December 24, 2004
Unfunny Surprises

The following table illustrates how some of the various county recounts produced surprising results. Specifically, I wanted to see how the newly added ballots in each count were distributed. (where "newly added" means previously unexamineed ballots, or former undervote/overvotes that were first awarded to one of the major candidates in the recount). A "surprise" is an outcome where the newly added ballots disproportionately favored one candidate or another, and in such an extreme ratio that cannot reasonably be attributed to either (a) random variation, or (b) a fair process used to correct random errors from the previous count. In other words, the outcome is such a statistical outlier that it should only be attributed to either (a) that the recount was biased in favor of one candidate, or (b) the recount corrected a bias against that candidate that existed in the earlier counting process.

For some reason, all of the surprises favored Gregoire. Again, that does not necessarily mean that the recount rules and implementation were biased in favor of Gregoire voters. It could also mean that the earlier counts were biased against Gregoire voters.

But it has to be one or the other and it's important to understand which one it was. It's not sufficient to simple say "the recount is biased", or "the recount is more fair accurate". We need to understand what errors and biases were either corrected or introduced. First, so we can understand whether all the recounting did it was supposed to do in this election, and whether the current result accurately reflects the decision of the voters. But more importantly, so we can improve the processes for the future and avoid another crisis of legitimacy that we're now facing.

Note that when I say "bias", it doesn't have to be the result of willful design. it simply means a process that gives an unfair advantage to one side or another. For example, it could be explained as innocently as in the machine recount the county discovered a box of previously overlooked ballots from one precinct that happened to strongly favor one candidate. Thus the resulting recount was a fair process that corrected an earlier bias.

With that, it's up to the public to demand that the auditors from the surprise counties to explain whether and how bias was corrected or introduced. King County in particular has a great deal to explain.

The surprises are highlighted in yellow. A complete explanation of the table columns follows below the table.

Machine Recount
Manual Recount
County % Change New G New R Bias Prob. % Change New G New R Bias Prob.
Adams 0.78% 12 27     0.10% 0 -5 -2  
Asotin 0.02% 1 -1 -1   0.18% 5 10 +1  
Benton 0.00% 1 2   0.01% 3 5 -1  
Chelan 0.01% 3 1 -2   0.00% 0 1    
Clallam 0.01% 1 1   0.02% 4 4  
Clark 0.01% 3 7 +2   0.04% 28 30 -1  
Columbia 0.05% 1 0 -1   0.05% 0 1    
Cowlitz 0.17% -29 -40 -6   0.01% -3 -2    
Douglas 0.05% 2 4     0.02% 1 1    
Ferry 0.16% 3 2 -1   0.00% 0 0    
Franklin 0.01% -1 1 +1   0.16% 10 15 -2  
Garfield 0.16% 1 1   0.00% 0 0    
Grant 0.28% 26 44 -4   0.03% -5 2 +4  
Grays Harbor 0.04% 6 5     0.07% 10 8 -1  
Island 0.01% 0 5 +2   0.03% 7 3 -2  
Jefferson 0.03% 1 -4 -3   0.08% 8 6    
King 0.11% 593 348 -38 0.67% 0.06%
358
179
-41
0.02%
Kitsap 0.03% 15 -19 -17 <0.01% 0.14% 72 82 +4  
Kittitas 0.29% 19 26 -1   0.00% 0 0    
Klickitat 0.01% 0 1     0.00% 0 0    
Lewis 0.02% 2 4     0.04% 2 9 +2  
Lincoln 0.02% 0 1     0.00% 0 0    
Mason 0.01% 1 2     0.09% 9 12 +1  
Okanogan 0.01% 1 -1 -1   0.10% 6 10    
Pacific 0.01% 1 0     0.00% 0 0    
Pend Oreille 0.14% 6 2 -3   0.03% 0 2 +1  
Pierce 0.17% 242 261 -1   0.14% 232 201 -24 1.07%
San Juan 0.01% 0 -1 -1   0.00% 0 0    
Skagit 0.32% 71 89 +3   0.04% -16 -5 +6 3.13%
Skamania 0.00% 0 0     0.08% 1 3 +1  
Snohomish 0.09% 131 130 -3   0.07% 119 75 -21 0.15%
Spokane 0.13% 121 134 -3   0.01% 8 15 +3  
Stevens 0.00% 0 0     0.01% 0 2 +1  
Thurston 0.01% 2 4 +1   0.03% 15 13    
Wahkiakum 0.05% 0 1     0.05% 0 -1    
Walla Walla 0.43% 59 37 -25 <0.01% 0.07% 2 13 +3  
Whatcom 0.00% 3 0 -1   0.04% 16 22 +3  
Whitman 0.08% -9 -4 +3   0.05% 7 2 -3  
Yakima 0.00% 0 0     0.08% 20 35 -1  
Totals 0.09% 1,289 1,070 -99   0.04% 561 569 -68  

