Saturday, April 30, 2005

Vote notes

(Incidentally, I tried to publish a vote fraud update a couple of days ago, but the cyber gods decided to kill it before delivery.)

Explaining the exits: A Kos diarist named Elizabeth Liddle, who writes under the nick "Febble," is a Scottish psychology specialist who, although initially a supporter of the fraud thesis, claims to have come up with a computational model which explains away the exit poll disparity. She offers her study as a response to the US Count Votes analysis, which is finally starting to acquire some attention.

I have not yet truly studied her paper -- and to be frank, from the looks of the thing, I'm not sure whether I would understand the math even after a close reading. However, a skimming indicates that she has not yet addressed what seems, to me, to be the most damning argument: The exit polls asked whether the responders had voted for Bush or Gore in 2000. Even though Gore won the popular vote, the 2004 exit polls indicated a strong pro-Bush 2000 tilt. This oddity provides us with prima facie evidence that the exit pollsters favored Bush, not Kerry.

This anomaly demolishes Mitofsky's beloved "reluctant Bush responder" thesis, which is, in the end, Liddle's bottom line. Everything comes down to this:

While we cannot measure the completion rate by Democratic and Republican voters, hypothetical completion rates of 56% among Kerry voters and 50% among Bush voters overall would account for the entire Within Precinct Error that we observed in 2004."
And if we accept malign elves as a hypothetical, then we can ascribe the disparity to their mischievous handiwork.

As one observer noted, a proper hypothesis should be (if it can be) put to the test. The "2000 question" provides one way to test the presumed 56%-50% response differential -- and that test does not support Liddle or Mitofsky. Once we toss out the "chatty Dem/tight-lipped Rep" scenario -- a scenario which, I might add, does not conform with real-life experience -- what have we left?

On Democratic Underground, "Truth Is All" provides a dizzying number of links which refute both Liddle and the "reluctant responder" hypothesis. Many of these links lead back to TIA's own writings, of which perhaps the most immediately relevant is this:

How can Mitofsky push a "Reluctant Bush Responder" theory, especially when analysis of the data in his own report AND THE NATIONAL EXIT POLL reveals just the opposite?

According the, Bush responders in Republican precincts were MORE inclined to speak to the exit pollsters.
I do not question Liddle's expertise or sincerity. But this is an issue she must address.

Sequoia: A good article on Paul Lehto's lawsuit against Sequoia appears in Real Change News, a journal largely devoted to advocacy for the homeless.

In what's being hailed as a simple but brilliant argument that could serve as a national model for legal action, Gordon's lawsuit, which was filed in King County, argues that it's unconstitutional to outsource voting to a company that keeps the process secret the way Sequoia and other makers of electronic voting equipment do.

Makers of touch-screen voting machines claim their software is a "trade secret" to protect. In the contract between Sequoia and Snohomish County, which purchased 1,000 machines for $5 million in 2002, the county auditor even agreed to defend California-based Sequoia from any subpoenas seeking information that, with any other voting method, are a matter of public record under the state Constitution, the Open Meetings Act, and the Public Disclosure Act.

Gordon, a Bellevue tort attorney known for his work in civil rights and labor cases, says that particular paragraph of the Sequoia contract that will go down in infamy as an "unholy alliance between government and a private corporation."

"We think there's a problem with a county entering into a contract with a private company which has the effect of letting the private company conduct elections."
As a result, Gordon, Lehto and a second plaintiff, John Wells, are asking the court to void Sequoia's contract and make the company give back Snohomish County's $5 million.

"A contract between two parties can't take away the rights of a third party, and that's what's happened in this case" and across the nation, says Ellen Theisen, executive director of Port Ludlow-based Voters Unite, a one-year-old voter advocacy organization. "It's taken away rights of citizens to an observable election."
Lehto will be appearing with Bev Harris at Seattle's College Club on May 4. Although I've had my problems with Harris, I urge anyone in that area to attend. (And maybe someone can ask Bev about those election tapes she picked up in Florida.)

Why hide it? The major voting machine companies all claim that their software is a trade secret. Why the secrecy? Australian law insists on open software. Insisting on transparency has not caused that nation to fall into the sea -- in fact, their elections have fewer problems.

