After the polls closed in Stoughton, Wisconsin last November 4, workers in three of the city’s four polling stations were surprised to see their voting machines had counted no votes for a municipal referendum. A different puzzle confronted the poll workers in the fourth polling station. In a precinct where 1,255 voters had cast ballots, the output tape indicated the machine had counted 16 votes (9 no, 7 yes.)
The main problem was rapidly diagnosed as a set-up error that sent all four optical scanners looking for referendum votes in a blank section on the back of the ballot, rather than in the ovals that voters had filled in. Had the error simply looked for 'yes' votes in the 'no' spot and vice-versa, the referendum would have failed and no one would ever have noticed the mistake under Wisconsin's current, ineffective post-election audit policies, However, because the error was so dramatic, a prompt hand count revealed the true results, which were certified on schedule.
But the mystery of the 16 votes remained. It was not possible that the optical scanners could have recognized 16 actual votes and ignored the rest. Those were phantom votes, cast by no one. What caused them?
Dane County Clerk Scott McDonell originally suspected heavily marked votes on the front had bled through to cause readable marks on the back, where the referendum was the only contest. Closer inspection of the ballot layout, however, revealed this was highly unlikely. The ovals that voters fill in are placed along the left edge of each column on the ballot, where the optical scanners are programmed to look for them. Any bleed-through from ovals on the other side would show up along the right edge. A very heavy write-in vote might bleed through to the left edge of a column on the other side but on this particular ballot the opposite side contained another yes-no referendum. No write-ins there.
Another possibility was suggested by the manufacturer’s representative, who told Stoughton City Clerk Lana Kropf that the 16 votes were likely a result of ballots being sent through the voting machine while the ink was still wet. This, he said, could have caused the wheels that feed the ballot through the optical scanner to transfer ink from a real vote to another area on the ballot.
Working with that hardly-reassuring suggestion, Kropf and I, working with another volunteer from the WGN Election Integrity Action Team, examined every ballot cast in that precinct, looking for evidence for either explanation. Could we find 16 ballots on which front-side votes had bled through to the area where the misprogrammed machines were looking for votes? Or could we find 16 ballots with evidence of wet ink from a real vote being picked up and redeposited into that area?
No, we could not. We saw about four or five ballots on which bleed-through seemed to be dark enough to be read as votes on the other side of the ballot, but the bleed-through was, as we’d expected, not in any target area where the voting machine would have been looking for votes.
We also saw three or four ballots with stray marks that looked like they could have been created by ink on a rotating wheel. However, the very darkest of these marks was in our opinion too faint to have registered as a vote and there were too few to account for all 16 phantom votes anyway.
Other types of stray marks appeared on perhaps a dozen or so more ballots—finger smudges, printers ink transfer from one part of the ballot to another when the ballot was folded, a mark made by a dropped pen, and an imprint left when a voter wrote on another document while the ballot lay underneath it on a desk. None of these were both dark enough to have registered as a vote and in the correct spot on the ballot to have been counted as a phantom vote.
What then? Is Dane County in possession of a voting machine with a will of its own, capable of generating votes on its own initiative?
I contacted Douglas Jones, a computer-science professor at the University of Iowa. Professor Jones has done academic work and research on electronic elections technology and has advised several states, the US Congress, Civil Rights Commission, and Federal Election Commission, and several foreign nations on elections technology issues, in addition to consulting with organizations such as the ACLU and the Brennan Center for Justice.
Jones’ research has been surprisingly thorough. He wrote, “I always test voting machines with a random handful of pens and pencils as well as testing them with the official recommended marking device. Glitter pen turns out to work well (if you let it dry properly before scanning!) but I would never recommend that voters vote with such pens.”
More seriously, he confirmed that he has “seen both bleed-through and ‘tractor tracks’ that end up being counted as votes.” Both, however, can be prevented.
Bleed-through. The right paper stock and marking pen can prevent bleed-through on ballots marked at the polling place, but Jones pointed out “No matter what kind of ballot-marking pen or pencil the polling place provides, some voter will pull a random pen or pencil out of their pocket and use it. This is particularly true in a crowded polling place when the official ballot marking pen or pencil goes bad (out of ink, needs sharpening) and also true for absentee voting.”
Next to heavy paper stock and special pens, ballot design is the most effective preventive measure: “When two-sided ballots are necessary, Jones recommends that clerks “hold a ballot up to a bright light (tape it to a window with the sun shining through it) and see how the voting ovals on one side line up with the ovals on the other. No oval on one side should be within 1/4 inch of an oval on the other side to be really safe, although 1/8 inch between ovals ought to be sufficient.”
