Prologue

Introduction

If this hypothesis is correct, the behavior of Democratic voters would result in these additional patterns in the voting data for Duval County:

1. Precincts with disproportionately high rates of "double punched" presidential ballots would have higher rates of "punch every page" voters, and would therefore show a disproportionately lower rate of "unvoted" ballots for less important races, such as Commissioner of Education.
2. The rate of invalid presidential votes would be higher in precincts with more Democratic voters, but the rate of invalid votes in other races would remain approximately the same (or lower) compared to predominantly Republican precincts.
3. The rate of invalid votes in non-presidential races would be similar to (or lower than) other comparable Florida counties.

This document analyzes Duval County precinct data to evaluate the validity of the "punch every page" hypothesis.

The raw precinct return data for Duval County, Florida, for the general election of November 7, is available from the web page of the Duval County Commisioner of Elections. This data is in a format which is not spreadsheet friendly.

The data has been reformatted into a standard CSV format file for this analysis (which you can download here).

Methodology

The first step of the analysis is to sort each Duval County precinct according to the percentage of Democratic voters in the November 7th general election (see "What is a Democrat?" sidebar). The precincts are placed in order, with "least Democratic" precinct first, and "most Democratic" precinct last.

The next step is to determine the percentage of missing votes for each precinct, for each of four races: U.S. President, U.S. Senator, Florida Treasurer/Insurance Commissioner, and Florida Commissioner of Education. "Missing" votes are determined by taking the sum of all votes cast for all candidates in a given race in a given precinct, and subtracting that figure from the total ballots cast in that precinct.

A "missing" vote can be the result of an undervote (failure of a valid selection to be read by the vote counting machine), of an overvote (an invalid selection of more than one candidate in a single race), or of an abstention (a deliberate omission of a vote for a candidate).

The resulting data can be plotted on a graph, as seen in figure 1.. Because of the "noisy" character of the data, and the relatively large number of different data sets being plotted on the same chart, it is difficult to visualize trends in this figure.

To facilitate visualization, a second data set is derived from the first set by taking a running four-unit average. The first element is computed as the average of the first, second, third, and fourth elements of the original set. The second element is computed as the second, third, fourth, and fifth elements of the original set, and so on. This running average technique has the effect of "smoothing out" local random variations in the data, so larger trends are more apparent.

The resulting data plot is seen in figure 2.

This organization of the data illustrates three types of information simultaneously:

1. The relative distribution of precincts countywide by different party composition: strongly Republican, Republican leaning, Democratic leaning, and strongly Democratic.
2. The absolute percentage and relative ranking of missing votes in the presidential, senatorial, treasurer, and education races.
3. The correlation between precinct party composition and missing vote percentages and rankings.

Figure 1 Explanation This is a split scale graph. The scale on the left, from 0% to 30%, applies to the percentage of missing votes in the four races. The four races themselves are indicated by the four colored lines: red for the presidential race, blue for the senatorial race, green for the treasurer race, and yellow for the education race. The scale on the right, from 0% to 120%, applies to the percentage of Democratic voters. The Democratic voter percentage is indicated by the thick white line. This line is reduced in vertical scale by a 4:1 ratio relative to the colored "missing vote" percentage lines. The 268 Duval County precincts are sorted from left to right in order of increasing Democratic party composition.

Figure 2 Explanation This is a split scale graph. The scale on the left, from 0% to 30%, applies to the percentage of missing votes in the four races. The four races themselves are indicated by the four colored lines: red for the presidential race, blue for the senatorial race, green for the treasurer race, and yellow for the education race. The scale on the right, from 0% to 120%, applies to the percentage of Democratic voters. The Democratic voter percentage is indicated by the thick white line. This line is reduced in vertical scale by a 4:1 ratio relative to the colored "missing vote" percentage lines. This graph is a "running average" data plot. Each position on the X axis of this graph represents an average of four adjacent sorted precincts, for all races and for party composition. The leftmost position represents the average of the first, second, third, and fourth precincts in figure 1. The second position on this graph represents the average of the second, third, fourth, and fifth position, and so on. This graph presents the same data as figure 1, but with local random variations smoothed out.

