| HUANG Wei,XIANG Keyan,YUAN Luoqi.Empirical Guide to Regression Discontinuity Design: Operational Specifications, Misapplications and Practical Extensions[J].The Journal of quantitative and technical economics,2025,(6):111-131 |
| 断点回归设计的实证指南：操作规范、应用误区与实践拓展 |
| Empirical Guide to Regression Discontinuity Design: Operational Specifications, Misapplications and Practical Extensions |
| |
| DOI： |
| 中文关键词: 断点回归设计 连续性框架 处理效应 |
| 英文关键词: Regression Discontinuity Design Continuity Framework Treatment Effect |
| 基金项目: |
|
| Hits: |
| Download times: |
| 中文摘要: |
| 断点回归设计作为因果推断的重要工具,在实证研究中得到了日益广泛的应用,相关理论研究也取得了显著进展。本文基于模拟数据与实际案例,系统讨论了断点回归设计的实现过程。首先,本文深入剖析了断点回归设计的理论基础与识别假设,并详细阐述了其实现过程中的关键环节,包括断点回归图的绘制、局部多项式函数设定、最优带宽选择、统计推断与有效性检验等核心步骤。其次,结合国际前沿文献,本文探讨了断点回归设计框架下的机制分析方法,并对时间断点回归与断点回归-双重差分等拓展方法进行了系统性评述。此外,本文深入分析了溢出效应对模糊断点回归中排他性假设的挑战,并提出了相应的改进策略。通过本文的讨论,研究者可以更好地理解断点回归设计方法的应用边界,规避常见的研究陷阱,从而提升因果推断结果的可靠性与准确性。 |
| 英文摘要: |
| Regression discontinuity design (RDD) is an essential tool for causal inference, and it is increasingly applied in empirical research and complemented by significant theoretical advancements in recent years. However, practical guidance on navigating common technical challenges and potential pitfalls remains relatively scarce in the existing Chinese literature. This study addresses this gap by bridging theoretical foundations with concrete implementation strategies. By integrating theoretical insights, illustrative examples, and detailed operational guidance, we offer researchers a clear, systematic roadmap for the effective application of RDD in applied research.We begin by outlining the theoretical framework of RDD in Section 2. Sharp RDD assumes perfect compliance with the treatment assignment, while fuzzy RDD addresses imperfect compliance through a local instrumental variable approach. We also highlight an emerging literature that explores conditions under which local treatment effects identified at the cutoff can be generalized more broadly.Transitioning from theory to practice, Section 3 presents a clear roadmap for the empirical implementation of sharp RDD. First, we discuss constructing effective RDD plots to visualize discontinuities, briefly covering practical aspects such as bin width, bin spacing, fit lines, axis scaling, and vertical lines. Next, we address essential choices regarding estimation methods and statistical inference. We advocate using local polynomial regression and emphasize careful selection of bandwidth, polynomial order, and kernel function to optimally balance bias and variance. Finally, we summarize key validity tests to indirectly assess continuity and local randomization assumptions, including density and placebo tests. We also highlight the donut RDD as a practical method to detect and mitigate data-heaping issues near the cutoff, strengthening the robustness of causal estimates.In Section 4, we review recent methodological advances in RDD. First, we highlight how recent studies use RDD not only to estimate treatment effects but also to identify and quantify causal mechanisms. We then discuss two extensions—regression discontinuity in time (RDiT) and regression discontinuity difference-in-differences (RD-DD). RDiT utilizes time as the running variable, making it particularly useful for evaluating policies activated at specific periods. When using RDiT, researchers must navigate some challenges. RD-DD integrates the strengths of RDD’s cross-sectional comparisons and DID’s temporal contrasts, offering enhanced causal identification when conventional assumptions may not hold. We illustrate these approaches with practical empirical examples and outline key estimation considerations. Finally, we address the challenge of spillover effects in fuzzy RDD, clarifying how such effects violate the exclusion restriction and bias local average treatment effect estimates downward. We recommend employing a sharp RDD to estimate intention-to-treat effects clearly and accurately under these conditions.This study combines theoretical foundations with a strong focus on practical implementation, offering a structured and accessible guide for applying RDD in empirical research. We emphasize operational procedures and demonstrate them using simulated datasets, helping researchers understand each step of the estimation process. A distinctive feature of this study is the attention to common pitfalls in empirical practice, including issues such as bandwidth selection, incorrect modeling choices, and misinterpretation under fuzzy designs. To further support applied work, we incorporate examples from recent applied literature to illustrate how RDD methods are used at the research frontier. Finally, we introduce key methodological extensions, such as RD-DID and RDiT, and discuss emerging topics such as external validity and spillover effects, providing readers with a well-rounded understanding of both foundational methods and ongoing developments in the RDD literature. |
|
View Full Text
View/Add Comment Download reader |
| Close |
|
|
|
