학술논문
베이즈망에 토대한 인과 모형 이론의 문제로서 심슨 역설: 충실성 조건에 대한 이영의 선생님의 비판과 이에 대한 응답
Simpson Paradox, Bayes Nets and Causal Modeling: On Young E. Rhee`s Criticism of Faithful Condition
Simpson Paradox, Bayes Nets and Causal Modeling: On Young E. Rhee`s Criticism of Faithful Condition
Document Type
Article
Author
Source
과학철학 / The Korean Journal for the Philosophy of Science. Nov 30, 2016 19(3):109
Subject
Language
Korean
ISSN
1598-754x
Abstract
이영의 선생님(2015)은, 심슨 역설이 대표적인 인과(구조) 모형 이론인 SGS(Spirtes, Glymour, Scheins 1993) 이론의 충실성 조건에 반례가 된다는 문제를 논의한다. 이 반례는 베이즈망에 토대한 베이즈 인과 이론에도 문제가 된다. 이영의 선생님의 주장은 카트라이트(Cartwright 2001[2003]) 등 SGS 이론에 비판적인 입장에 일부 함께 하고 있다. 필자는, 켈리(Kelly 2007; 2008; 2010; 2012)가 발전시킨 오캄의 단순성(경제성) 원리에 토대하여, 인과 추론의 방법론적 관점에서 SGS의 충실성 조건이 의미를 가질 수 있다는 것을 보여준다. 특별히 예측(발견)과 설명(정당화, 해명)의 구분을 통해 SGS 이론은 예측의 효율성에 초점을 두고 있으며, 충실성 조건도 그 맥락에서 유의미하다는 것을 보여준다. 이 논의는 SGS 이론과 이 이론에 대한 비판적 입장 모두를 함께 양립하게 할 수 있을 것이다.
In his outstanding book on Bayesianism, Young E. Rhee(2015) critically examines Bayes nets in terms of Simpson`s paradox that refutes the Faithfulness condition essential to SGS`s theory(Spirtes, Glymour, Scheins 1993). His criticism of the Faithfulness condition is in some ways aligned with the opponents of SGS`s theory such as Cartwright(2001[2003]). In this paper, I show that the Faithfulness condition of SGS`s theory has the methodological significance for causal inferences in non-experimental circumstances, while considering Ockham`s principle of simplicity, or economy Kelly(2007; 2008; 2010; 2012) has considerably developed. I show that SGS`s theory primarily concerns the efficiency of prediction, while noticing the distinction between prediction (discovery) and explanation (justification). I argue that, from the viewpoint of prediction, the Faithfulness condition is viable and is worth noticing. I expect my discussions on SGS`s theory to reconcile SGS`s theory and the opponents of it.
In his outstanding book on Bayesianism, Young E. Rhee(2015) critically examines Bayes nets in terms of Simpson`s paradox that refutes the Faithfulness condition essential to SGS`s theory(Spirtes, Glymour, Scheins 1993). His criticism of the Faithfulness condition is in some ways aligned with the opponents of SGS`s theory such as Cartwright(2001[2003]). In this paper, I show that the Faithfulness condition of SGS`s theory has the methodological significance for causal inferences in non-experimental circumstances, while considering Ockham`s principle of simplicity, or economy Kelly(2007; 2008; 2010; 2012) has considerably developed. I show that SGS`s theory primarily concerns the efficiency of prediction, while noticing the distinction between prediction (discovery) and explanation (justification). I argue that, from the viewpoint of prediction, the Faithfulness condition is viable and is worth noticing. I expect my discussions on SGS`s theory to reconcile SGS`s theory and the opponents of it.