A multivariate correction for attenuation

Abstract

A method is given for the joint correction for attenuation of all pairwise correlations among p variates, subject to the restriotion that the resulting correlation matrix be at least positive-semidefinite. The errors need not be assumed uncorrelated, but their covariance matrix must be positive-definite. The procedure is shown to yield maximum likelihood estimates of ‘true’ and ‘error’ component covariance matrices under the restriction.