RANDOMIZATION TESTS TO COMPARE MEANS WITH UNEQUAL VARIATION

摘要: SUMMARY. A randomization framework for testing significant differences between the means of two or more samples is proposed situation where may be from distribu tions with different variances. This based on concept that observed data arise a random allocation fixed set values to samples, followed by linear transformations are not necessarily same each sample. The null hypothesis that, respect distributions generated allocation, expected sample equal but variances equal. model leads in an obvious way test exact if parameters known. When these known (as usually case) three algorithms approximate tests proposed. properties have been studied simulation comparison Welch's test, usual randomiza tion F-test, and F-test using tables. It has found better than other four when true, uniform normal distributions. None perform well exponential distributions, one superior all under most conditions simulated.