« Sistani - Not insisting on Sharia | Main | Newspeak Update - Preserving is Destroying »

February 11, 2005




I presume that the court will want to move beyond the total numbers and look to see what happened at the precinct or polling place level.

I don't know what records would show how many absentee ballots were submitted by known, identifiable voters, but there is a law which indicates that there should be records at the precinct or polling place level:


"RCW 29A.44.280
Duties of election officers after unused ballots secure.
Immediately after the unused ballots are secure, the precinct election officers shall count the number of voted ballots and make a record of any discrepancy between this number and the number of voters who signed the poll book for that precinct or polling place, complete the certifications in the poll book, prepare the ballots for transfer to the counting center if necessary, and seal the voting devices."

If the records at the precinct or polling place level show that those 1800+ "voterless ballots" came from a relatively small number of polling places, it seems reasonable to conclude that something truly irregular was occurring at those few places.

If the records at the precinct or polling place level show that those 1800+ "voterless ballots" weren't in the ballot boxes or transport containers that contained the ballots from each polling place, then they apparently were introduced into the vote count after the close of the polls. Such a circumstance would certainly support a conclusion that something truly irregular caused those ballots to included among the legitimate ballots.

Until that more detailed evidence is presented and examined, I don't believe anyone could know whether the 1800+ "voterless ballots" are a reason to set aside the election or not.

Those ballots raise a big question about the integrity of the election result, but only a close look at the records might answer the question.

The claims made by King County elections officers that the post-certification process of crediting voters with having participated in the election isn't intended to discover such irregularities is really beside the point. They assert that the 1800+ "voterless ballots" aren't really ballots stuffed into the ballot boxes or inserted at the counting centers illegally -- but instead are simply the result of human errors in scanning the bar codes in the poll books, etc., during the crediting process.

However, their inability to produce a one-voter-one-ballot match as they went through the poll books raises the question: Is that because there were no eligible voters who lawfully cast those ballots at the polls?

Only a precinct or polling place level examination of the original records from which the election workers attempted to credit each voter with voting may answer that question.


Micajah, I think you're confusing processes. You're talking about the election night reconciliation, which is only for polling place votes. More votes were cast by absentee than at the polls. The discrepancy we're referring to represents all ballots, not just polling ballots.

I am expecting reply from the SoS about the status and documentation of election-night polling place reconciliations. The word from Democratic attorneys close to the case asset that they were done as normal. I'm waiting to verify that information.

I also think there's little chance King will have to execute a name-by-name verification for 3-400,000 ballots in the pollbooks. That's an unusual burden to apply without any evidence of need or wrongdoing.


Micajah: Your point about polling place records being the only true source of ballot reconciliation is spot on. I'll add that you also need to look at the logging system for absentee ballots because they are not processed at the polling place, but that's just an extension of your point. It's also the point I have been making lo these many weeks in various venues. The "discrepancy" obtained by looking at lists of currently registered, credited voters (a la Sharkansky) is dubious because the size of the discrepancy is likely to be on the same order as the size of probably errors in the list. It may be useful as a first order look, but it doesn't give a reliable estimate of the true size of the discrepancy.

To find out how many "voterless ballots" really exist, you need to know the number of voters who cast ballots and the number of ballots cast. The second part is easy -- just look at the canvassing report. To learn the first, you need to look at the polling place records and the absentee ballot log. These are the data that show which registered voters actually cast ballots.

It's possible, even in a completely fair election, that some errors occur at the polls. But the size of the error at any single polling place should be small (1 or 2 ballots) and many polls should have zero error. A large error at a few polling places would be worthy of further investigation.

In this respect, Sharkansky's analysis is not completely useless. His work would highlight polling places with large discrepancies and save the trouble of examining materials from all polls (King County has over 500 polling places and over 2600 precincts). Once problem precincts are identified, one could review the polling place records and determine whether the source of the error was unreconciled ballots at the polling place or errors in the credited voter list. Only the former case represents cause for concern.

It's truly frustrating to watch both sides of this dispute argue about the significance of reconciliation errors when the source of reconciliation is the credited list of currently registered voters. Sharkansky works to inflate this number as much as possible and imply that it is unique to King County, while KingCo works to whittle it down and then claim that the remainder is "typical" and no cause for alarm. Both arguments are vacuous because they are based on specious data.



statistical question.

what other correlations tests did you perform? what statitical significance test did you use on your correlation coefficient of .91. I assume it was a t test with n = 16.

Why didn't you perform a simple regression also. y = a + Bx = e. The a value might give you some info as to the critical county voter size at which vote errors should appear. Granted the sample size is small. You could also calculate how far the King county results varied from the regression estimate. I assume King county was a bit of an outlier, but it would be interesting to see how far it deviated from the regression estimate.

