Recent Perspectives on the Regression Discontinuity Design
The regression discontinuity design was originally proposed in 1960 as a powerful alternative to randomized experiments. It has been little used since. Over the past decade, however, the design has been increasingly and successfully employed by economists in a variety of studies. In this paper, the fundamentals of the regression-discontinuity are discussed. Recent advances are emphasized.
Even in these early formulations, the design was simple and powerful. However, there were few applications and apparently only four published studies with significant crime and justice content (Berk and Rauma 1983; Berk and de Leeuw 1999; Chen and Shapiro 2007; Berk et al. 2010). Over the past 15 years, a number of economists have extended the design (Imbens and Lemieux 2008b; Imbens and Kalyanaraman 2009) and applied it in a wide variety of settings (Imbens and Lemieux 2008a; Lee and Lemieux 2009). An account of how and why interest in the regression discontinuity design has varied over the years can be found in recent paper by Thomas Cook (Cook 2008).
In this chapter, the fundamentals of the regression discontinuity design are considered. Some recent advances are highlighted. The discussion begins with brief introduction to the ways in which statisticians think about causal inference. Then, the classic regression discontinuity design is examined. Newer material follows.
KeywordsResponse Function Average Treatment Effect Assignment Variable Estimate Treatment Effect Assignment Rule
Work on this paper was funded by a grant from the National Science Foundation: SES-0437169, “Ensemble methods for Data Analysis in the Behavioral, Social and Economic Sciences.”
- Berk RA (2008a) Statistical learning from a regression perspective. Springer, New YorkGoogle Scholar
- Berk RA (2008b) Forecasting methods in crime and justice. In: Hagan J, Schepple KL, Tyler TR (eds) Annual review of law and social science. Annual reviews, Palo AltoGoogle Scholar
- Berk RA, Brown L, Zhao L (2009) Statistical inference after model selection. Journal of Quantitative Criminology, forthcoming, University of Pennsylvania, Department of Statistics, Working Paper (under review)Google Scholar
- Berk RA, Barnes G, Ahlman L, Kurtz E (2010) When second best is good enough: a comparison between a true experiment and a regression discontinuity quasiexperiment. University of Pennsylvania, Department of Statistics. Working PaperGoogle Scholar
- Cameron AC, Trivedi PK (2005) Microeconometrics: methods and applications. Cambridge University Press, CambridgeGoogle Scholar
- Campbell DT, Stanley JC (1963) Experimental and quasi-experimental designs for research. Houghton Miffin, BostonGoogle Scholar
- Fan J, Gijbels I (1996) Local polynomial regression modeling and its applications. Chapman & Hall, LondonGoogle Scholar
- Freedman DA (2008) Diagnostics cannot have much power against general alternatives. http://www.stat.berkeley.edu/~freedman/
- Goldberger AS (1972) Selection bias in evaluating treatment effects: some formal illustrations. Madison, WI. Unpublished manuscriptGoogle Scholar
- Hastie TJ, Tibshirani RJ (1990) Generalized additive models. Chapman and Hall, New YorkGoogle Scholar
- Imbens G, Kalyanaraman K (2009) Optimal bandwidth choice for the regression discontinuity estimator. Harvard University, Department of Economics, Working PaperGoogle Scholar
- Lee DS, Lemieux T (2009) Regression discontinuity designs in economics. National Bureau of Economic Research: working paper #14723Google Scholar
- Neyman J (1923) Sur Les Applications de la Thorie des Probabilits aux Experiences Agricoles: Essai des Principes. Roczniki Nauk Rolniczych10:151. In PolishGoogle Scholar
- Trochim WMK (1984) Research design for program evaluation. Sage Publications, Beverly HillsGoogle Scholar
- Trochim WMK (2001) Regression discontinuity design. In: Smelser NJ, Bates PB (eds) International encyclopedia of the social and behavioral sciences, vol 19. 12940–12945, Elsevier, New YorkGoogle Scholar
- Westfall PH, Young SS (1993) Resampling based multiple testing. Wiley, New YorkGoogle Scholar