The Collinearity Problem in Regression-Discontinuity Models

The article addresses the issue of collinearity in regression-discontinuity models, as raised by a recent book review in this Journal. It examines the causes of collinearity between the constant term in the model and a dichotomous explanatory variable that indicated participation in a program being evaluated. It shows that removal of the constant does not solve the collinearity problem but in fact may bias the estimated program impact. Collinearity can truly be overcome only by collecting more data (of an appropriate kind).