As promised, analysis on the final recount. Here's the completed table:

Since the variances are so small, all the values in the table fall within probable statistical error. But they will still yield valuable information if we look inside the numbers for some comparison.

Of course, the King County numbers for Gregoire are troubling - remember that I mentioned in my earlier post that the variance between the Gregoire recount and the Rossi recount should be the same. Here's why: For statistical purposes, we need a representative sample other than King County to analyze. Let's give Gregoire the benefit of the doubt and take the highest vote count counties - without King County included for Gregoire votes. That would be Pierce, Snohomish, Spokane, Clark, and Thurston - in descending Gregoire vote order. We're going to look at these to compare them to King County. Another table is in order:

We can see that the total of the five counties is very close to the total Gregoire count for King County. We can see that the variance in the count for the five counties for both Gregoire and Rossi is 0.10%. For King County the variance for Gregoire is 0.12%. This is certainly not a large value but it is significant that it is the only one that differs from three other large samples.

Here's where the numbers become really significant: Look at the vote ratio between the total vote for each sample and the ratio for the recount for each sample. For the five county sample (all but one are red counties) Rossi leads 51.79% to 48.24%. What about the ratio of the 'new' votes that occurred during the recount? That ratio for the five county sample is: 51.79% to 48.21% - nearly identical to two percentage decimal places (that's ten-thousandths). It is what we would expect.

However, this is not true for King County. The total vote ratio for King County is a Gregoire lead of 59.05% to 40.97%. The ratio of the 'new' votes however, is a Gregoire lead of 63.02% to 36.98%.

Because of the variance calculation, it seems reasonable that the King County recount produced 348 votes for Rossi - it matches up with all the other numbers in the five county sample. But 593 votes for Gregoire does not. Statistically, that number should be between 499 and 503. During recounting of a large representative sample an undervote new tally is a relatively rare event. When they occur in such a large sample, it is reasonable to have the expectation that they will occur within a very close frequency.

What we are expected to believe is that, after collecting a group of samples in which there is almost exact correlation, a representative sample of nearly 1,000 is off by nearly 100 - roughly a ten percent error.

If you were doing a thesis, or writing a paper for a journal, or betting someone's life on a product design, this data would not be trusted.

I believe that it is a clear example of fraud perpetrated at King County.

UPDATE (12/1): There's nothing like new ballots that haven't been seen before showing up to be counted to skew a recount - 336 ballots to be exact.