27 February - 13 March 2014

Regression discontinuity designs

CO:STA presentation by Giuseppe Pietrantuono (Universität Mannheim)

Testing the causality of our claims has become more important over the recent years. But how to do so with a cross-section of observational data? One potential solution is a regression discontinuity design: RDD exploits a cutoff on a linear variable after which a treatment is assigned. An example might be a special training program that pupils who surpass a certain math grade are assigned to. The intuition is that values close to the cutoff are random, i.e. whether one just passes the cutoff or just misses to pass is not systematic. In consequence pupils with values close to the math grade cutoff are similar in all regards but the fact that some may visit the special program and others not. Hence we can estimate the treatment's causal effect. After a general introduction to RDD, Giuseppe uses it to estimate the causal effect of Swiss citizenship for immigrant integration. Since several Swiss municipalities used to decide on individual naturalization requests by closed ballot voting, a regression discontinuity design can be applied: Whether an applicant just passes the cutoff point of at least 50% approving votes or just misses it is arguably random.


Angrist, Joshua D.& Jörn-Steffen Pischke (2009): Mostly Harmless Econometrics. Princeton University Press: Chapter 6

Imbens G, Lemieux T. Regression Discontinuity Designs (2008): A Guide to Practice. Journal of Econometrics 142(2): 615-635.

Lee, David S. (2008). Randomized Experiments from Non-random Selection in U.S. House Elections. Journal of Econometrics 142(2): 675-697.