This document reviews the four voting machines displayed on March 15, 2005 by the Office of the Cook County Clerk and the Chicago Board of election Commissioners. The machines were chosen as finalists following a joint request for proposal by the two entities. This reviewer spent significant time with each vendor and attempted to cast a ballot using each of the machines. This review also includes published articles, reviews, and research on the machines in question.
Unfortunately, if any one of the four machines were to be deployed in Chicago or suburban Cook County as exhibited on March 15, many voters with disabilities, particularly blind voters, would not be able to cast a ballot independently and privately. This analysis was confirmed in a usability study conducted in the July 2004 issue of Access World, a publication of the American Foundation for the Blind. In the survey, 13 blind and low vision users attempted to cast ballots using the iVotronic from Election Systems and Software, the AVC Edge from Sequoia Voting Systems, the eSlate from Hart InterCivic, the Vote-Trakker from Avante International Technology, and the AccuVote TS from Global Diebold. All of the machines demonstrated on March 15 were part of the survey in Access World. The survey found that when the participants tested the audio voting features,
The likely cause for the difficulties experienced by blind users on these voting machines is the minimal and unsubstantive involvement of blind persons and those with disabilities in the design, development, and testing of the machines, their software, and their interfaces. Many of the problems encountered can be greatly minimized, reduced or eliminated through modification of existing audio instructions and audio prompts and in some cases adding prompts and instructions at certain key points in the voting transaction chain. However, this is not to say that on all of the machines script revision will largely correct inherent access and usability problems. These issues will be described below and the particular machines to which they applied.
Only the voting machine from Diebold could be used essentially out of the box as demonstrated. Even with this machine though, a blind end user can become stuck when changing a vote because none of the options presented offers the opportunity to continue reviewing the ballot in the final stage before casting a vote. A Diebold representative provided information about a key function not part of the audio script, which enabled an immediate return to the ballot review. While changing this prompt would be relatively simple to accomplish, the error still demonstrates that no matter which machine is chosen, some or a substantial amount of work will need to be undertaken by the election authorities in Chicago and suburban Cook County to analyze the vendor's product and ensure that the interface communicates effectively to people with disabilities.
Tactilely Discernable Controls
For many voters with disabilities, the operating controls should be able to be located and identified by feel. Of the four machines demonstrated, three had tactilely discernable controls. Only the eSlate from Hart InterCivic had controls that could not be located by feel. All of the controls on the eSlate with the exception of the selection wheel were flat flush with the fascia and of the same composition of surrounding material. With blind voters having difficulty locating various operating controls, problems in casting ballots or errors in voting will likely be the result.
Further, controls need to be able to be operated with one hand. The selection wheel was of a small enough size that it required fine motor control of the fingers and wrist motion to operate effectively. While an adaptation is available so that ballot items can be selected and cast without the wheel, the adaptation seemed not to be a permanent part of the machine design. Many would consider the adaptation to be assistive technology, making it a less desirable solution than those access approaches integrally designed into the machine itself.
The importance of tactilely discernable controls can be found in the accessibility guidelines for Section 508 of the Rehabilitation Act as amended in 1998. Congress enacted Section 508 to create greater accessibility with technology in federal programs and services and to broaden the market for accessible electronic and information technology. The 1998 amendments directed the United States Architectural and Transportation Barriers Compliance Board, also known as the Access board, to develop guidelines for technology and technology-based services that the federal government purchases. Leaders from the disability community, government, and premier technology companies worked cooperatively to develop standards and recommendations which were largely incorporated into the Section 508 accessibility guidelines adopted by the board. Hence, the guidelines represent to a significant degree the prevailing consensus on the means and feasibility of implementing technology accessibility for people with disabilities.
Electronic voting machines would be identified in the guidelines as “self contained, closed products,” which are defined as “products that generally have embedded software and are commonly designed in such a fashion that a user cannot easily attach or install assistive technology. These products include, but are not limited to, information kiosks and information transaction machines, copiers, printers, calculators, fax machines, and other similar types of products.” In the guidelines, Section 1194.25 pertains to self-contained closed products. Subsection (c) states that “where a product utilizes touchscreens or contact-sensitive controls, an input method shall be provided that complies with §1194.23 (k) (1) through (4). This subpart states:
A successful implementation of an audio interface for people with disabilities includes a script that provides prompts in appropriate decision points, provides all the information necessary to make a decision or complete a task, and that describes as well as states information.