Tractor trails. Jones wrote that he has never seen marks caused by rollers transferring wet ink from one part of a marked ballot to another, although “with many common ballpoints, sometimes a glob of ink gets loose and will dry extremely slowly. I can imagine a paper feed roller transferring such a glob down the ballot as extra marks.”
Nevertheless, he has seen such marks created by the wheels alone—without ink. When he used one stack of ballots to test several different machines, he noticed that “After 8 passes, the tread marks from the feed rollers began to be visible, after 16 passes, the tread marks were quite visible.” This obviously isn’t a problem in real elections, when ballots “are scanned just once, except when there is a machine recount, and then they get scanned twice.”
Again, the best preventive measure is ballot design. Jones wrote: “Find out where the paper feed rollers are in your scanners. They are extremely unlikely to be full-width rollers. Rather, the rollers are most likely 1/2 inch wide (or so) with from 3 to 5 rollers working across the width of the ballot. A well-designed ballot will never have a column of voting targets aligned with the feed rollers. The voting targets should fall between rollers. Following this design rule, even if the rollers leave black skid marks on the ballot, they won't interfere with the vote counting.”
My main question to Jones, however, was not about what we could see with our own eyes (bleed-through and tractor trails), but what we could not see: Any reason for the 16 phantom votes. I asked whether he knew of any malfunction that might cause an optical-scan voting machine to generate votes on its own accord--that is, for some reason other than readable marks appearing on a ballot.
Jones replied: “I can think of two things, one that remains a possibility and one that your hand search probably ruled out.”
Flaws in the ballot paper. “If you look at any commercial source of paper,” Jones wrote, “you will find occasional flecks of bark or other dark material embedded in it. Some art papers deliberately have many flecks. Most office paper has very few, and the ballot stock you get from (voting machine vendors) has even fewer, but every once in a blue moon, there will be a fleck.”
One precaution against this, Jones wrote, was used by the people who helped launch American Information Systems (the company that eventually became ES&S, which manufactured the Stoughton machines), who “in the early days…made a point of scanning all the blank ballots before election day, rejecting all the ballots that scanned as nonblank” as a result of printing defects or flecks in the paper.
Jones guessed that our visual inspection ought to have ruled this out, and he is correct—it did.
Dust bunnies and paper lint. “The possibility that remains,” Jones concluded, “is dirt. Just plain dust, really. In any scanner, there is the possibility that a hair or a piece of lint or something similar will lodge briefly in the scanner, being detected as a black spot. It might hang there for a while, scanning as a black streak from the top to the bottom edge of several ballots until the leading edge of some ballot wipes the dust away, or it might just show up for part of one ballot.”
The dust might come from outside the machine or from the paper ballots themselves. “All paper sheds lint, although the ballot paper I've worked with sheds less lint than most paper,” Jones explained. “A single fiber of paper lint would be too small to cause a problem, but if the fibers accumulate into a little ball of felt, this could be a problem.”
“The trouble with the lint hypothesis is that we'll never know,” he added. “If lint interfered with the counting of 16 ballots, it is long gone by now.”
Judicious use of compressed air can prevent dust bunnies and lint from voting in future elections, according to Jones. “If you look at a large absentee vote-counting center when there are…technicians on hand, you'll see them stop the count on each scanner after every few hundred votes and go in with a can of compressed air to blow out any dust.”
Jones helpfully provided a price check: “Canned air is pretty cheap. I just checked eBay, 12 cans for $55. I'd trust poll workers to blow dust with such a can, and the normal pre-election setup routine for each scanner ought to include a dusting by a trained technician.”
* * * * *
It’s unlikely we’ll ever know for sure what caused those 16 votes or—more importantly—how to prevent phantom votes in future elections. And unless we begin routinely to verify the output of our voting machines, we won’t know how many votes were cast by bleed-through ink, tractor trails, ink transfer, spotty paper, or dust bunnies the next time they decide to vote. We knew this time only because the set-up error removed all the real votes and left only the phantom votes to be counted.
It’s time we stop pretending that voting machines are any more reliable or secure than other computers or any better than the fallible humans who program, test, and maintain them. They are not. They are prone to the same sort of malfunctions and vulnerabilities as any other machine, and are at least as vulnerable to hacking as the computers operated by the likes of Anthem, Home Depot, and Target. We can never be sure we have anticipated and prevented every type of miscount that could affect our election outcomes, so we are imprudent fools if we do not routinely check to make sure they counted correctly after each election.