【摘要】 本文主要讨论流行病学观察性研究中几种常用的假定，介绍Simpson's悖论和Lord's悖论，逆回归问题，以及在不同假定下可能出现的相互矛盾的结果。探讨流行病学研究的因果推断中处理分配机制的假定，暴露作用评价中对照组的模型假定，以及不完全数据分析中缺失数据机制的假定。

    【关键词】 流行病学

    Several untestable assumptions in epidemiological observational studies:control models and mechanism of missing data GENG Zhi. School of Mathematical Sciences，Peking University，Beijing 100871，China.

    【Abstract】 This paper discussed several untestable assumptions which were often used in epidemiologi-cal observational studies，and some paradoxical results that may arise under different assumptions were showed.Simpson's paradox and Lord's paradox were introduced.Assumption of treatment assignment in causal inference of epidemiological studies，assumption of control group in evaluation of exposure ef-fects，and assumption of missing data mechanisms in incomplete data analysis were also discussed.

    【Key words】 Epidemiology (Chin J Dis Control Prev2006，10(1):1-4)

    流行病学的观察性研究中 ......