A Simulation Study to Assess the Performance of Three Normality Tests

Abstract

In this paper, Anderson-Darling (AD), Lilliefors (LI), and Jarque-Bera (JB) tests were compared for Type I error and for power of the tests. The simulation was run 100,000 times for different situations and for different types of departures from normality. For all different sample sizes and distributions, AD gave the most powerful results, followed by the JB test.