Well, what’s good for the goose is good for the gander. While errors tend to occur in every election, they tend to occur in a random distribution. This is illustrated by the county by county breakdown of the mishandled provisional ballots and votes cast by felons alleged by Democrats.
| County | Votes Cast By Felons | Unverified Provisional Ballots |
| Adams | 117 | |
| Benton | 25 | 37 |
| Chelan | 5 | |
| Clallam | 9 | |
| Clark | 35 | 1 |
| Cowlitz | 2 | 17 |
| Douglas | 4 | |
| Ferry | 4 | |
| Franklin | 4 | |
| Garfield | 1 | |
| Grant | 8 | |
| Grays Harbor | 1 | |
| Island | 5 | 27 |
| Jefferson | 2 | 16 |
| King | 31 | |
| Kitsap | 23 | |
| Kittitas | 25 | |
| Klickitat | 2 | |
| Lewis | 17 | |
| Lincoln | 2 | |
| Mason | 1 | |
| Okanogan | 7 | |
| Pacific | 1 | |
| Pierce | 49 | 164 |
| Skagit | 11 | 1 |
| Snohomish | 27 | |
| Spokane | 65 | 3 |
| Stevens | 1 | 400 |
| Thurston | 25 | |
| Walla Walla | 5 | 331 |
| Whatcom | 4 | 17 |
| Whitman | 1 | 783 |
| Yakima | 55 | |
| Total | 432 | 1939 |
That’s 2371 disputed ballots in all, most from precincts that heavily favored Dino Rossi.
Yes, most of the mishandled provisional ballots were legally cast, but then so were most of the mishandled provisionals in King County. If you are interested in more than just inflammatory rhetoric, the important number to look at is the votes cast by felons, as this is the heart of the Republican’s lawsuit. In the outside chance the court might accept a “proportional reduction,” the 432 illegal ballots alleged by the Democrats are more than enough to offset the GOP’s King County felons, and still maintain Christine Gregoire’s lead. And Democrats promise that there are more coming.
All in all, not a good day for the Rossi camp, though I doubt it was much of a surprise. Their research clearly cherry-picked irregularities in pro-Gregoire precincts, and it would have been silly to expect the Democrats not to fight fire with fire.
I should, however, note one difference between the way the Democrats have gone about this and the way the Republicans have. The Democrats have not released the names of the alleged felons to the media, only the numbers. They have issued subpoenas, seeking to determine if any of these people should be removed from their felon list, before their names are unfairly made public. For this they should be commended.
Show precinct break down or are you making accusations you can’t back up? Are you sure you didn’t mean Rossi leaning counties?
As Deadwood stated in the last thread the cities in eastern WA favor democrats, rurals favor Republicans. Unless you break it down by precinct you are just pissiing in the wind.
Goldy, it is good to see statistics, and I appreciate the table very much. But what do those numbers mean? Will they really help Gregoire all that much?
I will assume the Democrats had done a thorough job (not the least of reason is that they have had more time) and that all 432 of the felon voters are adult felons without their right to vote restored. If so, every one of these people cast an illegal vote.
(The earlier GOP list was almost 1/3 in error, including many juvenile convictions, and I basically called the authors of that list dumbasses for not being able to tell the extremely obvious difference in case number format and wording of adult vs. juvenile convictions in an earlier post on your website.)
However, simply because 1939 provisional ballots were improperly processed in other counties does not by any stretch of the imagination mean that these were 1939 illegal votes.
Take Benton County. 37 voters dropped their provisional ballot into “the can” (what people put a punch card ballot into in the local usage), instead of the envelope. However, only 8 of those 37 voter were not legally entitled to vote (i.e. not registered or had already voted, etc.). So that is only 8 illegal votes from that source in Benton County.
Take Adams, Stevens, Whitman and Walla Walla Counties. Not a single signature on provisional ballots was checked for matching and verification in any of those four counties. Does that mean that all 1333 provisional ballots in those four counties was an illegal vote? Of course not!
King County had the highest provisional ballot signature mismatch rejection rate in the entire state — just under 1%. And this was largely because the county failed to notify voters whose signature were rejected — contrary to the WAC regulations. It would have been nearly 3% rejections, but the parties were given the rejection list a couple of days before the deadline, and the Dems made a strong cherry-picking effort for likely Gregoire voters, while the GOP made a much weaker effort.
King County’s absentee ballot signature rejection rate — a process where rejected voters were notified — was also the highest in the state, but was still less than 0.2% after 566 or so of the misplaced “Larry Phillips” ballots were belatedly verified.
The Dems would have to get a handwriting expert to go over all 1333 provisional ballot envelopes from Adams, Stevens, Whitman and Walla Walla Counties, and compare the signatures with those on file for the actual voters. Given that King County had an initial provisional rejection rate of 3%, they might expect to find 40 or so voters that an expert would raise genuine questions about.
Those voters would then have to be identified and disclosed in discovery in the lawsuit. The GOP could then go to each and every voter that questions were raised about, and get them to sign an affidavit that they really did sign the provisional ballot envelope in question and voted back on November 2. After all, they would have had the same chance if their signature had been initially rejected in a verification process back in November. And the Dems have the burden of proving that a vote was illegal, if they are asserting that as a defense of some sort.
Most of the voters who couldn’t be located if there was a signature mismatch are probably Wazzu students — 783 of the non-signature verified provisionals came from Whitman County. And the student populated precincts around Wazzu probably were favored somewhat for Gregoire.
So I would really love to see the Democrats waste their efforts by trying to prove that some of the 1333 non-signature verified provisional ballots were actually illegal votes cast by someone other than the actual voter in question.
The final issue is the method of proportional analysis that will be used. If it is done at the county level, that is better for Gregoire. If it is done at the precinct level, that is better for Rossi. Illegal felon voters, whether in King, Pierce, Spokane, Yakima, or Benton, tend to be much more concentrated in low income urban precincts that Gregoire did better in than Rossi.
I have not received a precinct by precinct breakdown yet, but it’s a pretty safe bet that the Dems cherry-picked precincts, just like the GOP did. For example, the spreadsheet shows 31 additional felon votes in King County, and I have been assured that these are all from pro-Rossi precincts.
I mean… you really expect the Dems to offer evidence that helps Rossi? They’re not stupid.
Goldy @ 3
I understand that and wanted to show that your cherry picking comment/accusation on the last thread was a pretty dumb comment because both sides ARE doing it.
Goldy @ 3
It makes perfect sense for the Democrats (but not the Republicans) to cherry pick. If the Republicans have any hope of getting a “proportional adjustment,” they will need to show that all counties have undergone equal scrutiny (reasons are given here).
By cherry picking, the Democrats undermine the Republican case for “proportional adjustment”, but without actually doing all the work for Republicans. The Republicans would still have to go back and equally scrutinize each county (or precinct) in order to get an unbiased “proportional adjustment” for a statewide election.
I don’t think the identity of Dem-found illegal votes is very important at all. The point is more to show that election errors are a permanent feature of elections, and that King was not amazingly derelict in their processing of the votes. King had felons–so did most other parts of the state. King had provisionals not handled properly–so did most other parts of the state. King had people vote in the names of the deceased–so did many other parts of the state. Turning back to “it’s a mess” is a big loser for the GOP; there is no provision in the RCW to disallow an election result because “it’s a mess.” It has to be shown that the result is not correct–not that it’s not 100% accurate.
The proportional allocation really makes no sense–on what grounds will Rossi claim that felons vote like other people?
Deja-vu I think torrid might have said all this before.
Goldy, there’s another difference you didn’t mention: The Democrats aren’t paying for this with money that came from L & I taxes for injured workers.
all tools @ 7
I’ve had to repeat the major flaw in the felon study the right keeps plugging; it’d be no surprise if I repeated the words @6 before.
Just yesterday Rossi sent an update email with an attempt to show how felons were all Democrats, citing Florida’s 68% Democrat felon rate. Putting the veracity of the number aside, wasn’t it late November when the GOP was kiboshing fraud allegations in FL based on Democratic registration? Seems like many places in FL are heavy Dem registered, but vote solidly Republican–because the registrations are years to decades old. So 68% registrations definitely doesn’t mean 68% Democratic voters. (And of course it never meant 68% Democratic VOTES in any event).
