At roughly the same time that statisticians in academia were developing RLAs, the technologists who created Clear Ballot were focusing on how to build a voting system that could account for every ink mark on all the paper ballots cast. That is a different challenge than statistically verifying that the announced victor was correct -- with 95 percent certainty. The firm developed products to read ballot ovals, grade how voters filled them in (confidently or not, based on the ink mark's density), and to compile the results--and identify unreadable scans that needed further individual scrutiny.
Since 2016, Maryland has used Clear Ballot's image-based software to audit its initial electronic scan of paper ballots -- done by another manufacturer's scanners.
On the Friday after 2018's primary, the Maryland state Board of Elections (BOE) sent Clear Ballot CDs with images of 1.3 million ink-marked ovals and the giant spreadsheet with the most granular local results: the cast vote record. By Monday, it found about 1,100 ovals that didn't correlate with initial scanner reading. By looking at those images more closely, it was able to resolve voter intent on most ballots, and tell the state BOE where to look for individual ballots -- by precinct -- if further investigation was warranted. (The state has rules where it would examine those ballots by hand in close contests.)
"We don't think that technology can resolve an election," said a top official with Clear Ballot who didn't want to be named. "We believe the role of technology is to present the voter's intent in such a way that unbiased human judges can decide on behalf of their constituents, and preserve that judgment indelibly." An election is a giant accounting problem, and we are trying to get a very precise apples-to-apples comparison."
This executive said their process had a different goal than a statistical audit.
"We define an audit as not 'did the right winner win,' but as a comparison of two independently produced results that are derived from the same underlying data," he said, adding that using digital images eliminated the Achilles' heel of risk-limiting audits -- the difficulties surrounding ballot custody, followed by the need to bring everyone into a single room to randomly pull ballots following rigorous math.
"Does the technology of ballot images change the way the country should look at auditing?" he asked. "Take RLAs out of the question. We can go to an all-digital world where human behavior does not play a role. Or play a role in such a way that you can go back and check it. What will be the effects on statutes, etc., when audits can be done routinely, well within the certification period [formally declaring winners]? I don't mean sample audits. I mean complete, 100 percent, every contest, every ballot cast across every precinct. And do it before the certification window closes? And present the results on the internet, where the widest audience can look at it. Like we just did in Maryland."
Indeed, in the state's 2018 primary, there were four close contests where the state Board of Elections gave the candidates and campaigns access to Clear Ballot's image-based inventory so they could look for themselves at the votes and decide whether to pursue recounts. The images were persuasive -- and allowed the candidates in both sides to accept the result.
"We got a call from Nikki [Charlson, Maryland BOE deputy administrator]. She asked us to give them credentials. We did. We could see them pouring through the logs looking for votes. For us, it was a joy, to be able to provide that," he said. "People could sit in their offices. Not have to assemble someplace. Be able to look across the entire population of ballots that had already been reconciled to the primary voting system's results."
Needless to say, if this high-tech scenario seems too good to be true -- being able to verify millions of ink marks on paper before election winners are officially declared -- it is not good enough for some supporters of risk-limiting audits, and for one particular reason: hacking fears. Simply put, while critics of risk-limiting audits say their biggest flaw is ballots are sloppily handled in real life (vanishing, reappearing), upending the math, critics of ballot image audits say hacking and falsifying digital images is also a real concern (and can be seen in online political missives this week).
No Perfect System
"Nobody is going to balk at the efficiency there," Morrell said, of ballot images. "If I could get enough individuals that I talk to, to say that they would be comfortable with ballot images, and allowing the computer system -- and these algorithms -- to make those comparisons, I would be fine with it too, because it certainly would save a good deal of time. But having said that, let me just tell you, there's something about actual paper."
The EAC's Lovato said he once frowned on paper's inefficiencies, but he has changed his mind -- and it's not just a theoretical threat to him.
"One of the issues that arises, and one that you have probably heard from the computer scientists, is that you can't trust the images," he said. "I actually lean more toward that camp, just simply because there are voting system companies that do alter images. So, like the Dominion system, they produce what's called an audit mark on their [scanned] ballot image -- not on the actual ballot. So just by including that ballot mark, they are saying, 'we manipulated the image.' Whoever is viewing that can say that's a good thing or a bad thing, but the bottom line is they are manipulating the image. So if they are able to do that, how can I as a concerned citizen say, 'How do I know that they didn't do something else to the image?'"
The EAC has not taken an official position on risk-limiting audits versus ballot image audits, although it has not funded pilot projects involving ballot images.
"I used to be in the camp that it's not so likely," Lovato said, referring to hacked voting systems. "I'm now in the camp that it's more likely, at this point, when it comes to the cyber part of things. The other issue with Clear Ballot, as an audit usage, is the public perception side of the jurisdiction saying, 'Okay, we spent hundreds of thousands or millions of dollars on this voting system that can produce images, and we're going to rescan them on this other system?' And what happens when the results differ between the two? Whose system do you trust at that point? How do you explain that to the public, to lawyers, to reporters? If I'm a local election official, I don't want to get near that nightmare. But that is also a very real thing. I've seen that in the pilots that I have conducted with doing the different scanning -- the difference in results."
(Note: You can view every article as one long page if you sign up as an Advocate Member, or higher).
2
1
1
