An efficient algorithm for computing covariance matrices from data with missing values
Laszlo Engelman · Communications in Statistics - Simulation and Computation · 1982
An algorithm for computing covariance and correlation matrices from data with missing values is presented. In terms of the number of operations performed (hence CPU time used) this algorithm is more efficient than that used by most statistical computing packages. CPU time efficiency is attained without undue increase in the number of input/output operations or memory space requirements.