VoteTrustUSA - Lost Votes in the 2006 Election: A Look at Extraordinary Undervote Rates on the ES&S iVotronic

Nearly five months after the 2006 general election, the controversy over the undervotes in the Congressional District 13 race in Sarasota County continues. In recent days, however, an August 15, 2006 letter from ES&S, the maker of the iVotronic voting machines used by Sarasota, has refueled speculation about the cause of the problem and renewed efforts to determine if races in other counties were affected. Even though the iVotronic is the primary voting equipment for only 11 of Florida’s 67 counties, these include the state’s most populous counties in south Florida. Thus, problems with iVotronics have a huge impact on statewide races.

The newly revealed letter, along with evidence of undervote problems from the state’s analysis of overvotes and undervotes, and reports of undervote problems in other counties using the iVotronic, led us to initiate a review of undervote rates in top-of-the-ballot races in the eleven Florida counties that use the iVotronic for precinct voting.¹

Findings

Based on our preliminary examination of the data, we find the following:

During the 2006 general election, six of the eleven counties using the iVotronic experienced excessive undervote rates in at least one race.
These undervote rates are “excessive”² in three ways:

They are vastly higher than those experienced in the other counties,
They are vastly higher than experienced in other races in the same county, and
They are higher than experienced on absentee ballots in the same race.

None of the counties using the Sequoia Edge experienced these undervote spikes nor does it appear that any of the optical scan counties had such spikes, although we have not yet examined all races in those counties. It appears that the undervote spikes only occurred on the iVotronic.
Counties that experienced races with excessive undervotes had normal undervotes in other races and showed an otherwise normal increase in undervotes from the top to the bottom of the ballot, except for a conspicuous bulge in these races.
All non-iVotronic counties—that is, the other 56 counties in Florida—examined to date have shown a normal progression in undervotes from the top races on the ballot to the bottom races.
The most egregious example of excessive undervotes occurred in the attorney general’s race where five of the eleven iVotronic counties had excessive undervote rates.

In two counties—Charlotte and Sumter, one of every four ballotscast at early voting and on election day did not record a vote for attorney general.
In Lee County, one in five ballotson the iVotronics did not contain a vote for attorney general.
In the two most populous counties in the state—Miami-Dade and Broward—about one in ten ballots did not contain a vote in the attorney general’s race. In these two counties combined, more than 70,000 ballots did not record a vote for attorney general.
The Broward County undervote is particularly odd because it is the home county of one of the candidates, a situation that usually leads to low undervote rates.

Methodology and Terminology

First, we should explain the terminology. The ES&S iVotronic is a direct recording electronic voting machine that is usually referred to as a DRE or, more colloquially, as a touchscreen. Eleven Florida counties use the iVotronic as their primary voting system. Four counties use the Sequoia DRE. The remaining 52 counties use optical scanners made by Diebold Election Systems and ES&S. All counties use optical scanners to tabulate absentee ballots. Although the DRE counties account for only 15 of the state’s 67 counties, they are some of the state’s most populous, urban areas and thus account for about half of the votes cast statewide.

An undervote occurs when the ballot shows fewer votes than allowed. In most cases, that means no vote was recorded. Both optical scanners and DREs permit undervotes since voters sometimes choose to skip a race. On optically scanned paper ballots, when there are questionable undervotes, it is possible to examine the paper ballots to determine if the undervote occurred because of the voter’s choice or because of a tabulation malfunction. With Florida’s DREs, no such inspection of the ballot is possible. Inspection of the ballot images only confirms that the machine failed to record a vote, not that the voter intended to skip the race.

An overvote occurs when the voter selects more choices than permitted in a race. Both DREs and precinct-based optical scanners reject overvoted ballots; therefore, overvotes are usually confined to absentee ballots where the voter does not have a chance to correct his ballot. State law requires that overvotes be examined by the county canvassing board to see if the voter’s intent can be determined. If so, the vote is counted. The state has guidelines for determining voter intent in order to provide uniformity as required by Bush v. Gore. 531 U.S. 98, 121 S. Ct. 525, 148 L.Ed. 2d 388 (2000)

Undervotes, overvotes, and invalid write-ins comprise the residual vote—that is, votes that are cast but not counted.

