Tuesday’s trial of the process, called risk-limited auditing, was conducted by VotingWorks, a nonprofit dedicated to making elections simple and secure.
In essence, an audit looks at a statistically significant sampling of ballots to see if anything went wrong during an election, said Monica Childers, project manager at VotingWorks. Because samplings would, in a real-world audit, be taken from across the state, errors could be traced back to the source and fixed before the next election.
It starts with a 10-sided die such as one might cast in a table-top game, Childers said. A selection of citizens or election officials pair up and cast one die for each group. From there, open-source, pseudo-random computer algorithms assign each pair a sampling of ballots to check against the results recorded by ballot scanners.
Childers clarified pseudo-random to mean that the algorithms will always return the same results based on the number they are given. The number is completely random — produced by the ten-sided dice — and recorded in case the results of the audit need to be reproduced.
Importantly, the review panels check printed ballots against the machine-recorded results. By doing so, the process serves as a check on ballot scanners.
“It all works off what a voter is able to verify,” she explained.
This is all done with a mind toward giving voters confidence in the election system, Childers said. A risk-limited audit is not a full recount, but checking a specific number of ballots can ensure a certain percentage of accuracy.
For example, she said Colorado performs audits at a 4 percent margin, meaning an audit would catch 96 percent of errors. For most races, the difference between winner and loser is wide enough for that alone to be adequate. For others, a full recount may still be necessary to verify the results.
Public trust and transparency are the most important goals of the project, Childers said.
Childers said the original concept was created by a statistician with the University of California at Berkeley. It underwent some trials but did not take off statewide. That’s changed in recent years as states have more and more come to place high importance on ensuring the public has confidence in election results, said Childers.
Each state handles elections differently, so what works in Colorado likely wouldn’t work in Georgia. That’s where VotingWorks, and Glynn County’s elections office, comes in. The nonprofit has been helping the state develop an election auditing method that’s accurate and efficient enough to do after every election, Childers said.
Trials have been conducted since the 2019 municipal election after the Georgia legislature opted to overhaul the state’s election laws and purchase new voting machines, she explained.
Risk-limiting audits come in three forms — ballot comparison, ballot polling and batch comparison. Ballot comparison, by and large, offers accurate results with a smaller sample size than the other two, Childers said.
“I like to talk about it like ice cream flavors,” Childers said. “There’s rocky road, vanilla, mint. The flavor we’re using is ballot comparison.”
Officials with the Georgia Secretary of State’s office said the state has not committed to adopting any of the auditing methods it has attempted so far.
©2020 The Brunswick News (Brunswick, Ga.). Distributed by Tribune Content Agency, LLC.
