The 2000 Election, Reengineered

"It is embarrassing to America when technology fails and puts democracy to such a test as it did this month," David Baltimore, president of the California Institute of Technology, said during a video news conference this past December. He was referring, of course, to the 2000 presidential elections, plagued with butterfly ballots and pregnant chads, among other problems. But as outraged citizens accused both parties of trying to manipulate the vote, Baltimore and Charles Vest, his counterpart at the Massachusetts Institute of Technology, took a more scientific approach: they assembled a team of computer scientists, mechanical engineers and political scientists to reengineer the election process.

On July 16, the Caltech/M.I.T. Voting Technology Project published its findings, targeting three main problem areas in system: voting equipment, voter registration and absentee ballots. "Equipment is something that political scientist haven't thought about," says Stephen Ansolabehere, an M.I.T. professor of political science who co-directed the project with Thomas Palfrey, a professor of economics and political science at Caltech. "It's never been thought to have much importance until this election, so it was never clear how much the residual vote could be attributed to equipment." Residual votes are overvoted, unmarked and uncounted ballots that officials subsequently disqualify.

Problems with voting equipment are far from new: during the 1968 election, there were so many glitches that IBM, which developed the punch-card voting machines used at the time, soon after abandoned the business. And in 1996 Massachusetts eliminated punch cards after officials contested and then recounted the results of a primary election in the 10th Congressional District. Nor are the current problems with voting equipment isolated to Florida. In 2000, the situation was worse in several other places, including Georgia, Idaho, Illinois, South Carolina, Wyoming, New York City and Chicago.

According to the new report, equipment accounted for 1.5 to two million of the four- to six-million votes the Caltech/M.I.T team estimates were lost in this past election. "It's a very big number. We were surprised," Palfrey says. "We knew that quite a few [votes] were lost because of bad equipment, but we didn't realize that the polling place procedures were that bad." In fact, the team discovered that 1.5 to three million votes vanished in registration mixups, and another million went uncounted as a result of operational problems at the polls.

The researchers suggest technology changes could help cut the number of lost votes in half during the next presidential election. Now the U.S. employs five different voting systems: punch cards, paper ballots, optical scans, lever machines and electronic voting machines, also called Direct Recording Electronic (DRE) machines. Punch cards and optical scans are the most common, but punch cards also produce the highest percentage of residual votes. Only lever machines and DREs eliminate the need for paper. Punch cards are being phased out, but it's a slow process. "Once you have a punch card reader, there is no real reason to invest a few million dollars to get a different system," Palfrey says. "Much of it is just inertia."

Image: Caltech/M.I.T. Voting Technology Project PUNCH CARDS are commonly used for voting, but they are also very unreliable.

To reduce the other half of lost votes next time around calls for a new registration system, the scientists say. The Census Bureau estimates that more than 15 percent of all eligible voters had moved in the past year, and three million eligible voters did not vote in 2000 because problems with their registration. Because voter registration is decentralized, states find it difficult to track voters.

"We're a very mobile society, so people move a lot," Palfrey explains. "And as they move, they typically will register in their new place, but they won't necessarily close out their registration in the place they moved from." Michigan, for example, discovered that one million of its nine million registered voters were registered twice.

Also because the registration process is decentralized, it is open to fraud. An investigation sponsored by the Atlanta Journal and Constitution and quoted in the Caltech/M.I.T. report revealed that 5,400 dead people voted in Georgia in a 21-year period. "Registration is basically an honor system," Ansolabehere says. "The more we looked at it, the more troubling it appears. You're supposed to be a citizen to vote, right? Nobody ever checks your citizenship when you register. Many people have duplicate registrations, but the officers have no way of purging your name from the roll unless you call them up and tell them to do it."

Such holes in the registration process affect the polling places on Election Day. "The poll workers typically only have registration information for the voters that are supposed to be there," Palfrey says. "So if somebody shows up at the wrong place, they just don't know what to do and, in many places, they just tell the voter to go home." The huge number of different ballots in large urban counties makes matters even worse. "In L.A. County alone, counting all the different languages, there were about 4,000 different ballots in the last presidential election," Palfrey points out. "So it's a really big logistical problem."