For the top two statewide races on the ballot, the U.S. Senate race and the governor’s race, we used data from the state’s “Analysis and Report of the Overvotes and Undervotes for the 2006 General Election,” released January 31, 2007. After the problems of the 2000 general election, the state legislature mandated this report by statute in order to assess the performance of voting systems and election administrators in subsequent general elections. We also used the numerous data tables that accompanied the report.

For the attorney general’s race, we used results posted on county websites, including both summary and precinct-by-precinct reports, as well as data and reports posted on the Florida Division of Elections site.³ Other sources, such as newspaper articles, are referenced within the text.

Although we are not providing data or analysis for the Congressional District 13 race at this time, we are providing links to other reports and analyses of this race.⁴

State Analysis of Overvotes and Undervotes in the 2006 General Election

Table 1: Undervotes in the U.S. Senate Race for the iVotronics CountiesSince 2002, the state Division of Elections (DoE) has produced an analysis of the residual vote in each general election with the goal of monitoring the state’s progress in diminishing uncounted votes, evaluating the relative performance of voting systems and election officials, and determining the impact of new laws. This report is required by Florida Statute § 101.595. The statute requires analysis of the Governor’s race or the first race on the ballot, but the state report for the 2006 election examines both the U.S. Senate race and the Governor’s race. Oddly, it doesn’t examine other races with much worse undervote rates. In fact, the statute also lists a number of requirements that were not met by the report.⁵

While this year’s report by the DoE, released January 31, 2007, presents important information about residual vote rates, it has some significant limitations that tend to obscure or even distort problems.

First, the DoE report looks at only the top two statewide races—the Senate race and the Governor’s race. It compares the governor’s race both with the governor’s race in 2002 and the presidential race in 2004. Because the 2004 analysis did not look at the Senate race that comparison—which would have been the most obvious and useful—was not made.
The report only briefly mentions the Congressional District 13 race and presents no data or analysis of this race, despite the fact that the race had much higher undervote rates than the two races examined and is currently the subject of two lawsuits.
The only argument presented in the report concerning the CD 13 race is so clearly spurious as to be absurd. Instead of analyzing the 18,000 undervotes in Sarasota, it focuses on a sample of only 58 votes in Hardee County that were cast on a Diebold DRE with an entirely different ballot design! Despite this lack of evidence or discussion, the report assumes that the undervotes were solely the result of ballot design problems and makes recommendations on that basis.
In the examined races, the report only presents summary data, even though county-level data is available in the accompanying spreadsheets. By averaging the data and not mentioning counties whose performance is outside the norm, the report presents an inaccurate picture of performance.
Because of the emphasis on summary data and narrow focus, the report fails to mention the “elephant in the room”—spectacular failures at the county level that suggest specific problems with the iVotronics—most notably the double-digit undervote rates in the attorney general’s race.

Undervote Rates on the iVotronic

Despite these limitations and obfuscations, the data contained within the state’s report does present significant problems with the performance of the iVotronics. Based on this data, we found that the ES&S all-touchsceen counties had significantly higher undervote rates than any of the other precinct-based voting equipment used in Florida; that is, higher than all of the optical scanners and higher than the Sequoia DRE.

Summary Undervote Rates by Voting System

The U.S. Senate race experienced high undervote rates across all counties, thus indicating that many voters chose to skip the race; however, these rates varied significantly by the type of voting system. In the U.S. Senate Race, we see the following differences:

iVotronic—2.23%
Diebold optical scanner—1.00%
Sequoia—1.84%

The undervote rate on the iVotronic was an incredible 223 percent higher than the rate on the Diebold machine and 22 percent higher than the Sequoia DRE in the U.S. Senate race.

In the governor’s race, the undervote rates were lower, but the relative performance was similar.

iVotronic—1.02%
Diebold OS—0.62%
Sequoia DRE—.86%

The undervote rate on the iVotronic was 65 percent higher than the Diebold OS and 19 percent higher than the Sequoia DRE in the governor’s race.