Thanks for all your work.



I see you got the point I was making.

You may be right about absentee ballot logs -- they might contain enough information to show what came in and when. I'm not familiar with those records (while I've actually seen poll books when I vote), so I had to simply say "I don't know."


Can you take "yes" for an answer?

I'm not confusing anything. I specifically said that I don't know what absentee ballot records might be examined to see what's what. I specifically said that the poll books would allow an examination of source records to see what happened at the polling places.

I'm fully aware that the "voterless ballots" may have come from both absentee ballots and the polling places, but I'm also aware that Dean Logan said they might have been the result of issuing ballots at the polling places without requiring the voters to sign the poll books.

Since I'm familiar with the one kind of source document (poll books) that might answer the question that Logan has only speculated about, I offered only that as something to be examined more closely.

My whole point was in support of your point -- that Sharkansky's work raises a question but doesn't answer it. In other words, I was saying "yes," but you have a hard time breaking through your own argumentative nature and prejudice to see the validity in anything anyone says who isn't of the same political persuasion as you.

End of Dutch uncle's lecture.


first of all, how come none of the IE users complained about the horrid space before the table? It looked OK on Firefox, so I didn't notice until Zap told me. Sorry--it's a link now, like at Carla's.

Micajah, last things first. I run a blog with a guy whose political persuasion couldn't be more different than mine. I think I'm OK on battling prejudice.

I'm glad you say you agree with me, but when you said

However, their inability to produce a "one-voter-one-ballot match as they went through the poll books raises the question: Is that because there were no eligible voters who lawfully cast those ballots at the polls?"

begged not one but two questions: not just whether a failure to match votes to signatures has as its simplest answer "fraud" as a starting point; but we don't even know that they were unable to produce matches as they went through the pollbooks...because that's not what the 1,800 refers to. That's what I'm interested in hearing from Reed--what records does the SoS have on King's pollnight certifications?

Sorry if I misinterpreted you, but even on rereading it still sounds to me like you presume both a bad votes-to-signatures tally in many precincts, and that it's the result of something fishy.


chew2--thanks for the stathead question. We did a Spearman and a Kendall, because they're nonparametric and aren't as affected by the extreme outlier nature of King. It's way out there, on an absolute numeric basis. I think the Spearman was .6 something, and Kendall was .8. Both signficant to .01, at least.

I don't want to overstate those results. The effect of scaling of total votes on discrepancy rates, is hypothetical. But a common-sense look at the table sure makes that appear to be true, and the mathematical checks were very strong supporters of our eyes in this case.

north clark county

Not that I'm disagreeing with TJ and Micajah, but I do agree with what scottd says. The 1,800 or 3,800 or whatever the difference between ballots and voters credited may make for nice political slogans like "Every ballot should have a voter", but it isn't going to go anywhere in court.

Hopefully, when the discovery process is complete, the proper information will be released. That is the data that scottd and Micajah are referring to. And until that data is available, this issue is a loser for Rossi.


I agree with that NCC (which I think means I agree with everybody)--are they really just going into discovery to fish? If King didn't do a pollnight certification, and if the SoS and/or King don't have the forms from the King polling places, or they were suspiciously off--what possible reason is there for Rossi and his team to sit on that information? And if they have no known allegations to that effect, any attempt to force a precinct-by-precinct re-reckoning is going to be a fishing expedition by definition.

The polling place records seem much more important to others than to me, it seems. When they're only 37% of the total ballots, I can't see how a review of those would be at all dispositive of anything. (Sixty-three percent mailed it in this time--when will Washington give in to the wonderful vote by mail system?)

Our point here is that Rossi has put an awful lot of his case into King. What does he have?

Felon votes
Deceased votes
1,860 unresolved votes
Double votes
misfed provisionals
'enhanced' ballots

Felons are communicated by the states in an entirely incomplete fashion; that's the fault of the courts and thus unfixable without legislative change IMO (ie, get the courts to provide a statewide database). Deceased voters are the same way--the public health feeds them the data, and they have to work with it. The unresolveds are some sort of issue, but I think we've shown it's not really clear what. The double votes were miniscule. The misfeds were bad, but a) provisionals are new everywhere, and b) they rated out 72% valid, lower than the rest of the provisionals but leaving fewer than 90 illegal votes. Bad, yes. I think that's the only real case of poor King conduct that has a voter total behind it. On the other hand, there's only about 60 Gregoire votes at best, possible.

