This set of graphs shows the extent to which the election results for the 13th U.S. House District in Sarasota County, FL, fundamentally differ from those in neighboring counties. This analysis is based on the assumption that we can predict the number of votes a candidate will receive in a given precinct based on the number of votes received by a candidate of the same party in a different race.

Specifically, we use the Gubenatorial race; we assume that the number of votes received by the Republican candidate for Governor can help predict the number of votes received by the Republican candidate for U.S. House (the same applies for the respective Democratic candidates).

As a simple example, let us analyze this relationship in neighboring Charlotte County. For each precinct, on the X-axis we plot the number of votes received by the party's candidate for Governor, and on the Y-axis we plot the number of votes received by the party's candidate for U.S. House, 13th District. In addition, we perform a simple linear regression; this will quantify the relationship and the accuracy of our assumption.

We can see a nice, linear relationship. Within each precinct, the number of votes received by Republican House candidate Vern Buchanan is about 85% of the votes received by Republican Gubenatorial candidate Charlie Crist. Similarly, Democratic House candidate Christine Jennings receives approximately 122% of the votes received by Democratic Gubenatorial candidate James Davis. The quality of the linear regression -- measured by the R-squared value, in which values close to 1.0 indicate a good match -- is large for each candidate, indicating an extremely good fit for both predictions.

Notice that this analysis confirms the conventional wisdom of elections in the south. Many voters who selected Republican Charlie Crist for Governor also chose Democrat Christine Jennings for U.S. House. Democrats tend to fare well in legislative races, but not necessarily in executive contests.

Next, let us examine Manatee County. Manatee County is almost as large as Sarasota County and is split between U.S. House Districts 11 and 13. A handful of precincts straddle the line between 11 and 13. We remove results from those precincts to avoid any possible effects of gerrymandering, in which the lines have been drawn deliberately to partition the population in a way favorable to one party or another.

Once again, the relationship holds up and the quality of the predictions is still high. In the precincts in this county, Buchanan receives approximately 87% of the number of votes as Crist, and Jennings receives around 127% of the votes as Davis. The quality of the fits are still very close to 1.0.

This graph shows a close-up of the smaller precincts. Even at this level of detail we can see that the predictions are extremely close to the actual values. Only a dozen precincts (of the 122) are off by more than ten percent.

Now we move onto Sarasota County.

Two things jump out at us immediately. The first is that the slope of the line for each party is much lower than it was in the other two counties. Here, Buchanan is only getting around 78% of the number of votes as Crist (not 85% or 87%), while Jennings fares much worse at 109% (not 122% or 127%). The second is that the fit for Jennings is worse than it was for any of the other linear regressions. In other words, both candidates are getting fewer votes than expected and the quality of the predictions are declining.

A close-up of the smaller precincts in Sarasota. Notice how the grouping of points is more scattershot than in Manatee.

So what went wrong in Sarasota? Was it the machines, or something about the voters of Sarasota that caused them to hold back from voting in House District 13?

We can answer part of that question by looking at the voters in Sarasota who did not use the machines: absentee voters.

Removing the voters who used the touchscreens leaves us with those who used paper, and those who did use paper behaved nearly identically to their neighbors in Manatee and Charlotte. Buchanan picked up 83% of the number of votes as Crist, and Jennings received 121% of the number of votes as Davis. Additionally we have a better fit for this curve than when we lumped together all voters in the previous graphs.

But maybe it wasn't the machines. Maybe something happened in the closing days of the campaign that so disgusted voters about the District 13 race that they stayed away in droves? We can test this hypothesis by examining the behavior of voters who both voted before election day and used the machines: early voters.

Wow. Both Buchanan and Jennings are far below where they were in Charlotte, Manatee, and absentee Sarasota voters. Buchanan picked up only 74% of the number of votes as Crist, whereas Jennings failed to even pick up as many votes as Davis. Jennings is a full 25% off from absentee Sarasotans, and nearly 30% short of her levels of popularity relative to Davis in Manatee. While the quality of the predictions are still reasonable, what they are predicting is dramatically different than the model of absentee voter behavior in the same county.

What about Election Day, though? How did in-precinct voters behave?

While the number of votes received by each candidate is higher (relative to the number of votes received by the Gubenatorial candidate), the quality of the fit for Democrat Jennings is significantly lower. This would indicate a problem that is less widespread than the one that existed in early voting, but also less consistent.

Let's put all three types of Sarasota predictions together: in-precinct, early, and absentee voting.

We immediately notice that the support of each U.S. House candidate relative to their party's Gubenatorial candidate is highest when using absentee and lowest during early voting. Again, this would indicate that the errors during early voting were consistent, concentrated, and affected both parties.

Finally, let us put the Sarasota vote alongside that of their neighbors.

This graph compares the predictions (the linear regressions) for absentee and in-precinct voters in Sarasota, and for all Charlotte and all Manatee voters. The key provides the mathematical function for each line, which shows the support of each U.S. House candidate relative to the state's Gubenatorial candidate.

First, we see that the behavior of absentee voters in Sarasota mirrors almost exactly that of regular voters in Charlotte and Manatee. It is even possible that Buchanan suffered more undervotes in absentee voting than Jennings. Yet during election day it is likely that Jennings came off worse than Buchanan.

We hypothesize that a flaw exists in the machines -- whether it is in the ballot configuration or actual software itself -- that was present in some random number of machines. A significant number of the machines used for early voting had this flaw, resulting in the dramatic drop-off of relative support for both House candidates.

The flaw also existed in some number of machines used during election day, but in a lower total concentration than during early voting. The random distribution of these machines throughout the county contributed to a larger variation in undervote rates. It is possible -- and even likely, in our opinion -- that the flaw does not affect all machines. A uniform flaw would have lead to more uniform undervote rates in the precincts. Therefore, it will be necessary to test a wider collection of machines and ballot styles in search of the problem. Given the severity of the problem, though, it may not take long to replicate the flaw.

NOTE: A reader has pointed out that the lower (and erratic) undervote rate during election day when compared to early voting may be due in part to inconsistent attempts by pollworkers to alert voters of problems in the U.S. House race. This suggests that the flaw may be transient within a machine, or may otherwise be negated by exceptionally sharp-eyed voters.

In conclusion, the voting anomaly in Sarasota County affected both candidates for U.S. House. By basing our predictions solely on the number of votes a candidate received relative to the number of votes received by a member of the same party within each precinct -- and having this relationship be constant across the three counties -- we reduce or eliminate the need to understand the factors that may or may not have brought supporters of certain candidates out to the polls, or how voters felt about the candidates at certain times during the campaign. We see a statistically signficiant difference in the behavior of voters that used paper and those that used touchscreens in Sarasota County. The lack of any tangible form of voter intent -- combined with the flaw affecting both candidates -- means that the true victor of this contest will never be known with any certainty.