Evaluasi Kinerja Uji Normalitas pada Ragam Distribusi dan Ukuran Sampel

Abstract

The normal distribution is a fundamental assumption in many parametric statistical methods. Therefore, testing for data normality is a crucial step prior to further analysis. This study aims to evaluate the performance of three widely used normality test methods: Kolmogorov-Smirnov (KS), Anderson-Darling (AD), and Shapiro-Wilk (SW), across various distributions (standard normal, exponential, and t-student with degrees of freedom 1, 20, and 100) and sample sizes (n = 20, 50, 100, 200, and 500). Data were generated through simulation with 1000 iterations for each combination. The results show that the KS method performs well on standard normal and t-student distributions with larger degrees of freedom. The AD method proves to be more sensitive, especially in detecting deviations from normality, though it is less stable for small sample sizes. Meanwhile, the SW method demonstrates optimal performance with large samples. These findings provide practical guidance in selecting appropriate normality test methods based on the characteristics of the data.