An alternative to the rejection of observations

Abstract
A method is given for dealing with sets of observations, appreciable fractions of which are affected by abnormal errors.