As mentioned in the introduction, the machine from Diebold offered the best script and prompts. After each selection, the user was read back the choice and told what key to press to confirm that choice. If the selection was not the end user’s choice, the machine also identified the key to change the choice. This system rhythmically moved the end user of the audio ballot through the voting process to the end where casting a vote was simple and easy.
Although much less developed, the iVotronic from Election Systems and Software has much potential if audio prompts were added in some key areas and some others were re-written. For example, after the end user has cast a vote in a particular contest, the system confirms the vote but it fails to instruct the end user as how to use the machine to advance to the next contest and cast a vote in that contest.
The logic of such a design typically is based on the assumption that a thorough orientation is sufficient to effectively communicate how to use the machine. My experience has found that this is not enough for blind users. Orientations effectively provide information about location of machine elements and controls. They also do well in helping the end users with a disability understand how a transaction will be conducted and the various keys used to conduct basic and frequent transactions. Unlike transit fare card machines or automated teller machines, voting machines will be used at most twice a year and likely on fewer occasions by many voters. This makes the specific details, such as key sequences and functions, unfamiliar to most voters with disabilities, requiring continual reiteration of information throughout the voting process.
The Hart InterCivic machine had more systemic issues with missing scripts, omitted information about the location of controls and a lack of prompting after a voter had voted in a contest so the voter knew what to do next to advance to the next race. The machine designed assumed the user was familiar enough with the interface that it would seem obvious what the next option was. This highly minimal approach was explained by the Hart representative as one intended to not load the blind end user with too much information. Yet, the machine design allows for interruption of the audio when the end user presses a key or moves the selection wheel, so if no more information is needed the end user can easily and quickly move on without waiting to hear the full audio message.
When the representative was asked about usability studies and how this belief was developed about the needs of the typical blind person in the community, it turned out that the conclusion was drawn from speaking with a handful of blind persons who were extremely sophisticated and talented users of technology. By contrast, a 2003 study by the national Library Service for the Blind and Physically Handicapped, which provides Braille and talking books to people with disabilities, found that only about half of all users have ever heard speech synthesis with a much smaller number having used an adapted computer.
The Section 508 guidelines for self-contained, closed products also contain provisions that audio prompts need to be interruptible. All of the machines did this, so messages that end users do not want to hear can be skipped at any time by pressing a key to advance the voting process. Further, this reviewer had the impression that changing audio prompts on the Hart InterCivic machine would require permission and approval from the manufacturer as to their form and content. Representatives from the other three manufacturers said that local election authorities can modify or add prompts at their own discretion.
If the position described by the representative from Hart InterCivic is accurate, it represents a significant loss of control by local government of the voting process to a private entity that is unaccountable and may be unresponsive to community needs. Currently, local and state election authorities have control over the design, layout, format, and content of ballot books for elections. Local election authorities should have the same control over ballot content and presentation of information as they do with the paper ballot book so adjustments can be made to best meet the needs of the community. Assertions of intellectual property have no place when simply changing a few words of an audio prompt is involved so voters with disabilities can vote independently without assistance or intervention.
The AVC Edge from Sequoia Voting Systems had significant scripting problems as well. The voting process actually could not be initiated by this reviewer because the instructions failed to inform the audio end user to press a certain key twice. As with the ES&S and Hart InterCivic machines, many prompts and instructions were either totally inadequate or omitted crucial information about how to conduct a particular function. An additional problem with this machine was too much information in a frequently played prompt. After the voter had cast a vote in a particular race, the machine played a long and complicated message that could be highly simplified for much greater comprehension.
