Pelz comes out against proposed new Diebold system for King County

[eridani]

My letter thanking him for this--


I observed the current ballot duplication process in 2005, and it is very well controlled. The central scanners reject ballots for obvious reasons like visible stray marks, and sometimes for no obvious reason at all (possibly printing registry problems or non-obvious stray marks). All such ballots are duplicated by hand, and the number of the old and the new ballots recorded by hand in a bound notebook. Two people do the duplication; one reading the votes and the other marking the ballot. Then they trade, and the other person reads the duplicated ballot which is checked against the original. If they make a mistake, that is recorded and the spoiled ballot saves as well.

This may be tedious, but it is CHECKABLE, unlike the proposed new system which stores images electronically, and voter intent changes would wipe the original image and substitute a new one. This is a fundamentally unauditable process, and should not be allowed.

I'm glad that you are stepping up here. We don't have problems with our current system that can't be fixed by adding more machines and buying more memory for the current tabulation software. In addition, the new system is ILLEGAL--we are not supposed to be using equipment that has not been certified and that has not been used in other elections.

The worst part of all is that the few users who have tried the new system report that it DOESN'T WORK WITH BALLOTS THAT HAVE BEEN FOLDED AND MAILED. The Diebold demonstration only used ballots that had not been folded.

I sympathize with advocates of hand counting, but with ballots as complex as we have there is no alternative but optical scanning. Unauditable direct recording is flat out unacceptable--there is no reason why touchscreens used by disabled people could not be designed to print ballots that can be scanned along with everybody else's.

That doesn't mean you should trust the process--that's why we need random hand count audits. I depend heavily on computer software as an analytical chemist, but with every sample set I submit a hand calculation as a check against bit rot. Usually everything checks out, but last year I discovered a random error that occasionally multiplied my final results by ten and had to wipe my drive and reinstall everything. As Alastor Moody always says "Constant vigilance!" BTW, Boeing and Weyerhaeuser CEOs never whine about "Why don't you trust us?" when their shareholders continue to demand independent auditing of their books even though no hankypanky was uncovered last year. The Elections Department could learn from this.

That scanning and hand counting are very different methods is in itself a check and balance. Paleontologists who count tree rings and then submit samples for carbon-14 dating are very happy campers indeed when dates derived by these two different methods are in reasonable agreement.

As you have correctly pointed out, the 2004 recount was the ultimate audit. Given the outcome, naturally the Repubs are still whining and encouraging the ignorant jackasses that form their voting base to keep whining loudly and often. They've made a lot of political hay out of the fact that their base doesn't understand anything about method error.

The fact is that we don't know (and cannot in principle know) who "really" won, because you can't measure the difference between two tallies when the difference between the two final numbers is equal to or less than the method measurement error. Hand recounting has been tested and found to have an error rate of 0.1%. Optical scanning has a slightly higher error rate when a single race is counted, though it pulls ahead when three or more races are being counted at once. The difference between Gregoire and Rossi in all three counts was less than the error rate and about equal to the percent difference in the counts.

So we were confronted with a problem analogous to measuring to 1/32" accuracy with a ruler having only 1/4" divisions. The hand count was analogous to a ruler with 1/8" divisions. Since no available tool could have done the job, the law says use the best system you have (in this case hand counting) and live with the result.

The proposed new system is NOT checkable in this way, and should not be allowed. It's not worth jack shit that it's faster than our current system. The slow step is waiting for all the ballots to come in, and the slow step always determines the speed of the overall process. Thanks again for having the state Democratic Party take a stand against this untried system.