교육학 연구에서 성향점수를 이용한 인과효과의 추정

Abstract

In estimating the effect of an educational program or policy using observational data, drawing causal inferences is challenging because the treatment and control groups in such studies are almost always not directly comparable. Rubin’s causal model provides a valuable conceptual framework for defining causal effects and clarifying the assumptions that must be met in order to draw sound causal inferences. In conjunction with this framework, propensity score methodology (Rosenbaum & Rubin, 1983a) provides an important tool in efforts to estimate causal effects when working with observational or quasi experimental data. Propensity score approaches provide the basis for forming comparable subgroups given the estimated probability of being assigned to treatment. The purpose of this paper is to introduce Rubin’s causal model (RCM) and various propensity score methods for estimation of causal effects in educational program evaluation, and to provide detailed discussion of the critical issues that should be considered when implementing these methods in educational settings. Data from the Early Academic Outreach Program (EAOP), which was implemented in a large U.S. school district, are used for illustrative purposes.