Regression Discontinuity Applications with Rounding Errors in the Running Variable

Many empirical applications of regression discontinuity (RD) models use a running variable that is rounded and hence is discrete, e.g., age in years, or birth weight in ounces. This paper shows that standard RD estimation using a rounded discrete running variable leads to inconsistent estimates of treatment effects, even when the true functional form relating the outcome and the running variable is known and is correctly specified. This paper provides simple formulas to correct for this discretization bias. The proposed approach does not require instrumental variables, but instead uses information regarding the distribution of rounding errors, which is easily obtained and often close to uniform. The proposed approach is applied to estimate the effect of Medicare on insurance coverage in the US, and to investigate the retirement-consumption puzzle in China, utilizing the Chinese mandatory retirement policy.