How many Diebold voting machines failed in Pennsylvania?

What follows is a post written in the week after the election. Because it concerns an anomaly in election results that were then provisional, I delayed publishing it. As it turned out, the final results erased that anomaly – but in a way that suggests my original concerns may have been on target. An update at the end explains what I mean by that.

Are you certain that your vote was recorded and counted properly on November 4? If you live in Pennsylvania your answer may be "not at all".

On primary day in April, I happened to hear a poll worker comment that the Diebold AccuVote TSx machines at our precinct in Lehigh County (PA) aren't very credible. So I asked whether she trusted paperless DRE (Direct Recording Electronic) machines. She didn't. As it turned out neither did any of the other poll workers there - nor the other voters waiting in line. Nobody seems to trust them. That in itself is reason enough to get rid of these damnable black box systems. Without trust, what is left of democracy?

Ultimately there's no way of knowing whether those machines counted votes accurately in the April primary, or whether those votes were tabulated correctly. The very technology makes accountability impossible. As if that's not bad enough, the (ironically named) AccuVote has a history of flawed performance. What's more it can be hacked rather easily.

In fact, I see evidence that on Election Day one or more of these touch-screen machines in a local precinct may have failed to record as many as a few hundred votes in the presidential race.

The problem in question concerns the rural township of Lowhill, PA, where 767 voters (52%) are registered Republicans, 489 are Democrats, and 215 are Independents. Turnout on Election Day was at least 80%. The (still unofficial) vote totals in Lowhill for all the state-wide and local races ranged between 1117 and 1178. Even a bond referendum generated 1023 votes in the town.

However only 950 votes are recorded for president in Lowhill. That's 228 fewer votes than were cast in the (non-competitive) Congressional race. Thus the undervote in the presidential race in this town was extremely high (over 19%). By contrast, in neighboring rural towns with similar Republican majorities the normal pattern prevailed – that is to say, the presidential race received more votes than any of the down-ballot races did. I've not found any other precincts in the region where the presidential race this year generated fewer votes than down-ballot races, so the result in Lowhill is highly anomalous.

It looks like the entire undervote there is to be associated with John McCain. Clearly something's amiss with McCain's vote total, since he got only 15 more votes than Barack Obama in Lowhill (Obama 461, McCain 476). Democratic candidates simply do not come within 15 votes of winning Lowhill. In this century only a single Democratic candidate has lost by fewer than 120 votes in Lowhill Township (Ed Rendell in 2006, who lost to the hapless Lynn Swann by just 34 votes). To judge by the Democratic vote tallies in other races in Lowhill (past and present), and by the results this year in neighboring towns, 461 votes for Obama is about what you'd expect. However McCain's 476 votes in Lowhill is about 250 votes below what I would have anticipated based on these same factors.

Significantly, the estimated 250 vote deficit for McCain in Lowhill is also the entire amount of the undervote in the presidential race. Thus any 'missing' votes in Lowhill would have come from McCain's column.

It's possible that the peculiar vote tally in Lowhill was due to clerical error. Perhaps McCain's 476 votes should instead have been 746? A few days after the election I called the Lehigh County clerk's office to alert them to the anomaly. They refused to comment until official results have been certified. I've heard nothing from them since then.

It's also just barely possible that in this single town, for some unknown reason, vastly fewer Republicans supported McCain than in the area generally. The median income in Lowhill is higher, and it's true that national exit polls show McCain's support fell off by 3-4% as voters' income rose above the $200,000 mark. However that factor isn't remotely sufficient to cause the truly stark discrepancy between the recorded and the projected vote for McCain in this town. And in any case why would so many people cast no vote at all for president?

The Diebold DREs used in Lowhill are error-prone. Elsewhere in Pennsylvania, the same type of machines also malfunctioned so badly that an hour after the polls closed Judge Charles Saylor ordered them to be impounded – at the urging of both Democratic and Republican officials. Wired Magazine's Kim Zetter explains why all 185 of one county's AccuVote machines were impounded.

Touch-screen voting machines used yesterday in Northumberland County, Pennsylvania, have been impounded on a judge's order after voters experienced problems on the machines when trying to vote a straight-party ticket.

County poll workers discovered around 7:30 am Tuesday morning that voters who chose to vote a straight-party ticket could not see their selections on the summary review screen. The summary review screen allows voters to verify that the machine has registered their selections accurately before they cast their ballot.

Northumberland County uses AccuVote paperless touchscreen machines made by Premier Election Solutions (formerly Diebold Election Systems), which are supposed to display the chosen candidate's name on the review screen. But voters who voted a straight-party ticket could see only a message saying they had voted a straight-party ticket.

