Outlier Detection in the 1996 Florida Election

The 1996 presidential ballot in Palm Beach County was not of the butterfly type. If the county also had excessive Reform Party vote share in 1996, compared to other Florida counties, this could provide a plausible alternative explanation for the apparently excessive Buchanan votes in 2000. The idea would be that support for the Reform Party has often been unusually high there. On the other hand, if the Reform Party vote share was exceptionally low in Palm Beach County in 1996, then the anomality of the 2000 vote share could be exaggerated. The 1996 data support neither of these alternatives.

The presidential election of 1996 featured three main contestants, one of whom was Reform Party leader Ross Perot. The other candidates on the ballot included Democratic incumbent Bill Clinton and Republican challenger Senator Bob Dole. In the overdispersed binomial model of equations (1), (2) and (3) for an analysis of Florida county data in 1996, $y_{i}$ is the count of votes received by the Reform Party candidate in the 1996 presidential election in county $i$. Our regressors are the proportion of votes officially received by the Republican candidate in the 1992 presidential election in county $i$ ($x_{1i}$), and the proportion of votes officially received by the Reform Party candidate in the 1992 presidential election ($x_{2i}$).

Figures 6 and 7 show the 1996 Florida county residuals. These contrast sharply with those shown for the 2000 residuals. In the 1996 data there are three outliers, all with negative residuals: Pinellas County ( $\tilde{r}_i=-4.4$), St. Lucie County ( $\tilde{r}_i=-4.0$) and Okaloosa County ( $\tilde{r}_i=-3.9$). Palm Beach County is in the tail of the distribution, with a residual of $\tilde{r}_i=-2.2$, but it is not an outlier. Certainly it is not a positive outlier. This result undermines a potential critique based on historically high support for the Reform Party.

Palm Beach County was not a negative outlier in 1996. In terms of raw rank position, the Reform Party vote in 1996 was not exceptionally low compared to 1992. In the 1992 presidential election Palm Beach County had the 11th lowest Reform vote share among counties in Florida and in the 1996 presidential election it was the 7th lowest. As a counterfactual thought experiment we consider how large a vote share in 1996 would have been required for the 2000 vote share not to be an outlier. A simulation exercise shows that the 1996 vote share of 8 percent would have to have been 18 percent in order to bring the 2000 studentized residual below 4.0. A value of 18 percent in 1996 would have put Palm Beach County among the five highest counties in terms of support for the Reform Party's presidential candidate in the state.