Wednesday, November 16, 2005

hack the vote: ohio '05 edition

bradblog links to this piece in the columbus free press about the "impossible outcomes" in ohio's election last week. the final "results" for 4 of the 5 ballot initiatives were drastically different from the results of a columbus dispatch poll. the poll, which historically has been extremely accurate with a high level of confidence, was eerily accurate on issue 1 but off by ridiculously large numbers on the other four issues. statistically this suggests either an enormous error or voter fraud.

another thing: this year, 41 of ohio's 88 counties had just started using new diebold electronic voting machines. the state of california had decertified use of that model voting machine after a shameful 20% of the machines failed in a test. why would ohio adopt machines that california had such good reason to reject? was it simply a problem of ignorance, or something more?

the polling for issue 1 was dead accurate, as the free press explains:

A poll run on the front page of the Columbus Dispatch on Sunday, November 6, showed Issue One passing with 53% of the vote. Official tallies showed Issue One passing with 54% of the vote.

The polling used by the Dispatch had wrapped up the Thursday before the Tuesday election. Its precision on Issue One was consistent with the Dispatch's historic polling abilities, which have been uncannily accurate for decades. This poll was based on 1872 registered Ohio voters, with a margin of error at plus/minus 2.5 percentage points and a 95% confidence interval. The Issue One outcome would appear to confirm the Dispatch polling operation as the state's gold standard.

But Issues 2-5 are another story.

The November 6 Dispatch poll showed Issue Two passing by a vote of 59% to 33%, with about 8% undecided, an even broader margin than that predicted for Issue One.

But on November 8, the official vote count showed Issue Two going down to defeat by the astonishing margin of 63.5% against, with just 36.5% in favor. To say the outcome is a virtual statistical impossibility is to understate the case. For the official vote count to square with the pre-vote Dispatch poll, support for the Issue had to drop more than 22 points, with virtually all the undecideds apparently going into the "no" column.

The numbers on Issue Three are even less likely.

Though again opposed by the Christian Right, Issue Three drew an extremely broad range of support from moderate bi-partisan citizen groups and newspapers throughout the state. The Sunday Dispatch poll showed it winning in a landslide, with 61% in favor and just 25% opposed.

Tuesday's official results showed Issue Three going down to defeat in perhaps the most astonishing reversal in Ohio history, claiming just 33% of the vote, with 67% opposed. For this to have happened, Issue Three's polled support had to drop 28 points, again with an apparent 100% opposition from the previously undecideds.

The reversals on both Issues Two and Three were statistically staggering, to say the least.

The outcomes on Issue Four and Five were slightly less dramatic. Issue Four meant to end gerrymandering by establishing a non-partisan commission to set Congressional and legislative districts. The Dispatch poll showed it with 31% support, 45% opposition, and 25% undecided. Issue Four's final margin of defeat was 30% in favor to 70% against, placing virtually all undecideds in the "no" column.

Issue Five meant to take administration of Ohio's elections away from the Secretary of State, giving control to a nine-member non-partisan commission. Issue Five was prompted by Secretary of State J. Kenneth Blackwell's administration of the 2004 presidential vote, particularly in light of his role as co-chair of Ohio's Bush-Cheney campaign. The Dispatch poll showed a virtual toss-up, at 41% yes, 43% no and 16% undecided. The official result gave Issue Five just 30% of the vote, with allegedly 70% opposed.

how is it possible for a poll that was so accurate on issue 1 could be so totally wrong on the other issues? if all the results were wrong, you could blame the poll, attributing it to sampling error or the like. but the accuracy on issue 1, well within margin of error, suggests that the sample was indeed representative.

the free press concludes that something very wrong must have happened:

Though there were glitches, this year's voting lacked the massive irregularities and open manipulations that poisoned Ohio 2004. The only major difference would appear to be the new installation of touchscreen machines in those additional 41 counties.

And thus the possible explanations for the staggering defeats of Issues Two through Five boil down to two: either the Dispatch polling---dead accurate for Issue One---was wildly wrong beyond all possible statistical margin of error for Issues 2-5, or the electronic machines on which Ohio and much of the nation conduct their elections were hacked by someone wanting to change the vote count.

the free press points out that the GAO released a widely-ignored report on electronic voting, which concluded that the machines are easy to hack and full of security problems. bradblog posted this GAO report last month when it came out:

Voting System Vulnerabilities Identified by GAO:

  • Cast ballots, ballot definition files, memory cards, and audit logs could be modified.
  • Supervisor functions were protected with weak or easily guessed passwords, and memory cards that allowed individuals access to voting machines were inadequately protected.
  • Systems had easily picked locks and power switches that were exposed and unprotected.
  • Voting machine vendors had weak security practices, including the failure to conduct background checks on programmers and system developers, and the failure to establish clear chain of custody procedures for handling software.

Voting System Failures Have Already Occurred During Elections

In addition to identifying potential vulnerabilities, GAO identified a number of cases of operational failures in real elections. These examples included:

  • In California, a county presented voters with an incorrect electronic ballot, meaning they could not vote in certain races.
  • In Pennsylvania, a county made a ballot error on an electronic voting system that resulted in the county's undervote percentage reaching 80% in some precincts.
  • In North Carolina, electronic voting machines continued to accept votes after their memories were full, causing over 4,000 votes to be lost.
  • In Florida, a county reported that touch screens took up to an hour to activate and had to be activated sequentially, resulting in long delays.

sounds like something is rotten in the state of ohio.

No comments: