2004 Election Anomalies in Florida

Was There a Major Panhandle Democrat Defection?

Update!

After learning of more recent voter registration data for Florida, I re-ran my calculations on the minimum number of Democrats to have voted for bush. The result is that I can make fewer assumptions while doing the calculations, and it turns out that even more Democrats had to have defected than I previously calculated. The new map can be seen below.

Overview

The particular debate I'm referring to here is the fact that a number of the counties in Florida had presidential vote results that were quite different from what you'd expect if you looked at the declared party affiliation of registered voters - and not only that, but the fact that these anomalies appeared predominantly in counties that used optically scanned ballots, rather than electronic voting machines. The suspicion is that clever hackers avoided the closely watched counties with electronic voting systems and instead focused on the central vote tabulating machines of counties with optical scanning systems that people were less concerned about. I found information on the issue at Common Dreams News Center, Democratic Underground Forums, UsTogether.org, and Truth Out.

UsTogether.org has the most analysis of the issue that I have seen, as of this writing (9 Nov. 2004). The issue of computer security and the election is discussed in the Votergate video that can be downloaded at www.votergate.tv

Cautions

From the outset, the analysis does not appear too problematic: look at voter registration and compare to voting results, with and without electronic voting, and see where they fall out on the map. However, one issue not really discussed on the pages cited above is the fact that the voter registration data were rather old - from 5 March, 2004 - and a good amount of voter registration efforts took place since that date around the country. How much of the difference in votes vs. voter party affiliation can be attributed to the changes in registered voters between March and November? Can you really draw any conclusions with such data? I came up with one analysis that turned out some interesting results.

Data Sources

I made an effort to get data for my analyses from as close to official sources as possible:

Florida County Vote Totals: USA Today election website, copied into an Excel spreadsheet 4 November, 2004
Voter Registration, by County: Florida Division of Elections, PDF revised 5 March, 2004
Voter Turnout: Florida Division of Elections, first set of unofficial returns - total registered by 2 November, 2004, and voter turnout per county
County Boundaries and Population Stats: U.S. Census Bureau provides GIS data for county boundaries and 2000 census statistics
Voting Method: UsTogether.org - My efforts to Google official information about which counties used electronic voting failed, so I went with what UsTogether.org reported

Data from these sources were compiled using a combination of Excel and ArcMap. All the data entry was double checked, so I am relatively confident that the tables these maps were based on are accurate. Currently the data exist in multiple tables that I join in order to create the maps. I hope to get a clean single table of my data and calculations on this website soon to document my work.

Verifying Initial Allegations

Since you can find a lot of crazy stuff on the internet, I decided to first compare the relationship between registered voter party affiliation and election outcome for each county, using the data I'd assembled. These graphs show simply the correlation between percent of votes for Bush/Kerry and percent of registered Republicans/Democrats per county. The first only uses percents based on the total Bush/Kerry votes (ignoring votes for other candidates) and on total major party affiliated registered voters to focus on the basic left-right split in this country. It is in part by necessity because the USA Today site only gave election results for Bush, Kerry and Nader for each state, not all candidates on the ballots. However, I include the second graph to show how calculating percent Democratic/Republican party affiliation changes if you calculate percent based on total registered voters of all inclinations. The points shift down slightly along the Y axis, but the overall pattern remains similar.

The "E" and "O" in the legend stand for electronic voting and optically scanned ballots, respectively.

As you can see, the counties with electronic voting fall quite closely along the line representing a 1:1 ratio between the two sets of numbers. It's the counties with optical scanning that deviate the most from what you would expect if party affiliated voters tended to vote along party lines. Conclusion: This stuff wasn't made up. I was able to replicate the observed pattern with "official" data readily available online.

Mapping Results

Election Results: Percent votes for Bush based on the total number of votes for either Bush or Kerry (ignoring votes for other candidates). This was in part so that you could simply reverse the numbers and get the corresponding map for Kerry. Counties with electronic voting are outlined in bright green, as indicated.

Voter Registration: Percent of registered voters that declared party affiliation as Republican. The first map shows percent of major party affiliated voters (Democrat+Republican) and the second shows percent of all registered voters per county. Remember that these are percentages as of 5 March, 2004.

Percent Difference in Registration vs. Vote: To get an initial feel for how the differences were distributed throughout the state, I simply subtracted the percent registered Republicans (based on the fraction of total Democrat+Republican registered voters) from the percent of votes for Bush (based on total Bush+Kerry votes). The result is the number of percentage points higher (or lower) than expected that Bush got than what would be expected from straight party-line voting. The map for Kerry would be the inverse of the numbers on this map.

Evidence of a Panhandle Democrat Defection?

There is no way to reconstruct who voted for whom. What can we do with the data that are available? Not only do we not know how many of the registered Republicans and Democrats each made it to the polls on election day, but we don't know how many may have voted for the opposite candidate. On top of that, there was an average increase of 8.5% in registered voters from March to November, and I don't yet have data to indicate whether these new voters leaned more or less to one party.

My approach was to ask this: If we assume the absolutely best outcome for the Republicans, are there still counties where large numbers of Democrats MUST have voted for Bush, based on the final vote count? Here's how I calculated this:

100% of Republicans registered by 5 March, 2004 were assumed to have made it to the polls and voted for Bush
100% of all other non-Democrat voters registered by 5 March, 2004 were assumed to have made it to the polls and voted for Bush
If there was an increase in registered voters from March to November, 100% were assumed to have gone to the polls and voted for Bush
If there was a decrease in registered voters from March to November (3 counties), 100% of the reduction was assumed to have come from the registered Democrats.

So, for all but three counties, the maximum number of votes for Bush excluding Democrats is the total number of non-Democrat registered voters on 5 March, 2004, plus all the new voters added by November. For the three counties that saw declines in voter registration, the maximum possible number of votes for Bush excluding Democrats is the number of non-Democrats that were registered as of 5 March, 2004. To generate this map, I simply found the counties where the total number of votes for Bush was greater than the maximum possible non-Democrat votes, and calculated the difference. (The percent was calculated as the minimum number of Democrat defections divided by the total number of Democrats registered in March, times 100.) The result is the minimum number/percent of Democrats who had to have switched sides and voted for Bush in order for the total number of votes for Bush to add up.

According to my analyses, there were no counties where reverse was true. If you assume the best possible outcomes for the Democrats, you don't need to have any Republicans voting for Kerry to get the results seen.

Update

I was recently made aware that much more recent numbers on voter registration party affiliation were available from the Florida Division of Elections. The 4 October numbers exactly match those reported as the final voter registration numbers at the time of the election, so we can use these numbers to come up with a much tigher estimate of the minimum number of registered Democrats to vote for Bush.

To get these numbers, I simply took the final total number of registered voters and subtracted out the number of declared Democrats. This gave the maximum number of non-Democrats who could have voted for bush. Then, for each county where the number of registered non-Democrats was greater than the number of votes for Bush, I subtracted the number of registered non-Democrats from the number of Bush votes. The result is the absolute minimum number of registered Democrats who must have voted for Bush. The percent is calculated by dividing this number by the total number of registered Democrats and multiplying by 100.

As you can see, no new counties appear as requiring Democrat defections, but for those that do, the numbers generally go up. There are now five counties that require 40% or more of their registered Democrats to have voted for Bush, and another five that required 30% or more. And remember that this minimum is based on the assumption that 100% of all registered non-Democrats voted on 2 November, and they all voted for Bush. Most likely, some of them did not, which would increase the percent number of Democrat defections required for Bush's vote total to add up.

Population Density

Because these maps only show percent votes and not total numbers, having a picture of population density will help understand where the greatest weight in the state-wide vote count falls. (Data from 2000 Census)

Conclusions

These graphs don't prove that voter fraud occurred in Florida. However, if the data I've been using are correct, then they do suggest that Kerry lost a significant amount of support in the Panhandle region of Florida. I don't know enough about the social and political situation in that area to say how plausible that is. I leave it up to you, the viewer to make your own guesses.