Mr. Cynical wrote just a few days ago that. . .
You know for sure that felon votes in Rossi County’s are going to even be considered?
Your brilliant lawyers may have screwed up bigtime dumbass.
Watch for the rulings on Monday.
The Dems may have some serious timing problems.
That will make you LEFTIST PINHEADS go absolutely nuts won’t it??
God, I’d love to see the look on your “puckered” ugly face on Monday. I WILL RUB IT IN!
Ahhhh, haaaaa, haaaaa. . . .
Oooooohhhhh, hooooooo, hooooooo. . . .
It is not even Monday, yet, Cynical. What do you have to say for yourself now?
(snicker)
TorridJoe @ 9
For the 11/02/2004 election, Florida had 10,301,290 people registered to vote.
http://election.dos.state.fl.u.....nParty.pdf
3,892,492 or 37.79% of these were registered as Republican. 4,261,249 or 41.37% of these were registered as Democrat.
George Bush received 52.1% of the Florida vote, while John Kerry received 47.1% (also from Florida Secretary of State website).
While there is no exact match between party registration strength and party voting strength in a given race, there is still a pretty high correlation. On average, Bush was 14% stronger than GOP registration, while Kerry was 6% stronger than Dem registration.
So if 68% of the illegally registered felons in Florida registered as Democrat, then is likely that 74% or so of them voted Democrat.
richard, it depends on where they’re from. Many “Democrat ” registered counties went Republican on the ticket in 04, and were already represented by Republicans.
You also miss the point that registration does not necessarily dictate behavior among felons.
LMAO@10–
Let’s just wait and see what Monday brings, shall we?
These GOLDY numbers are meaningless.
No names is a good thing?
Goldy@3
I’m happy at long last you realize this is an election CONTEST. There will be WINNERS and LOSERS. The Judge will define the rules and we ought to all live with it. I said months ago that I would live with whatever Judge Bridges decides. No LEFTIST PINHEADS would agree with that. WHY? They want the political State Supreme Court to make the decision AFTER Judge Bridges sets this election aside.
I’m sure the PINHEADS are excited to actually see something come from the Dems after nearly 4 months of hacking at the R’s.
Torridjoe @ 12
There are some small rural counties in northern Florida that are extremely heavy in Democrat registration, but vote very heavily Republican for national office. For example, Liberty County is 3597 D to 320 R (in fact highest D reg% in entire state), but voted 1927 to 1070 for Bush over Kerry. (Although the losing Dem for US Senate narrowly carried Liberty and the incumbent Dem US Congressman won by almost 2-1.)
However, in the larger counties in Florida, where most of the voters live, there is a much better correlation between party registration and party voting patterns — also much less ticket splitting than in the smaller rural counties.
I would think that improperly registered felons in Florida would tend to be more concentrated in the areas where more people are living, and not in the smaller and quirkier ticket-splitting rural counties.
And probably these felon voters are more concentrated in the lower income precincts, which in Florida tend to have very high percentages of both Democrat party registration and voting for candidates of the Democrat party.
In Washington, the main indication of how a felon is likely to have voted is by how the people in his or her neighborhood (i.e. precinct) voted. We have no party registration here, and the last time public preference could have been expressly was in the February 2000 presidential primary (in which maybe 20% of the total registered voters actually publicly signed for a party ballot).
It is nice to try and minimize the Florida statistics. But I think a proportional analysis of voting at the precinct level would closely correlate with the party identifications these people selected when they registered.
richard, what on earth do you base this statement on?
“In Washington, the main indication of how a felon is likely to have voted is by how the people in his or her neighborhood (i.e. precinct) voted.”
If you’ve got some research to cite, bring it out.
There’s also no evidence for this last statement:
“But I think a proportional analysis of voting at the precinct level would closely correlate with the party identifications these people selected when they registered.”
Because?
Nice to see the democrats are finally trying to put this effort into this thing instead of sitting there and being attacked by the BIAW.
They are wisely making a case that:
1) There is no evidence the felon vote gave Gregoire her victory.
2) Even if a proportional analysis was used based on the voting record from each county, which it shouldn’t be, Rossi still would have lost.
Richard @ 14
I find many of your posts insightful and informative. But I agree with torridjoe @ 15. There is no empirical evidence to support the idea that ex-felons vote like those in the neighborhoods they live in. This kind of reasoning is a classic ecological fallacy.
We have a halfway house for convicts in my neighborhood. To suggest their voting behavior can be established by looking at the votes of the homeowners, middle managers, professionals, and retirees living in my neighborhood is ludicrous.
All tools here and yet there are still screws loose @ 7
Does repeating the ‘truth’ cause harm to your feeble mind?
Mr. Cynical @ 13
I don’t think Goldy’s numbers are meaningless at all, at least with respect to the number of felon voters. These apparently illegal voters will have to be verified, and considered by the voter in whatever method of analysis is determined to be legally appropriate. If precinct level proportion is used, I think that the number of felon voters will be sufficient to erase Gregoire’s 129 vote lead, even when adding the Democrats’ apparent 432 additional felons from mostly pro-Rossi counties.
TorridJoe @ 15
Since Washington doesn’t have party registration (unlike Florida), the only objective evidence as to how an illegal felon voter might have voted would come from precinct level proportional analysis. Except for probably less than 20% who might have been registered in 2000 or 1996 and publicly selected a party ballot in the presidential primary.
Erik @ 16
If you take the Republicans’ felons list and the Democrats’ felons list, and analyze this at the county level, it probably would not overcome Gregoire’s 129 vote victory margin. At the precinct level, it probably would overcome the margin. It would be interesting to see both methods of analysis, with the Democrats’ new information considered as well.
DJ @ 17
You will never have totally convincing evidence, since every person’s ballot is secret, and we can never tell for sure how a felon (or anyone else) actually voted. I think the most logical way to deal with an illegal vote (or excess vote — i.e. ballots exceeding number of voters) is to randomly eliminate an appropriate number of ballots from the given precinct. The proportional analysis method has the same expected mean result, without the whole number uncertainty (i.e. you deduct 0.2 net votes for Gregoire vs. 60%-40% chance of pulling out a Gregoire ballot).
Alan @ 18
I will give you $10.00 for every convicted felon registered at your local halfway house who voted in the November 2004 election without having their voting rights restored. Just get the Democrat and GOP lists, and see which felons end up being verified as illegal voters.
Most halfway houses are for recently released convicts still under Dept of Corrections supervision. While those folks are not yet eligible to vote again, probably not too many do something so obvious as illegally registering under those circumstances.
To get voting rights again, a felon must (1) finish with Dept of Corrections supervision,(2) pay off all fines and restitution, and (3) have a court order entered based on # 1 and # 2 to restore civil rights. Most illegal felons register when they finish with DOC supervision, but still owe money on fines and restitution.
I would think most illegal felon voters share the lifestyle of their community, and are not some kind of eccentric weirdo. You will have mostly low wage earners (since we are talking about folks who can’t pay off their restitution) living in a community of cheaper apartments with folks of similar income levels. I don’t think you will find too many illegal felons in higher income precincts, whether these precincts be in nicer neighborhoods of Democrat Seattle or Republican Kennewick.
But if you found a felon living in low-income housing in a high income neighborhood, then a good argument could be made for the ecological fallacy in that particular case.
Richard @ 20
“Since Washington doesn’t have party registration . . . the only objective evidence as to how an illegal felon voter might have voted would come from precinct level proportional analysis.”
and later
”I think the most logical way to deal with an illegal vote (or excess vote – i.e. ballots exceeding number of voters) is to randomly eliminate an appropriate number of ballots from the given precinct.”
But, such an analysis disenfranchises some voters in a big way. Here is why. Suppose you find that 30 white male ex-felons voted in a precinct that is 50% women and 30% Asian. The random draw method (like the proportional method) would be expected to eliminate the votes of 15 females, five who are Asian females. Since these 15 females probably do not have the same voting preferences on average as our male ex-felons, we have disenfranchised women and Asians. This is why any proportional method (including the random draw method) seems unlikely as a way to settle this lawsuit.
