Mississippi is getting a lot of attention, with its razor-thin margin in the still (as of this writing) not yet fully counted Senate contest and the surprising lack of a recount provision in its election law. Whoever wins, the loser is virtually guaranteed to feel unsatisfied by a process that denies them a safeguard that's available in some form in every other state.

It seems plausible that this situtation will serve as a catalyst for the Mississippi legislature to add a recount trigger to its election law. But I think going down that obvious path would represent a missed opportunity. We can do better.  

"Recounts", as implemented in almost every state, were designed as a safeguard against small-scale human error, dating from the days when ballots were hand counted (sometimes by candlelight) on election night. It was natural to expect that people would make mistakes.

Recounts are generally triggered when the margin of victory is within the expected margin for error of the counting process that existed at the time the law was enacted.

Unfortunately, the assumptions that underlie recounts are mostly obsolete.

The problem is that traditional recount triggers assume that any errors in the tally will be fairly small. The precise margin that triggers a recount varies from state to state, but almost every state's recount rule is based on this assumption.

But modern automated ballot tally systems are vulnerable to different kinds of errors than those committed by fatigued (or perhaps even dishonest) human counters. Current systems might make large errors just as easily as small ones.

Modern ballot tally systems, including optical-scan ballot readers found in polling places, are controlled by software. Software is notoriously unreliable and subject to unexpected bugs, even after rigorous testing, and the software used in ballot tally systems has proven to be no exception. But worse, software is designed to be replaceable, which means that it may be possible for an adversary to replace the tally software with malware that reports any result the attacker desires, and then erases its tracks.

Fortunately, there's a reliable, rigorous, and efficient safeguard against tally system errors and malware, and it works for almost any voting system that has reliable paper ballots. It was invented about 20 years ago by Berkeley statistician Phillip Stark. The method is called "Risk Limiting Audits" (RLAs), and is regarded as the gold standard for protecting elections against tabulation errors or malice.

In a Risk-Limiting Audit, a random sample of ballots is compared (by human judges, generally including candidate representives) with how the automated tally system interpreted them. As we sample more ballots that were interpreted correctly, we gain more confidence that the automated system reported the correct overall outcome. Some relatively straightforward statistical math lets us quantify the level of confidence we achieve (called the "risk limit") and when we can stop sampling.

(I'm glossing over the details here, including the fact that there are different RLA methods used depending on the precise nature of the voting system. But this is the basic idea.)

The good news here is that if the automated tally system was operating correctly, the number of ballots that have to be sampled and compared to achieve a very high degree of confidence (a low "risk limit") is surprisingly small.

The even better news is that RLAs achieve essentially the same assurances that traditional recounts do, but with the added feature that they also protect against catastrophic software errors or malware in the tabulators. They render recounts an obsolete approach. And they're generally much more efficient. To achieve this, RLAs should be done after every election, not just those with tight margins.

So as Mississippi looks in the (hopefully near) future to amend its election law to include "recount" safeguards, it will have an opportunity to start from a clean slate and adopt a proven, state-of-the-art statistical safeguard that works against the kinds of errors to which modern voting systems are actually susceptible. And other states with obsolete recount rules should follow suit.

Don't think recounts. Think Risk-Limiting Audits.