Jamie Hester, Intern, AEI-Brookings Election Reform Project
May 7, 2008
Election officials in Union County, New Jersey, are learning the hard way that the widespread implementation of technology can present numerous difficulties for administrators and poll workers alike.
On the evening of February 5, the day of New Jersey's primary, Union County election officials collected the printouts from their AVC Advantage direct recording electronic (DRE) machines, and noticed discrepancies in the vote totals. On one machine's printout,
for example, Barack Obama had 182 votes and Hillary Clinton had 179, for a total of 361 votes. This total should have perfectly matched the machine's "Operator Switch Total," a count of how many times each party's ballot had been activated; instead, the machine indicated a total of 362 ballot activations for the Democratic Party, begging the question of whether one vote had
disappeared. The Republican Party totals also had a similar discrepancy of
one vote.
Puzzled and wary of falling victim to the voting machine malfunctions that have plagued other states, the election officials contacted Edward Felten, a computer scientist at Princeton University, to review the results and conduct his own investigation of the source of the error. Sequoia Voting Systems, the vendor of the AVC Advantage machines that produced these incongruent results, quickly announced that it would not tolerate the independent review, claiming that it would be a breach of the company's contract with Union County and would infringe on its
intellectual property rights by requiring the disclosure of protected information. The company pointed to its cooperation with third party reviews by two separate test laboratories, one of which is
accredited by the Election Assistance Commission, as evidence of its reliability. Union County officials rescinded their request and the independent review did not take place.
Sequoia argues that
poll worker error is partly to blame for the inconsistent vote totals. Each political party was assigned one of twelve numbered buttons on the operator panel. For each voter who arrived in the polling place, a poll worker pushed one of the buttons to activate the appropriate ballot. In Union County, the Democratic Party was
assigned button number six while the Republican Party was assigned number twelve. Poll workers pushed either the six or the twelve on the panel for each arriving voter based on that individual's party affiliation (New Jersey has a closed primary). However, if the poll worker pushed the incorrect button and then corrected the mistake, the Sequoia voting machine would not properly classify the vote as "Democratic" or "Republican." Sequoia maintains that the votes for the particular candidates were recorded correctly, and, thus, that the problem was with the total number of votes matching a sum of the candidates' votes and not with the
overall accuracy of the vote totals.
However, without an independent review, it is difficult to be certain that this is the full explanation. New Jersey, which uses the AVC Advantage in 18 of its 21 counties, has
no documentation of this potential poll worker error occurring previously, much less in as many as five different counties during the same primary. According to Professor Felten, as many as nine voting machines in Union County alone showed
incompatible totals.
If Sequoia's explanation is to be fully accepted and improperly pressed buttons are responsible for the inaccurate ballot activation totals, the focus of attention shifts, but blame remains with the vendor; the problem is now simply one of usability, rather than security. The poll workers may have made errors, but a well-designed machine should not be flummoxed by a simple error like a poll worker pressing the wrong button. Whether this flaw is inherent in the source code of the software, the placement of the operator panel, the default settings of the software, or all of the above, the voting machines failed to perform at an acceptable level of functionality.
To avoid similar problems in the future, states should strongly consider vendors' public transparency before purchasing voting machines. Why does Sequoia, or any large voting machine vendor, eschew public scrutiny of their source codes? The company did submit its voting machines and source code to two independent reviews, but those
reviews were done by contractors solicited by the company. It would behoove the industry to allow more independent examination of its products to assuage public concern over the accuracy of the vote counts and prove the reliability of its own voting machines. The more comfortable the voters and election officials are with the security and usability of voting machines, the more competitive that particular vendor will be in the larger market. New Jersey officials might have discovered the cause of vote discrepancies in the February 5 primary, but that does not mean that the underlying problems of lack of transparency and machine functionality have been sufficiently addressed.
Jamie Hester can be reached at politicalcorner@aei.org
Viewpoint is an occasional feature analyzing various election reform issues.