”The proportional analysis method has the same expected mean result, without the whole number uncertainty”
Indeed it does. I think either method can be used in a circumstance where, for example, malfunctioning machines were occasionally double-voting. If we know there are 500 double votes throughout the precinct, then randomly selecting 500 votes prior to certification works for me. (This situation is different from the ex-felon one because the improperly votes are randomly drawn from all votes in the precinct).
After the election has been certification, however, the situation is completely different. In this lawsuit, the Republicans must show that Gregoire owes her win to these errors. Objections over the non-representiveness of ex-felon voters aside, suppose Judge Bridges does decide that 1500 invalid votes must be randomly removed in the precincts where the votes were counted.
Rather then randomly removing the 1500 invalid votes just once, the exercise should be repeated some great number of times like 100,000 times (this is a trivial exercise with computers). Now we actually get a distribution of wins for Rossi and Gregoire. We might find that Rossi wins 60% (60,000 trials) and Gregoire wins 40% (50,000 trials). Should the election be overturned? Probably not, because it means there is a very high probability (40%) of improperly overturning the election—in other words, we cannot say with any certainty that Gregoire owes her win to the errors. Elections should not be overturned on such flimsy evidence. Judge Bridges would be in a situation where he must pick a probability that clearly demonstrates that Gregoire owes her win to the errors.
He might decide that Rossi would be required to win 95,000 of the 100,000 trials to provide the appropriate level of certainty. This still means there is a 5% probability that he is improperly overturning the election. An even higher statistical certainty might be deemed appropriate.
dj–
“elections should not be overturned on such flimsy evidence”.
It “appears” you are over-complicating what the definition of “appears” is. What does appear mean to you? How is it defined in the law? Appear does not mean PROVES does it? Just like FRAUD is not required either.
I could go back to January where I stated repeatedly that this will come down to the definition of APPEARS. Still does.
DJ @ 21
Someone on here did that statistical math a few weeks ago when we were discussing the random ballot elimination experiment.
The 95% probability model generally means two standard deviations from the mean. Since there are two tails to the Bell Curve, we would actually be talking about a situation where 97.5% of the outcomes would have to show a Rossi victory in order to meet your test.
I wonder what the standard deviation would be in the random ballot elimination experiment. Somehow I think it would maybe be 10 to 20 votes, and that Rossi would be a 97.5% or better winner of this experiment even with a mean expected margin of 20 to 40 votes.
Actually, DJ — look at your post # 30 on this thread:
http://www.horsesass.org/my-co.....5&c=1
You did random ballot elimination at a constant 59/41 ratio of Gregoire to Rossi (i.e. like King County as a whole).
725 ballots eliminated produced an expected one vote win by Rossi. Basically, just over 50% probability of Rossi winning.
800 ballots eliminated (75 more than one vote margin) produced a 99% probability of Rossi winning a series of experiments — according to your calculations.
But with 800 ballots eliminated at 59%/41%, this would be a median value of Rossi winning by only about 14 or 15 votes.
So I think the standard deviation in this experiment is very small, and anything more than a couple dozen vote margin is beyond a reasonable doubt statistically.
Richard @ 21
“Most illegal felons register when they finish with DOC supervision, but still owe money on fines and restitution.”
Keep your money (or donate it to Goldy’s beer fund), but how do you know this? Wasn’t a good deal of the felon voting problem a result of not canceling their registrations when they were convicted?
Mr. C @ 22
“Appear does not mean PROVES does it?”
“Appears” and “proves” are two different things, but it’s important to understand that BOTH are required; i.e., the GOP must prove the facts that make it appear Rossi won.
I can’t tell you how Judge Bridges will define “appears,” because nobody knows. How, in the context of the statute, this probably doesn’t mean what you think it does. You might look out your window and say, “It appears to be raining.” For a court, this isn’t good enough; it’s either raining or it isn’t. Black’s Law Dictionary defines “appears” as, “To be in evidence; to be proved.” In other words, the evidence must create a likelihood that Rossi won the election. Speculation won’t do. Thus, statistical methods that are inherently speculation won’t pass judicial muster. The Repubs can’t guess who those votes were for; they have to show who they were for, by some reasonably reliable method.
Ladies and Gentlemen: If 1 felon voted, it’s one felon too many. If 130 total felons voted, we have an illegally certified election, more illegal voted than the election difference. Follow the RCW’s people. Butt, I am amazed on how fast the Democrats identified these felons. How did they know, inside information, hmmm….
Pudster
Puddybud @ 26
“If 130 total felons voted, we have an illegally certified election, more illegal voted than the election difference. Follow the RCW’s people.”
Sorry Buddy, this simply ain’t true. The RCW says that the invalid votes must actually change the outcome. There is a VERY low probability that 130 ex-felon votes would change the election outcome.
If felons vote proportionately the same as other voters, then it makes no difference if felons vote.
So dj @ 26 & headless one: You don’t care people illegally voted? Proves my point again. As long as Democrats win who cares how it happened. If the shoe was on the other foot, you would be screaming bloody murder and you know it.
Pudster.
Richard @ 23
”The 95% probability model generally means two standard deviations from the mean. Since there are two tails to the Bell Curve, we would actually be talking about a situation where 97.5% of the outcomes would have to show a Rossi victory in order to meet your test.”
With the Monte Carlo simulation described in post 21, this makes no difference whatsoever. Judge Bridges could choose whatever probability makes him comfortable for turning over the election (95% certainty, 97.5% certainty, 99.9% certainty, etc.). There is no parametric assumption (i.e. bell curve) in the Monte Carlo method—you just take the straight probabilities.
”Actually, DJ – look at your post # 30 on this thread: http://www.horsesass.org/my-co.....5&c=1 You did random ballot elimination at a constant 59/41 ratio of Gregoire to Rossi (i.e. like King County as a whole).”
Close, what I did there was a “proportional analysis” by statistical adjustment, not using the Monte Carlo simulation described @ 21, but by making use of some quick and dirty asymptotic statistical assumptions (technically: the normal approximation to the binomial distribution). The method gives approximate results. The Monte Carlo simulation gives, essentially, exact results.
“725 ballots eliminated produced an expected one vote win by Rossi. Basically, just over 50% probability of Rossi winning. 800 ballots eliminated (75 more than one vote margin) produced a 99% probability of Rossi winning a series of experiments – according to your calculations.”
That sounds right.
”But with 800 ballots eliminated at 59%/41%, this would be a median value of Rossi winning by only about 14 or 15 votes.”
I don’t really follow you on this. I think Rossi wins by much more with 800 eliminated ballots.
So, after reading your post, I decided to do the Monte Carlo simulation. I used the election totals for KC, including the votes for Bennett (sorry, I didn’t have data by precinct or for other candidates or write-ins). I then simulated ballot removal just for errors in KC. The ballots are randomly removed from each of the 3 candidates proportional to their votes. I did each simulation 100,000 times, each time randomly removing votes from the candidates, and then computing a new winner. Here are the results:
bals: 500, D_win: 96045, R_win: 3658, ties: 297, D_win: 96.0%, R_win: 3.7%
bals: 600, D_win: 82955, R_win: 15835, ties: 1210, D_win: 83.0%, R_win: 15.8%
bals: 700, D_win: 55280, R_win: 43091, ties: 1629, D_win: 55.3%, R_win: 43.1%
bals: 800, D_win: 28357, R_win: 70213, ties: 1430, D_win: 28.4%, R_win: 70.2%
bals: 900, D_win: 12635, R_win: 86926, ties: 439, D_win: 12.6%, R_win: 86.9%
bals: 1000, D_win: 5605, R_win: 94241, ties: 154, D_win: 5.6%, R_win: 94.2%
bals: 1100, D_win: 1748, R_win: 98071, ties: 181, D_win: 1.7%, R_win: 98.1%
bals: 1200, D_win: 556, R_win: 99313, ties: 131, D_win: 0.6%, R_win: 99.3%
bals: 1300, D_win: 121, R_win: 99863, ties: 16, D_win: 0.1%, R_win: 99.9%
This analysis is better than the one I did before because it has fewer statistical assumptions and includes the possibility of votes being removed from Bennett’s total in addition to Gregoire and Rossi’s. This analysis suggests a tie at something over 700 ballots (i.e. 50-50 chance of improperly invalidating the election). Furthermore, something over 1000 bad ballots must be removed if the certainty level is 95%. At 99.9%, 1,300 bad ballots would need to be removed.
