It's unclear to me if alternatives to FA for generating starting values for the latent variables have been considered, but:
I have added PCA for starting value generation, when n=p, and when Factor analysis fails (which my simulations have pointed out, happens very often for Gaussian and gamma distributed responses), as alternative. For large samples at least, PCA seems to perform quite well in generating starting values for the latent variables.
It has been suggested that (PCA/)CA/DCA could be used instead of FA (when n
p as well), though in my experience CA/DCA is inconsistent in generating quality starting values (i.e. FA is superior).
Some minor bugfixes.
It's unclear to me if alternatives to FA for generating starting values for the latent variables have been considered, but: I have added PCA for starting value generation, when n=p, and when Factor analysis fails (which my simulations have pointed out, happens very often for Gaussian and gamma distributed responses), as alternative. For large samples at least, PCA seems to perform quite well in generating starting values for the latent variables.
It has been suggested that (PCA/)CA/DCA could be used instead of FA (when n
p as well), though in my experience CA/DCA is inconsistent in generating quality starting values (i.e. FA is superior).