Pennsylvania’s risk-limiting election audit, explained
As part of what will be a routine effort to verify Pennsylvania’s election results, the Department of State has asked counties to perform what’s known as a risk-limiting audit.
The Department of State tested that kind of ballot check in 2020, when it ordered counties to review 45,000 randomly-selected ballots. The audit that year showed evidence that the results were accurate – despite efforts by some Republican state leaders to amplify former President Donald Trump’s election fraud lie.
Election audits came into sharp relief following the 2020 election, when activist groups began calling for them, believing they would reveal widespread impropriety. That year’s risk-limiting audit and another routine audit of two percent of ballots in all counties showed that results were accurate. Numerous other investigations by experts of both parties confirmed that finding.
This year marks the first time the Department of State is checking ballots using a risk-limiting audit on a full statewide scale. It plans to do one every year from now on. Department of State Acting Secretary Leigh Chapman has called them “ the gold standard of robust election audits.”
14 other states, including swing states like Georgia and Michigan, check their election results using risk-limiting audits.
What is a risk-limiting audit, or RLA?
They’re designed to reduce the chance that an election is certified for the wrong candidate. To do that, election officials have picked a race to review – in this case, the one for Governor between Attorney General Josh Shapiro and state Sen. Doug Mastriano (R-Franklin) – and on Thursday, they selected groups of ballots at random that each county will have to review.
A group of agency employees rolled ten-sided dice to generate a 20 digit “seed” number, which tells a computer software program which groups of ballots counties will have to check.
“Teams of auditors in each county will manually review and hand-tally the votes cast for Pennsylvania governor from the ballots in those batches,” Deputy Secretary Jonathan Marks said. “Once all counties have reported their results of their audit, the Department of State will compile and evaluate the statewide results to determine whether the risk limit has been met.”
In other words, whether the result matches what the audit found.
Marks added that because this year’s audit will only review the race for Governor, each county will only have to check a relatively small number of ballots.
“The number of ballots required by the RLA is race-specific and depends on the margin between the winning candidate and the runner-up. The smaller the margin, the more ballots that need to be audited.”
Shapiro won the race by more than 14 percentage points, but an audit still has to be performed to make sure that result is accurate across the state.
Why use an RLA over any other type of audit?
Every election, counties check at least two percent of all ballots they reviewed for errors. In some smaller counties, at least 2,000 are checked. But that small sample of ballots isn’t always the most representative of a county’s election result – and reviewing that group alone isn’t always the most effective way to root out any potential problems.
Pennsylvania’s election system generally delivers accurate election results, but stray ballot errors or voting record mismatches happen from time to time.
A risk-limiting audit is designed to catch issues like that and investigate them to determine whether they’re widespread.
“If the reported outcome is deemed incorrect, an RLA will continue to expand and may lead to a full recount of the contest,” Marks added.
In other words, if counties find any problematic ballots in the first batch of ballots they check, they check another group to see if they find any more. More errors leads to more checking, which is supposed to stop bad election results before it’s too late.
Risk-limiting audits work especially well in close races: by design, election workers have to scrutinize more ballots from the start, which increases the chance they’ll catch counting errors.