Outliers sensitivity in EDF goodness-of-fit statistics: an exploratory analysis

Abstract

This paper deals with the study of the behavior of EDF goodness-of-fit tests in the presence of outliers. The approach is based on the use of data analytic techniques for analyzing changes in the values of the test statistics caused by extremes observations and groupings in finite dimensional samples. After developing systematic criteria for defining sets of ideal observations for the stylized representation of continuous distributions (profiles), the behavior of each test statistic is considered when introducing some controlled variations into these samples. Stylized sensitivity curves and resistance to rejection are obtained for EDF test statistica and used to compare their attitudes towards outlying and grouped observations.