If you’ve spent any time in election administration circles over the past several years, you’ve probably heard the phrase “risk-limiting audit” come up more and more often. It’s a mouthful, and the statistics behind it can look intimidating on paper, but the underlying idea is refreshingly simple: audit harder when the race is close, and audit less when it isn’t.
We touched on the basic mechanics of post-election audits in an earlier piece, but risk-limiting audits deserve their own explanation because they solve a real weakness in older audit methods. A traditional fixed-percentage audit — say, always hand-counting 1% or 2% of ballots — treats a landslide race and a nail-biter identically. That’s inefficient at best and misleading at worst: a small fixed sample might be more than enough to confirm a 30-point win, but nowhere near enough to responsibly confirm a result decided by a few hundred votes.
Risk-limiting audits flip that logic around. Instead of starting with a fixed sample size, they start with a target: how much risk are we willing to accept that a wrong outcome slips through uncaught? From there, statisticians calculate how large a hand-counted sample needs to be to meet that risk threshold, given the actual margin in the race. A landslide might only require a tiny sample. A near-tie might require counting nearly every ballot by hand. As the nonpartisan election-verification group Verified Voting has explained, this approach is now considered something close to a gold standard among election security researchers.
We find the underlying logic appealing for a fairly simple reason: it scales scrutiny to where scrutiny is actually needed. Resources aren’t unlimited, and a state that pours the same auditing effort into every race regardless of margin is, in effect, underprotecting its closest and most consequential contests while overspending on its least competitive ones.
That said, risk-limiting audits aren’t free of trade-offs. They require statistical expertise that not every county election office has in-house, software tools to calculate sample sizes correctly, and public communication that can make a fairly technical process understandable to reporters and voters who just want a yes-or-no answer about whether the count was right. States that have adopted RLAs successfully have generally done so gradually, piloting the process in a handful of counties before scaling statewide.
We think that gradual, tested rollout is the right model to follow elsewhere. Risk-limiting audits aren’t a silver bullet, and they don’t replace the underlying need for paper records and solid chain-of-custody practices we’ve written about before. But as a more precise, statistically grounded way of deciding how much checking is enough, they represent a genuine improvement over the blunter methods many jurisdictions still rely on.