Given the incestuous linkages between the major voting machine companies, secrecy has little purpose. It's not as though these guys are competing with each other. I advise the reader to check out this Democratic Underground page.

Fitrakis says it all. Sorry I didn't mention this earlier: Bob Fitrakis has a killer article on Blackwell's perfidy in Ohio. The most noteworthy section:

The Free Press is printing for the first time a hand-drawn map from an employee of the Warren County government. The employee, who wishes to remain anonymous for fear of retaliation, communicated to the Free Press thoughts on what happened on Election Day in the county that received national attention by declaring a "homeland security alert" while the votes were being counted.

With the media and independent election observers "walled off," as the Cincinnati Enquirer described, the employee claims that "some ballot boxes were taken to the holding area" where they were not monitored by election officials. The Warren County employee referred to the person supervising the unauthorized warehouse as a "Republican Party hack."

The employee is concerned that it would have been easy to "stuff" the ballot boxes or that "signatures could have been forged in the unauthorized holding area."

The anonymous employee told the Free Press that testimony would be provided if subpoenaed. A list of questions that remain unanswered in Warren County was also supplied: Which precinct ballot boxes were taken to the unauthorized holding area? Were officials from the Board of Elections present in the holding area? When were the ballots taken from the holding area to the check-in tent that was erected temporarily to count ballots? If Warren County was under a state of emergency, then why weren't any metal detectors engaged? The FBI has denied that there was any homeland security threat on Election Day.
Koehler: The Chicago Tribune (a notoriously reactionary paper) refused to run columnist Bob Koehler's powerhouse piece on vote fraud (available on the web -- here.) But the paper did see fit to print a piece which attempts to debunk Koehler. The article, by Don Wycliff, does not even begin to address the many outrages detailed by John Conyers' committee.

Josh Mittledorff, of the Philadelphia Inquirer, wrote a column on the Nashville conference that deserves your attention:

I met David Griscom, a retired physics prof who spent months with colleague John Brakey poring over election tapes, signature rosters and "consecutive number registers" from Brakey's Tucson home precinct.

They audited and verified, one by one, the 895 votes in the precinct and found: 12 innocent and unsuspecting voters who had their names duplicated on the roster and their votes for Bush counted twice. Twenty-two "undervotes" where the machine had failed to register a preference for president, and these had been dutifully and meticulously converted to 22 votes for Bush.

The "Republican" and "Democratic" co-directors of the polling place were a local fundamentalist preacher and his wife. Thirty-nine of their parishioners from another precinct had cast provisional ballots, which were (illegally) converted to regular ballots and passed through, all 39 for Bush.

I met Richard Hayes Phillips, a geologist from New Hampshire who was invited to Ohio to study the integrity of the vote, and realized that a complete inventory of lost and miscounted votes was needed. To date, Phillips has analyzed 15 of Ohio's 88 counties, and by his most conservative estimate has found 101,000 uncounted Kerry votes - 136,000 is the margin by which Bush officially defeated Kerry.


Anonymous said...

Re the last sentence: I believe 118,000+ was the official final vote difference in Ohio.

One more thing re exit polls: how was it that in Colorado the polls for US Senator were accurate, but for president there was a large swing to Bush?

Febble said...

A response:

I have not attempted to "explain away" the exit poll disparity. I have simply demonstrated that one inference made by USCV from the data, namely that the data we have indicate "Bush strongholds have more Vote-corruption" is not justified. My math is "fraud-neutral". My concern is that the E-M report dealt with a measure of error (the "within-precinct-error" or WPE) that is mathematically confounded with vote margin. This means that any analysis of WPE by vote margin (i.e. how much of a Bush or Kerry stronghold a precinct was) is confounded by the fact that WPE is itself a function of vote margin.

The tables (given in the E-M report) of mean and median WPE values by precinct category, are therefore suspect, as are any inferences by USCV based on them. By modelling the kind of error likely to be introduced into the E-M tables by the confound, I concluded that, contrary to the USCV claim, that tables are also consistent with over-estimates of Kerry's vote that are randomly distributed across all precinct categories.

This remains true whether the cause of the over-estimates is fraud or differential non-response.