[NOTE: for the purpose of this table and discussion, all of the numbers pertain to votes for Rossi and Gregoire only, excluding minor party, write-in, and blank and spoiled ballots.]

For each recount, we have the following columns

% change: total number of votes that changed in the recount as a % of the previous count
New G and New R: new votes for Gregoire and Rossi in that recount, respectively
Bias: Difference between the expected number of changed votes to fall to Rossi and the actual number of changed votes that went to Rossi. For example, if the previously counted ballots in a county went 60% Rossi, 40% Gregoire, and there were 10 new votes in a recount, one would expect these to fall 6 Rossi, 4 Gregoire. If the actual breakdown were 7 Rossi, 3 Gregoire that would be noted as +1. If the actual breakdown were 4 Rossi, 6 Gregoire, that would be noted as -1.
Prob.: The probability of the given outcome based on the county's returns in the previous count and under the assumption that a random selection of new ballots is expected to be distributed proportional to the previously counted ballots in the county. i.e. the binomial distribution, calculated by Excel using =binomdist(L,L+W,p(L),1), where L is the number of votes for the losing candidate in the newly added ballots, W is the number of votes for the winning candidate, and p(L) is the percentage of the previously counted ballots that went to the losing candidate. [suitable adjustments made when L and/or W are negative]

A few additional notes:
If there is no Bias, i.e. Rossi got as many new votes as expected rounded to the nearest vote, the Bias cell is blank.
I only display the Prob. value if (a) it is <= 5% and (b) the absolute value of the Bias is >= 5. All of the other biases are too small to be considered outliers and not worth cluttering the table with the probabilities.
I only calculate one Prob. for a county recount that favored Rossi -- Skagit's manual recount. It's not really an outlier, as with 39 counties it's not unexpected to get one data point that has a probability of 3.1%, but that is the closest thing to a surprise that favored Rossi and I leave it in there for comparison sake. Posted by Stefan Sharkansky at December 24, 2004 02:07 PM | Email This

Comments
1. These statistics ignore the fact that machines do make mistakes, and the votes in King County were counted correctly. Apparently Kitsap's error in the machine count was corrected in the manual recount (Rossi -34 in machine, +10 in manual).

This assumes that any new ballots resemble that which was already cast. THat assumption cannot be validated until the ballots are opened in a very public and transparent way.

Also, a voter migth have crossed out one vote, the machien recorded a vote for the unintended candidate, and in the manual count it was reviewed and found not to be a valid vote.

Posted by: John Slyfield on December 24, 2004 03:01 PM
2. Stephan great job. Some thoughts

1. In county after county, precinct after precinct ballots will read the same thru the machine counters the same. Extra votes are added when ballots do not read correctly. There are cresses or tears or bubbles marked or stained. At this point the tyranny of the 2-1 vote takes over. Yes each ballot is examined on its merit, but the underlying logic is this. The Gregore voter is not as bright or able to fill out the ballot as a Rossi voter. The Gregore voter needs a little extra push or help to enable their vote to count. Unjust or unfair. Do a little research on under and over votes. Democrats insult democrats.

