How to be “confident” in post-election audits

I attended an interesting talk by Professor Philip B. Stark, a statistics professor at Cal Berkeley, called “Urning Voter Confidence.” His talk focused on ways to develop an efficient and effective post-election auditing rules, drawing on insights from probability and statistics.

The talk title is a bit misleading–Philip did not talk about voter confidence in the way many of us think about it, e.g. voter confidence that their ballot was counted correctly or voter confidence that the election result is accurate. (For the purposes of this post, I’ll call this latter quantity legitimacy.)

Confidence in the context of the talk, I think, means drawing on the insights of statistics to generate a confidence interval, and then not wasting our time on auditing procedures when there is no reasonable chance that the election result (even if there was miscounting or fraud) could be overturned. I suspect that Philip will argue that legitimacy is related to appropriate post-election auditing, but he didn’t make that link in the talk.

The math is pretty forbidding, but his main point is intuitive: in a close election, say one decided by 1%, you may have to do a much more extensive audit to make sure that any miscounts could not change the outcome. On the other hand, in a lopsided election, say 80-20, you can do a much less extensive audit. Philip provides a methodological solution, based on random draws from an “urn”, which gives guidelines for auditing.

Why is this important? It’s important because most of the auditing procedures that are in place (e.g. California) or are being proposed (e.g. Holt Bill) don’t take confidence intervals into account. They just establish some number (although Stark points out that the Holt bill at least adjust the size of the audit based on the closeness of the race.)

I was also surprised to find out that in California, the rules don’t specify what you do if (as is almost certain to happen), the 1% audit shows a discrepancy from the initial tally.

Stark’s method establishes very specific steps to follow in the post-election audit. If enough errors are found, and the election is close enough, ultimately you end up completely recounting the election, but this is very unlikely. On the other hand, if your 1% audit does not give you enough confidence, surely you need some guidelines about where to go next.