The enhanced ballots I've seen no evidence on. King claims to have over 4,000 duplicated ballots, used to preserve the original markings. So if they've got something there, they're not letting anyone in on it.

north clark county

While I don't disagree with your conclusion, there are a few nits to correct in your presentation.

"When they're only 37% of the total ballots, I can't see how a review of those would be at all dispositive of anything."
They might show something, as has been discussed above.

"1,860 unresolved votes"
The number is larger than this. First this is the net number from just King County. It is more accurate to use the sum of absolute values from each precinct. Second, even using the net numbers, your table shows that there 5,594 in the 15 counties you presented.

"provisionals are new everywhere"
Actually they aren't. We've had them here in Washington for years. They were originally called "questioned ballots", and then "special ballots". In fact, the HAVA requirements for all states to provide provisionals was modeled on Washington.

Also, on the issue of felons and deceased, SoS Reed discussed this matter at the Vancouver meeting of the Election Reform Task Force. HAVA requires states to develop a statewide voter database with unique voter identification numbers. This statewide database is due to be in place by the end of the year and will coexist with the county databases. There will be a number of advantages to this. First, if a voter moves from one county to another, he files a change of registrations rather than a new registration. Second, the state is developing procedures with the courts and with Vital Statistics to all a faster and coordinated data sharing to exclude the felons and deceased.

Reed also indicated that there is some hope of getting INS information on legal aliens, so that they can be excluded should they somehow be registered.

Given that HAVA requires drivers license or last 4 of SSN when registering now, the new database will have better ability to screen for duplicates and to validate that the person is real.


I'm not sure there _was_ a by-precinct comparison of ballot count to voter credits done after certification.

The "net discrepancy" idea so far exists only in Sharkansky's addled dataset.

What would polling place discrepancies show? A discrepancy, I'm thinking. What's the relief for that? None.


TJ: State law requires a reconciliation of ballots and voters for each polling place on Election Night, _before_ the ballots are sent to the central collection point. If that wasn't done, KingCo has a problem. I've seen no evidence one way or the other on this matter, but my default assumption is that they followed this basic procedure.

What would polling place discrepancies show? I'm surprised you asked. For one thing, it might replace the endless blather over meaningless statistics with actual, relevant _data_. That data might show that nothing is amiss in King County and help restore public confidence in the election process. Or, it might show that Sharkansky, et al were right and put some wind in Rossi's sails. Either way, I'd like to know.



It's my understanding that there is specific reconciliation paperwork that is completed by each county on election night, according to Washington State law. There's not been a question that any county, including King, completed that paperwork and turned it in. In order for cert of an election to be completed in Washington State, that paperwork would have to be turned in to the SoS.

I highly doubt Reed would do an election cert without the requisite paperwork from each county.

IMO that meme is DOA.


Carla: I think you misunderstood me. I have no reason to doubt that the required paperwork was filed. I think King County could be more open in reporting this fact as a means of dispelling rumors to the contrary. It would also be helpful to reveal the results of this reconciliation as a means of diverting attention from more specious accounting.

One of the first rules in PR is to get out in front of rumors by flooding the zone with facts. KingCo has done a terrible job at this. Their denials and vague responses make it look like they have something to hide, although I believe this is not the case.

Public officials have a duty to build or maintain confidence in the processes they manage. This goes beyond their obligation to merely administer according to the law.


TJ and Carla,

Statistical question.

You are claiming that "variance" or errors are a function of county size, the number of voters in the county. The GOP are claiming the King county errors are so large they must have been due to something else, like fraud.

One way to test this is to run a simple regression on your 16 county data.

y = a + Bx

where y = predicted variance and x = number of voters.

using this model you can get a predicted value for variance for King county based on its number of voters.

The regression will also give you an estimated standard error for your estimate y.

You can then test to see how many standard errors away the King county variance of 1800 is from its predicted value. If this is statistically significant, then one could argue that your model is incomplete and something else, like fraud or special incompetence, was at work. The same could be done for Spokane.

If the King and Spokane observations are too influential, you could also try dropping King and Spokane from the model to get the true underlying relationship.

This is what I remember from my stat classes, but I don't have any way to do the tests now.



You're right, I misunderstood. :)

You're also dead on correct with the PR thing. The Dems (both in WA state and nationally) need to learn that lesson, IMO.

north clark county

You have to wonder why KC doesn't do better. It took them too long to reply to the military ballot mailing question, leaving some to doubt in their response. It's the age-old problem of the accusations on page one and the retraction on page eight. When you are facing page one accusations, you've got to get your reply on page one, and quick. Both political parties have a lot to learn in this regard.


scottd--I know they have to do a pollnight reconciliation between ballots cast and signatures. That's not what I was talking about--it's why I said by-precinct comparison of ballot count to voter CREDITS done AFTER certification. That's what yielded the 1,860, not what was done on Election Night. The pollnight reconciliation only took care of roughly 40% of the votes. Even if it were 100% correct, all that would mean is that the discrepancies occurred within the counting of absentees.

