Data Analysis with R, Second Edition - Second Edition

Pairwise deletion

Also called available-case analysis, this technique is (somewhat unfortunately) common when estimating covariance or correlation matrices. For each pair of variables, it only uses the cases that are non-missing for both. This often means that the number of elements used will vary from cell to cell of the covariance/correlation matrices. This can result in absurd correlation coefficients that are above one, making the resulting matrices largely useless to methodologies that depend on them.