corybrunson
commented
2 months ago Thanks @jbgracey6 for the note.
As a reminder to myself, at least two things must be checked here: first, that the internally handled "inertia" is indeed the inertia of the data matrix, not its square (the diagonal matrix of the eigenvalue decomposition); and second, that the conference correctly reflects how it is distributed in the recovered matrix factors.
In R/methods-stats-factanal.r, the definition of recover_conference.factanal() implies that half of the inertia is distributed to the rows and half to the columns, but doing an eigendecomposition of a covariance matrix as factanal() does, we see that the full inertia is distributed to both rows and columns.