By Daniel P. Tokaji
Voting technology issues continue to loom large in the 2006 election season. This is largely the result of the Help America Vote Act of 2002 (HAVA), which requires that new equipment be in place this year in states that accepted federal funds to get rid of their old punch-card and lever machines. HAVA also mandates that at least one disability-accessible unit be at each polling place in this year’s elections. While HAVA was intended to remedy some of the problems that arose in the 2000 Florida recount, the equipment that’s been put in place in a number of states raises new questions about what would happen if a close election led to another recount. Security Concerns and the VVPAT The introduction of new voting equipment has prompted security concerns on the part of some computer scientists, who argue that paperless electronic voting machines are susceptible to fraud and error. This has in turn led some advocates to demand that electronic voting machines produce a contemporaneous paper record, or “voter-verified paper audit trail” (VVPAT), which may be used in the event of a recount. According to electionline.org, 22 states have now enacted such laws, 17 of which use electronic voting equipment. In some of those states — including Ohio — the paper ballot is the official ballot of record as a matter of law. ORC 3506.18. A problem that’s not been given sufficient attention is how a recount would actually work with a VVPAT electronic voting system. In Ohio and most other places that have VVPAT electronic voting machines, the paper record is printed on a roll of paper tape adjacent to the touchscreen interface. That paper tape is behind a transparent screen, so that the voter can see but not touch it, as depicted here and here. A closer examination of this type of VVPAT system reveals difficulties, both legal and practical, that could arise in the event of a recount. On the practical side, an in-depth study of the equipment used in Cuyahoga County, Ohio’s May primary election showed that 10% of the VVPAT records were in some way compromised. Among the problems were blank VVPAT tapes, accordian-style crumpling, destroyed VVPATs, printing anomolies, and missing text. To be sure, there were many other problems found in Cuyahoga County’s primary election, as itemized in today’s story from Wired.com. Those include human errors that could result in the electronic records being compromised. The problems relating to the paper records are particularly troubling, however, because they could mean trouble in the event of a recount. We don’t know for sure whether the problems observed with the paper records in Cuyahoga County also exist in other places that are using VVPAT electronic voting machines. That’s because no comparably detailed study has been done, in Ohio or elsewhere, to study the functioning of VVPATs in a real election. As Professor David Dill of Stanford puts it in the Wired.com story: “I suspect that Cuyahoga County may be below average (in terms of how well it ran its election), but if you lift up the rock and look at election administration across the country, you’ll see the same thing elsewhere.” If there are similar problems elsewhere, it will open up a new possibility for post-election litigation over recounts. Ohio’s Recount Law Let’s assume, for the moment, that the VVPAT problems witnessed in Cuyahoga County’s primary are widespread. How might this play out in the event of a close election? To answer this question, it’s necessary to take a closer look at state law. A statute passed by the Ohio state legislature in 2004 states:
This statute would seem to suggest that, if there’s a recount, the paper record would trump the electronic record in the event of a discrepancy. This is complicated, however, by Standards for Voter Verified Paper Audit Trail that were issued by the office of the Secretary of State to implement the statute. These standards include “Design requirements” which provide that: “In the case of a difference between the electronic record and the paper record copy, the paper record copy shall govern, unless there is clear evidence that the paper record copy is inaccurate, incomplete or unreadable as defined in the system procedures.” On the face of it, there appears to be some tension, if not conflict, between the statute passed by the legislature and the standards promulgated by the Secretary of State. Although a statute is a higher source of law than administrative rules, courts are generally supposed to accord deference to rules promulgated by agencies like the Secretary of State’s office. That deference isnt’s warranted, however, if the agency’s interpretation is “unreasonable or conflicts with a statute covering the same matter.” State ex rel. Celebrezze v. National Lime & Stone Co., 68 Ohio St. 3d 377 (1994). If there were a close election in Ohio that led to a recount, one of the questions that could emerge is whether the Secretary of State’s standards are a reasonable interpretation of the statute