Ted Selker,

MIT director, Caltech/MIT Voting Project

20 Ames St., Cambridge MA, USA 02139

selker@media.mit.edu

617-253-6968

1/11/2008

Regarding the January 6, 2008 New York times Magazine article on voting technology commentary.

Voting machines preside over the most powerful of decisions, the direction of our government. No wonder we care about making sure they operate perfectly. Vendors of election equipment have to make dependable devices that serve a steady stream of hundreds of concerned voters a day so that back-end counting computers can add these votes to a river of hundreds of thousands of in a district. As Clive Thompson points out in his New York Times Magazine piece, "Can You Count on Voting Machines?" (January 6, 2008), the devices have to conduct this process with the dependability of a banking system while maintaining the privacy of each individual voter.

Clive Thompson works hard to point out some of the problems inherent in voting systems, and he brings together many valuable perspectives about the state of voting technology right now. However, in many cases, his story draws unfounded conclusions based on information that's one-sided or sometimes flat wrong.

For instance, Thompson disparages voting systems that display votes on a screen (known more technically as Direct Record Electronic voting machines) in favor of optical scan voting systems. Both types of systems have more than a third of the market share each in the US so voters might be familiar with them. In the first system, you make choices by selecting regions of a large computer screen. In the second, you mark a piece of paper that is scanned by computer to capture each vote. Thompson's main point has to do with the handy paper copy accompanying the optical scan process. Optical scan systems produce a fat stack of paper ballots, the actual pages marked by voters, to audit, recount, or verify.

But there is a problem with verifying votes with paper if you dare to look deeper. In the US we use so-called "mark-sense" scanners, which don’t check the candidates' names printed on the ballots. The ballot mark positions are matched with a separate ballot module. When the computer compares the paper ballot with the electronic ballot module, things can sometimes go awry. A ballot with minor folds can throw the system off, adding phantom marks or changing the position so that the selection becomes invalid or goes to the wrong candidate. Additionally the ballot may be scanned by a machine using the wrong ballot module, or a ballot module may itself have an error in how it interprets ballots. Yes, a stack of pages at the end of an election does serve as a good record of voter intention. But essentially, we're left with the same dilemma as with the touch screen systems -- Regardless of what the page says, did the voting system ultimately mark the correct vote on the memory card? The stacks of paper get processed by human hands, often without enough supervision. In many recounts, such as the famous governor's race in King County, Washington in 2006, new ballots were miraculously “found” at each recount.

There is a solution. An optical scan system used in several countries takes optical scan one step further, capturing the actual ballot itself with the names of all candidates and races using Optical Character Recognition (OCR). Such systems often save an image of the ballot, so that when the ballot is viewed again, any possible changes to it can themselves be found. Until we use such equipment in the US, the scans will not be able to be audited and will continue to be prone to many types of problems (counting folds, misalignments of timing marks, and allowing malicious or accidental ballot module mistakes to throw the count).

Although Thompson's reporting has lead him to believe that optical scan systems are superior, he also writes that all voting systems -- including optical scan -- are problematic because of their underlying reliance on computer tabulation. He summons quotes from technologists saying that computer tabulation can cause unpredictable errors, and ones on an entirely different scale than human hands. It's not clear how to argue this point. For anyone aware of common practices in the industries of banking, supply chain, ticket sales, inventory, and government in the last 60 years, computers are the most dependable means of tabulating a large set of items. And on a scale that voting systems regularly deal with -- oftentimes millions of votes in one day -- errors in hand counts are indeed predictable -- predictably large.

So computers will probably continue to be a part of all massive tabulation efforts. With procedures such as parallel testing, election officials can routinely check equipment before during and after elections to prove that it is not loosing or adding votes. The real question is which front-end interface is best for collecting votes from voters. Optical scan systems have complications of having to scan each ballot correctly, in the right orientation, and against the correct ballot module. These systems also cause difficulty among voters in making their marks count; stray marks, incomplete marks, and ovals filled in too hastily can nullify a vote.

