A generally robust approach for testing hypotheses and setting confidence intervals for effect sizes.

摘要: Standard least squares analysis of variance methods suffer from poor power under arbitrarily small departures normality and fail to control the probability a Type I error when standard assumptions are violated. This article describes framework for robust estimation testing that uses trimmed means with an approximate degrees freedom heteroscedastic statistic independent correlated groups designs in order achieve robustness biasing effects nonnormality heterogeneity. The authors describe nonpara-metric bootstrap methodology can provide improved control. In addition, indicate how researchers set confidence intervals around effect size parameter estimate. online supplement, use several examples illustrate application SAS program implement these statistical methods.