"Waiting for Life to Arrive": A History of the Regression-Discontinuity Design in Psychology, Statistics, and Economics (WP-07-03)
Thomas D. Cook
This paper reviews the history of the regression discontinuity design in psychology, statistics, and economics. The design was invented by Donald T. Campbell in 1958. He and a group of Northwestern University colleagues in both psychology and statistics worked on the design and its analysis until the early 1980s, with Campbell's student William Trochim then carrying on the work. Trochim explored variants of the design in terms of their validity, implementability and analysis. But by 1995, no psychologists could be found working to improve the design, only to popularize its use—at which they had only modest success at best. In statistics, the design has never had a high profile, perhaps because it is deemed unexciting to be able to model selection when the selection process is completely known, perhaps because the design leads to causal inferences whose generalization is limited to the cutoff point, and perhaps because its interpretation depends on functional forms that cannot be directly observed. The first formal proof of unbiased causal inference resulting from the design came from unpublished papers in economics—by Goldberger in 1972. However, economists were then pursuing a broader causal agenda and interest in regression-discontinuity lapsed. It was not used in economic applications until the mid-1990s. However, it has widely caught on since them among younger labor economists and econometricians in both the United States and Europe. Has life now arrived for this 50-year-old design, invented in 1958 and rarely used until the beginning of this century?