Hernán M A
Department of Epidemiology, Harvard School of Public Health, Boston, MA 02115, US.
J Epidemiol Community Health. 2004 Apr;58(4):265-71. doi: 10.1136/jech.2002.006361.
Estimating the causal effect of some exposure on some outcome is the goal of many epidemiological studies. This article reviews a formal definition of causal effect for such studies. For simplicity, the main description is restricted to dichotomous variables and assumes that no random error attributable to sampling variability exists. The appendix provides a discussion of sampling variability and a generalisation of this causal theory. The difference between association and causation is described-the redundant expression "causal effect" is used throughout the article to avoid confusion with a common use of "effect" meaning simply statistical association-and shows why, in theory, randomisation allows the estimation of causal effects without further assumptions. The article concludes with a discussion on the limitations of randomised studies. These limitations are the reason why methods for causal inference from observational data are needed.
估计某些暴露因素对某种结果的因果效应是许多流行病学研究的目标。本文回顾了此类研究中因果效应的正式定义。为简单起见，主要描述限于二分变量，并假设不存在因抽样变异性导致的随机误差。附录讨论了抽样变异性以及该因果理论的推广。文中描述了关联与因果关系的差异——为避免与仅表示统计关联的“效应”的常见用法混淆，全文使用了冗余表述“因果效应”——并说明了为何从理论上讲，随机化无需进一步假设就能估计因果效应。文章最后讨论了随机化研究的局限性。这些局限性正是需要从观察性数据进行因果推断方法的原因。
