錯誤發現率控制之模糊置換法

Abstract

生物醫學研究者進行研究時，經常會遇到大p小n的情境—個案數很少，但測量變數卻很多。我們提出模糊置換法，以解決小樣本情況下的多重檢定問題。此方法引進模糊量至標準置換法中產生隨機p值，而後轉換成q值以控制錯誤發現率。蒙地卡羅模擬顯示，我們的方法在資料是常態或非常態的情況下，都有良好的統計性質。我們分析一筆實際資料，作為範例。大p小n的情境下，我們推薦使用本研究提出的模糊置換法。

Biomedical researchers often encounter the large-p-small-n situations—a great number of variables are measured/recorded for only a few subjects. The authors propose a fuzzy permutation method to address the multiple testing problem for small sample size studies. The method introduces fuzziness into standard permutation analysis to produce randomized p-values, which are then converted into q-values for false discovery rate controls. Monte-Carlo simulations show that the proposed method has desirable statistical properties when the study variables are normally or non-normally distributed. A real dataset is analyzed to illustrate its use. The proposed fuzzy permutation method is recommended for use in the large-p-small-n settings.