Comparisons of Confidence Interval for a Ratio of Non-normal Variances Using a Kurtosis Estimator

Abstract

In this study, the author investigated a confidence interval approximation for a ratio of variances in bivariate non-normal distributions by Bonett [2], of which the first model was performed to replace a trimmed mean by a median in estimating kurtosis while the second model was carried out with unbiased kurtosis estimators proposed by Brent [4] in a confidence interval approximation for a ratio of variances in bivariate Non normal distribution. When the third and forth model were performed to modify the kurtosis estimators for a mean proposed by Brent [4] with a median and a trimmed mean respectively, the fifth, sixth and seventh were also applied to modify the pool kurtosis introduced by Layard [6] in order with a median and a trimmed mean in a confidence interval approximation procedure. As a result, the efficiency comparison determined by coverage probabilities was implemented to test the performing models.