Nonetheless, those endorsing RLAs said there were great benefits to be gained by doing them, saying, once learned, they brought a heightened diligence to the ballot custody and vote-counting process. They urged those observing RLAs for the first time to be patient and suspend judgment.
Many Questions Were Raised
In the follow-up discussion, Karen McKim, a Wisconsin election integrity activist, recounted how she had used Stark's instructions, found online, to conduct RLAs with county clerks in 2014 and 2016. But she used digital images of the ballot cards (made by optical scanners) instead of randomly pulling the paper. McKim said she liked the speed and visual impact of using images, but said the number of ballots to be examined was so small that some voters did not have "emotional confidence" about the process. She asked Stark whether it would hurt to keep pulling ballots until skeptics were satisfied.
It would not, Stark replied, saying RLAs were a minimal standard for how much auditing to do, amid their wider benefits.
Walter Mebane, a University of Michigan political scientist, Washington Post columnist and "election forensics" expert, told Stark that he required his advanced students to read Stark's papers on RLAs. But picking up on McKim's concern about giving the public "emotional confidence" about verifying votes, Mebane said RLAs rely on "voices of authority" amounting to every statistician saying they would do the same thing.
"This is fine when this is not [an] important" election, Mebane said. "But we are in an era now when expertise is no guarantee of anything. If you're important, then our president, for example, is happy to denounce expertise, and I can guarantee you there will be experts on either side with their presumably bogus arguments. But what is the way to explain this? I get frustrated. I say we have to explain to people what statistics is in some fundamental sense" I don't think arguments from authority can survive a sustained form of [propagandistic] attack."
Stark replied the math behind RLAs was available and as such was "transparent," even if one needed "to learn two years of math to do it." He then offered an analogy, saying that pulling random ballots was like tasting a sip of well-stirred soup. "If you stir it well, and take a tablespoon, that's enough. If you don't stir it, all bets are off. It's the stirring and then taking a tablespoon that amounts to a random sample... I don't expect everyone to be able to check the theorem. I do expect everyone to be able to check the calculations based on the theorem... An astute fourth-grader could do the arithmetic."
After other questions, Stark returned to McKim's comment and made a crucial point to RLA's promoters. He emphasized that vote count audits were not the same as recounts, which would be a new counting procedure that would have to resolve the closest contests.
"The decision whether to do a recount is different from the decision whether to audit more; you need to decouple those," Stark said. "The other thing I have to interject is I don't view auditing from [digital ballot] images as a risk-limiting audit, for reasons I think everyone here understands. Because it's not the artifact that the voter looked at; there are lots of things that can cause there to be a different number of images."
Looking Beyond Factions
This exchange, while deep in the details of vote verification techniques, revealed key decision points facing election officials and schisms among experts about solutions. The MIT summit showed there are new ways to verify the accuracy of vote counts as many states are poised to buy new gear that likely will be used for years. Indeed, the push by Florida Supervisors of Elections to use a second ballot image-based system to audit their initial tabulations is additional evidence of this overall vote-verification trend.
But what was glossed over at MIT's summit was the fact that not every audit technique could handle all of the stages in progressively verifying vote counts. RLAs, despite their complexities, can be efficient in affirming outcomes in races that are not close -- as Mebane put it, "not important." As the margins tighten, those efficiencies vanish.
That turnaround raises big questions, considering RLAs' backers have been successful in getting some state legislatures to require them before certifying election outcomes. What are those states going to do when the public sees election officials start an RLA, but have to shift to another start-to-finish process to recount? What happens when this vaunted verification process can't take closest races across the finish line?
That is not conjecture. Similar dynamics were just seen in Florida's three simultaneous statewide recounts. There was not a smooth transition between the way ballots initially were handled and initially tabulated and subsequent two-stage recount. Some counties that had carefully indexed their ballots saw those libraries turned into veritable haystacks as they followed Florida's current process: a coarser sorting and counting, followed by manually examining tens of thousands of certain categories of questionable ballots.
Some of Florida's biggest counties could not finish the recounts due to their machinery, protocols and state deadlines. That was partly why its election supervisors, meeting a few days before the MIT summit, endorsed a resolution urging their legislature to allow them to use a second independent ballot-scanning system with visually oriented software for future recounts. That news was not mentioned at MIT.
There are explanations why not -- namely many mathematicians, cryptographers and election integrity activists distrust anything involving computer records and analytics. But there's a bigger point beyond parsing their beliefs. Every auditing approach has strengths and weaknesses. The verification choices -- RLAs, other statistical audits, image audits, hand counts -- differ in what stages of the process and scale they can handle well. Their goals, estimating an election's accuracy or a fuller accounting, also vary.
(Note: You can view every article as one long page if you sign up as an Advocate Member, or higher).
