The much-delayed report from the November 2, 2004 election in Milwaukee contains evidence of the same HAVA violation. The complete report can be found here.

On page 37 of 67 the report, the investigators documented that the OPTECH IIIP Eagle scanners failed to scan ballots.

The exact figures reported are:

In Ward 43, the scanner failed to scan 18 of 1199 ballots.
  An error rate of 1.50%.
In Ward 44, the scanner failed to scan 7 of 1954 ballots.
  An error rate of 0.36%.
In Ward 98, the scanner failed to scan 10 of 660 ballots.
  An error rate of 1.52%.

The unsupported speculation offered by the City of Milwaukee Election Commission of how a ballot was fed into the front of the scanner, deposited in the locked box below the scanner, yet the scanner failed to register a ballot had been cast, was that the ballot somehow the ballot "jammed".

These three error rates are all substantially above the maximum error rate allowed by HAVA (1 in 500,000).

For those who are interested, here’s how I calculated the numbers and percentages in this article.

1. Count the total number of choices on a single ballot to get the Ballot Positions Per Ballot.

2. Multiply the number of Ballot Positions Per Ballot by the number of ballots cast to get the Total Ballot Positions.

3. Count the number of choices a voter can record on a single ballot to get the number of Votes Per Voter. (If a voter may only vote for one candidate in each contest, the number of Votes Per Voter is the same as the number of contests.)

4. Count the number of Votes Per Voter that are recorded incorrectly or not at all to get the number of Ballot Position Errors. (When entire ballots are lost, as in Butler County, this is the number of lost ballots multiplied by the number of Votes Per Voter.)

5. Divide the number of Ballot Position Errors by the Total Ballot Positions to get the Error Rate as a decimal number. Multiply by 100 to get the Error Rate as a percentage.

6. Divide the decimal Error Rate by 0.000002 (the maximum allowed by law --- 1 error in 500,000 ballot positions). This Violation Factor is the number of times higher the Error Rate is than the rate allowed by federal law.

Note: I reported a range of Error Rates and Violation Factors for Butler County, since the Republican ballot was over twice as long as the Democratic ballot, and I wasn’t able to determine how many of the lost ballots were for each party. If all ballots had been Democratic, the Violation Factor would have been over 400. If all ballots had been Republican, the Violation Factor would have been over 900. A combination of both party ballots would place it in between.

(Multiply the Violation Factor by 100 to get the percentage that the Error Rate is higher than the 0.0002% allowed by federal law.)

Please explain why a one-vote difference amounts to a 60,000 times legally allowed error rate? This is very confusing:

New Jersey election officials discovered discrepancies in the 2008 primary election summary reports of 60 of their Sequoia Advantage touch-screen e-voting machines. Sequoia blamed the discrepancies on poll worker error, but subsequent discoveries disprove the company’s explanation.

One of the Sequoia Advantage vote summary tapes from Englewood Cliffs, District 4, in Bergen County, New Jersey showed 106 votes cast by 105 ballots. In addition, the automatic counter on the machine showed only 105 voters. (At least one other machine reported the same types of discrepancies.) Princeton computer science professor Ed Felten offers more details and photos of the tapes here.

Since the machine counter matches the total number of Republican and Democratic ballots cast, it appears that the machine added one vote to the totals for one of the candidates. In this case, the error rate on the Englewood Cliffs Sequoia Advantage machine was over 60,000 percent higher than the maximum error rate allowed by HAVA. [HUH?]

Please explain in the main piece this issue.

Art,

Good question.

The 84 (or perhaps 83) Democratic ballots had 8 ballot positions each (8 candidates) for a total of 672 (or perhaps 664) ballot positions.

The 22 (or perhaps 21) Republican ballots had 7 ballot positions each for a total of 154 (or perhaps 147) ballot positions.

That gives us a grand total of either 819 or 818 ballot positions (depending on whether a Democratic or Republican vote was added by the machine). To be conservative, I use 819 (more ballot positions means a lower error rate).

One error in 819 is 0.0012. Note: I rounded down to be conservative.

The allowable error rate is .000002 (five zeros). So I divided the actual error rate (.0012) by the allowable error rate to get 600. This means the actual error rate is 600 times the allowable rate (or 60,000%).

It's a bit more obvious if you compare "one error in 819" to "one error in 500,000." Then you can easily see how far it exceeds the allowable maximum.

The reason a one-vote error makes such a huge difference is that the total number of ballot positions on the 105 ballots wasn't even close to 500,000. The folks who wrote the standards were attempting to ensure complete accuracy, so they made the allowable error rate exceedingly small (one vote in 500,000 ballot positions).

Consider what one vote per machine means if a county has 5000 machines. Bear in mind, also, that if a machine can add a vote out of thin air, why should we think it hasn't subtracted some from one candidate and added them to another --- even non-maliciously. A miscalculation of one on a computer reveals a severe, severe bug.

Art,

PS. The reason I didn't explain all this in the article itself is that, as you see, it's kind of a complicated explanation.

The answer is PAPER BALLOT, HAND COUNTED. Anything else is a waste of time.

My prediction: the voter turnout will plummet this fall as millions of Americans now understand that the elections are being stolen every time now.

Although I don't read your site regularly I have seen postings elsewhere and usually your work is up to snuff. But, sorry folks, this post is ridiculous. A friend and I compared votes in the two NJ counties we live in, Ocean and Warren, on opposite sides of the state. In the Republican primary McCain received votes in every town or township in both counties with a deviation of just 4%; same with Ron Paul. In every single town in these two counties McCain's vote was 53%-57%; Ron Paul's was 4%-7%. This is prima facie statistically impossible. Sure, they both could have achieved these votes overall; but the votes in the individual towns, in two widely separate, demographically diverse counties, on either side of the state, should show much wider deviation. Instead, what they show is computer manipulation. The only thing your analysis proves is that the machines were inefficient, not even that they were rigged, which they obviously were. So while you prattle away the entire election was stolen in front of you. It is said of some dogs like Golden Retrievers, that they make great watchdogs: they just sit and watch the theft happen because they haven't got a clue to what's going on. The owner comes home, the dog starts yapping about "allowable error rate," but of course the family jewels are long gone.

Ellen is correct that the error rate is based on ballot lines not ballots. My math above is incorrect because of this oversight. the correct error rates (rounded to two decimal places) for the three location listed above are:
0.43% for Ward 43
0.10% for Ward 44
0.43% for Ward 98

These error rates are 2100 times and 500 times the maximum allowable error rates.

Here is a detailed breakdown of the observed error rates from 4 locations in southeastern Wisconsin from the Novmeber 2, 2004 election.

Dear Mr. Forbes:

I think you misunderstand the point of this thread. Yes, this is an arcane part of the Help America Vote Act. But, it is arcane because it has been systematically ignored by those charged with enforcing this provision of HAVA. But, it is a part of HAVA and it is a requirement upon which a voting system can be declared as not HAVA compliant.

HAVA compliance is critical to a voting system.

If a voting system is not HAVA compliant, then that voting system cannot be used in a federal election. The Help America Vote Act is absolutely clear on this.

It may be true the NJ election was stolen. But, statistical anomalies are not evidence. Observed data on error rates IS evidence.

Assume you are correct and the election was stolen using the Sequoia equipment. If the Sequoia AVC Advantage had been properly determined to violate the HAVA accuracy requirement, then the AVC Advantages would not have be available as tools with which to steal the election.