The 1,860 represents total votes vs total credits issued, done AFTER certification. As I said, I'm not sure that job was done at the precinct level. And I'm not sure I understand why that's not actual, relevant data.

chew2--I had already surmised that King was probably far enough away from the group to possibly skew a parametric test, so that's why I did the two non-parametric tests. I toyed with the idea of running a regression analysis on it, but opted not to at this point.



You should run the regression tests anyway. You can't tell whether King or Spokane are statistically significant outliers unless you run the tests, or am I wrong about that?

Frankly I think Spokane is a bigger outlier.

Plus, if I recall correctly, the fact that King is an extreme value doesn't necessarily violate the assumptions of a regression model. Its the error term or residuals that have to be random and normally distributed.

As you noted, a Pearson correlation is a parametric test, just like a regression, and assumes a linear relationship also. But you presented it as probative. If the correlation was statistically significant, the regression should be also.


Spokane's a bigger outlier in terms of ratio, but this was pure variance. On an absolute basis, King sticks way up in the top right corner. It doesn't violate the principle, but it does skew the effect.

I'm not attempting to make a firm scientific statement. What I did was a little validation of what seems relatively clear from visual inspection. Without making the firm claim that elevated ballot levels are specifically responsible for a higher discrepancy rate, I think it's enough to note the very strong relationship between the two that exists.


NCC: I couldn't agree with you more...

TJ: Sorry if I haven't been clear, but my point has little to do with whether reconciliation is done at the precinct level or over the entire county as a whole. My point is more related to the source of data used in the reconciliation.

The 1,860 number tells you nothing of value, because its source data are incomplete -- and KingCo has never claimed otherwise. That number relies on a list of registered voters who were credited with voting in the election. However, that list is incomplete for a number of reasons. Since it was created from _current_ voter rolls at some date many weeks _after_ the election, it's likely that some of the Nov. 2 voters had already been removed (moved, died, convicted, whatever). It's also likely that there were some errors in the crediting process. This shouldn't be surprising. The purpose of that list is to help maintain voter registration rolls (by inactivating the voters who haven't been voting) and to provide information for public use such as political campaigns -- it doesn't exist as an accounting check to help detect election problems.

Normally, errors in the credited voter list would be of little consequence, so it's likely that the procedures for updating that list aren't especially rigorous. An error rate of only 0.2% on that list would account for the entire "discrepancy". That's why I've claimed that any analysis based on that list is irrelevant. Based on the likely margin of error in creating that list, you just can't say that there are 1,860 "voterless" ballots. The number may be considerably lower.

If you want a reliable count of voterless ballots, you need something that at least purports to be a _complete_ count of everyone who voted. You can get that count by adding the total number of pollbook signatures to the number of validated absentee/provisional ballots. Any other measure is meaningless. Fortunately, you don't have to examine every pollbook -- the election night summary sheets should include the number of pollbook signatures. Doing this on a by-polling-place basis isn't necessary, but it would give you an idea of the distribution of errors.

BTW, I don't mean to impugn the work you and Carla have done. I respect your reporting efforts. However, it's frustrating to see the undue attention the blogosphere and media pay to this meaningless statistic.

I've taken enough of your bandwidth on this issue. If you want to talk about it more, feel free to send email.


Thanks for your comments, scottd. Don't worry--I'll wear you out long before you use up all our bandwidth. :)

I don't think it's entirely true that they are using a registration file from weeks later. They have a file that was used to create the pollbooks themselves, and while I agree that the board has since had the task of correcting addresses and typos (about 100 didn't make the pollbooks on Election Day because of simple typos, apparently), they should actually have a reasonably complete voter roll frozen just before the elections. Complete, if not accurate. And you're absolutely right that it's not the directed purpose of the file to make 1 to 1 accountings. Carla and I did make that point in Part One, that trying to cobble together his own file was folly on top of folly for Sharkansky. It's part of the reason I didn't go "all the way" as chew2 indicated, and do a regression on the relationship. I don't think it's "meaningless," however, and what we're trying to do here is go so far as to accept Sharkansky's premise that discrepancies are a serious problem as stated--and then point out that zeroing in on King may not yield what they think.


