A1913
Title: RDD without a discontinuity: Correcting for Berkson measurement error in the running variable
Authors: Junlong Feng - The Hong Kong University of Science and Technology (Hong Kong) [presenting]
Leonard Goff - University of Calgary (Canada)
Abstract: Measurement error is a pervasive phenomenon in empirical research. In the context of the regression discontinuity design (RDD), measurement error in the running variable can be especially problematic. Individuals who are marginal with respect to the treatment assignment rule might be spread across values of the mismeasured running variable, completely eliminating any observable discontinuity in the conditional mean of the outcome at the cutoff. When measurement error in the running variable takes the so-called Berkson form, that is when it is independent of the observed running variable, it is still possible to identify the typical sharp RDD estimand. The identification procedure requires that binary treatment status be observed without error, for the support of the measurement error to be suitably rich, and an assumption that the measurement error is not informative about potential outcomes once one conditions on the true running variable.
