We compute the maximum likelihood estimates of a principal component analysis on the logit or probit scalem using a majorization algorithm that
computes a sequence of singular value decompositions. The technique is applied
to 2001 house and senate roll call data and compared with other techniques for
roll call analysis.