The arguments for the latter are not in my paper, but in the E-M evaluation, which reports that the over-estimates were greatest were sampling protocol was poor. This is a strong finding, as there would be no reason to suppose that fraud would be greatest where the pollsters were messing up. However, because of their use of a confounding variable, I believe these analyses should be re-done, using my variable.

I would also like to see whether a plot of my variable against vote margin confirmed or refuted the idea that overestimates were greatest in Bush strongholds. I am not saying they weren't. I am saying that it cannot be inferred from the data we have.

Regarding response rates. It is not correct to say that "Bush responders in Republican precincts were MORE inclined to speak to the exit pollsters." The E-M report states that there was no significant difference,although they do not give details of the statistical test. However, it is true that the "differential response" hypothesis predicts that response rates should actually be lower in Bush strongholds than in Kerry strongholds - because allegedly, Bush voters don't like responding.

This is an important argument. However, statistically, a small difference is only detectable if you have enough statistical power. The predicted difference is small (in theory you should get 56% response from an all-Kerry precinct, and 50% in an all Bush district, so the maximum difference is 6 percentage points). Statistical power depends on two things - plenty of data and good signal-to-to noise ratio.

Here the numbers are small where it matters - in the extreme precincts where the differences should be maximised. There were only 90 "high Kerry" precincts and only 40 "high Bush precincts". So the data would have to be pretty clean. We know it was not, as the E-M report gives as a value called the "absolute" error, which, without going into too much detail, tells us how much noise there was in the data.

When I did my model, I compared the amount of noise in my model data with the amount of noise in the real data. I had less, and even so, statistical significance of the lower responses in my "high Bush" precincts was only borderline. I estimated that if I increased the noise so that it matched the real data, the difference would be lost in the noise. It only takes a couple of stray data points to lift the response rates at the "high Bush" end, or lower them at the "high Kerry end" to destroy the effect.

So I concede that the lack of evidence of lower response rates in "high Bush" precincts than in "high Kerry" precincts is an argument worth considering. However, the way statistics works that what you test, statistically is a "null hypothesis". If you cannot show that an effect is unlikely to have occurred by chance you "retain the null". If you "retain the null", have not proved that the effect was not there. You simply have not proved that it was. It is the statistical equivalent of "presumption of innocence".

Unless E-M provide us with better data, I do not think we can reject the null hypothesis that over-estimates of Kerry's vote were randomly distributed amongst precincts of all degrees of partisanship.

Whatever the reason for the over-estimates.

DemFromCT said...

Thos of us who are impressed with febble's work and wish it to be disseminated for more peer review do not disagree with checking out voting irregularities, suppression etc, and do not disagree with the concept of needed voter reform. Blackwell, along with Kathryn Harris, remain models of what Secretaries of State should NOT be. Local insistance of nonpartisan SoS is at least as important as paper trails, etc.

As for the difference between senator and president in exit polls, there is always a huge amount of ticket splitting that confounds polling. The discrepancy is not proof of anything.

Anonymous said...

demfromct: The ticket splitting itself isn't the confounding thing. The confounding thing is that the results for senate, as predicted by the exit polls, were accurate. But the results for president as predicted by the exit polls were off the mark.

DemFromCT said...

cmg, see comments here. The gist is that the noise in polling makes Senate-President an unhelpful comparison; bias in presidential polling does not automatically translate into senatorial polling, even within the same state.

However, there needs to be more work done on this topic, as I haven't seen (or can't recall, anyway) a definitive treatment of this.

Anonymous said...

I just ran across "Vote notes" while conducting some research on various Instant Detective sites. Even though it wasn't really what I was searching for, I think the people who visit your site might still be interested in reading our reviews of various sites including Instant Detective. I started creating these review sites to help people determine what's good and what's bad on the net. As you know, the more information people have before spending their hard earned money, the better. Your readers can even add their opinions for others to read. By the way, I enjoyed reading the posts on your site.

Anonymous said...

Hey Blogger, I am conducting some research on various Detective sites and ran across this page. Even though it wasn't "exactly" what I was searching for, I think the readers of "Vote notes" still might be interested in reading our review of Detective sites and other similar websites. As you know, the more information people have before spending their hard earned money, the better. Sorry for the self-promotion, just trying to get the word out while looking for more sites to review. On to the next -- Thanks!