In the real world, this analysis would be done at the precinct level, and include invalid votes uncovered in all counties, as well. In addition, I believe some adjustment must be considered for felon voting preferences (but Richard does not feel this is necessary). I don’t have precinct level data, but if anybody can provide a link or the data, please let me know.
Puddy:
It’s not a question of caring whether felons vote illegally. The issue is whether the fact that some felons voted illegally in this election (as they most certainly have in every recent election) is reason to overturn the election. The statutes require more than you seem to believe, the illegal votes have to “appear” to change the result, not just number greater than the result. I can’t imagine you would think it appropriate to overturn the election if Rossi had won by 129 votes and the Democrats pulled out 130 illegal felon votes from King County. That would be a ridiculous basis to overturn an election, but by your procedure, that’s what would happen. It is apparent the legislature wasn’t out to make it easy to overturn an election. I think Goldy is correct that felons who have served their time should have their voting rights restored. On the other hand, if the law is not changed, we should try to prevent felons from voting illegally. That doesn’t mean we overturn an election based on the issue.
There was a lot of Republican complaining when Gore tried to cherry-pick in 2000 by having a recount in only a few Democratic counties. I thought it was a mistake at the time. It is equally inappropriate for the Republicans to cherry pick illegal votes from Gregoire counties and then request proportional adjustment.
As far as assuming that felons will vote generally the same as the people in their precinct, it is probably accurate in part. However, the felons are likely predominantly men. In general, men vote Republican in higher percentages than women do. It would not be accurate to assume that male felons voted for Gregoire in the same percentage as the entire population of a precinct did. I don’t think proportional analysis works unless the pool of illegal voters has generally the same makeup as the precinct at large. I don’t see how that can work though.
Pudster @ 29
“You don’t care people illegally voted?”
Yes, I do care. I want the election to be conducted at the highest level of integrity that is humanly, technically, and fiscally possible. I want no ineligible voters to vote, no dead people to vote, and no double voting. I also want voting machines that are designed ergonomically so that the voter’s will is registered with certainty and there is almost no chance of people making mistakes because the machine is stupidly designed (I have a hunch this is the largest source of error in the last election). And, I want voting machines that leave a clear, verifiable, paper audit trail.
But, I cannot get overly vexed about ineligible ex-felon voters in the last election because they probably did not change the outcome. I suspect that if we could tally all these votes, they would largely cancel each other out. Remember, ineligible ex-felon voters likely voted for Rossi, Gregoire and Bennett. And some may not have voted for Governor at all.
I would bet the Democrats cherrypicked, just as the Republicans did. I suspect they picked illegal votes in Rossi precincts primarily. However, the Republicans can’t call them on it. How do they make the argument?
Rossi’s attorney: “The D’s cherrypicked their illegal votes from Rossi precincts! It’s not fair.”
Judge Bridges: “How come you didn’t find any of these illegal votes? You seemed to do a much more thorough job in King County. Why is that?”
Rossi’s attorney: “You’ll have to ask the BIAW. I mean … the elections a mess! You can’t tell who won!”
Judge Bridges: “Challenge dismissed.”
Wayne–
This is an Election CONTEST. It was not the R’s responsibility to go around the State. PERIOD. Just like Judge Lean Dum ruling allowed the Dems to go out and cherry-pick provisional ballot affadavits…remember that? Was that fair?
Alan–
It all eventually rolls back to the issue of what is reasonable relief when ILLEGAL BALLOTS are already comingled with LEGAL ones. It is impossible to physically pull them out. Does that mean no relief? I believe the Judge WILL allow some form of proportional analysis….but why 95%+ certainty??? Is that what appears means?
Stay tuned for tomorrows rulings.
dj at 32: “And some may not have voted for Governor at all.”
But with the voter intent thing going on in this state, the people who look at your ballot will decide that you just “forgot” to mark the correct box. If you voted straight ticket for everyone but Governor they will ASSUME you WANTED to vote for that party’s candidate, and mark it accordingly. Voter intent. I will never leave another box blank again, ever. I’ll write in a neighbor’s name. Many times I do not vote for an office if I feel I do not have sufficient information to make an informed choice. But when I learned that some one somewhere will decide what I meant to mark, I was appalled. So let’s get rid of voter intent too. If the box isn’t marked, it isn’t marked. That’s pretty simple.
Read this AM’s Seattle Times.
David Postman does another fine job of laying it out.
Keith Ervin has done a fine job too.
As opposed to Greg Roberts @ the P-I and his editors who have not been timely or comprehensive in their puny efforts. In fact, the HorsesTass (err….I mean P-I) has been downright irresponsible. Read the Journalistic Code of Ethics dudes.
Rae # 35
“But with the voter intent thing going on in this state, the people who look at your ballot will decide that you just “forgot” to mark the correct box. If you voted straight ticket for everyone but Governor they will ASSUME you WANTED to vote for that party’s candidate, and mark it accordingly. Voter intent.”
No. . . it does not work that way. Nobody fills in undervotes.
“Voter intent” means that the voter clearly marked her choice. This means the vote is valid even if the machine cannot read it. For example on the optical scan forms, a voter might check the ovals instead of darkening them in. Another error is circling the candidate’s name.
“But when I learned that some one somewhere will decide what I meant to mark, I was appalled.”
I am curious, Rae, where did you learn this?
“If the box isn’t marked, it isn’t marked. That’s pretty simple.”
Yep! The voter intent is clear: the voter does not want to vote in that race.
Wow….this argument back and forth is terrific reading. So much logic on all sides — with one common theme:
GET (KEEP) ROSSI (GREGOIRE) IN OFFICE!
This should be a great court discussion — (undoubtedly to be appealed by the losing party). The hardest part would probably to be a lawyer. Seems like the number of arguments (on both sides) published in these fora are huge — and the losing attorney will be brandished for not following some previously suggested line of logic that was outlined here.
Seems like our democracy is live and well — this discussion shows it.
Does anyone know if the hearing Monday will be covered on cable TV?
The statute says, and the judge has agreed, that the GOP must show it “appears” that Rossi got more votes after removing the illegal ballots cast for Gregoire and the ones cast for Rossi. All illegal votes, not just the ones brought to the Court’s attention by the GOP. If the judge finds that the GOP did cherrypick, and is convinced that there are significant number of illegal votes outside King County that were intentionally not discovered or brought to his attention by the GOP, he could not find that it “appears” that Rossi got more votes after deleting illegal ones. Therefore, either the judge has to count the ones found by the Dems, or he waves off the GOP’s challenge on that score as “appearing” devious and unreliable.
dj@37
But, if you simlpy had the pencil setting on a canidate, contemplated then determined that none of the canidates were worthy of your vote, you have made a mark, not readable by the machine however an election official could wrongly determine your “intent”.
But somehow “proportional analysis” is better at “divining” intent.
HAH…
Chuck @ 41
“But, if you simlpy had the pencil setting on a canidate, contemplated then determined that none of the canidates were worthy of your vote, you have made a mark, not readable by the machine however an election official could wrongly determine your “intent”.”
This is one reason for observers! If a tiny dot is made, it seems unlikely that any vote would be credited, as the voter’s intent is not clear.
With any such system, there will be “close calls,” however the vast majority of the cases are probably ballots where the machine cannot understand the intent but the voter’s intent is crystal clear to a human.
DJ @ 30
Your analytical approach is also better than what the one of the GOP’s experts is using. I refer you to a supplemental report filed by Professor Jonathan Katz of Caltech on 04/29/2005:
http://www.secstate.wa.gov/documentvault/361.pdf
Professor Katz uses a statistical prediction with mathematical formulae, instead of the 100,000 calculation computer test that you used. Since 100,000 or 1,000,000 computer tests can easily be done, and would correspond almost exactly to what the statistical formulae come up with, this method should be preferable to less mathematically gifted folks, such as most judges and lawyers (not to mention the general public).
