Always Applicable Sequential Randomization Tests for One-Way Anova That Emphasize the More Recent Data

摘要

The data are independent observations which, under the null hypothesis, have the same unknown (arbitrary) distribution. Observations occur in sets of given sizes, with a maximum total number available. An overall test is a succession of subtests with significance when at least one subtest is significant. All observations are re-used until a stated totality occurs. Then, a specified number of these observations are chosen by randomization (all possibilities equally likely). Data for re-use now consist of the observations chosen by randomization and newly obtained observations. New observations are taken until the totality for re-use reaches a given value. Then, a specified number of these observations are chosen by randomization and constitute the data for re-use at this stage. Further new observations are taken, etc. Exact significance levels are obtainable, by use of appropriate randomization models and special kinds of subtest statistics. Subtests are such that the significance level of a new subtest is independent of prior subtest results. The overall test terminates when a significant subtest occurs (thus saving time and expense). Some applications to quality control are outlined. (Author)