2. Now as to the handcounts. We know extra votes were added in Pierce, King and Snobomish counties. Canvas boards added in votes not previousily counted. In addition to those votes ballots were judged by boards as to whether a valid vote was cast or not. In King county 47 plus for Gregore and 12 minus for Rossi. The marks, smudges, stains, and spellings seem to have been overwhelming for Gregore. I call this the tyranny of the 2-1 vote. Gregore is being elected by the canvas boards of three or four counties and not the electorate. The vote is 2-1.

Posted by: Ralph on December 24, 2004 03:17 PM
3. This analysis exists at a higher bandwidth than I possess. I'll say that errors are always made when people are involved.

They become suspicious when they add up in one direction and are self-serving, as is the case in King County.

Posted by: South County on December 24, 2004 03:20 PM
4. Stefan - You're our man. Merry Christmas!

Posted by: CP on December 24, 2004 03:51 PM
5. We can overdo statistics. It would be more informative to look at all of the ballots that were enhanced or duplicated on the original count and machine recount, as well as those which were decided by the county canvassing board on the manual recount. Historically, that is what has happened when courts or legislatures (especially legislatures) have had to decide election contests. At best, statistics can point out likely problem areas. At worst, you can start to sound like the Ohio conspiracy theorists who think that Kerry's victory in that state was somehow stolen.

The other statistic that would be important is simple math -- addition and subtraction. Compare the individual precinct ballot totals in every county, and by category, for each of the three counts. King County has had wide swings that are unexplained -- +336 in the machine recount, and +59 in the manual recount. AS you said earlier, +336 was a net figure, since about 1100 ballots appeared in certain precincts and 800 disappeared in other precincts. In Kitsap County, 16 or so ballots disappeared in the manual recount.

Also, count the number of actual people from the voting lists in each precinct, and compare that with the ballot count.

If we have more than 130 ballots appearing from people who didn't even vote, more than 130 ballots disappearing from people whose votes were actually accepted, or any combination thereof, then this election must be nullified and rerun.

Posted by: Richard Pope on December 24, 2004 03:55 PM
6. this is another, sadly to say, King county election that typically rules all of Washington state. The voters have to worked twice as hard to nullify this county that seems to shape all the policies of washington, and not for the better of this state!

Posted by: Janet on December 24, 2004 04:42 PM
7. I dont see anything out of the ordinary, considering the variables.
This 2-1 tyranny you suggest, actually only involved 20 votes.
In other words, only 20 ballots were not unanimous, and several were in rossi's favor i believe.
I also believe calling the Supreme Court,liars, as Vance is doing, will not further your cause.
Many repubs. claim there has been a re-canvassing, when in all actuality, these votes were simply left out of the canvassing to begin with.
This pile of votes, was only matched to a computer database, and not to hard copy records as were all other ballots including ones from repubs.
These votes were then set aside in the middle of being verified as rejects.
Maybe it was the lone republican you speak of that actually didn't finish doing his job properly, but I digress.
I would happily accept another election, were it necessary, but as we can easily see if we choose truth, all votes were now checked and re-checked, and determined to be valid by quite a diverse group.
Why, by the way, do you only worry about the 2-1 ratio of dems to repubs in king county?
Chelan county had no dems on their canvassing team.
Who knows what they did.

Posted by: D.D. on December 24, 2004 04:54 PM
8. Let me try interpreting some of these numbers:

Those biased hand counters in King, Kitsap, and Walla Walla sure have some 'splainin to do! Especially those lefties in Kitsap and Walla Walla. Less than a .01% probability and 42 votes STOLEN from Rossi.

And what's with the bias in the machine recounts in King, Pierce, and Snohomish. You guys kept telling me those machines were supposed to be SO much better.