If one looks only at the summary data in both races, the inescapable conclusion is that the general performance of the iVotronic is inferior to the Sequoia Edge, the Diebold optical, and the ES&S optical scanner. But the summary numbers obscure the main reason for the higher rate—highly anomalous numbers from one or two specific counties.

While these percentages may seem low at first glance, the difference between the undervote rate on the ES&S iVotronic and other voting systems represents thousands of lost votes.

An Examination of County-Level Undervote Rates

Senate Race.

An examination of the county-level data reveals a different story. In the Senate race, if Miami-Dade is removed from the Senate race calculation, then the iVotronic undervote rate falls to 1.74 percent, which is actually lower than the average undervote rate for the Sequoia Edge (1.84%), although still substantially above the rates for the optical scanners (1.00 to 1.04%).

In fact, Miami-Dade’s undervote rate in the U.S. Senate race is highly anomalous at 4.37 percent—more than twice the state average and four times the average for optical scanners. Because Miami-Dade is such a large county, this undervote translates into a disproportionately large number of uncounted votes and skews the summary data. You will note in the following chart that Miami-Dade accounted for nearly half of all the undervotes on the U.S. Senate race in the iVotronic counties. Even more surprising is the finding that Miami-Dade accounted for nearly a quarter (23%) of all the U.S. Senate undervotes statewide during election day and early voting:

Table 1: Undervotes in the U.S. Senate Race for the iVotronics Counties

County	Precinct Tabulator	Undervotes	Undervote Rate
Broward	iVotronic	6886	1.88%
Charlotte	iVotronic	560	1.22%
Collier	iVotronic	2024	1.44%
Lake	iVotronic	934	1.24%
Lee	iVotronic	1510	1.16%
Martin	iVotronic	546	1.23%
Miami-Dade	iVotronic	15,110	4.37%
Nassau	iVotronic	230	1.17%
Pasco	iVotronic	1849	1.63%
Sarasota	iVotronic	1394	1.16%
Sumter	iVotronic	347	1.26%
Total		31,390	2.23%

Table 2: Undervotes in the U.S. Senate Race for the Sequoia Edge Counties

County	Precinct Tabulator	Undervotes	Undervote Rate
Hillsborough	Edge	3,917	1.62%
Indian River	Edge	747	1.98%
Palm Beach	Edge	5,133	1.62%
Pinellas	Edge	5,619	2.30%
Total		15,416	1.98%

Table 3: Undervotes in U.S. Senate Race for Large/Urban Diebold Accu-Vote OS Counties*

County	Precinct Tabulator	Undervotes	Undervote Rate
Brevard	Accu-Vote OS	1,073	0.73%
Duval	Accu-Vote OS	2,159	1.22%
Polk	Accu-Vote OS	1,088	0.95%
Seminole	Accu-Vote OS	952	0.98%
Volusia	Accu-Vote OS	1,028	0.90%
Total			1.01% (all Diebold counties)

* Absentees removed to make data comparable with iVotronics and Edge.

Governor’s Race

Overall, one would expect the governor’s race to have the lowest undervote rates, and it does. In fact, the governor’s race is the one race at the top of the ballot in which the iVotronics are not grossly out of sync with the other precinct tabulators. But even in this case the difference is substantial.

Again, as with the U.S. Senate race, when the two counties with the high undervote rates—Sarasota and Miami at 1.40 percent and 1.35 percent respectively—are removed the calculations, it is easy to see that remaining counties have undervote rates in line with other DREs. The average with Miami-Dade removed from the calculation falls to 0.99 percent; with both Miami-Dade and Sarasota taken out the average is 0.81 percent. The average undervote rate for the Sequoia DREs is 0.86 percent. In the following chart, six of the remaining nine counties have undervote rates below that average.

Again, we note that Miami-Dade accounts for a large percentage of the total undervotes on the iVotronics on this race—more than a third. Together, Sarasota and Miami-Dade account for a little less than half (46%) of the iVotronic undervotes in the Governor’s race.

