(some of this is part of a paper I'm thinking of submitting to WOTE'06... let me know if you want to work with me on it!)
At last week's CFP 2006 conference we held a "Primer on Electronic Voting" tutorial meant to get people from zero to sixty on the many issues surrounding electronic voting technologies; no small task. Besides the fact that this material will be out of date in a few months, we felt that there is a gap in tutorial-like material for people that would like to be able to work in or get acquainted with this emerging field.
I was surprised at how much I learned myself.
Case in point: Larry Norden from NYU's Brennan Center for Justice presented on their threat analysis report that will be released in three weeks. Their report examines about 120 attacks on three types of electronic voting technologies (PCOS, DRE and DRE with VVPR).
Larry pointed out evidence that would be necessary to detect DRE+VVPR technologies "switching votes" -- where the DRE is maliciously or erroneously programed to register a vote as being for a choice other than what the voter intends. Specifically, it would seem reasonable that you'd see statistically significant differences in voided VVPRs in systems that were misbehaving. Of course, for each model of these kinds of machines, we'd need to establish a baseline of voided VVPRs. There seem to be a few reasons why a VVPR would be voided (in order of what I would consider increasing likelihood):
- If the voter made a mistake and didn't detect it until the end of the process.
- If the contents of the VVPR didn't match the contents displayed on the screen.
- If the voter wanted to void their VVPR for fun; that is, just to see what happens.
- If the voter voids their VVPR on accident.
What we want to separate out is item 2.
Unfortunately, while there are 26.5 states* that require their voting machines to produce some sort of VVPR and 12 of those states take the very necessary step of doing a 1-10% randomly-selected manual audit of the VVPRs against the electronically-stored vote data, I know of no state that requires the number of voided VVPRs to be reported in the manual audit; the statutes contemplate only reconciling the paper and electronic records.
It would seem, from the perspectives of policy, auditability and security, that reporting the number of voided VVPRs would be instrumental in detecting cases where some anomaly had the side-effect of causing the contents of the VVPR to be corrupt. This might also be a useful proxy to measure in a more significant manner the amount of voters that actually read their VVPRs (which anecdotal evidence and some new research shows that is a low percentage of voters). Thoughts?
* The half-state is Arkansas where rural counties are required to have VVPRs but the more populous counties had their non-VVPR equipment grandfathered into the statutory requirement.