In short, the AccuVote machines that had been certified ready for use were instead malfunctioning in a consistent pattern that cannot be attributed to voter error. It's unclear so far whether the DREs in Northumberland County tallied any votes incorrectly, or perhaps caused frustrated voters to 'correct' their vote in such a way as to nullify their intent. What matters in the end is that the machines are demonstrably unreliable – and, since they lack a paper trail, any malfunctions are difficult if not impossible to reverse after the fact.

And let's not forget that a study by computer scientists at Princeton has demonstrated that the AccuVote machine is easily hacked.

We analyzed the machine's hardware and software, performed experiments on it, and considered whether real election practices would leave it suitably secure. We found that the machine is vulnerable to a number of extremely serious attacks that undermine the accuracy and credibility of the vote counts it produces.

Computer scientists have generally been skeptical of voting systems of this type, Direct Recording Electronic (DRE), which are essentially general-purpose computers running specialized election software. Experience with computer systems of all kinds shows that it is exceedingly difficult to ensure the reliability and security of complex software or to detect and diagnose problems when they do occur. Yet DREs rely fundamentally on the correct and secure operation of complex software programs. Simply put, many computer scientists doubt that paperless DREs can be made reliable and secure, and they expect that any failures of such systems would likely go undetected.

Not to put too fine a point on it, if you aren't concerned yet about the integrity of elections in the US, then it's about time that you were. Paperless DREs are subject to any number of problems that make their vote tallies suspect. Diebold in particular has worked hard to earn every bit of its public notoriety.

Indeed, in October the elections officials in Ohio learned that the PES (Diebold) vote tabulators used in conjunction with their unreliable DREs are also prone to malfunction. Kim Zetter again:

Several election watch dog groups have sent an advisory to election officials warning them about a problem with Premier Election Solutions' vote tabulating software that could cause the system to lose votes.

Premier (formerly called Diebold Election Systems) disclosed the problem in August after officials in Butler County, Ohio, discovered that 150 votes were dropped from a memory card during the state's March primary. Ten other Ohio counties discovered their system had dropped votes as well when vote totals on the memory card were uploaded to a county server. The problem occurred when officials tried to upload multiple memory cards at once.

All of the votes were recovered, but Ohio officials had to expend considerable time and energy to retrieve them and make sure all were accounted for.

The flaw is in Premier's Global Election Management System (GEMS), which is used in at least 31 states. GEMS software sits on a computer system at a county's election headquarters and is used to tabulate votes cast on both touchscreen voting machines and optical-scan machines. Premier said the flaw was in versions 1.20.2 and earlier of the software, though other versions may be affected as well.

After denying for some time any responsibility for the manifest problems with their vote tabulators, and getting sued in August by the Ohio Secretary of State, Premier finally admitted that it's software doesn't work properly (PDF). As a result, an alert went out in the hopes of containing the problem.

The Brennan Center for Justice at New York University School of Law sent an advisory to election officials Thursday -- along with Common Cause and Verified Voting -- urging them to follow the procedures and to conduct a rigorous post-election manual audit to reconcile vote totals on memory cards with those in the tabulation system.

Well, I hope that elections officials in Pennsylvania are conducting a rigorous audit to catch these malfunctions. Perhaps that will also suffice to correct the suspicious anomaly in Lowhill's presidential tally.

And after they're done trying to catch and correct all the mistakes caused by Premier Election Solutions' malfunctioning machines, I hope they'll have the good sense to ditch the damnable DREs, make a clean break, and find a more reliable technology to count votes - one that voters can reasonably place their trust in. At a minimum, we desperately need to adopt a voting technology that allows officials to audit and recount the vote with confidence. It's not good enough to rely upon DRE voodoo.

Update: This week Lehigh County posted on line its certified results here. In nearly every race in all the local precincts I'd examined earlier, including the races in Lowhill, the final tallies for every candidate were within 5 votes of the preliminary tallies posted immediately after the election.

However the vote tally for John McCain in Lowhill Township rose by 254 votes, from 476 to 730. The new total is approximately where I suspected it should be. Obama's total in Lowhill rose by only 4 votes (461 to 465).

So what happened? I'd hoped it was just a question of clerical error in transcribing or entering the completed vote tally. But given the McCain numbers before and afterwards, that doesn't appear a likely explanation. A problem with one or more of the 5 AccuVote machines used in Lowhill seems much the more likely explanation for the huge initial undervote.

If so, it may seem reassuring that elections officials were able to catch and correct a loss of that many votes. But on the other hand, when I discovered the anomaly I did press a County election official to investigate it carefully. So it's possible that their normal post-election certification process would not otherwise have caught a problem of this type.

In any case, the correction of the undervote does not reassure me that the technology in use is credible. Quite the opposite, it strengthens my concerns about Diebold's AccuVote machines.

If I'm able next week to get the Lehigh County clerk to explain how they discovered the error and what caused it, I'll post the information at this site.

crossposted at unbossed.com