The Regression Discontinuity Design (RDD) is one of the emerging methodology for recent policy impact assessment area. But based on the unique nature of RDD, its assumptions for practical application and points to remember for the result interpretation are really different from other policy impact assessment methodologies. For now, such points are not well understood by the users and some case seems to be inadequate and inappropriate as a policy impact assessment.
This paper aims to explain assumptions, key points of result interpretation and limitations for practical applications of RDD based on related major preceding papers.
First, this paper shows that most of the limitations comes from the nature of RDD that it is a kind of cross section analysis and estimates local average treatment effect. Secondly, due to these natures of RDD and its assumptions, RDD has vulnerability for the relational index data, vulnerability for partially missing and concealed data, vulnerability for dependent and internal treatment selection, vulnerability for unequal distribution of "compliers" and vulnerability for stable unit treatment value related problems. Those vulnerabilities requires special care for the result interpretation. Thirdly, this paper introduces some measures to close those issues and recent new approaches to fix the problems.
This paper aims to contribute to the development and fair application of RDD as a policy impact assessment methodology through above discussion.