Vote “Flipping” on Hart InterCivic eSlate Systems

In the news, we’re seeing a number of reports concerning “vote flipping.” The story, as typically reported, is that voters are attempting to vote for one candidate but observe that the machine “flipped” their vote to the other candidate. For touch-screen voting machines, the most likely cause of this issue is miscalibration of the screens (or, perhaps, a voter who is significantly taller or shorter than the person who did the calibration, since different angles of view require different calibrations). I wrote a detailed explanation of this issue two years ago.

UPDATE: Barbara Ballard, a usability expert with Little Spring Designs offers some excellent advice about how to configure touch screen button layouts to minimize or eliminate the parallax issues that seem to induce or exacerbate vote flipping.

UPDATE 2: Matt Blaze writes how miscalibration of touch-screen voting machines could be used as a mechanism to disenfranchise voters.

A related issue concerns reports of vote flipping on the Hart InterCivic eSlate voting machine. These machines do not have touch-sensitive screens, so therefore poor calibration cannot explain the voter confusion. Since my home county uses eSlates, I went to vote early, this morning, and paid careful attention to how the user interface works. For those unfamiliar with eSlates, the voter’s primary interface to the machine is a dial-wheel and an “Enter” button, which operates in a manner that would be quite familiar to users of Apple’s iPod. You turn the wheel and it highlights successive entries. You press the Enter button and it indicates your selection graphically. Using some HTML tables, I’ve attempted to recreate the salient details below.

First, here’s an approximation of what a normal voter will see when they start off. Note that I’ve simplified a few things, and I’ve replaced the actual instructions with some fake text. It’s sadly safe to assume that most voters won’t bother to read the instructions, anyway. For purposes of explaining the user interface, I’ve left out our Libertarian candidate for U.S. Senate, Yvonne Adams Schick, and any write-in boxes. My apologies, but this is already pretty complicated. (Also, I didn’t debug this against every web browser. If you’re using Firefox or Safari, you should see everything as intended. Hart also has an online Flash demo, slide show, and videos that broadly show how an eSlate works, although they don’t include straight-ticket voting.)

The blue background highlighting tells you where you are, and when you start, you’re on the instructions. As you turn the wheel, it highlights “straight ticket” and then “Republican party” and so forth. Let’s say our voter wants to vote straight-ticket Libertarian.

Several things just changed. Every race with a Libertarian candidate now has that candidate selected with a dark red box, and the other candidates have their visual contrast reduced. Also, the blue highlight block has now advanced to the header for the presidential race. Also of note, if there’s a race without a Libertarian candidate, such as my simplified U.S. Senate race, those candidates will still have the full black&white contrast. At this point, the voter would most commonly scroll past the presidential race and indicate preferences in other races. If the voter decides to “split the ticket” and change their presidential preference, perhaps by selecting John McCain, the whole screen goes blue and displays a message in large, white letters indicating that they just split their ticket. (“Press Enter to Continue.”)

Okay, now that we have a basic understanding of the eSlate UI, how could this be causing voter confusion? The Houston Chronicle reported:

In response to reports from other locations that electronic voting machines were reversing votes cast in the presidential election, [County Clerk Beverly Kaufman] said, “We tested those aspects of the system in a formal way with party officials and that did not occur. I still cannot urge people enough to check your work and keep your focus on what you’re doing.”

Her office was informed early today that some of the first voters had cast straight-ticket Democratic ballots and then discovered that the electronic machines listed them as voting for John McCain in the presidential election.

Hector de Leon, spokesman for the County Clerk’s Office, said problems such as those reported with the identification scanning system likely reflected poll workers’ inexperience in operating the new equipment.

Ahh, inexperience. Let’s try to sort out what might have happened. Let’s assume that some of the voters in question didn’t read the instructions. Let’s further assume that these voters are not regular computer users, so the visual cues to indicate choice selection (the dark red square, the low-contrast colors for non-selected candidates) do not speak to their experience. After our example voter cast a straight-ticket Libertarian vote, they might not understand that they had selected Barr already and would miss the graphical cues that confirm this. Instead, they would scroll down to Barr and press the Enter button. This will generate the full-screen warning message, deselect Barr, and leave the blue highlight on Barr for the voter to make a subsequent choice, including re-selecting him. This sort of confusion is entirely feasible on an eSlate.

Two years ago, we heard concerns about voters “bouncing” on the Enter button, and problems which may have resulted, so I paid careful attention this time. After the voter makes a selection, the highlighted area advances to the subsequent race title (“President”, “U.S. Senate”, etc.), which cannot actually be selected. This, at least, gives us some confidence that a voter who accidentally presses the Enter button multiple time will not accidentally indicate multiple selections in a series of races.

As an aside, some eSlate units support a “Disability Access Unit,” adding headphones that work in a relatively intuitive fashion. The voting machine reads out whatever text is highlighted. In this way, a visually impaired voter will always know “where they are,” and navigation is strictly one-dimensional, going forward or backward from one area to the next. It’s reasonable to recommend that voters with perfectly good vision but who have limited computer experience also use the headphones. It may well help reinforce the navigation and selection model of the eSlate and reduce voter concerns about “vote flipping.”

Of course, we could also implore voters to pay attention to the summary screen. For my ballot, which I cast this morning, the summary screen was actually three separate pages listing my selections. Races in which I abstained to offer a selection were highlighted in red text. Will this help voters detect mistakes? It will for some voters, but many voters won’t carefully read their summary screen. At Rice, we did some experiments on this, building a voting machine that introduced deliberate errors into the summary screen. Would voters notice? Unfortunately, our research shows that as many as 63% of voters fail to notice errors on the summary screen.

In the future, there are number of ways Hart InterCivic could improve this system. They could replace the dark red square, as an indication of a voter’s selection, with a symbol that is clearer to non-computer users, perhaps a large, green check-mark. The EAC’s Effective Election Design document has a number of good recommendations along these lines.

Likewise, many voters will simply not understand the concept of straight ticket voting. We encourage states that require straight ticket voting, such as Texas, to eliminate the practice altogether. This would be particularly helpful in a state like North Carolina where, by law, the straight ticket voting option affects every race except the presidential race. That’s not at all intuitive.

In conclusion, while we have no reason to believe that voting machines are actually flipping votes, the issues underlying these concerns are very real. We should not blame voters when the real problem lies with poor usability engineering of electronic voting machines, whether that means calibration and angle-of-view issues on touch-screens, or non-intuitive navigation and selection models on eSlates.

[The author thanks Mike Byrne, Marti Hearst, Doug Jones, Sharon Laskowski, Peter Neumann, Whitney Quesenbery, Dan Sandler, Pamela Smith, and Dave Wagner for their comments on drafts of this article.]