Professor Katz does use precinct level data (at least that is what he claims) to do his analysis. However, his analysis is somewhat flawed, since it allocates all invalid ballots to either Gregoire or Rossi, based on their percentages in the given precinct.
Katz admits this is a flaw, since 2.7% voted for Bennett and 2.6% for undervote, etc. in the given precincts on average. Katz says this should balance out between the candidates. But since the precincts in question (weighted by the number of illegal ballots) voted 61.09% to 38.91% for Gregoire over Rossi, the expected 63 ballots out of 1183 cast for other than Gregoire or Rossi would deduct proportionately from each candidate’s total, rather than 31.5 for Gregoire and 31.5 for Rossi.
So I think that Katz overstates the impact of the 1183 illegal ballots that he considered by about 14 votes against Gregoire — i.e. 1183 X .053 X (.6109 – .3891)
Your model avoids some of the Katz problem, since you have included Bennett votes. If you had the undervote data, I am sure you would have included it as well.
Katz used a 95% confidence interval — literally 95%, since he went 1.96 standard deviations from the mean on either side. Most 95% confidence intervals — such as in polling results — actually use 2.00 standard deviations from the mean in either direction, and are in fact a little bit over 95% confident.
Based on 1183 illegal votes — GOP cherry-picked to include 947 from King County and 236 from elsewhere (all 174 illegal provisional ballots — illegal as opposed to merely irregular processing — were from King County!), Katz came up with results based on precinct analysis.
Katz expects these 1183 illegal votes would have a mean result of 722.64 Gregoire to 460.36 Rossi — which is a net change of 262 votes, and would give Rossi a predicted mean victory margin of 133 votes.
As best as I can determine, Katz came up with a standard deviation of 15.89 votes — dividing the changes on each end of his 95% confidence interval by the 1.96 standard deviations he appears to have used.
Since ignoring the 5.3% Bennett, undervote, etc. overstated the impact on Gregoire by approximately 14 votes, I will adjust Rossi’s predicted mean victory margin down to 119 votes. I don’t think the standard deviation would be impacted significantly by this 5.3%, and would — if anything — be reduced slightly from this apparent 15.89 votes.
So for Gregoire to win under the Monte Carlo simulation (i.e. 100,000 or multi-millions of computer experiments), it would require a z-score of -7.49 standard deviations (after subtracting 14 votes for Professor Katz’s erroneous analysis).
The probability of overcoming a z-score of -7.49 standard deviations is almost zero. I can’t find anything easily on the internet to calculate the probability of such a low z-score.
What I do find, the odds of being X standard deviations below the mean are:
2 standard deviations: 1 in 44
3 standard deviations: 1 in 741
4 standard deviations: 1 in 31,574
5 standard deviations: 1 in 3,486,914
6 standard deviations: 1 in 1,014,713,328
The website in question simply refused to calculate a z-score for anything more than six standard deviations from the mean. I would think that the probability of 7.49 standard deviations would be substantially in excess of a trillion to one.
So if you had all the data that Professor Katz had about these 1183 illegal ballots (by precinct) and had the full vote results (include Bennett, undervote and other — by precinct), you could run the Monte Carlo experiment by computer simulation 100,000 times, and every outcome should be a Rossi victory. In fact, you could run the 100,000 test computer simulation 100,000 times (a total of 10 billion tests), and every single one of these 10 billion tests should show a Rossi victory.
On the other hand, if we take the 432 additional felons that the Democrats found, and also consider the actual illegal provisional ballots (as opposed to merely being processed irregularly) cast in the other counties that the Democrats have pointed out, I think the probability of Gregoire prevailing in a statistical analysis or Monte Carlo simulation will be significantly greater than trillions to one.
Of course, if there are unexplained phantom ballots that cannot be traced to any voter (legal or illegal), such as the apparent unexplained surplus of ballots over voters in King County, those would increase Rossi’s odds if they are disproportionately found in King County.
When I said “I think the probability of Gregoire prevailing in a statistical analysis or Monte Carlo simulation will be significantly greater than trillions to one” —
What I meant was perhaps millions, thousands, or dozens to one, as opposed to quadrillions, quintillions, or centillions to one.
I.e. one in a million is a much better chance than one in a trillion.
Puddinhead @ 29
“You don’t care people illegally voted?”
That’s not what she said, dummy! Can’t you do even simple math? Here, I’ll spell it out for you.
Headless lucy said, “If felons vote proportionately the same as other voters, then it makes no difference if felons vote.”
As a simple example (we need to keep things simple, so your simple mind can grasp them), let’s say that in a hypothetical election, 1020 people voted for Candidate A and 980 people voted for Candidate B.
Step 1 – Candidate A got 51% of the total vote (1020/2000 = .51).
Step 2 – Candidate B got 49% of the total vote (980/2000 = .49).
Step 3 – Now let’s say 100 voters were illegal felons, and they voted for A and B in the same proportions; i.e., Cndidate A got 51% of the illegal votes and Candidate B got 49% of the illegal votes.
Step 4 – Candidate A got 51 illegal votes (100 x .51 = 51).
Step 5 – Candidate B got 49 illegal votes (100 x .49 = 49).
Step 6 – Now let’s subtract the illegal votes from each candidate’s total.
Step 7 – Candidate A now has 969 votes (1020 – 51 = 969).
Step 8 – Candidate B now has 931 votes (980 – 49 = 931).
Step 9 – There are 1900 legal votes (969 + 931 = 1900).
Step 10 – Candidate A still won with 51% of the votes (969/1900 = .51).
Step 11 – Candidate B still lost with 49% of the votes (931/1900 = .49).
Step 12 – Therefore, the illegal felon votes MADE NO DIFFERENCE to the outcome.
This is a purely mathematical exercise that contains no subjectivity or value judgment. Likewise, headless lucy’s statement was a purely mathematical observation that contained no subjectivity or value judgement. These calculations have NOTHING TO DO WITH caring or not caring whether the felons voted illegally. Headless lucy may care very deeply about felons voting, but her statement doesn’t tell you that one way or the other.
Get it now, dumbass?
Wayne @ 31
You’re probably wasting your breath. Puddinhead is too simple-minded to understand something as complicated as the election contest statute. He probably didn’t bother to read it.
Mr. Cynical @ 34
“Alan–
It all eventually rolls back to the issue of what is reasonable relief when ILLEGAL BALLOTS are already comingled with LEGAL ones. It is impossible to physically pull them out. Does that mean no relief? I believe the Judge WILL allow some form of proportional analysis….but why 95%+ certainty??? Is that what appears means?”
You’re probably right about what Judge Bridges will do. I asked one of the GOP attorneys what the legal basis is for proportional analysis, and he replied there are U.S. Supreme Court precedents on the subject. I didn’t research it myself (I have other things to do) but I’m willing to take his word for it.
So, the answer to your question “what is reasonable relief?” would seem to be: Some type of statistical analysis.
Rest assured the Democrats won’t even bother to argue the answer is “no relief.” They (and you) can count on Judge Bridges (and the State Supreme Court) to say, “The Legislature didn’t enact the election contest statute with the intention that no relief to illegal votes is possible, because that would render the statute superfluous, and there’s a presumption that legislative bodies do not enact superfluous laws.” The courts will craft a means for the plaintiff (the GOP) to demonstrate whether illegal votes changed the result, even though it’s impossible to identify the illegal ballots and pull them from the stack.
The unresolved question is, how?
In attempting to predict how the courts will answer this question, it’s important to keep in mind that lawyers (and, by extension, judges) are not rocket scientists or math whizzes. They’re accustomed to dealing with approximations. Every day, they round off monetary damages to the nearest thousand, ten thousand, hundred thousand, or million dollars — they DON’T get hung up on decimal points nor do they work very hard at trying to understand statistics. Standard courtroom procedure goes like this. An expert witness sets up his charts and graphs, and spends two hours testifying and waving a pointer. Then the defense grills him for another hour. At the end of which the judge says, “OK, so … who won?” And the expert says, “Gregoire” or “Rossi” and the judge says, “Oh, okay.” And that’s it. The judge isn’t going to sit in his chambers poring over the transcript trying to figure out what the hell the guy said. If it sounds good, he’ll go with it. Period.