Only the Diebold and the Hart InterCivic machines provided a visual display with the audio ballot. A Sequoia representative said that the feature is not available in currently shipping machines but will be available in the next release. ES&S only delivers a dark, blank screen with its audio ballot and the representative at the demonstration said there are no plans to offer a combined visual and audio display. Only the Hart InterCivic incorporates the audio ballot and visual display together as an integral part of the unit. Diebold and Sequoia require poll workers to enable this function. There is no prompting of end users asking them if they want the screen turned off. The Sequoia machine though prompts audio users to choose a language, either English or Spanish. A Sequoia representative could not explain the basis for the different design decisions that led to the differing approaches.
People with disabilities must either know about this feature in advance and specifically ask for it or be asked directly by a poll worker if they want the screen on or off during the audio ballot. As people age, there is an increasing likelihood that they will develop a disability, with incidence of disability rising sharply with the aged. Many older persons are self-conscious about perceived declines in health or ability and are likely not going to publicly make accommodation demands for ballot presentation or engage in extended conversation about their needs beyond that of requesting use of the audio ballot.
This analysis led Greg Vanderheiden to develop a model voting machine that is accessible and provides audio output to all. Dr. Vanderheiden is the director of the federally funded Trace research and Development Center at the University of Wisconsin at Madison. His center is the only one in the federal network of rehabilitation engineering and research centers that focuses on access to information technology. His model designs for accessible fare machines, talking cell phones, voice guided ATMs, and kiosks with audio output have been utilized by leading companies and institutions, including Wells Fargo, Bank of America, Citibank, Bank One, Cingular Wireless, the Smithsonian, and the United States Postal Service. His groundbreaking accessible design of a touch screen that can be used by the blind without a keypad is at 69 W. Washington as the Citibank talking ATM on the lobby and concourse levels, where the two election authorities in Cook county are located. In the Vol. X No. 2 December, 2004 issue of the Journal Information Technology and Disabilities, Dr. Vanderheiden describes his development of the model voting machine interface:
Additionally, Diebold is fully aware of the mainstream benefits of the dual interface. Diebold piloted three ATMs in New Orleans during 2001 and 2002 that dispensed money orders. The July 11, 2002 issue of ATMmarketplace.com said the machines feature audio prompts that guide consumers step-by-step through the unfamiliar money order transaction. The director of business development for Diebold, Tonya Myers-Jordan, explained "we're finding that they're listening to the audio instructions more than they are looking at what's going on at the terminal." Transaction volumes were four times the amount anticipated. Ms. Meyers Jordan said the pilot “has wildly exceeded our expectations.”
Clearly then we have one of the voting machine companies (Sequoia) offering a lower standard of flexibility and definability to its interface for people with disabilities when compared to those with limited English proficiency and another company (Diebold) who seems not to have incorporated successful market tests into its designs.
Poll Worker Assistance
A common refrain from the representatives from the voting machine companies when errors and omissions were found in the interface was that in an actual election “a poll worker would help.” The experience of many people with disabilities in receiving accommodations from mainstream workers is that beyond routine, basic or standard services, detailed, complicated, or highly individualized accommodations often are performed poorly, not at all, or in a way that does not lead to a desired result. Whether the service provider is a bus driver expected to call stops, branch bank employee expected to read out loud deposit and loan agreements, or a restaurant worker expected to provide different seating options or waive delivery charges, trying to compensate for inaccessibility or inadequate systems for delivering and providing access rarely yields adequate and sufficient accessibility to people with disabilities. Beyond turning on a voting machine and orienting a disabled user to the machine’s controls, troubleshooting and problem diagnostics may be beyond the level of training and preparation of many poll workers.
Further on three of the four machines demonstrated, the default audio ballot display is that of a black screen, making assistance from a nondisabled poll worker effectively unfeasible. Even if poll workers could view the screen or listen to the audio output, in instances of high voter turnout, such as the November 2004 election, the attention and time of poll workers may be limited. Voting machines should be viewed as self-contained stand-alone systems to be used without the assistance or support of a nondisabled person. Poorly written scripts or audio prompts, bad interface design, or an inadequate physical layout of controls should be addressed and the voting machine made fully accessible and usable by people with disabilities rather than the current satisfaction of machines being partially effective and relying on poll workers to bridge the accessibility gap.
