Agenda-Setting and Tax Referenda: Implications for Regression Discontinuity Identification Strategy Using Election Outcomes

Election outcomes frequently have been used as a source of identification in regression discontinuity (RD) designs by political scientists and economists. This paper tests the validity of using this type of RD identification strategy in county, municipality, township, and school district tax referendum elections where the taxing authority has some discretion over the referendum structure and timing. Using optimal bandwidth specifications, we find discontinuities at the majority vote threshold for numerous referendum and election characteristics that correspond with referendum aspects over which taxing authorities have discretion, and we do not find discontinuities in characteristics over which taxing authorities have less discretion. In addition, when considering the distribution of election closeness, we find a change in the probability across the majority vote threshold for these types of referenda. These findings bring into question the validity of the RD identification strategy when the taxing authority has agenda-setting ability and the bandwidth selection is such that the estimated discontinuity uses election outcomes that are not extremely close to the cutoff.