DRE voting administration must be carefully set up to test for mistakes and fraud. But as Thompson's article points out, printer reliability has not been up to the task of creating reliable paper trails. And even if the paper trail systems were better designed, would more busy voters take the time to double check over dozens of selections on paper? Thompson’s article agrees: voters are not able to appreciably assist in this audit because few voters look at and fewer can interpret the paper trail to verify their vote.

With or without paper trails, DRE systems have come to provide the clearest ballot presentation and best feedback for voting systems. The devices alert voters when they attempt to vote for two candidates in the same race or when they leave a race blank. (With apologies to Thompson, the biggest problem in Palm Beach County in 2000 was not the sensationalized 609 hanging chads, but some 19,000 ballots containing marks for two presidential candidates and some 10,000 containing no marks for the office.) DRE systems have the advantage of a large, clear surface on which to design ballots. Well designed electronic ballots break down large elections into smaller sections; they orient voters as to what actions to take; they differentiate races clearly as voters move through the ballot; they give the best second-chance feedback of any voting equipment when voters deposit their ballots. Because races can be better laid out, and candidates and their respective selection buttons moved closer together, we have found that voters make fewer errors in which a candidate is picked adjacent to the actual choice. Well designed DRE experiences give the voter a chance to review choices at the end using carefully laid out review screens. In our experiments, more than .5 percent of voters flip their votes on all voting systems we have studied. Yet they tend to notice and fix them with the active feedback of a touchscreen display. In an analysis of missing election results by political scientist Charles Stewart in 2004, Stewart shows that with optical scan machines as much as .9 percent of voters did not a make selection for president. In the four states using computer feedback interfaces, less than .4 percent of voters left the same selection blank. This .5 percent difference across the country represents more than 500,000 votes that could have been added with the help of the structured touchscreen interface and feedback that electronic voting systems provide.

Thompson underestimates the potential of DRE interfaces, but he makes a few glaring factual errors in the article that make it all the more dubious. For example, the article states that electronic voting started in 2000 after the problems in Florida and other states. DREs were first introduced starting with the Video Voter in Illinois in 1975, and DRE use expanded to almost 27 percent by 2000.

Instead of chicken-little arguments calculated at fueling an exciting article, it is better to take the time to think through the tough and technically challenging issues around voting. Vulnerabilities exist in all voting systems whether they be hand-counted ballots, optical scan ballots, or touch screens. And all voting technologies need the proper administration to succeed -- thorough software testing, thorough training, two-person oversight, auditing, and planning. DRE system concerns have been glamorous to report on in the last few years, while greater problems of more mundane approaches -- like mail-in voting, optical scan, Inkavote, punch card, and hand count -- have flown under the radar. In Thompson's article, it's not clear why Ohio’s inability to train pollworkers or test their incoming machines becomes a statement about Diebold's poorly written manuals. Maryland achieved a .3 percent residual with the same machines years earlier.

There are many improvements that can be made to all these technologies: optical character recognition can improve security, accuracy, and auditabilty of optically scanned ballots; better practices around not folding ballots in the postal mail would ensure that more optically scanned ballots are recorded correctly; simply offering voters a wooden ruler to place under each line as voters mark optical scan ballots would save significant votes. And, yes, better DRE architectures, interfaces, and verification methods would ensure seamless DRE voting. Many voting places have changed technology twice since 2000. This isn't a time to swap out voting machines again; it's a time to improve the way we check our registration systems, design our ballots, and run our polling places -- currently, the top three ways that votes are lost in United States elections. Oh, by the way, one primary cause of low voter turnout every year is strong skepticism from voters that their votes will matter. With steady improvements in voting systems, voter turnout will improve too. I believe that in 2008 we are all much more vigilant than ever and will have the most accurate election ever held.

Sincerely

Ted Selker