Does the Running Variable Matter? A Second Look at Discontinuity Designs for Evaluating Regional Economic Development and Business Incentive Policies

In recent years, a growing number of studies have applied regression discontinuity designs (RDDs) to the analysis of prominent regional economic development and business incentive policies. As a result, impact evaluations of these policy interventions are being increasingly commissioned under the assumption that local linear regressions have superior impact identification properties in the cases in which an eligibility boundary, geographical border or application cut-off is identifiable as a source of treatment assignment. This paper shows that in a number of frequently encountered cases, however, economic development and business incentive programs pose peculiar impact identification conditions in terms of violations of main assumptions for the validity of RDDs and of running variables that have little or no influence of the outcome variable. Under these circumstances, that bear generalizable relevance also for the analysis of other public programs at large, no adequate testing to support the internal validity of the applied discontinuity design is often performed in the literature and the use of local linear/polynomial regressions, or cross-border comparisons, becomes an inadequate choice. When these conditions apply, adopting instead other quasi-experimental methods to achieve an actual adequate treatment-control balance of the relevant baseline characteristic will often be a better approach.