MICRO-COMPUTER PROGRAM FOR STATISTICAL POWER

点击翻译按钮获取中文摘要

Traditional tests of statistical significance focus on whether or not the null hypothesis is viable. This approach leads researchers to confuse statistical significance with substantive significance, and to misinterpret the absence of a significant effect as evidence that no effect exists. Confidence intervals by contrast focus attention on the magnitude of the effect, and as a separate issue on the precision with which that effect is estimated. We are developing a series of computer programs that will enable the researcher to plan and interpret studies within the framework of confidence intervals. The program will enable the researcher to report the magnitude of an effect in one format and then have the defect displayed in multiple formats (odds ratio, attributable risk, proportion variance explained, etc), which will allow for a clear picture of the substantive value of the effect. The program will display the confidence intervals about the effect, and how these vary as a function of sample size. This will enable the researcher planning a study to ensure that the precision of the planned study will be appropriately high, and the researcher analyzing a completed study to present the results comprehensively.