线性回归分析中的两个悖论（英文）

Two Paradoxes in Linear Regression Analysis

概述:回归是应用统计学中最受欢迎的工具之一。然而,回归分析结果的误用和误解在生物医学研究中是常见的。本文运用统计理论和模拟研究来说明有关这种普遍使用的统计方法的一些悖论。我们还特别指出在顶级医学期刊发表的很多文章中广泛使用的模型选择程序事实上是错误的。模型选择使用哪一种步骤化程序需基于可靠的统计理论。

Summary： Regression is one of the favorite tools in applied statistics. However, misuse and misinterpreta- tion of results from regression analysis are common in biomedical research. In this paper we use statistical theory and simulation studies to clarify some paradoxes around this popular statistical method. In particu- lar, we show that a widely used model selection procedure employed in many publications in top medical journals is wrong. Formal procedures based on solid statistical theory should be used in model select-ion.