Now I can't trust machines OR humans.

I know, we should just let the commenters in this blog chose the next Governor! That'd be unbiased.

Posted by: Anonymous Reader on December 24, 2004 05:03 PM
9. Stephan,

Nice work. It clearly shows that the outrage should be concerning Adams County, King was one of the best. Oh, wait, Adams didn't produce Gregoire votes.

The chart dispels the conspiracy theories about the Western counties.

Posted by: Erik on December 24, 2004 07:30 PM
10. What happened to "count every vote"? Guess that refrain is dead now that the Dem. machine in King County found the right amount of votes. We must keep up the pressure in the national media and talk radio/bloggers to redo the military votes. Who better qualified than the guys who risk their lives so we can be free?

Keep up the good work, Shark!

Posted by: Judi on December 24, 2004 09:49 PM
11. I heard a great suggestion tonight. Someone outside of WA needs to be investigating this. The statistical analysis alone is grounds for a Federal investigation into vote fraud.

I really think that the only chance that either side will ever consider this a fair election is if it is reviewed from the outside. But someone (preferably with a little clout) needs to make a formal complaint to get the Feds to look this way.

Has anyone made such a complaint? Has the FBI or Atty General commented on whether they have jurisdiction?

I'd lose my job if I made the complaint, have any of you done it? Somebody needs to poke their noses into this mess.

Posted by: Just Askin on December 24, 2004 10:06 PM
12. Tyranny of the 2-1 and finding just enough ballots from the outcast and downtrodden ballots. What about registering 500 or so bums to the nearest county building to sweep a bunch of votes for Gregoire.

Does any one, even Goldy have any doubt that these people are not really voting as much as parroting for whom they are told to choose by their handlers.

Democrats should be ashamed. Even if the results stand as valid, and the Repubs can't produce enough rejected ballots or enough evidence of fraud, honest people know in their hearts that Gregoire did not really win this election. She merely ended up with just enough numbers in the returns to tip the scale in her favor. Calling these real "votes," or calling many of the Democrats "voters" is a stretch. The bums and uneducated minorities who voted for Gregoire are behaving much more like primates who are taught how to pull the lever for Democrats.

Posted by: Jeff B. on December 24, 2004 11:38 PM
13. Your anaysis proves the dems. point.
There was nothing wrong with king county election.
Satterburg, said the election board in King county was run very professionally.
He said he felt the election was fair and had nothing but praise for the election process in King county.
In case you didn't know, Dan Satterburg is the republican member of the canvassing board in King county.
I guess you can start some other rumors now.
Go ask Vance why he is calling the Supreme court justices liars.
He'll get far doing that.

Posted by: D.D. on December 25, 2004 12:08 AM
14.
Hello, I am writing this on Christmas Day from Eastern Europe:

The Democrats in Washington State are _worse_ that the most corrupt former Soviet croonies we have in my part of the world.

I am shocked, absolutely shocked, that you as Americans would allow this blatant sttempt to STEAL the election by Gregorie to take place.

I would never accept it. Get in the streets. Protest every day. Don't let it happen.

We are going to have an election in Ukraine tomorrow. The same kind of people who poisoned our candidate is the kind of people who are trying to steal the Election in WA for the Democrats.

Don't let it happen.

Posted by: Oleg on December 25, 2004 02:33 AM
15. This election recount may possibly have been conducted fairly, but thanks to the one party institutionalization of King County, public confidence in administering close elecions has seriously been eroded.

Posted by: JPO on December 25, 2004 10:16 AM
16. So there you have it:
"The Democrats in Washington State are _worse_ that the most corrupt former Soviet croonies we have in my part of the world."

Well, I guess we should just call it quits, disband this corrupt Democratic party.
After all, Dubya's cousin Oleg the NAZI here thinks we should.
The republicans that were actually involved in the election, (not the losers outside the door like Vance and on down the food chain from there), said the election was very professional, and had no doubts about gregoire being the new governor.
Sour grapes there Oleg Shecklgruber.

