Subject: Re: FALSE IMPRESSIONS -- PRESS 09152006 Suit: Ban computer voting
In addition to what Al Kolwicz says, all of which I agree with, it is doubtful that the DRE machines anywhere have actually been successfully tested in an election scale event. The reason this is basically impossible is that there is no way to accurately test vote tens of thousands of votes into a collection of DREs... short of possibly using a bunch of robots to do it, and even that might have accuracy problems. Any county which has honestly attempted to conduct a LAT on the DREs must be aware of this.
The potential for error in the DRE election tabulations is not a trivial consideration. It is not impossible that there are bugs within the software which compile the votes from the various DREs into larger tabulations (such as the JBC in the case of the Hart system) which could by tabulation register rollover or similar accidental faults modify the results in a particular range of count, for example. Unless the election scale tests are performed, we would not necessarily know about such faults which could be unintended or otherwise.
It would be possible to have an election scale test in which the VVPATs are counted and compared to the election tabulation. This should most certainly be done, and I suspect it has never been done on any realistic scale with any DRE system. To avoid human error in the counting of paper ballots, it is unquestionably an advantage to sort before counting. For this purpose it would be necessary to cut the individual VVPATs from their sequential positions on the paper roll. Of course, we all are depending on the possibility of this full scale hand recount being done for the safety of all our elections, but I doubt if anyone has demonstrated how to do it successfully in an election scale with real VVPATs.
On the other hand with paper ballots it is possible to sort and count and recount test paper ballots until a group of human counters are certain about and agree in spite of partisan differences to the expected result. And this group of test ballots can be entered into the counting machines as many times as necessary to gain confidence that they always produce the same result (or fail to do so). Paper ballots provide an entirely different and more testable situation, although again, it is doubtful that proper testing has been done in the actual ITA or state certification tests, including ballots incorporating questionable voter intent.
Ocasionally there may have been a hand recount of a given race in a real election, but the results of this recount are rarely if ever compared against the machine count with the intention of reconciling the differences to improve the machine accuracy. In almost every case, such a comparison is made with the expectation of improving the human count accuracy. Election failures are not collected into a single repository for the purposes of learning how to improve our voting process or even which devices are more prone to failure. Each county learns these things on their own, or fails to do so.
The effect of whatever causes questionable voter intent on a paper ballot has the very different and troubling result of a false impression of certainty on the DRE. This is not a technical advantage of the DRE, this is an irreparable fault of the DRE. The DRE unavoidably hides the condition that the voter either has difficulty voting, or is uncertain about the choice for some reason perhaps intentional. Apparently we are no longer concerned about the mechanization of society, so there seem to be few people arguing the need for humanistic interfaces nowadays, unfortunately. Many nations still vote entirely on paper and use human counting and still do this without difficulty.
It is time for us to be realistic about the need for and the lack of testing of our voting machines. The salient issue is not only securing our machines against fraud. The additional security may have the unfortunate side effect of a preventing public oversight of our election processes and this could actually lead to additional potential for fraud or accidental error.
Also it is critical that our post election audits actually perform an audit at a sufficient level of accuracy, and that they consist of an actual audit, not a recount retest. This will require some careful planning and may require the hand counting of more than the prescribed number of ballots as it is crucial to count to a number which is a real subtotal in the real election results. The accuracies resulting from the 2005 habit of recounting 100 ballots are embarrassingly low. In 2006 in most counties more than 100 ballots were counted in the audit, but this number should consist of at least several percent of the ballots in the election. Then the county should be prepared to hand count all the VVPAT and paper ballots in case a reported race result comes too close for the accuracy of the statutory tests (LAT or post-election audit). We should always have a paper ballot hand count in each close election.
Following these guidelines, in addition to making sure that all voters are aware of the significance of what is printed on the VVPAT, would give us a reason to trust our voting machines' results. Then there is also the process of determining whom is allowed to cast a vote to consider (elsewhere).
Chair, Eagle County Democrats
Member, Eagle County Canvass Board