Issues using ATLAS (and possibly other Lapack versions) have been solved. The
problem was a non-unique Q matrix from a QR decomposition (so the results
were not wrong just different).
Functions are added to compute marginal figures post estimation. See
?mmcif_pd_univariate and ?mmcif_pd_cond.