Risk-Limiting Audits
In the past decade, two differing approaches to answering that question have emerged and evolved. The first to surface is what's called a risk-limiting audit (RLA). Jerome Lovato, now an election technology specialist with the U.S. Election Assistance Commission, was present at the start of developing and implementing RLAs a decade ago when Colorado hired him to improve their audit process. Colorado had been sued for a lack of transparency surrounding its testing and certification after buying new machines in 2006. Back then, Colorado -- like many states today -- grabbed and examined hundreds of ballots after every election to see if they matched the announced winners.
"We were looking at something more statistically based, that said, 'Did the winner win?' 'Did the loser lose?' And providing that statistical confidence to the public," Lovato said. "And that also gave us what we didn't have, which was what happens if we have a discrepancy. Because the way we did audits before, a county [office] would say, 'We have a discrepancy. Our counts aren't matching. We ran 150 ballots and our hand count says 149 and the machine says 150.' We [the state] would say, 'Okay, try it again. Do it one more time. Make sure the hand count was accurate, because that always tended to be the case"' But we didn't have any kind of calculation that said, 'Look at X-percentage more ballots' or anything."
That void and guesswork was filled by what's now called a risk-limiting audit. While there are several ways to do these, the basic idea is that it is statistically possible to randomly count a small number of paper ballots -- if all of the ballots have been carefully handled and assembled (a big if) -- and determine with 95 percent probability, or more, if the correct winner was announced.
This past summer, Fairfax County, Virginia, piloted two types of risk-limiting audits in a race where 948 ballots were cast. In the first, it wanted 95 percent certainty the correct winner was chosen and, given the math, only needed to pull 69 paper ballots. That "ballot comparison" process assumed that all of the ballots were initially scanned, collected and compiled in a precise order -- so they could be subsequently traced by individual ballot. That controlled environment mimics vote centers -- a central site where people vote.
The second, a "ballot comparison" audit, is designed to verify counts of paper ballots that fall as a disorganized jumble into bins below the scanners at local precincts. Seeking 90 percent certainty, it required officials to examine 260 unique ballots in its initial round. After pulling that many randomly selected ballots from their storage boxes and seeing how their ink marks compared to the reported outcome, the math found there was a "53 percent chance that the audit would have identified an incorrect outcome," a Virginia Department of Elections report said. "In a true RLA, election officials would have selected a second round of sample ballots and completed the process again, repeating until either the risk limit was achieved or it was determined that there was a need to proceed to a full recount."
In other words, when precinct voting is involved -- which is the way most of the country votes east of the Rocky Mountains -- double-checking the vote this way can be laborious, although somewhat more efficient than a full manual count, or reexamination, of every ballot. Virginia's Department of Elections compared risk-limiting audits to political exit polls -- where some voters are questioned to assert larger trends. It's not a full accounting process. It is a statistical analysis to confirm, or question, whether the announced winner is correct.
"The whole premise behind an RLA is to provide the statistical confidence that the outcome is correct and it is based on the risk limit that has been set," said Jennifer Morrell, an ex-election official in Utah and Colorado who now leads the Democracy Fund's new Election Validation Project. "So the tighter the race, the more ballots will need to be retrieved and examined. I am sensitive to that, having run elections. I'm definitely sensitive to [the fact] that you might be asking, in a really tight race, an election administrator to have to audit more ballots than they would normally have to. What always comes up in states is how do we fit all of this in[to the workloads], the audit procedures? Because we always have to recount; we always have tight races."
A handful of states will be conducting risk-limiting audits this fall. Some will be done before the winners are officially certified -- like Colorado. Others are more akin to the pilot project in Virginia, an exploratory exercise. Compared to the vote verification a decade ago, they impose structure on a previously haphazard process. Proponents like Morrell, who oversaw RLAs in Colorado's Arapahoe County, said they prompt everyone involved to be more meticulous about handling ballots. And she asserted their primary virtue, for public confidence, came from examining actual paper ballots.
"I see so many pluses that came out of having to pull the paper ballot. It forced me as an election official to really refine and think about and improve the way that we organized and tracked and stored and handled our ballots," she said. "But having said that, let me just tell you, there's something about going to the actual piece of paper marked by the voter. There is some level of confidence there."
But in regional or statewide races, and especially those with photo finishes, the random ballot-pulling and statistical analysis can get complicated.
"That is part of the challenge," said Patrick. "I think RLAs have a great purpose. But I also think that there has been a little bit of a bait and switch. Everyone has been told they are easy, they are simple, the math could be done: 'Anyone can figure this math out. It's not complicated.' But in reality, the math is complicated, particularly when you are talking about a statewide race across multiple jurisdictions, and you also have to have the infrastructure in place to make sure you have your ballot manifests [inventorying and handling all the ballots], you have your cast vote record CDRs [compact discs with the spreadsheet of every vote]. And all of these things have to be in place, so in fact you can go and pull the 73rd ballot from the 489th batch run on machine number two. If you don't have that in place before, you're not going to be able to do an RLA."
Ballot Image Audits
This mix of ballot-handling concerns, methodological complexity and statistical math that many people do not understand leads to the hardest challenge facing those pushing RLAs -- the fact that they are not easily explained. As Morrell said in a long interview, "Anyone who listens to all of this conversation would get lost. How much information does the public need to convince them that it was done correctly?"
That very question is not a trifle -- not in our political era where truth, facts, objectivity and unpopular election results all are under attack, as evidenced by the president's tweets. Not a day goes by without reports in newspapers like the New York Times about how social media is distorting political debates in harmful ways. Indeed, some of the most astute new political books are deconstructing and bemoaning this dark cultural trend.
As a professional culture, the world of elections -- like the world of legislating and governing -- often lags behind popular culture and technology. When it comes to vote counts, this also is true. In the past decade, there has been only one new voting machine maker accredited by the U.S. Election Assistance Commission. That firm is Clear Ballot, which relies on digital ballot images -- capturing each side of every card that's scanned while processing an individual's overall ballot -- to count and verify votes.
(Note: You can view every article as one long page if you sign up as an Advocate Member, or higher).
2
1
1
