Investigation of procedures of test of statistical hypotheses used in quality management problems. Russia. Russia

Statistic distributions and the power of normality tests have been investigated by statistical modeling methods. Advantages and disadvantages of different criteria have been shown. The considered criteria have been compared by power with goodness-of-fit criteria for testing normality. The tables of percentage points has been obtained for Grubbs´s test statistics in case of testing for discards of three maximal (or three minimal) values simultaneously or minimal and maximal values simultaneously in a sample. It has been shown that in testing hypothesis about mathematical expectations the usage of classical results of F-test turns out to be correct even when an observed law essentially differs from the normal distribution. The tables of percentage points of Bartlett´s and Cochran´s statistics, which can be used for correct testing hypotheses when observed laws are described with the exponential distribution family, have been obtained.