or, alternatively, whether they conflict with the law passed by the legislature. Going back to the Cuyahoga County problem, this means that there could be a dispute over which records govern — the electronic records or the paper ones — in the event of an election in which the VVPAT records were compromised. One side might claim that the electronic records would govern in a recount, citing the Secretary of State’s standards. The other side could respond that the VVPAT records should be treated as the official ballot of record, arguing that the Secretary of State’s standards are inconsistent with the statute. My point is not that electronic records are foolproof. Without proper procedural safeguards, they too are subject to fraud and error. But paper records aren’t foolproof either. In making the paper record the official ballot, a new set of legal and practical problems has emerged. Recounting a Close Election Could the discrepancy between paper and electronic records really make a difference? In a close enough election, it might. If voting equipment and associated errors were randomly distributed throughout the state, then one would not expect it to matter whether paper or electronic ballot records govern in the event of a recount. Even if 10% of votes are thrown out, it wouldn’t affect the election result so long as the errors are random — that is, so long as they affected all candidates equally. In reality, however, voting equipment isn’t randomly distributed, at least in Ohio. To illustrate the point, I’ve created this spreadsheet, which shows the type of voting equipment used in each of the 88 Ohio counties, according to the Secretary of State’s website.* The counties labelled “DRE” are ones using direct record electronic voting equipment that, under state law, must generate a VVPAT. It also shows the results of the 2004 presidential election, also taken from the Secretary of State’s website, including which candidate (Bush or Kerry) received more votes in each of the counties. Overall, Bush won in Ohio by a margin of 118,601 votes, receiving 2,859,768 votes to Kerry’s 2,741,167. Using this data, I have attempted to measure how the use of VVPAT records as the official ballot might have affected the outcome, if the current equipment had been used in 2004. To do this, I determined the number of votes cast for Bush and Kerry in the 57 counties that are now using VVPAT electronic voting machines (as shown on the “DRE Stats” tab of the spreadsheet). It turns out that those counties, on the whole, tended to lean toward Kerry over Bush. Overall, the counties that are now using VVPAT electronic voting equipment cast 1,972,006 votes for Bush versus 2,021, 895 votes for Kerry. Next, I attempted to measure the effect that equipment differences could have on the outcome, in the event of a recount where 10% of all the VVPAT records were unusable (as shown on the “-10%” tab of the spreadsheet). Assuming that these compromised paper records would furnish the official ballot of record, that would mean that 10% of the votes in the counties using an electronic VVPAT system would be lost. This figure is not precise, since some votes were also lost due to the punch-card card equipment that was used in most of those counties in 2004, but it at least furnishes an approximate baseline by which to measure the effect of compromised VVPAT records. In this scenario, both candidates would have lost votes — but Kerry would have lost more. Bush would have wound up with 1,774,805 votes in the affected counties, while Kerry would have wound up with 1,819,715 votes. On a statewide basis, making compromised VVPAT records the official ballot of records would have made Bush’s final margin of victory 123,590 votes rather than 118,601 votes. Bush would thus have picked up about 5,000 net votes. My point here is not to suggest that we’re on the brink of another election meltdown, but simply that compromised VVPAT records could make a difference in a close election. While a 5,000 vote swing in either direction wouldn’t have been enough to affect the outcome of the 2004 Ohio presidential election, it could affect closer elections — such as Florida’s 2000 presidential election or Washington’s 2004 gubernatorial election. What’s ironic is that Democrats have, for the most part, been the ones pushing for laws to require the VVPAT and to make it the official ballot of record. The above analysis shows that, in Ohio, Democratic-leaning counties would stand to lose more votes than Republican-leaning counties, if VVPAT records are compromised. The problems in implementing present-generation VVPAT systems should counsel hesitation in making paper the official record in the event of a recount. In the event of a close election, states that have chosen to make the VVPAT the official ballot of record — as well as the advocates who supported such laws — could be in for a rude surprise.
*Irene Mynatt of the EL@M team provided indispensible assistance in creating this spreadsheet, though any errors are the author’s alone.