On the power of the extreme spread

Abstract

A variety of techniques are available for testing whether a random sample might reasonably be assumed to have come from a normal population. If, however, the random sample is a bivariate random sample, the number of choices is greatly reduced. The purpose of this note is to suggest a new statistic for testing the hypothesis of bivariate normality in the case of known σ which is easily computed and whose power seems to be far superior to the classical chi-square test, and in addition, may be used with small sample sizes, where the chi-square test is inappropriate