In an article published yesterday, reporters from the Miami Herald provided data from hand counts that they conducted in several northern "dixiecrat" counties in Florida. They concluded that nothing was awry, and in fact, went so far as to title the article "No flaw in Bush's state win."
Here's a link to a reprint of the story that doesn't require registration
Here are the tallies for Union County:
Bush original: 3396
Bush hand count: 3393 (-3)
Kerry original: 1251
Kerry hand count: 1272 (+21)
Net change: Kerry +24
Here are the tallies for Lafayette County:
Bush original: 2460
Bush hand count: 2452 (-8)
Kerry original: 845
Kerry hand count: 848 (+3)
Net change: Kerry +11
It's a bit more complicated for the third county they looked at, Suwanee County, because they only report the totals for a hand count of "almost 60%" of the ballots.
Bush original: 11153
Kerry original: 4522
Bush hand count: 6140
Kerry hand count: 2984
In the original count, 71.2% of the votes cast for Bush or Kerry (n=15675) went to Bush. In the hand count, this drops to 67.3%. That is a significant drop. Let's translate that into numbers. If you take the percentages from the hand count and extrapolate, here's what you get:
Bush = 15675 x .673 = 10549 (loss of 604)
Kerry = 15675 x .327 = 5126 (gain of 604)
Net change: Kerry +1208
A switch of 1208 votes in a county with less than 16K votes cast is obviously huge. Now maybe there's a very large percentage of Bush votes in that remaining 40% that they didn't count, but we can't know that because they didn't count them. Which begs the question...why did they stop counting in Suwannee County when their tabulation of 60% of the ballots deviated so much from the original total? And without actually counting those remaining ballots, how can they possibly report that nothing is amiss when the data they have so far suggests a possible problem?
They conclude that there's "no flaw in Bush's state win." Sorry, but what I see is a possible gain of 1243 votes for Kerry from three small counties in which only 23627 ballots were cast. That represents about 0.3% of the ballots cast for Bush and Kerry statewide. If Kerry gained votes at the same rate statewide, he picks up nearly 400,000 votes and wins Florida.
Thanks Miami Herald, you just revealed to us in your hand counts that there's a possibility that Kerry won Florida.
update: 12/02
This post keeps resurfacing, so I thought I'd give an update.
The on-line Miami Herald article did not give all of the data needed to examine Suwannee County, the county where there were the biggest questions. In particular, they did not tell us which precincts were counted, nor did they give us the certified vote tally for the precincts that were counted. Given this, we didn't have the ability to directly compare the hand counts to the corresponding certified counts from the precincts they counted. I therefore had to extrapolate from the numbers they gave us, and this suggested that there *could* be a problem--not that there definately was a problem, but that there could be a problem (particulary if their 60% sample was at least somewhat representative of the county).
We now have additional information, and it shows that the hand counts in the precincts they chose to count did not substantially differ from the official counts. The vote totals changed only a little bit, just like in the other two counties.
So what about the extrapolation, then? Again, I did that only to show that based on the information they provided, there could be a problem, and that we needed more information (or a full hand count). Upon getting additional information, it turns out that the Herald reporters didn't follow standard practice when you're sampling a population, which is to either randomly sample from different precincts or choose precincts in a way that represents the full county. Instead they chose to count those precincts where Kerry faired best and not count those precincts where Bush won by the biggest margin. As a result of this choice of precincts, the 60% sample was skewed toward Kerry (and the uncounted votes are presumably skewed toward Bush, such that the numbers for Bush potentially could "catch up" to the official results).
So in a nutshell, the totals from the hand counts do NOT wildly deviate from the official counts. At the same time, there are some lingering issues:
--Why did they choose those particular precincts in Suwannee County? Their sampling of precinct was in no way logical or scientific, and seems to focus on those precincts least likely to show problems.
--Why didn't they count all of the precincts, and given that they left out those where Bush won by the biggest margin, what happened there?
--How can they possibly generalize the counting of 0.3% of the vote in small, northern, dixiecrat counties to the whole state, and make the ridiculous claim that their hand count shows "no flaw in Bush's state win"?
All in all, although the hand counts didn't reveal big problems, the write-up itself is still complete garbage in that it is no doubt politically motivated. Yeah, lets only hand count in those counties that are known to be dixiecrat counties, and that statisticians have already said aren't a big problem. And lets cherry pick precincts so as to stay away from those where Bush won by the biggest margin. And lets include a bunch of rhetoric and snide remarks from partisan election officials. And lets generalize from 0.3% of the vote in three small, dixecrat counties to the entire state. And lets ignore the counties that statisticians have pointed out to be a problem, like Miami-Dade and Broward. And lets ignore the counties where very suspicious activity has already been seen, and vote tallies do not match, like Volusia County.