GRAPHICAL METHOD ON AN OMUNIBUS TEST FOR NORMALITY

Abstract

We give a new graphical method to test for normality based on an approximate joint probability density function of the sample skewness and kurtosis. This approximation is initiated by Shenton & Bowman (1977). For Fisher's iris data, our examples are shown in comparison with qqplot and Shapiro-Wilk test. Computer simulations are conducted as follows : Type one errors are evaluated in order to show the validity of the proposed reject region. Powers are calculated against the samples drawn from contaminated normal distributions. We show our method has the same power as Shapiro-Wilk test excluding the case of some parameters. Therefore, our method is advantageous to make graphical interpretation in testing for normality.