因果推断应用中协变量问题的研究.pdfVIP

摘 要 统计可以根据当前拥有的数据来推断变量间的关系，进而利用统计规律为 决策提供依据。这种统计关系可以分为相关关系和因果关系两种，而推断和利 用因果关系是人类推理和决策的核心。在经济学，医学，教育学等领域中政策 或方案的决策者们都希望在实施前知道政策或方案的效果。而推断因果关系可 以通过预测变量在给定条件下的结果来指导决策行为，故而在现实生活中有着 广泛应用。 本文针对在因果推断中协变量的问题进行了研究并通过实际例子进行了验 证。本文首先基于 Rubin 因果模型来介绍因果推断的基本理论，重点说明了因 果推断的应用条件。其次，本文探讨了协变量的含义、重要性、引入条件，以 及引入协变量后因果效应的估计等基本问题；进一步地，本文还对协变量的选 择方法进行了研究，采用了逐步回归法和偏最小二乘法。最后，本文通过实际 例子对在引入和不引入协变量的情况下因果效应估计进行了比较，结果发现， 引用协变量后可以对因果推断的估计做出更加可靠的评价。 关键词：因果推断；协变量；偏最小二乘；数据挖掘 1 Abstract Statistics can infer the relationship between variables according to the current data, furthermore by use of statistical law we can provide scientific basis for decision-making. This statistical relation can be divided into correlation and causation, while the inference of causal relationship is the core of human reasoning and decision-making. In economics, medicine, education and other areas policy or project decision-makers want to know the effects of policy or project before the implementation. Since the causal inference can guide decision-making behavior by predicting the results of variables in a given condition, they have a wide application in the real life. This paper aiming at the research of the introduction of covariate variables to causal inference and a practical example is given to verify. At first, based on the Rubin Causal Model the basic theory of causal inference is introduced in this paper, focusing on the conditions of the application of covariate variables. Secondly, this paper explores the meanings, importance and the introduction conditions of the covariate variables, the estimate of causal effect