Volume Control and Normalization
Only the machines from Diebold and Sequoia offered the end-user volume control. The Section 508 accessibility guidelines for self-contained closed products specify user adjustable volume control in part 1194.25(F). The only machine showing difficulty producing adequate volume was the eSlate by Hart InterCivic. With all the machines, there are two kinds of audio files: interface files used by the machine from election to election and election files that are unique to the particular election. The various audio files therefore will be recorded in different environments, with different kinds of equipment at different times. Even when a speaker uses the same microphone and recording system in the same room but is just three inches closer or farther away from the microphone, the difference in sound level can be quite noticeable. To correct for the varying volume levels in a recording a process called normalizing is employed. Audio normalization is a process of determining the average volume level of several audio files and then increasing or reducing the levels of the individual files so each is heard at the same volume. The eSlate from Hart InterCivic cannot normalize files which is why there were noticeable differences in audio output levels during the demonstration. The differences were so significant that some of the audio information could not be understood. A company representative said normalization functionality will be incorporated in a future release.
The ease of confirming and reviewing choices is an important consideration in ensuring voting access for people with disabilities. Of the four machines, the eSlate from Hart InterCivic performed the best. Ballot review following voting selections was linear and audio output was fully controllable by the end user. On the single page the race was listed followed by the selected candidate. Unfortunately, when end users change their votes in ballot review, they are left in the original voting screen and need to scroll all the way to the bottom of the ballot to exit the voting screen and return to ballot review. A Hart representative said a different more user friendly presentation is usually on the machine but it was changed right before the Chicago demonstration.
The Diebold and ES&S machines confirmed vote selections as the audio end user voted, making the need to review a completed ballot nearly a redundancy. When the audio end user wishes to change a vote in ballot review, the end user is taken to the voting interface to the race to be changed and then must scroll through all of the races to the end of the ballot again in order to cast the ballot, a tedious and complicated necessity that could be better designed. As described earlier, the Diebold system works well for ballot review but the audio prompts are incomplete by not informing the end user on how to return to reviewing the ballot.
The voting machine from Sequoia functioned poorly at ballot review. During the voting process, audio end users had no control over the rate of movement through the ballot as with the eSlate from Hart InterCivic and the IVotronic from ES&S or confirmatory prompts as with the machines from Diebold and ES&S. After pressing a button, the Sequoia machine immediately advanced to the next contest with no information about what one exactly voted for and the ability to change one’s vote in the event of error. In Sequoia’s ballot review scheme, the Sequoia machine presents the voter’s selections in a continuous audio file at the end of the voting process. Computer scientist Noel Runyan, who is a trained electrical engineer, said the lengthy reading of the ballot in its entirety took 23 minutes during the November 2004 election in Santa Clara County, California.
“It took me 30 minutes to work my way through the ballots and make my selections. After that, I had quite a bit of trouble getting into the review mode, to get a full list of all my selections. When I did, it went on and on, for 23 minutes, like a long uncontrolled drink from a firehouse. The review function read each item, and then, at the very end, said what my selection was for that item. It even threw in the details of what the fiscal impact would be, and took forever,” he writes on the votersunite.org website.
He adds that “when I did find that I had made a mistake in my selections, I had to wait until the end of the whole review process to correct it, instead of being able to stop, make the change, and then continue with the review where I left off. I did not want to abort the ballot verification review, to make a correction, and then have to start the 23 minute review all over again.”
This review and those conducted by the American Foundation for the Blind, Manhattan Borough President C. Virginia Fields with The Center for Independence of the Disabled in New York, and a blind computer scientist and electrical engineer all have found that while the electronic machines represent a significant advance in accessibility from the current poll worker assistance system they often fail to effectively communicate the voting process to audio voters or are physically designed in a way that does not meet the current consensus on accessible design as crafted by the technology industry, the disability community, and leading national governmental institutions.
