Election Integrity Audit Calculator
In the ongoing debate over electronic voting, most people appear to agree that sound procedures and effective audits are necessary to ensure secure and transparent elections. The real debate has been over how best to promote these values, while at the same time safeguarding others such as equal access for all citizens.

The most frequently discussed auditability mechanism is for electronic voting machines to generate a contemporaneous paper record, or "voter verifiable paper audit trail" (VVPAT). In the most common configuration, this entails a paper printout of the electronic ballot printed behind a transparent screen, so that the voter can see but not touch it -- and thus check his or her recorded choices -- before that vote is cast.

The problems with the VVPAT raised by its critics, including me, can be broken down into two basic categories. The first has to do with its workability. This includes getting a system in place that will actually function as it's supposed to. This has recently been shown to be a significant problem in Cuyahoga County, Ohio, where almost 10% of the VVPAT records were compromised in some way. The other problem has to do with the efficacy of the system -- that is, whether it will actually fulfill its intended function of making elections safer and more transparent. This partly hinges on whether voters actually the VVPAT records. It also depends upon conducting routine audits of those records, without which they are effectively useless.

An issue that's received insufficient attention, as I've noted here and here, is exactly how many paper records have to be recounted in order to ensure a sufficient level of confidence. This applies, by the way, not only to the VVPAT but also to other types of audit records -- such as the "voter verified audio audit trail" that Ted Selker has proposed. One of the few studies to examine this issue is this one written by Andy Neff of VoteHere. He found that, for a California congressional race, a 1% recount has only about 40% chance of catching a switch of 10% of the votes -- more than enough to alter the result in a reasonably close race. Unfortunately, there's been very little else written on what percentage of paper records should routinely be recounted in elections of different sizes.

US Count Votes has now taken a step in the right direction, however, with this analysis entitled "The Election Integrity Audit." A short description may be found in this press release. Although I've criticized USCV's work in the past, this seems to me a very useful contribution to the debate. Particularly helpful is this Election Integrity Audit Calculator, which purports to calculate the number of paper records that should be recounted to obtain a particular level of confidence, given the size of the election and the margin separating candidates.

I'm not mathematically sophisticated enough to assess whether the calculations generated by the USCV site are correct. At first glance, it appears to arrive at different results than what Neff had previously estimated would be necessary. Nevertheless, this is precisely the sort of analysis that needs to be conducted, in order to figure out exactly how to audit elections with the VVPAT or other verification systems. I hope that others will pay attention to this, and check on whether the calculations the USCV site generates are accurate.

Assuming so, the next question is this: What do we do if the audit results don't match the electronic vote totals? It could mean that the electronic records are compromised. Or it could be that the paper records are compromised, as in Cuyahoga. I don't have a good answer to this, but it's a problem that also needs to be considered. - posted by Dan Tokaji @ 10:25 AM