Posted by: D.D. on December 25, 2004 01:37 PM
17. Gregoire's comments in the PI are shameful:

"I think we have been a model to the rest of the nation and the world at large about how an election system, as close as this one is, can be done with the highest of quality ... This is the biggest display of democracy I have ever seen and I am proud of it and I think it's an inspiration."

King County's rejected ballots were given preferential treatment over the 38 other counties rejected ballots. This is a race that was decided by partisan canvassing boards, NOT by the voters.

Would this fall under any kind of equal treatment law violation? So far, the only arguement the Dems are using is the letter of the law--saying other counties certified their results and therefore can't examine their rejected ballots. This is VERY inconsistent with the dems mantra that "every vote should count". They're lying through their teeth, and hiding behind legalese IN THE GUISE of fighting for the disenfranchised.

After all, according to them, some disenfranchised voters are more equal than others (thank you George Orwell).

Posted by: noel on December 25, 2004 03:19 PM
18. No sour grapes - just plain old cynicism - one party institutional rule - it makes no difference which pary it is - the results is either incompetetance or corruption or both.

Posted by: JPO on December 25, 2004 05:08 PM
19. The mathematical bias can result from three sources:
1) Fraud
2) Unintentional Bias of the canvas board
3) A real bias in the ability of democrats to complete a ballot v republicans

1) So far no evidence of intentional fraud has been presented (though incompetence abounds).
2) But there seems to be some bias by the canvas boards, as they try to “make every vote count” they went out of there way to call the “clear intent” of questionable ballots (the Christine Rossi and the stray mark ballots for example).
3) But mostly it seems clear in FL, OH, AK, and WA that recounts find Democrats are much more likely to screw up their ballots than Republicans: Democrats register later and can’t find their precincts, so they use provisional ballots more; they mess up their absentee ballots more often, notably by forgetting to sign them; they do everything but fill in the bubble correctly.

Rossi would do best by conceding and simultaneously calling for voter reform to reduce fraud and better define what can be counted as a legal and valid vote

Posted by: zorkpolitics on December 25, 2004 06:46 PM
20. It has been well-documented that democrats are typically "worse" voters than republicans. Republicans vote more consistently, and, until recently, dominated the absentee votes in most states. One consequence of this fact is that democrats tend to screw up their votes more than Repubs (not filling in ovals, not reading the directions). It's natural that recounts tend to favor dems, because machines are more likely to reject these votes than humans.

I don't see the real point of this here - of COURSE the changes in the recount favored Gregoire, we knew that when she went from -261 to -42 to +10 to +130.

There is NO reason to believe that a binomial distribution is applicable as described, but it's a fun game to play.

Posted by: Anonymous on December 25, 2004 07:00 PM
21. D.D., you are on par with Paul Berendt. Who are you trying to convince? ..I feel sarcasm coming on.. Oh, I see the light, D.D. has shown me the error of my thinking! It's clear now, the process has lead to an honest outcome. How could I be so wrong?

OK. Feeling better now. Go away Dirty Democrat!

Posted by: Splatter on December 25, 2004 08:44 PM
22. Ya splatter, you disgruntled repubs. sound like you really have your finger on the pulse down at election central!
Kicking and screaming out "FOUL, OUR GUY WAS SUPPOSED TO WIN!"
BULLS**T!
Your guy Satterberg, came to the conclusion the election and recount went flawlessly.
He had nothing but good things to say.
I guess they musta paid him off right?
Oh "SH*T!" You guys will have THAT in the paper tomorrow.
I guess I shouldn't give you any ideas.
Are there people following you also?
Probably dirty democrats.
Dummy, I'm no democrat, I'm one of the new republicans come to infiltrate your corrupt organization, and introduce some diversity!!
LMAO!!