Fortunately, implementing fully accessible and highly usable voting technology for people with disabilities in order to cast a private, independent vote can be accomplished rather simply and easily through the modification or addition of audio prompts on some of the voting machines. Only the demonstrated machines from Diebold and Election Systems and Software are of this nature. In ranking these two machines, the Diebold machine would rate highest from this reviewer. This preference is based on the fact that the Diebold machine offers volume control and a dual visual/audio display. The ES&S IVotronic displays only a black screen in audio mode, which causes difficulty and problems in voting for elderly and low vision users as identified in the AFB, New York, and Trace Center reports. While the interface design is good with the IVotronic, the lack of a visual display with the audio ballot, nor any plans for such an enhancement, is responsible for ranking this machine at the very bottom in the ranking of the four machines demonstrated. If the IVotronic from ES&S had a dual visual/audio display and volume control, the machine would rank second.
In further regard to the Diebold machine, it needs almost no modification of existing audio prompts and no addition of prompts in various parts of the voting transaction chain would be anticipated. By contrast, the IVotronic from ES&S would likely need much more script revision to ensure full understanding and clarity of the interface. In addition, new audio prompts would need to be added to help users of the audio ballot take the next step in progressing through the ballot.
Ranking second is the eSlate from Hart InterCivic. The dual visual and audio display and the straightforward, easy to use interface positively reflect on this machine. The normalization problem described earlier will be corrected in an upcoming version of the software. Adhesive pads could be placed on the button based operating controls to make their shapes tactilely distinctive. The representative at the demonstration said the review screen could appear after the end user changes a vote in review mode, which was the design at the previous demonstration. Contractual agreements could be put in place to ensure that county and city election officials can change and modify the content of system prompts to meet community standards and expectations. While the issues seem numerous, all of them can be resolved through a planned upgrade, using a different presentation employed somewhere else, minor modification, and negotiation. The reason the eSlate would rank third if the IVotronic had a dual audio/visual display is the numerosity of the issues of the displayed machine and the extensive scripting and prompt creation necessary to effectively communicate machine operations to blind end users.
The Sequoia machine ranks third because of its complicated interface. The audio end user needs to select a contest, open the contest, and make a selection. There is no confirmation of the selection. Instead, the end user is prompted to advance to the next contest. The opening of contests and single and double button pushes adds to the complexity of the machine. Further, ballot review is impractical in Cook County with a long list of judicial candidates. An additional factor worth considering is that voting with the audio ballot with this machine for all practical purposes needs to occur sitting down. This is because the unit uses a detachable control box which cannot be attached to the machine itself. The audio end user needs to place the control box on a flat surface such as a table or counter or on their lap. The box is too large to easily hold in one’s hand like a remote control for an electronic device or a cellular telephone. The machine itself occupies most of the flat surface area in a voting booth. Further, the current shipping unit does not offer a dual audio/visual display. Even if it did, as future upgrades will, the other issues with this machine in addition to the extensive audio prompt revision necessary rank it behind other machines.
Appendix A: Darren Burton and Mark Uslan. “The Ballot Ballet: The Usability of Accessible Voting Machines.” AccessWorld, July 2004.
Appendix B: Darren Burton and Mark Uslan. “Cast a Vote by Yourself: A Review of Accessible Voting Machines.” AccessWorld November 2002.
Appendix C: § 1194.25 of the accessibility guidelines for Section 508 of the Rehabilitation Act (Self contained, closed products).
Appendix D: Gregg C. Vanderheiden. “Using Extended and Enhanced Usability (EEU) to Provide Access to Mainstream Electronic Voting Machines.” Information Technology and Disabilities, Vol. X, no. December 2, 2004 issue.
Appendix E: Ann All. “Sssh, not so loud - Money Order Pilot a Success.” July 11, 2002, ATMMarketplace.com.
Appendix F: “Noel Runyan’s experience in voting with a Sequoia voting machine in November 2004” as reported by votersunite.org.
Appendix G: “Voting Technology For People With Disabilities.” March 2003: A Report on Disabled Voters' Experiences. Virginia C. Fields and The Center for Independence of the Disabled in New York, Inc., April 3, 2003
for the difficulties experienced by blind users on these voting machines
is the minimal and unsubstantive involvement of blind persons
and those with disabilities
in the design, development, and testing
of the machines, their software, and their interfaces.
~ Kelly Pierce