Treatment of outliers in samples of size three

Abstract

A reading that is a long way from most of the others in a series of replicate determinations is called an outlier. A particular procedure for rejecting outliers, and a lso a particular procedure for modifying outliers , are considered for samples of size three, s upposed drawn from a common normal population except that one of the three readings may have an added bias . Numerical results are given illus· trating the effects of the procedures on es timation of the location paramete r. The calculations support a tentative general conclus ion that estimation by least squares shou ld us ually be tempered by suc· cessive application of both a rejec tion rule and a modification rule.