Robust Detection of Outlines in Univariate Data with Skewed Distribution

Abstract

This research aimed at investigating the problem of outlier detection in univariate data when the distribution is skewed to the right . A new nonparametric method is proposed that involves a more conservative rule of detection. The performance of the suggested method is compared through a simulation experiment, to that of four widely used nonparametric methods. The comparison is made under three distributions , namely the chi- square , beta and log normal distribution. Small, medium and large sample size are used in the experiment. The result of the simulation, confcrmed the superiority of the Proposed method to the other methods when testing for outliers in the right tail of the distribution . This superiority increase with increase in sample size. The Performance of the method in the left tail is relatively poor.