Evaluation Of The Comparative Methods Of Independent Samples

Object Evaluation of the comparative methods of independent samples: Mest, u-test, Satterthwaite method, Cochran-cox method and Wilcoxon rank-sum test, offers some information for statistical analysis in practice. Methods All experiments was simulated by SAS software with different population distribution, different parameters and different sample sizes. Using the type I error, the type II error and test efficiency of five methods as the index of evaluation. Results The t-test is sensitive to unequal population variance, u-test is appropriate to large samples comparative. Satterthwaite method and Cochran-cox method are sensitive to skewed distribution with small sample sizes. When two population's shape are same or two population are L-distribution, the type II error of Wilcoxon rank-sum test is smallest than other tests. The power of Wilcoxon rank-sum test is a little less than that of Mest, Furthermore it is higher than Mest sometimes. Conclusion The experiments express that the power of parametric test is not higher than nonparametric test at any time. According to different situation to select different statistical analysis methods in order to interpret data reasonably in practice. When sample size is large enough, Mest, u-test, Satterthwaite method, Cochran-cox method and Wilcoxon rank-sum test are all recommended. The author give some suggest for sorting data and selecting the comparative methods of independent samples.