A Simple Univariate Outlier Identification Procedure Designed for Large Samples

Abstract

A simple univariate outlier identification procedure is presented for the detection of multiple outliers in large and moderate sized data sets. This procedure is a modification of the well-known boxplot outlier-labeling rule. Critical values are easy to obtain for the large sample case for a variety of useful distributions, including the normal, t, gamma, and Weibull. Simple adjustment formulas and graphs are provided for handling smaller samples. Basic probability properties are obtained mathematically and through simulation. Two data sets illustrate the procedure's application as a simple and effective screening tool for both moderate and large-sized univariate samples.