The Evaluation And Comparisons Of The Normal Test Statistics Based On Confidence Bands Theory

In this paper,based on the construction theory and method of the confidence bands,we pay attention to when the distribution function of random variable interval is limited,the test of normality effect how to change.Under the given amount limit range,based on the different types of twelve non-normal distribution simulated calculation of power function,comprehensively compared to normal inspection.In this paper,seven normality test statistics are discussed including five graphical tests:D,D_M,D_e,D_bi,D_be and two non-graphical tests:AD,SW tests.According to the characteristics of the test statistics respectively design limited the symmetric and asymmetric intervals,and under three confidence level for seven normality test are compared.Finally we concluded that the limit of the interval has an influence on the normality test,and the test of normality becomes worse as the interval of the random variable becomes smaller;For seven normal tests,compared with the other four graphics test,D in most cases the power is the worst,and for the distribution of random variables on the asymmetric interval D_bi’s is the best,but for the distribution of random variables on（-∞,+∞）symmetrical interval,D_e’s effect is better;With the increase of the confidence level,normality tests effect become good.