An estimation of Phi divergence and its application in testing normality

Abstract

In this article, a new goodness of fit test for normality is introduced based on Phi divergence. The test statistic is estimated using spacing and the consistency of the test is proved. Then with replacing some special cases of Phi divergence, the efficiency of each test statistic is analyzed by Monte Carlo simulation against some competitors (based on Phi divergence using kernel density function and also some classical competitors). It is shown that each special case of Phi divergence based test is the most powerful in each group of alternatives (depending on symmetry or support).