Rossi v. Gregoire: Recount number-crunching
My optimism is returning. This morning’s pessimism was based largely on a misapprehension that the 20,000 or so King County undervotes that I had heard referred to were actually uncounted votes for Governor, rather than all ballots without counted votes for governor. Because of the sheer number of people who voted only for President, this should not necessarily work out to be that many more actual votes coming in from King County. Thanks to Nathan of Pajama Jihad for setting me straight in the comments.
Sound Politics has kept up with the recount very well all day. At the moment, with 25% of the vote counted, Rossi has gained 17 votes. The big question, though, is what will King County’s votes look like?
According to this page (which, admittedly, may be out of date), King County uses GES Global Accuvote vote counting machines. Chelan, Klickitat and San Juan counties also use this same brand. The net chance in these counties has been +1 for Gregoire. The actual total changes were +3 for Gregoire, +3 for Rossi, -1 for Rossi, for a total of seven votes changed out of 47,841, or .015%. If King County, which counted 874,266 votes, has similar numbers–which it has no reason not to–then they should change 128 votes. Even if every single change went for Gregoire, which would be absurd, that still wouldn’t catch up to Rossi’s lead of, at the moment, 286 votes. If significantly more votes are changed than 128, however, then there will plenty of reason for suspicion.
For those who are interested, there have been 166 total changes in vote in the state so far, out of 709,442 total votes, or .023%. If King County follows that pattern (which it has no reason to) there will be 201 total changes. If the state as a whole follows this pattern, which it may well, there would be 482 changed votes.
However, the vast majority of those changed votes–112–came from counties using the ESS Opscan system (Adams, Columbia, Douglas, Ferry, Garfield, Kittias, Pend Oreille and Wahkiakum). Out of 49,242 votes, that’s .23%. (We can be thankful King County doesn’t use this method–that would project a total of 2,010 changed votes!) There are 5 counties left to count that use this system–Cowlitz, Gray’s Harbor (ugh), Skagit, Skamania and Walla Walla–with a total of 148,122 votes. This gives an estimated total vote change of 337. If the vote changes break the same way they have so far in Opscan counties, which it almost certainly won’t, Rossi will get 225 of these, Gregoire 157 and Bennett 18.
The counties that use neither ESS Opscan, GES, nor electronic voting (Yakima, which reported no change, and Snohomish, which should do the same) had a total of 47 changed votes, out of a total of 542,166, .0087%. (Interestingly enough, many of these are punch card counties, which appear to be more accurate than optical scans!) The rest of the 780,285 votes in the rest of the counties like this would thus be projected to add another 68 votes.
Overall, things still look good for Rossi. Votes are generally breaking in his direction, which isn’t surprising because these are most–but not all–good counties for Rossi. That being the case, some Gregoire counties gave votes to Rossi and vice versa. And, the numbers aren’t changing so much as to cause as drastic a shift in the numbers as Gregoire needs.
However, all this number crunching should be taken with several spoonfuls of salt. My “total votes changed” number may well be off by a large margin–if, say, in Chelan County, Rossi gained a dozen votes, and lost 11, that would be significantly more vote changes than the “1″ reported to the Secretary of State. Additionally, we apparently cannot be certain that everyone is counting votes in the same way (the drastically high changed numbers from Adams and Kittias insinuates that everyone is not). Last, the page where I’ve gotten my voting machine information appears to be somewhat out of date (if anyone has information that contradicts it, please let me know).
Still, while this number-crunching–like that I did before the recount–cannot be taken as a true prediction per se it does give us a good understanding of what we can expect if nothing wierd happens. We can expect a total of about 700 total changed votes. We can expect most of these to come from ESS Opscan counties. We can expect King County to change about 130 votes. If we start seeing different numbers than this, we’ll know that something wierd is happening. That, I think, is useful information.
And, for old time’s sake, here’s the spreadsheet I used for all this. Don’t expect much, I just sort of threw it together just now. But enjoy!
UPDATE: Scott of Living in Babylon has directed me to the Secretary of State’s list of voting systems, which is, of course, up to date. It doesn’t change much, but it does change some, and I’ll apply those changes to the spreadsheet and to the above post momentarily.
UPDATE: Fortunately, the only changes were in nomenclature, and they’ve been made. Thankfully, this takes one of those spoonfuls of salt away.
You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site. To comment, please register. No profanity, please. Non-family-friendly words will be changed to amusing euphemisms as suits the author.
November 23rd, 2004 at 12:32 am
Thanks! Please update more often if you can.
Bye,
A.
November 23rd, 2004 at 8:17 am
There is a list of which voting systems each county uses (complete with model numbers) on the Washington Secretary of State’s website, here:
http://www.secstate.wa.gov/elections/votingsystems.aspx
There is a wealth of information here, and I’m going to use it to update the spreadsheet on my blog soon.
November 23rd, 2004 at 11:14 am
That’s me nervously hitting ‘refresh’ on your blog. Thanks for your work on this!
November 23rd, 2004 at 11:44 am
I would strongly argue against the post you made here, where you are attempting to extrapolate mathematical formulas to the recount. That logic would work, and probably does work, when attempting to predict final results from partial results. However, in a recount of an election, I would argue that extrapolation anticipation is fatally flawed.
