Monday, November 17, 2003[posted by jaed at 6:44 PM]
On touchscreen voting machines
Volokh has a number of posts this week on the issues surrounding touchscreen voting machines, counting, etc. Eugene Volokh seems to think giving a receipt to the voter is being proposed; I've never heard this suggested, but if it has been, the suggestor should be slapped with a large fish into which has been carved the phrase "Secret Ballot". Diebold's management should be slapped with a similar fish that says "Open Source", or perhaps "Security by Obscurity Doesn't Work". (Depends on the size of the fish.)
Design criteria for a reasonable touchscreen voting system:
1. System is coded to prevent overvotes and inadvertant undervotes. The user interface should not let you overvote for an office. If you undervote (don't vote in a contest, or vote for fewer candidates than you're allowed to [for such things as at-large city councils]), it should ask "Are you sure?" in some fashion, to confirm that your undervote is deliberate.
2. System should, when the voter is finished, print out a ballot that's both human-readable and OCR-able, and ask the voter to confirm that these are in fact the right choices. Voter deposits this ballot in a box at the polling place before leaving, and it's used as a double-check on the computer tally.
(I say OCR-able, rather than having two sets of data - machine-readable and human-readable - because if the voter can't read the machine-readable part, it's not really a check; if the two sets are different, due to either a bug or software sabotage, it would be difficult to tell.)
3. The initial numbers are taken from the computer tally. If a candidate challenges the results, the printed ballots are used to generate either a machine count (faster, cheaper) or a hand count. You can toss the printed ballots after the time for a challenge has passed.
4. For the first N elections (where N is some small integer, 1 or 2 or 3, say) after the machines are installed in a county, a machine count is done as a matter of course. If the machine count is more than some delta different from the computer tally, consequences are triggered (these might include comprehensive software review, manual recount, and penalties due from the machine's manufacturer). This is repeated for the N elections after a software change.
As far as I can tell, not one of these provisions is being used. I wonder why not? They all seem fairly obvious and fairly reasonable. They solve the problems I've heard about with touchscreens, up to and including the theory that "Diebold is being paid by Bush to rig the machines!", as well as the problems that touchscreens are supposed to address (overvotes, ambiguous undervotes, slow counts). So what's up?