That, Mr. Cynical, is how litigation works in the real world. But then, being a CPA who possibly has testified in court a time or two, you already knew this, didn’t you?
In short, litigation is a dice roll.
That’s why you hire the best lawyers you can find and pay them whatever they ask.
Hi Richard,
Thanks for the link. Indeed Professor Katz used the same approach that I previously used for the quick & dirty analysis (except that Katz does it at the precinct level and covers more than one county). He assumes a binomial distribution and then uses a normal approximation to the normal for computing statistical significance.
There are two major flaws with Professor Katz’s analysis: First, as you point out, he adjusts votes for only 2 candidates. The results change quite a bit if you also allow ballots to be statistically removed from the entire pool of ballots (about which, more below).
The second flaw is that the normal approximation to a binomial is not good when getting areas under the extreme tails of the distribution.
“Since 100,000 or 1,000,000 computer tests can easily be done, and would correspond almost exactly to what the statistical formulae come up with, this method should be preferable to less mathematically gifted folks, such as most judges and lawyers (not to mention the general public).”
Yes, but the Monte Carlo approach is much better because it gets rid of the bad behavior at the tails of the distribution when trying to correct things statistically. It yields an empirical distribution that needs no statistical interpolation of any type.
So, I did find the total number of votes in KC and re-ran the simulation with three categories: Gregoire, Rossi and All Other. Adding in the extra 2.5 % “other” ballots makes some difference:
Bals: 500, D_win: 96497, R_win: 3361, tie: 142, %D_win: 96.5, %R_wins 3.4
Bals: 600, D_win: 84938, R_win: 14076, tie: 986, %D_win: 84.9, %R_wins 14.1
Bals: 700, D_win: 60299, R_win: 38131, tie: 1570, %D_win: 60.3, %R_wins 38.1
Bals: 800, D_win: 32816, R_win: 65657, tie: 1527, %D_win: 32.8, %R_wins 65.7
Bals: 900, D_win: 15469, R_win: 83790, tie: 741, %D_win: 15.5, %R_wins 83.8
Bals: 1000, D_win: 7131, R_win: 92407, tie: 462, %D_win: 7.1, %R_wins 92.4
Bals: 1100, D_win: 2431, R_win: 97361, tie: 208, %D_win: 2.4, %R_wins 97.4
Bals: 1200, D_win: 864, R_win: 99074, tie: 62, %D_win: 0.9, %R_wins 99.1
Bals: 1300, D_win: 350, R_win: 99613, tie: 37, %D_win: 0.4, %R_wins 99.6
Bals: 1400, D_win: 116, R_win: 99868, tie: 16, %D_win: 0.1, %R_wins 99.9
Now it requires roughly 1050 ballots to overturn with 95% confidence, and a little over 1400 (instead of 1300) ballots to overturn at 99.9% confidence. The only difference is that I included ballots other than those who voted for Rossi, Gregoire, and Bennett.
”So if you had all the data that Professor Katz had about these 1183 illegal ballots (by precinct) and had the full vote results (include Bennett, undervote and other – by precinct), you could run the Monte Carlo experiment by computer simulation 100,000 times, and every outcome should be a Rossi victory.”
That would be an interesting exercise. Are the raw data available (precinct level voting information and the number of illegal ballots by precinct)? I would be happy to do the same analysis using those data—and with and without the county-level data that Goldy gave.
“In fact, you could run the 100,000 test computer simulation 100,000 times (a total of 10 billion tests), and every single one of these 10 billion tests should show a Rossi victory.”
I doubt it—the extreme results you calculate are because the asymptotic assumptions break down at the extreme tails of the distribution. I sure wouldn’t mind taking you up on that challenge, though! Got data?
Correction @ 50
He assumes a binomial distribution and then uses a normal approximation to the binomial distribution for computing statistical significance.
Rae @ 35
Your statement is incorrect. If the ballot contains no markings for an office, no vote is counted. Election officials refer to this as an “undervote.”
The question of intent comes up only if the ballot is marked but it’s not clear which candidate is marked. A typical example is a partially filled in oval next to one candidate’s name, with stray ink marks next to another candidate’s name. By law, the canvassing board is responsible for deciding if the voter’s intent can be determined, and what that intent is. If they can’t decide which candidate the vote is for, it’s an undervote and is not counted.
This decision is NOT made by the election workers who count the ballots. In the machine recount, ballots that had to be “enhanced” or “duplicated” because the machines couldn’t read them were inspected by Democratic, GOP, and Libertarian party observers; and if anyone disputed which candidate the vote was for, it was sent to the canvassing board. In the hand recount, every ballot was inspected by representatives of both parties, who could send any ballot to the canvassing board they wanted to, no questions asked.
In no case was a vote imputed to a ballot that had been left blank.
Chuck @ 41
Then you erase your mark, or draw an “X” through it, as the instructions tell you. Or turn in the spoiled ballot and request a new one.
If you turn in a ballot with a mark you didn’t want counted, you have no one to blame but yourself.
dj @ 43
The more interesting ballots are the ones on which the voter, in addition to marking his preference for a candidate, writes his opinion of the other candidate. Sometimes in crude language.
Alan @ 54
“The more interesting ballots are the ones on which the voter … writes his opinion of the other candidate. Sometimes in crude language.”
I’m sure that’s just to help establish their intent :)
DJ @ 50 & 51
I would assume that the tail-end problems are due to the normal distribution only being an approximate to the binomial distribution. So the probability of a result six standard deviations from the mean are presumably greater than .000000001 (i.e. one in a billion) with a binomial distribution, than with the normal distribution — i.e. you are likely to get that result more than once in a billion tries in a binomial distribution. Also, that such outlying results are even more common when the sample size of a binomial distribution is smaller.
I am not sure how much the tail end results of a sample size of 1183 would vary from the normal distribution. But presumably the further away from the mean in any binominal distribution, you will see the low probability outliers relatively more often than you would in a normal distribution — and even more so with a smaller sample size.
And with numbers for Bennett and Other, we would at least have to use a trinomial distribution, if not a tetranomial distribution, instead of a binomial distribution. Probably a trinomial distribution would suffice, since there are only three important outcomes — Gregoire, Rossi or Other — and the distinction between Bennett and Undervote/Other is not important.
I would imagine there is some formula — not the normal distribution, standard deviation formula — that can be used with binomial, trinomial, and other multinomial distributions to determine the likelihood of outcomes at given distances from the expected mean.
Of course, the Monte Carlo simulation repeated often enough will approximate this pretty well. You can also computed the standard deviation of each sample size, and see what number of results lie within certain standard deviation range, and compare these numbers with what would be expected in a normal distribution.
By the way, how long does it take to run 100,000 tests on a sample size, let’s say of 1,000 ballots? Is that almost instant on a laptop/desktop with 3 GHz of processing power, or does it take a while? Seconds? Minutes? Hours? How much longer would it take if each of these 1,000 ballots was assigned its own probability (i.e. coming from hundreds of different precincts), instead of having the same probability? Does adding Bennett and Other significantly increase the processing time?
I am not sure where to get the raw data that Professor Katz relied upon. Presumably, the state GOP had to give the raw data to the Dems and the SecState as part of discovery in the lawsuit. Opposing parties are certainly entitled to know the raw data on which the other side’s expert witness has relied.
The numbers that Goldy presented, of course, would also be more useful at the precinct level. As for provisional ballots, the gross numbers of irregularly processed ballots would be meaningless. We would have to have the exact numbers cast by illegal voters or by someone other than the actual voter (i.e. for the 1333 ballots in the four counties that did not verify signatures). Professor Katz has used only the irregular provisional ballots from King County that were not cast by legal voters (and was not supplied with any information on those kinds of provisional ballots from other counties).