Table 4: Undervotes in the Governor’s Race for the iVotronic Counties

County	Precinct Tabulator	Undervotes	Undervote %
Broward	iVotronic	3231	0.88%
Charlotte	iVotronic	376	0.82%
Collier	iVotronic	490	0.70%
Lake	iVotronic	787	1.04%
Lee	iVotronic	954	0.73%
Martin	iVotronic	342	0.77%
Miami-Dade	iVotronic	4,684	1.35%
Nassau	iVotronic	164	0.83%
Pasco	iVotronic	913	0.80%
Sarasota	iVotronic	1,673	1.40%
Sumter	iVotronic	248	0.90%
Total		13,862	1.02%

Table 5: Undervotes in the Governor’s Race for the Sequoia Edge Counties

County	Precinct Tabulator	Undervotes	Undervote Rate
Hillsborough	Edge	1,288	0.54%
Indian River	Edge	423	1.12%
Palm Beach	Edge	3,399	1.07%
Pinellas	Edge	2,069	0.85%
Total		7,179	0.86%

Table 6: Undervotes in Governor’s Race for Large/Urban Diebold Accu-Vote OS Counties*

County	Precinct Tabulator	Undervotes	Undervote Rate
Brevard	Accu-Vote OS	884	0.60%
Duval	Accu-Vote OS	1,221	0.69%
Polk	Accu-Vote OS	552	0.48%
Seminole	Accu-Vote OS	566	0.60%
Volusia	Accu-Vote OS	623	0.55%
Total		3,846	0.62% (overall)

* Absentees removed to make data comparable with iVotronics and Edge.

Conclusions

So by noting county deviations from the norm, we reach a much different conclusion about the iVotronic problem than does the state report on these two races. We see that the two types of DREs used at the precinct performed comparably in most cases, but there are spikes in undervote rates by county and race that suggest situation-specific machine malfunction, failures, or other problems with the iVotronics that require further investigation. Further, we note that these spikes accounted for a large percentage of the total undervotes on the iVotronics in these races.

Attorney General’s Race

While the above discussed undervote spikes in the Senate and gubernatorial races are troubling, they pale in significant compared to the undervote problems in the attorney general’s race. We are at a loss to understand why the state’s report did not address this conspicuous problem.

A preliminary examination of undervotes in the attorney general’s race shows that most counties experienced rates that were consistent with expectations; that is, undervotes increased as one went down the ballot. In most counties, undervotes in the attorney general’s race were more than those for the governor’s race just above it and less than those in the chief financial officer’s race just below it. In addition, undervote rates in the optical scan counties did not appear to vary significantly by mode of voting—that is, undervotes for absentee ballots were about the same as those for election day and early voting. In the DRE counties, absentee ballot undervote rates were consistently lower than election day and early voting rates, but were consistent with the overall performances differences between DREs and optical scanners.

Five iVotronic counties, however, had horrendous undervote rates in the attorney general’s race that were far out of line with other counties, out of line with other races in those counties, and far higher than absentee ballot rates in those races. As seen in the following table, undervote rates in the attorney general’s race in these five iVotronic races were highly suspicious—ranging from two to ten times(200 percent to 1,000 percent) higher than the rates in the other iVotronic counties, whose undervote rates were in line with counties using other precinct tabulators.

Table 7: Undervote Rates in the Attorney General’s Race for iVotronic Counties*

County	Precinct Tabulator	Undervotes	Undervote Rate	Polling Undervote	Early Voting Undervote Rate	Absentee Undervote Rate
Broward	iVotronic	35,743	8.68	11.02	7.88	1.99
Charlotte	iVotronic	11,647	20.88	24.90	24.41	2.69
Collier	iVotronic	2,844	3.24	*	*	*
Lake	iVotronic	3,110	3.57	*	*	*
Lee	iVotronic	27,642	17.74	21.01	19.16	2.38
Martin	iVotronic	1,673	3.14	*	*	*
Miami-Dade	iVotronic	35,456	8.62	9.64	7.01	5.91
Nassau	iVotronic	573	2.53	*	*	*
Pasco	iVotronic	5,855	4.50	*	*	*
Sarasota	iVotronic	6,217	4.36	4.77	4.45	2.91
Sumter	iVotronic	6,655	21.76	24.96	24.63	3.12
Total		137,415

* Data by mode of voting were not available from some counties’ websites.

