fa_g = FactorAnalyzer(rotation=None,
                                   is_corr_matrix=True,
                                   method=self.method_fa_g,
                                   n_factors=1)

I have recompiled the .pyc and tested this. It should be a straightforward fix.

P.s. - Thanks for your work on this project, I've included your citation in our upcoming report. All the best.

[rafaelvalero]

Thanks let me try locally and update version on pip

[rafaelvalero]

Thanks, new version ready: https://pypi.org/project/reliabiliPy/0.0.36/

[rafaelvalero]

Done

[pmdscully]

Great stuff.
Btw, I noticed that the factor model fitness metrics (indices - i.e. CFI, TLI, RMSEA, GFI, etc.) are not calculated by the FA library (their issue 99 raised this already, but no implementation yet).

As with the Psyche R library implementation for Omega, I think it is important to show underlying model fitness before presenting an Omega result; so I will look at adding those fit indices (similarly to semopy) into the FA library. Then, from reliabiliPy, those functions can be called on the underlying fa models.

It all seems to depend on how easily a 'null-model' (flat) can be implemented and evaluated by FA. Let's see..