By the way, if you want to see Professor Katz’s original report, which uses only a county level proportional analysis (and fewer numbers of illegal ballots), it is at:
http://www.secstate.wa.gov/documentvault/267.pdf
Hope you have a higher speed modem, since this one is about 6.8 MB. Not for dial-up people :)
The state GOP has two experts: Anthony Gill of the UW and Jonathan Katz of CalTech. Both have PhD’s in political science and are on the political science faculties of their respective universities. Katz, however, has a bachelor’s degree in applied mathematics from MIT, and seems much more statistically oriented than Gill.
The analysis by Gill in his report seems to be based solely on normal spreadsheet math (add, subtract, multiply and divide). Gill does have a lot of numbers in his report, and he does include Bennett, undervotes, write-ins, overvotes, etc. I am sure that Gill does his math correctly. But there is no higher-level math or statistical analysis, and nothing about 95% or other confidence levels in his report. Gill’s undergraduate degree is also in political science.
Alan@48–
After months of rasslin’ around in the mud for yucks…
After months of pokin’ and jabbin’
After months of just plain being a couple of meanass goofballs..
I think it pretty much boils down to what you said @ 48.
I have testified in Court on more than a few occassions.
I can remember one case in a battle between insurance companies on a compensation issue that the other side had an “expert” lawyer testify and the lawyer got into trying to be a numbers expert. I watch the Judge (I think his last name was Tui) sit and listen to this “expert”. When his brow became permanently furrowed with confusion, we knew we had won the case.
The side that makes the simplest, easiest to understand methodology that seems reasonable, will prevail. While I can certainly appreciate the tremendous efforts of some of our resident statisticians here @ HA, I think y’all got to simplify thigs a bit more. While I’m not arguing your methodology, you still must convince the Judge and he must understand clearly.
Judge Bridges WILL NOT approve some methodology that he doesn’t clearly understand.
It will be fascinating tomorrow, won’t it Alan?
re: #29 by the Pudwhacker. That’s correct. That’s how the game is played on both sides. Except we won this one, not you– and since Repubs. have bet the farm on winning at any cost, this does not bode well for the Repub. future. It sets a bad example. You can be beat at your own game. That’s why reversing this thing is so important to you. The pendulum has now begun to swing the other way. Tom “the leper” DeLay is on the way out and you will never see a privatized SS account EVER. And Rossi is looking more amnd more like a petulant joke.
just a comment to thank all of the posters on this thread. The most readable thread I have seen here in a very long time. Kudos to all on both sides of the contest.
Eat shit and die jpgee…you stupid cocksucker!
nice quality post Mr. Cynical @ 61, but then again, I figure the previous posts were from an imposter. Nice to see that you are still on life support…..
fsc–
No sense of humor dude.
jpgee is our resident pansy-ass and self-proclaimed arbiter of what a good post is..(even though he is known for his foul, angry posts.)
Richard @ 56, 57
”I would assume that the tail-end problems are due to the normal distribution only being an approximate to the binomial distribution. So the probability of a result six standard deviations from the mean are presumably greater than .000000001 (i.e. one in a billion) with a binomial distribution, than with the normal distribution “
Yes, the normal is an approximation of the binomial. As p, the probability of a “success,” moves away from 0.5, the binomial becomes skewed (the normal is perfectly symmetrical). Additionally, the binomial is restricted to the range 0 to N, where N is the maximum number of “Bernoulli trials” (in this case, bad votes). The normal distribution has mass on the entire real number line. I afraid I don’t have a good intuition of exactly how the tail probabilities get messed up for different combinations of p and N.
”And with numbers for Bennett and Other, we would at least have to use a trinomial distribution, if not a tetranomial distribution, instead of a binomial distribution. Probably a trinomial distribution would suffice, since there are only three important outcomes – Gregoire, Rossi or Other – and the distinction between Bennett and Undervote/Other is not important.”
This is exactly correct. A trinomial will do the trick.
”I would imagine there is some formula – not the normal distribution, standard deviation formula – that can be used with binomial, trinomial, and other multinomial distributions to determine the likelihood of outcomes at given distances from the expected mean.”
There are no closed solutions for either the Normal or ___nomial distributions. Instead, you simply add up the areas under the ____nomial distribution from one limit to another to get areas under the distribution. In fact, you can use areas under the trinomial instead of the normal approximation.
”Of course, the Monte Carlo simulation repeated often enough will approximate this pretty well. You can also computed the standard deviation of each sample size, and see what number of results lie within certain standard deviation range, and compare these numbers with what would be expected in a normal distribution.”
Yes, it will give the same mean result, but a more accurate estimate of certainty. One can mess with comparing results against the normal approximation, but again, I am less interested in approximate methods. The simulation is the real deal (oxymoron intended :-) ). It also allows you to add subtle complexities like, what do you do with tied elections? You can draw straws, ignore them, or pretend that trial would have gone to the legislature and mimic the behavior of the legislature. You can also do subtle, second-order improvements like re-adjust the probabilities within a county/precinct each time an illegal ballot is removed in the simulation. Finally, you don’t have to confuse a judge with the statistical mumbo-jumbo beyond percentages.
”By the way, how long does it take to run 100,000 tests on a sample size, let’s say of 1,000 ballots? Is that almost instant on a laptop/desktop with 3 GHz of processing power, or does it take a while?”
The simulations I ran before (100,000 KC ballot removal exercises x 10 different bad ballot sizes from 500 to 1400) took 23 minutes on an AMD 2600 running Linux. I wrote the program in an interpreted statistical programming language—it would be 5 to 15 times faster if I had written it in a compiled language.
”How much longer would it take if each of these 1,000 ballots was assigned its own probability (i.e. coming from hundreds of different precincts), instead of having the same probability? Does adding Bennett and Other significantly increase the processing time?”
Modifying the program to loop through actual county or precinct data increases the time per simulation, but with actual bad-vote data, you no longer need to run it at multiple “bad ballot” sample sizes. I did a county-wide simulation with 100,000 “trials” and it took 4 minutes.
”I am not sure where to get the raw data that Professor Katz relied upon. Presumably, the state GOP had to give the raw data to the Dems and the SecState as part of discovery in the lawsuit. Opposing parties are certainly entitled to know the raw data on which the other side’s expert witness has relied.”
Well, if you or someone else finds the data, I have a program ready to make use of the data at the precinct or county level. I’ll put the county-level analysis (using a preliminary data set) in my next post.
”As for provisional ballots [from Goldy], the gross numbers of irregularly processed ballots would be meaningless. We would have to have the exact numbers cast by illegal voters or by someone other than the actual voter (i.e. for the 1333 ballots in the four counties that did not verify signatures). Professor Katz has used only the irregular provisional ballots from King County that were not cast by legal voters (and was not supplied with any information on those kinds of provisional ballots from other counties).”
Good. . . I dropped all of the Democratic “unverified provisionals” from Goldy’s list and kept all the provisionals from the Republican list in my county-wide analysis.
”By the way, if you want to see Professor Katz’s original report, which uses only a county level proportional analysis (and fewer numbers of illegal ballots), it is at: http://www.secstate.wa.gov/documentvault/267.pdf ”
Thanks! Although, that linked to Professor Gills’ report. I was able to get a (somewhat obsolete) county-wide breakdown of the ineligible ex-Felon votes from that, and update them with information from Professor Katz’s report. Where is the search engine for finding these documents?
Okay, so here is a preliminary analysis using Monte Carlo simulation to statistically remove bad ballots. The analysis is preliminary because I don’t have all the data. Any corrections to the raw numbers are most welcome.