A recount, by definition, looks explicity for ANOMALIES, not for predictability. If you wanted predictability and rational outcomes based on projections, there would be no reason to do a recount, by definition. Washington state law, by ordering a recount when a race is decided by less than 0.25%, by definition, assumes that vote-counting has an error rate at least that high in an initial go-around. Thus, with the 2.9 million votes tallied, one should absolutely expect that errors, on a statewide basis, be at least over 7,000 votes. They could be in any direction, but that should be the number of errors.
Thus, counties showing that kind of an error rate are thus doing a credible and diligent job in their recount. Those counties, like Yakima that show ZERO errors on its volume of votes obviously did not take the recount seriously. There is no probability factor that would say — in almost any endeavor — that in a universe of over 70,000 units of anything (let alone the vagaries of multiple types of votes and human handling of those) the original count was precisely accurate! If I was the Secretary of State and saw counties come back with no changes, or changes of just a tiny handful of votes, I would immediately issue a challenge and order a hand recount.
Logic dictates that.
Note that I am not claiming that one side or another should gain any advantage from this. I am just commenting on what should be a logical (and illogical) error rate in the counting.
November 23rd, 2004 at 2:33 pm
You guys have way too much time on your hands
November 23rd, 2004 at 3:50 pm
Nelson,
I believe that your logic is also flawed, in that the law of large numbers indicates that although the chance of a single outcome may be small in one instance, when you apply many instances, ie counties, that given multiple outcomes some or many will come out w/ the “improbable”.
2nd I think that rather than indicating counties that dont take it seriously, I believe it may point to counties that took the voting much more seriously in the first go around. (thereby making a match more likely) Now if King county were to come up the same w/ all of there slop and loose accountability then it would point more to dishonesty. (although they certainly won’t in this case because they have been seen to have people inside trying to “come up with” gregoire votes. ie-”enhancements”, thrown out counters, etc.
Although some of your other stat comments were on the mark, the comment that recounts are triggered because the state or stats suggest ‘at least that much variability’ is also completely wrong. The state has never had a recount show that kind of variability that you suggest. The recount is triggered because of the “possibility” not the “probability” or “expectation” of a change.
November 23rd, 2004 at 4:47 pm
Nelson, you bore, I’m not even sure where to begin with you. Among many (many) other things, Yakima is a punchcard system, King uses an optical scanner that scans ballots completed by hand. Your comparison of the “error” rates (a misnomer) of those two counties is like, um, comparing apples to oranges.
Note also that the regs governing manual review of undervotes are “permissive. ” That is, officials in King are free to manually review a ballot that appears not to have a vote for governor — and based on their review of the ballot, complete an oval for the candidate they think the voter “intended” to vote for. In other counties they may or may not do the same, or may or may not decide to complete an oval that has a check mark in it (for example). And so on.
That process does not exist in Yakima, again which uses punch ballots, not ballots with ovals completed by hand. See above re: apples and oranges.
November 23rd, 2004 at 8:07 pm
Actually, Yakima now has an electronic voting system–which is even more of a reason for no one to be surprised that there were no changes. It’s downright impossible for there to be changes in the electronic voting during a recount. The only changes in Snohomish County came from those 200+ new ballots they found, and presumably from some other absentees that were miscounted originally.
Scott’s right, though. It’s silly to say that no changes means something fishy is up, when it could just mean that the first count was very careful. Or, as is the case in Yakima, very electronic.
November 23rd, 2004 at 9:23 pm
First of all, the two of you obviously missed my point completely because of your partisan bias. I scrupulously left any partisanship out of my original comments, focusing only on logical and probability. Again, logic and probability — the only rules that truly count — argue that in a unit of 70,000 instances, the fact that two totally independent counting actions can reach precisely the same conclusion strains credulity.
It could only be possible on a “garbage in, garbage out” scenario, if all entries were made on a single electronic machine and someone just looked twice at a single window on that machine and said, “yep, same numbers.”
Which is what you’re arguing about the e-voting. But we all know that that is never the case. There are mail-in ballots, absentee ballots, provisional ballots, etc. by the thousands. And all of them are handled by totally fallible and error-prone human beings.
Again, the State law requiring the recount makes the realistic provision that the system is sufficiently imperfect that a finite, if small, error rate WILL absolutely occur. As I stated earlier, nobody knows in which direction those errors will be made — nor does it matter which direction they occur.
The whole idea of a recount exercise is to make every effort to get it right, on the assumption that if you do something twice, you will be very likely to find — and correct — any error that was made the first time.
Ever count a deck of cards and end up with 51? Of course, everyone has. What do you do? You re-count it to make sure. If the second count is still 51 you then look for a missing card to once again re-confirm the accuracy of the count. But if the second count is 52, you then feel comfortable dealing out a hand.
All I’m saying here is that errors are absolutely certain in this system. It defies logic that there would be virtually no errors made in the first count. All it can mean is that the second count was equally flawed.
November 23rd, 2004 at 11:06 pm
Partisan? Where’d you get partisan. Personally, I’m just defensive. Were I partisan I’d say, ‘yeah, you’re right, that dirty no good Demmmycrat auditor in Yakima is trying to steal votes from Rossi!”
But I’m not. What I am saying, though, is that it’s silly to say that “errors absolutely WILL happen.” Probably, yes. But they’ve got lots of checks to make sure errors don’t happen. So, sometimes they won’t. Saying anything “absolutely will happen” is ridiculous, especially when there are two or three or more layers of oversight trying to make sure said thing doesn’t happen.
November 30th, 2004 at 6:37 am
[…] lse analogy that I’ve heard twice now, in a couple different forms–once in the comments, and once on the radio. In the comments, Nelson said Ever count a deck […]