TJ: Thanks for your patience. Let's try another round :-)

When King County released the voter list that represented their latest reconciliation effort (Jan 7), they noted that the list had been updated to include "additional registrations, cancellations, and changes to individual voter records." Other county auditors have been quoted in the press saying that public releases of voter lists represented snapshots of a "living" database that was constantly being updated -- including cancellations.

The point is that the credited voter lists being used for these various "reconciliation" efforts don't represent a snapshot of the voter rolls on Election Day -- so they are incomplete and any reconciliation done from these lists is, at best, an estimate. Any estimate should be accompanied with an analysis of its range of uncertainty. If that range is as large (or larger) than the estimated value, then the estimate must be regarded as unreliable.

My contention has been that estimates of ballot v. voter discrepancies have been unreliable because the range of error is likely to be close to the value of the estimated discrepancy -- hence, my characterization of these estimates as "meaningless".

In principle, I suppose King County (and others) could reconstruct an Election Day roll from their records. Without knowing the details of their data systems, I couldn't say whether this would be easy or difficult. In any case, KingCo has said that they haven't done this. Why should they? It represents an extra expense and there is no legal obligation for them to do so. Their primary obligation is to maintain an accurate list of _currently_ registered voters so they can administer the _next_ election. Secondarily, they need to make certain those current registrations are properly credited with voting in previous elections. The voter lists they have released represent the fruit of those efforts.

You said that your main point is that zeroing in on King County seems unwarranted, but you are willing to accept the premise that discrepancies are a serious problem. I would accept that the latter is a problem insofar as the public confusion being raised by various parties affects confidence in the election process, but I would point out that no one has presented reliable evidence regarding the size of the discrepancy or whether a significant discrepancy even exists.


scottd--thanks for continuing to play. :)

King released three file sets to the public, if I'm not mistaken: Nov 1, Dec 29 and Jan 7. I agree that if you try to reconstruct King's reconciliation from the Jan 7 or Dec 29 files, and compare them to Nov 1, you're going to run into trouble with changed information for those credited with voting. And if you use just Nov 1, you're going to run into trouble with records that hadn't been properly updated yet.

So I agree--as we pointed out to Sharkansky in Part One--that trying to work with one or more of these files to replicate King's analysis was a failed mission from the start.

But I think it goes too far to say that King would be similarly hamstrung. And in my conversations with Egan at King, it's my impression that while they haven't tried to construct an "Election Day file," they HAVE attempted to find every voter who cast a ballot, and give them credit in their data.

Here's why I think King can do it at the county level, but Sharkansky can't at the precinct level. Say person A votes at polling place X, but then moves to polling place Y. Person A retains a single record in King's database, but there would be two versions: the record with his old address, and the changed record with his new address. I agree that King does not maintain a history record of person A, showing his old address. However, it DOES hold a record for Person A, including his NEW address, and with credit for voting Nov 2.

Sharkansky's problem is that if you try to reconcile at the precinct level using the new address, you're going to credit polling place Y with a voter, but no ballot was cast there by him. At the same time, polling place X has a ballot, but no person to tie it to. Where King would have no error at the county level, Sharkansky (if unable to match the two records properly) would have not one but two discrepancies.

Now he's said he tried to account for these, and maybe he was dilligent enough to catch most of them. But without the pollbooks that King can rely on to investigate and resolve discrepancies, Stefan is at the mercy of errors he cannot fix, and errors he's not even aware he's made. And he makes some of his own, as you pointed out in his original thread: imputing the machine recount absentee/provisional breakdown for the manual recount totals (because King didn't provide the latter) invites potential for a whole slew of counting errors.

Which was much of our point--Sharkansky is doing a precinct-by-precinct analysis using public voter files (minus birthdays, I might add) and ballot counts. King benefits from having date of birth, and also can rely on pollbook entries to investigate discrepancies.

I can't speak for Carla, but I wouldn't say I'm willing to accept that discrepancies ARE a problem, only that they MAY be a problem. And I would accept that on an ongoing basis, voter databases would be better served by utilizing history records to track changes in the electorate...at least between pre-election lockdown and certification. But as for whether I consider the fact that King can't match ballots to voters at the county level any better than 99.8% to be a "problem" for this election, I don't think so.

I've been promised by the SoS's office that they'd respond to my fairly detailed questions this week. While I may not receive access to polling place accountability forms, I should be able to determine how the pollnight certification process worked, whether SoS maintains those records, whether they believe they are complete, etc.

The comments to this entry are closed.

April 2006

Sun Mon Tue Wed Thu Fri Sat
2 3 4 5 6 7 8
9 10 11 12 13 14 15
16 17 18 19 20 21 22
23 24 25 26 27 28 29


Blog powered by Typepad