多變量分類的平均數與比

Abstract

學界對於將連續型變數分類(categorization)後，所造成之問題爭議已久。但因其在結果呈現解釋上之方便明瞭，此方法在流行病學領域仍舊普遍。為了修正這些不可信賴的統計方法，我們將 Huang (2013)的研究從單變數延伸到多變數分類。我們探討了母體分類平均數(population categorized means)與分類比(categorized proportions)在有母數及無母數兩種情況下之估計量及其統計理論。而模擬結果更顯示這些估計量有利於建立有效推論，值得未來受到關注及進一步的研究。

Although the categorization of continuous variables have been criticized for a long time, it is still popular for its convenience in presentation of final results. To improve these untrustworthy statistical methods, we extend the univariate approach of Huang (2013), and we focus on developing suitable methods do deal with multivariate categorized variables in order to come up with meaningful evidence of association between continuous variables. Population categorized means and categorized proportions are studied with both parametric and nonparametric estimations. The simulation results show that both estimations perform well, and they may be used to construct efficient inferences. These well-performed estimations deserve attention and further investigation in the future.