The effect of influential outliers on parameter estimation in regression analysis.

A comment is made on an article recently published in this journal by Borgeat, Elie, and Castonguay (1991). It is noted that the raw data contained an outlier which is shown to have had a large influence on their estimation of the regression coefficient. Analysis of the data using a nonparametric statistic that is optimal for long-tailed distributions showed that the true regression coefficient is probably smaller than that reported. Other approaches to the analysis of data containing influential outliers are discussed.