After the meltdown in the Congressional District 13 race, several newspapers published stories about high undervote rates in the attorney general’s race in three counties—Charlotte, Sumter, and Lee. The stories mentioned undervote rates of 17 to 20 percent; however, the numbers are even worse when only the votes on the iVotronic are examined; that is, election day and early voting combined. On the iVotronics, these two counties had unbelievable undervote rates of 25%. Nearly one in four ballots cast in Sumter and Charlotte counties did not contain a vote for attorney general. In Lee County, one in five ballots (20%) did not contain a vote for attorney general. Yet, these same voters voted in other, down-ballot races such as commissioner of agriculture, retention of judges, and amendments. Of all the numbers discussed thus far, these are the most seriously anomalous. The newspaper stories suggest that ballot design was the problem in these races, but it is hard to imagine a ballot so poorly designed that one in four voters would miss such a high profile race. In addition, it is hard to imagine that voters would not have complained about the problem.

Two other iVotronic counties—Broward and Miami-Dade—also had very high undervote rates in the attorney general’s race, even though the rates for these counties were far below those for the three given above. As seen in the above table, however, these two counties accounted for a very large number of undervotes—more than 71,000—and more than half of the total undervotes in this race in the iVotronic counties. (Miami-Dade also had the unwanted distinction of being the only iVotronic county to have high rates in the attorney general’s race on its absentee ballots.) What makes the undervote rates in these counties so serious is that they are the two largest counties in the state and the home territory of one of the two candidates. So votes lost in these counties most likely had a partisan bent and thus a disproportionate partisan effect.

We have begun a precinct-by-precinct investigation of the votes in the seriously anomalous counties to see if the undervotes can be traced to particular precincts and if there is any evidence of partisan or racial bias. In addition, we are checking with the counties to find out about the ballot layout since this has been alleged as a possible cause both in the attorney general’s race and in the CD 13 fiasco.

So far it seems as though the undervote problem was widespread—that is, it seems to have affected most, if not all, precincts in Charlotte, Lee, and Sumter counties, although some precincts had much higher undervote rates than others—up to 35 percent in some precincts. In several precincts, the number of undervotes exceeded the votes for one of the two candidates; in at least one precinct, “no selection” beat out both candidates. It does not seem to have affected absentee balloting in those precincts.

We did find an interesting case in Sumter County. In precinct 104, only four people voted during early voting. Leaving out the attorney general’s race, three of the four voted a straight Democratic ticket from top to bottom—beginning with the U.S. Senate race and going through the county commission races. The remaining person voted a straight Democratic ticket with the exception of the governor’s race. But in the attorney general’s race only one vote was recorded—the other three were undervotes! These voters not only voted for every other Democrat on the ballot, they also voted in every race on the ballot, all the way down through the amendments, except for one undervote in the commissioner of agriculture race. The chart is attached. Whether this precinct is simply an interesting anomaly or represents evidence of a larger trend remains to be investigated.⁶

Conclusions and Recommendations

The high undervote rates on the iVotronics represented a major failure in the 2006 general election. Certainly, the state Division of Elections should have investigated the problems to determine the probable cause or causes of the problems. Yet, the state report is silent on the performance of the iVotronics. Indeed, it doesn’t mention voting system performance on election day or early voting at all. It does mention some other problems, but offers neither explanations nor plans for additional investigation.

Yet the state division of elections had a statutory mandate to examine these issues. See Florida Statute § 101.595. Aside from requiring an analysis of the Governor’s race or the first race on the ballot, it also mandates the following:(2) The Department of State, upon receipt of such information, shall prepare a public report on the performance of each type of voting system. The report must contain, but is not limited to, the following information:

(a)  An identification of problems with the ballot design or instructions which may have contributed to voter confusion;

(b)  An identification of voting system design problems; and

(c)  Recommendations for correcting any problems identified.

