<![CDATA[The normal distribution is an important assumption for many statistical methods. The t-distribution is similar to the normal distribution, but there are differences in variance and free degree depending on the sample size. Normality testing usually uses Kolmogorov-Smirnov, Anderson-Darling, and Shapiro-Wilk tests. Simulating was performed on data derived from normal distributions, t-distributions, and exponential distributions. In the N(10,2) generation data, it was found that the Shapiro-Wilk method was better than other methods, while in a large sample, it was found that the Anderson-Darling method was better than the other methods. The data for the generation of the distribution of t(1) shows that the cumulative value of rejecting is close to 100%, meaning the data is not normally distributed. In a near-normal t(20) distribution, but not a normal data gain, Anderson-Darling and Shapiro-Wilk test performance showed good results. Meanwhile, in the t(100) generation data, the result was obtained that the most consistent cumulative value was the Anderson-Darling method. Furthermore, the Exp(1) data generator produces a cumulative value of rejecting close to 100%, with the most consistent method being Kolmogorov-Smirnov. So it can be known that the normality test selection depends on the number of samples and the data distribution.]]>
Copyrights © 2023
