Important notesΒΆ

  1. It is possible that factor_analyzer may return the loading for a factor that has all negative entries whereas SPSS/R may return the same loading with all positive entries. This is not a bug. This can happen if the eigenvalue decomposition returns an eigenvector with all negative entries, which is not unusual since if \(v\) is an eigenvector, then so is \(\alpha * v\), where \(\alpha\) is any scalar (\(\ne 0\)). Additionally, signs on factor loadings are also kind of meaningless because all they do is flip the (already arbitrary) interpretation of the latent factor. For more details, please refer to this Github issue.

  2. When using equamax rotation, you must compute the correct value of \(\kappa\) yourself and pass it using the rotation_kwargs argument. This is different from SPSS which computes the value of \(\kappa\) internally. For more details, please refer to this Github issue.