Max-imum Likelihood from Incomplete Data