A Power of Comparison of Goodness of Fit Tests for Exponential Distribution With Grouped Data

Abstract

This study compares the powers of Goodness of fit tests for grouped data from the exponential distribution. These tests include the Weighted Kolmogorov- Smirnov statistic, Anderson-Darling statistic and Cramer-von Mises statistic. The sample sizes were 30, 50 and 100. These samples were grouped in intervals of 6, 7 and 10. The specified significance level for the tests was 0.05. Each sample of data was generated by simulation and each combination of sample size and interval was run 1,000 times. The simulation studies show that the best test for control the type l error is the Weighted Kolmogorov-Smirnov statistic. Next in rank is Anderson-Darling statistic followed by the Cramer-von Mises statistic. The Weighted Kolmogorov-Smirnov statistic has the highest power of test followed by the Anderson-Darling statistic then the Cramer-von Mises statistic.