Earlier this week Dr. Alex Shvartsman of UConn sent us a revised, Version 2, of the Statistical Analysis of the Post Election Audit Data, November 2007 Elections <read>
This report is based on the added investigations of the discrepancies reported in the initial audit data reported in the Observation Coalition Report <read> followed by the original UConn Report.
While I have no reason to question the revised data or results, overall I see little reason to change the earlier Coalition conclusion that:
The audit statistics and observations leave us without the information necessary to vouch for the accuracy of many of the hand-counting results, whether those results indicated discrepancies or agreement with the counts obtained by the optical scanners.
I find no obvious flaws in UConn’s portion of the work. Nor do I question the integrity of the Secretary of the State’s Office. However, the report suffers from the legendary computing problem of “garbage in, garbage outâ€:
1. Since the discrepancy review, as I understand it, was conducted by a single person, no matter how trustworthy and reliable, it lacks the necessary transparency that would support credibility.
2. Since the review was conducted with ballots no longer under seal, such a count could use ballots that could be easily altered by insiders after the fact. Of course, modification is unlikely to occur in cases where the count was simply done inaccurately. However, modification after unsealing would be a very important part of covering up a programming error or fraud.
(Connecticut law is unclear if ballots need to be sealed after 14 days, while post election audits cannot begin until 15 days after the election. In addition to chain of custody concerns before and during the post election audits, most municipalities did not maintain ballots under seal after the initial audit count was conducted)
3. Our understanding is that many of the explanations reported in the review were not based on a review of actual ballots but a review of numbers reported looking for reasons that would provide differences that would approximate the difference between the machine and the hand counts, and attributing any such errors to inaccurate hand counting.
While hand counting errors might often be the actual case, if there were machine errors they could easily go undetected: Many such errors in programming could be expected to be approximately the same size as those attributed to human counting errors, since in both cases discrepancies would likely be caused by the same errors made consistently.
4. An investigation and report concluded some five months after an election falls far short of time-frames necessary for serious action based on significant reported discrepancies that might indicate errors, fraud, or the possibility incorrect election results. (As we have indicated occasionally, the motivation and paradigm of the current Connecticut audit law seems to be based on finding general flaws in voting machines, rather than anticipating, reacting to and correcting individual race results)