Analysis

For Republican precincts (those with less than 50% Democratic voters), the plot of missing Commissioner of Education votes is unexceptional. It is only a few percent greater than the comparable data for Lee County (figure 3). The overwhelming majority of these missing votes are almost certainly "abstentions": votes not cast for a race few care about. However, the rate of missing education votes does not increase or decrease as Democratic participation rises from 10% to 50%. This suggests that Democratic voters in the Republican precincts were punching or not punching selections for Commissioner of Education at approximately the same rate as Republican voters.

However, in the same range, the percentage of missing presidential votes rises in direct proportion to the percentage of Democratic voters, ostensibly due to the influence of "every page" voters. One might argue that the Republican voting population has the same proportion of "every page" voters as the Democratic population, but that the Republican "every page" voters never mistakenly double-punch the presidential race. This argument, however, does not withstand the evidence of the Democratic precincts, where the percentage of missing education votes increases significantly in proportion to the increase in Democratic composition.

The increase of missing education votes in Democratic precincts indicates that the "every page" voting behavior of Democratic voters in Republican precincts differs from that of Democratic voters in Democratic counties. Yet the increase of missing presidential vote indicates that Democrats have the same "every page" voting behavior regardless of precinct party composition.

The "every page" hypothesis only provides an explanation for how presidential votes could go missing. One would expect, in consequence, that the missing vote rate for non-presidential races in Duval County would be similar to comparable counties. However, by comparing these races with the plots in figure 3, one notices that Duval County missing vote rates in the senatorial, and treasurer races also show an anomolous trend upward with increased Democratic participation, albeit not to the same extent as the presidential race.

Figure 3 Explanation This figure is the same as figure 2 but is based on precinct data from Lee County, a Florida County that uses the same voting mechanism as Duval County (see " Comparison of Precinct Return Data between Duval County and Lee County, Florida") for more information). The corresponding raw data graph for Lee County is available here.

Conclusion

What is a Democrat? There are many ways to define a Democrat (or Republican, for that matter), but most are too ambiguous for the purposes of a statistical analysis. A voter can be a registered member of the Democratic Party, but vote for Republicans more often than not. Another voter may not have a party affiliation, but vote straight Democratic tickets election after election. Which is the Democrat? For the purposes of this analysis, a Democrat is someone who, all else being equal, will vote for a Democratic candidate over a Republican candidate. This definition, though, then raises the question of how to determine "all else being equal". Generally speaking, the less a voter knows about contending candidates, the more "all else is equal". A cross-correlation analysis of all the races in the Duval County general election bears this out. The minor races for Treasurer and Commissioner of Education are the most partisan of all the countywide races. That is to say, Democratic and Republican votes in these races are most strongly correlated with votes for Democratic and Republican candidates in other races. Also, Republican votes in the Treasurer race have the least correlation with Democratic votes in the Education race (and vice versa, both ways). Treasurer Education Gallagher Cosgrove Crist Sheldon Bush 0.997 0.036 0.998 0.080 Gore 0.244 0.979 0.215 0.985 McCollum 0.998 0.046 0.998 0.089 Nelson 0.295 0.972 0.265 0.981 Gallagher 1.000 0.084 0.999 0.128 Cosgrove 0.084 1.000 0.054 0.997 Crist 0.999 0.054 1.000 0.097 Sheldon 0.128 0.997 0.097 1.000 Correlation Coefficients of Candidate Votes In Duval County The definition of "Percent Democratic" used in this analysis is based on an interpolation between percentage of votes cast for Treasurer that were for Cosgrove, and the percentage of votes cast for Commissioner of Education that were for Sheldon. The formula is as follows: Percent Democratic = (Cosgrove + Sheldon) / (Gallagher + Cosgrove + Crist + Sheldon)