To improve the situation, the researchers propose that voters should be allowed to check their registration before Election Day. The information might be posted in a local paper, mailed out on postcards or, as practiced in several counties in North Carolina, posted on the Internet. Officials should also assign voters an I.D. number, such as their driver's license number. Two John Smiths then wouldn't be able to register in the same district—and one of them couldn't register in two places.

The scientists also suggest compiling voter-registration data on CD-ROMs, which could be distributed to polling places. This method has already eliminated most registration problems in some locations. So-called provisional ballots might also help voters with disputed registrations weigh in on Election Day. Once the provisional ballots were collected, officials could verify registrations later on and, depending on the outcome, count or dismiss the votes.

Image: Caltech/M.I.T. Voting Technology Project ANSA is a framework that the researchers propose as a guideline for the development of future voting equipment. CLICK IMAGE TO ENLARGE

Provisional ballots may also help to alleviate another problem: long lines at the polls. According to the U.S. Census, 2.8 percent of registered voters in the U.S.—or one million voters—did not vote in 2000 because they had to wait too long or the hours during which they could vote were too short. Absentee ballots have alleviated long lines in some states. In Oregon, for example, the entire election was held by mail in 2000. But, as Ansolabehere says, "remote voting has a fundamental problem, which is that your ballot is not secret." The researchers recommend that only those voters who truly cannot vote in person cast their ballot via mail.

In conclusion, the researchers state that a secure and accurate election system must involve many people, require a separation of responsibilities, take place in the public eye, and allow for redundancy. Some of that might seem obvious, but the scientists point out that newer voting technologies don't always meet these criteria. Voting machines, for instance, take over the counting process, eliminating a certain degree of openness. And the absence of physical ballots means later audits are impossible.

The team proposed not one particular technology as a solution but rather a framework-what they called A Modular Voting Architecture (AMVA). Within this framework, they break the voting process down into six steps: a voter signs in; he or she generates a vote; he or she confirms the selection; the vote is cast; the vote is counted; and the vote is audited. To separate the steps requires some kind of physical medium—a paper ballot, a memory card or something else—on which voting information is recorded. The researchers arbitrarily call it a FROG. At sign-in, voters would receive FROGs that contained information about the voting precinct and the kind of ballot cast, but not themselves.

One key characteristic of AMVA is that steps two, three and four—often combined in electronic voting systems—remain separate and independent. By separating vote generation from vote casting, officials can continually improve the vote-generation interface without affecting the actual vote-casting, which needs to be standardized. The separation makes it easy to customize the interface for different individuals—perhaps making it multilingual to accommodate new citizens or adapting it for blind voters. "The current user interfaces are pretty bad," Ansolabehere says. "Because IBM, Microsoft, Xerox and others have stayed out of this industry, very little of the interface design ideas that have happened over the last 10 years have been imported into this business. It's a little like voting on 'Pong.'"

Another benefit of the separation is that election administrators would be able to purchase equipment from competing manufacturers, making them less dependent on one company. The authors suggest that in the case of an electronic system, the source code for the counting machinery should be freely accessible, allowing manufacturers to develop different interfaces for the vote-generation process. To give for-profit companies financial incentive, the software for the user interfaces could be proprietary, ensuring a return on investment.

So what might such a multipurpose voting machine look like? The scientists say that a regular PC—equipped with specialized software and peripherals to read FROGs—would suffice. The PCs would be affordable so that voting districts could purchase new equipment every four years. And after the election, manufacturers could refurbish the PCs for use in schools or public administration. "Right now the equipment is bought and stored and sits, so it's this very infrequently used equipment," Ansolabehere says. "And most electronic equipment that's coming on the market is just Windows-based machines with a touch screen attached to it, so it's really a waste to let that equipment sit and not let school kids use it afterwards." —Harald Franzen