Posted by: D.D. on December 25, 2004 10:06 PM
23. It's hard to argue with rational analysis like that...

Posted by: South County on December 26, 2004 12:11 AM
24. I felt the level of professionalism used, met the requirements by law of an intelligent response to a disgruntled republican's wail.

Posted by: D.D. on December 26, 2004 10:24 AM
25. As a stastician I feel it is my responsibility to point out that your analysis is completely worthless. I don't care who wins btw, I am just upset when people abuse math, especially for their own partisan and incredibly biased purposes that you do.

Your major failing is the assumption, without ANY valid reason to support it, is that newly discovered ballots should be randomly distributed and therefore likely to be proportional to the previously counted results. In order for this to be true and your conclusion to hold, you would need to make a case that the ballots that were added were somehow randomly chosen. BUT THEY WERE NOT. In the case of King Counties notorious 700+ ballots they were all absentee. This alone means that the expectation for these ballots should, at the very least, be based on the absenee ballots. In addition if ballots were uncounted say, because a tray was misplaced, that tray is unlikely to be a random collection of ballots but ballots from a specific polling place AND from a specific time period. Your conclusion that the ballots should fall the way you claim they should and the probabilities you assign for the actual count fall apart on this point alone, the lack of provable randomness.

If that weren't enough you ignore a SECOND key point of stastics. Is the sample size large enough to make an analysis accurate. In the case of the added ballots we are talking about fractions of percents of the total ballots cast. The idea that this small number of ballots would conform to the larger sample is shaky as well.

Statistical accuracy relies on appropriate sample sizes. For example. Based on probabiliy if you flip a coin it has a 50/50 chance of landing on one side or another. In order to verify this statistically you need to flip the coin enough times for the sample to be valid. If you flip a coin twice you actually have a 50% chance that you will have the same side both times. if we were to conclude that this is a valid use of statistics we could conclude that the coin was an unfair one. But this would be the wrong conclusion to make. In order to be sure of our results we must look at a large sample size. If you flip the coin 100 times the chances of it favoring one side or the other go down. Flip it 100,000 times and the chances of a fair coin showing a bias is even lower.

So next time you try and make a conclusion using statistical analysis make sure you're argument has met the basic requirements. Otherwise a team of ninja commando statisticians will kidnap you and force you to watch old Barney the Dinosaur tapes for a week.

Posted by: MathMan on December 28, 2004 12:43 AM
26. Well, "MathMan",

if you understood my post you would know that I was setting up randomness as a strawman, and that I allowed explanations such as "a misplaced tray from one particular precinct".

Also, the sample size for the identified surprises is large enough to be illustrative in some cases hundreds of ballots.

Based on this and your other comments to other posts under different phoney email addresses, I conclude you are merely a screaming troll.

Posted by: Stefan Sharkansky on December 28, 2004 09:04 AM
27. The names and phoney e-mails were for the sake of entertainment :) I don't put my actuall e-mail on most internet related things these days. As for the sample size, until you provide an argument that the small number of ballots, relative to the total number is still a valid sample size your arugment is still mathematically worthless. I'm not saying that there wasn't fraud, mind you, I'm just saying that you need to proivde a stronger argument before we should believe it.

My biggest problem with both political parties is how quickly they cry fraud when they don't like the results (Gore in Florida, Kerry in Ohio, Rossi here at home). Its quite sad really. Its like the boy who cried wolf. Maybe someday they will proivde the proof first and then make accusations, rather than making them and hoping they can come up with the proof later.

Posted by: MathMan on December 28, 2004 06:56 PM
28. I just posted an Election Recount Odds Calculator over at my blog, which you may be interested in for future use. I was able to build it using the "instructions" from your site and also (mainly) from the Pull On Superman's Cape blog. Thanks for the idea!

- Ironman

Posted by: Ironman on December 29, 2004 10:56 AM
Post a comment
Name:


Email Address:


URL:


Comments:


Remember info?