An analysis of each type of voting system was not presented in the state report nor were ballot design or voting system design problems addressed, despite the high undervote rate in certain races on certain voting systems in certain counties. Since the problems are not identified, the report’s recommendations fail to address the problems.

This failure to investigate known problems, coupled with its knowledge of the vendor’s letter about slow responses on the iVotronics, suggests that the division of elections is not interested in any investigations that might reveal issues with the machines that could have compromised the results of the election or could influence the current debate over whether these machines should be retrofitted with printers. The state’s report seems determined to find no problems that cannot be attributed to voter choice or voter error. The Division of Election’s failure to fulfill its responsibilities to examine these problems should be the subject of a full investigation.

With the legislature considering bills to mandate optical scanners across Florida and counties set to buy new equipment, it is imperative to get to the root of this problem and determine its impact on the 2006 election. We hope to have a more complete report on these excessive undervotes on the iVotronics available no later than May 1, 2007.⁷

[1]The August 15 letter has been around on an election reform website for quite some time; however, its existence was not known to the attorneys in the lawsuits or the general public until recently. The attorneys had requested all correspondence with ES&S so the letter should have been produced during discovery; however, neither the state nor the Sarasota Supervisor of Elections included the letter with discovery documents.

Also, please note that the numbers used by the state in its analysis are sometimes slightly different from those given on the county’s official certified results. In most cases, the differences are so small as to be insignificant.

[2]Although we have not formally defined “excessive,” the undervote rates in these counties are so much higher than the norm that their aberration is obvious.

[3]While the state division of elections website has summary data, it does not break down vote totals by mode of voting (absentee, early voting, or polling) or by precinct.

[4] For a discussion of the excessive undervotes in the CD-13 race in Sarasota, see Walter Mebane and David Dill, “Factors Associated with the Excessive Undervote in the 2006 General Election in Sarasota County, Florida,” January 23, 2007. http://macht.arts.cornell.edu/wrm1/smachines1.pdf. Another investigation of the undervotes in the CD-13 race is found in Lauren Frisina et al., “Ballot Formats, Touchscreens, and Undervotes: A Study of the 2006 Midterm Election in Florida,” February 13, 2007, http://www.dartmouth.edu/~herron/cd13.pdf . Other articles and reports on CD-13 are referenced in these two papers. For additional analysis of the Sarasota undervote, see Steve Runfeldt, “Most FL-13 Undervotes Came From Democratic Voters,” http://www.justaskthem.com/Sarasotavote

[5]Our final report will provide a more in-depth discussion of the state’s analysis and report and its many shortcomings.

[6]Mebane and Dill in their examination of the ballot images in the Congressional District 13 race in Sarasota County found that voting a straight Democratic ticket increased the chances of having an undervote in that race. (Walter Mebane and David Dill, “Factors Associated with the Excessive CD-13 Undervote in the 2006 General Election in Sarasota County, Florida,” January 23, 2007.)

Note that we are able to determine conclusively how the four people in Precinct 104 voted in every race except the attorney general’s race since they all voted the same way. This illustrates perfectly what we have been saying for quite some time—that reporting of results by precinct from early voting will violate the privacy of people’s votes. In this case, I can get a disk containing the names of those people voting in precinct 104 during early voting. It is a public record that anyone can get. Consequently, these people have lost their right to a private vote because they voted during early voting. This is happening across Florida because of this incredibly bad law that has been passed simply for the benefit of the political parties.

⁷ To obtain the final report, contact Kitty Garber at or go to our website at www.FloridaFairElections.org. That document will contain a summary of the various investigations of the

CD-13 race as well as a more detailed analysis of the attorney general’s race in Charlotte, Lee, Sumter, Broward, and Miami-Dade Counties. Florida Fair Elections Center, and its lobbying counterpart, Florida Fair Elections Coalition, are the two arms of a nonpartisan, nonprofit grassroots organization that serves the public good as an election watchdog, educational and research group, and advocate for fair, transparent, accessible, secure, and auditable elections throughout Florida.

Comment on This Article

You must login to leave comments...

Other Visitors Comments

There are no comments currently....