WA________ Dem Rep Rep Gregoire Rossi Other
County_____ Illegal illegal prov’l Hand Hand Ballots
Adams______ 0 0 0 1529 3481 5010
Asotin______ 0 0 0 3530 4914 8444
Benton______ 25 16 0 19834 44895 64729
Chelan______ 5 0 0 10077 18438 28515
Clallam_____ 9 0 0 16230 18836 35066
Clark_______ 35 33 0 72828 85924 158752
Columbia____ 0 0 0 671 1371 2042
Cowlitz_____ 2 0 0 20204 20045 40249
Douglas_____ 4 0 0 4360 8667 13027
Ferry_______ 4 0 0 1278 1900 3178
Franklin____ 4 0 0 4977 10634 15611
Garfield____ 1 0 0 428 840 1268
Grant______ 8 0 0 7821 17431 25252
Grays_Harbor 1 0 0 13729 13457 27186
Island______ 5 0 0 16895 20000 36895
Jefferson____ 2 0 0 10650 7295 17945
King_______ 31 773 219 506194 351306 857500
Kitsap______ 23 6 0 56236 57775 114011
Kittitas _____0 0 0 6125 9567 15692
Klickitat____ 2 0 0 3919 4767 8686
Lewis______ 17 7 0 10247 20851 31098
Lincoln_____ 2 0 0 1850 3686 5536
Mason 1____ 0 0 11797 12519 24316
Okanogan___ 7 0 0 6107 9460 15567
Pacific______ 1 0 0 5210 4730 9940
Pend_Oreille_ 0 0 0 2567 3368 5935
Pierce 49___ 28 0 145431 157905 303336
San_Juan____ 0 0 0 5872 3660 9532
Skagit______ 11 1 0 23250 27219 50469
Skamania___ 0 0 0 2233 2525 4758
Snohomish__ 27 26 0 139189 145628 284817
Spokane_____ 65 20 0 90581 105584 196165
Stevens_____ 1 0 0 6992 12295 19287
Thurston____ 25 37 0 58970 49426 108396
Wahkiakum_ 0 0 0 993 1099 2092
Walla_Walla 5 2 0 8008 14290 22298
Whatcom___ 4 13 0 44072 42000 86072
Whitman___ 1 0 0 7722 9365 17087
Yakima____ 55 30 0 24755 46079 70834
Note that there are at least 17 Republican ballots (14 deceased and 2 multiple voter ballots) that I could not include because I don’t know which county they go with. Also, I eliminated ALL 1939 “unverified ballots” from Goldy’s post. This is because most of these will end up being verified. But, again, better data would be welcome. Leaving out the Dem’s unverified ballots will offset the missing 17 Repub’s ballots—for now.
Don’t get pissed off at me if you think I missed some data. But, do send the data to me (or a link) and I will update the results. I am trying to do this exercise as impartially as possible. Disclaimer: As you know, I think the proportional analysis is deeply flawed because it assumes Felon voters vote like their precinct (or county) neighbors. But, this is the only likely remedy that Judge Bridges would choose (short of Republicans demonstrating fraud). Therefore, I am curious how the numbers flesh out.
Here is how it works: I went through each county in turn, removing bad ballots in proportion to how many were cast overall for each candidate. The net gain or loss was computed and, after all counties are processed, deleted ballots for both candidates is subtracted from the 127 vote victory margin. The election was classified as a tie, as a win for Gregoire or a win for Rossi. This was repeated 100,000 times, and the percentage of wins for Democrats and Republicans was computed.
I did it two ways: once using all ballots and the second time using D & R ballots only (this is not, IMHO, the right way to do it, but it is how Professor Katz did it). I ignored ties in computing the percent wins—other options are reasonable. Here are the results:
All ballots: D_win: 72808, R_win: 26239, tie: 953, %D_win: 73.5, %R_wins 26.5
D&R ballots: D_win: 68931, R_win: 29107, tie: 1962, %D_win: 70.3, %R_wins 29.7
In words, with the data set given above, it appears that Gregoire wins 73.5% of the trials where illegal ballots are removed. The percentage goes down a little to 70.3% if other candidate’s ballots are not considered for random removal.
DJ and Richard,
Thank you both for the incredibly informative dialectic! I don’t understand a lot of it (as Stefan will tell you, I’m not up on the vocabulary of statistics) but I’m learning a lot.
It’s very gratifying to see the threads used for more than just name calling.
Hey DumberAss Alan @ 46: Vitriol. Standard Democrapic playbook. Name calling. Standard Democrapic playbook. Your assumptions about statistics don’t fool me. I took it in college and in graduate school. Your assumption of the balance being the same for the felon voting doesn’t wash. It will fool the simple AssesHorse democraps but not me. All you did was apply simple ratios and naturally by the laws of mathematics the ratios never change so your outcome never changes. You should have tried a 70/30 ration for democrapic felons voting. Then multiplied the values and subtract differently. You assumption regarding the breakdown is probably way different than that of other states. I rest my fingers on you. Why would Sam Hunt offer bills to allow felons to vote, because on the inside the democraps know where the felon breakdown lies. You all on theis website don’t but I bet you they do in Olympia. Otherwise the bills from your side would freeze out felons. Get it?
Otherwise, why would Hilary and John Kerry say let them vote. Why would Kerry pay send > $250K to the election, because you all know felons have voted and helped your cause forever. But that will soon change.
Hey HeadinAss Lucy @ 59: That’s how the democrapic game is played. If you read the democratic newspapers, the Republicans don’t want felons voting, BUTT you do. There is no game here. It’s called fairness, and vote disenfranchisement by illegal means. My vote has be disenfranchised by an illegal felon, GET IT? Probably not that you would being headless and all.
Oh yeah, about Tom Delay. Did you see the latest in the democrapic newspapers like the WA post? Probably not since you are headless. But if you see Nancy Pelosi, Jesse Jackson Jr, and others have family members on their payroll. Jesse Jackson’s wife has been working for him since he started. Strike #1 off of Delay’s pick sheet. Regarding the use of lobbyists, better watch out when they start looking at the democrapic side and see much worse, like the lady from Ohio. Strike #2 from Delay’s pick sheet. And last butt not least, your great friend, Jim McDumNuTTs. Remember him, he illegally sent a taped phone conversation to teh NYT. Well wasn’t he convicted of that. Strike #3 from Delay’s sheet. Exonerated or light slap on teh wrist, just like the WA Post said. So headinass one. Stop the Delay stuff. You democraps will lose another in the light of day no less.
Pudster
DJ @ 64 & 65
You have done an incredibly impressive analysis. I wish I had the precinct level data to give to you.
If the county level proportional analysis is used, it looks like Rossi loses — at least assuming that the numbers given by the Dems and the GOP are correct.
Rossi will probably lose by more, when the true illegal votes are considered from the 1939 irregular provisionals. For example, Benton County has 8 illegal provisionals out of the 37 irregular provisionals (i.e. ballots dropped “in the can” versus placed in envelopes) based on an e-mail I got a few months ago from the Benton County Auditor. I don’t expect any illegal provisionals to be proven at all from the 1333 non-signature verifieds from Adams, Steven, Whitman and Walla Walla Counties. But we can probably find a few more illegal provisionals from the other 569 ballots (I already know the answer for Benton!) from the other listed counties.
On the other hand, a test that shows Gregoire winning only 70.3% or 73.5% of the time does significantly narrow her mean expected victory margin. Probably not enough to overturn the election under the relevant legal standard, but it sure does raise a lot of doubts. We certainly wouldn’t think beyond a reasonable doubt that Gregoire was the true winner.
Precinct level analysis may tell a different story, and may result in a convincing win for Rossi. This would be the case, for example, if illegal felon voters tend to be concentrated in lower income, higher minority precincts that more strongly favored Gregoire than the given county did as a whole. The analysis by Professor Gill tends to suggest that this may be the case.
On the other hand, the GOP could very well have cherry-picked its felons list, and excluded illegal felons that voted in strongly pro-Rossi precincts. A comparison of the number of GOP-list felons and Dem-list felons from each county, versus the Gregoire-Rossi totals in each given county, seems to support this hypothesis.
(It may be unfair to say that the Dems also cherry-picked. If the GOP took all the Gregoire cherries and toss-up cherries from the tree, then there were only Rossi cherries left when the Dems got around to sending in their own pickers to get what was left on the tree.)
By the way, that Document 267 PDF file link on the SecState website contains BOTH Professor Gill’s report and Professor Katz’s original report. Katz’s original report is at Pages 71 to 80 of the PDF. Katz’s CV (resume’) is at Pages 82 to 87, and again at Pages 88 to 93 of the PDF. (The GOP attorney included TWO copies of Katz’s CV in the PDF!)
I hope you started with a 129 vote victory margin for Gregoire. Your post included a reference to a “127 vote victory margin”. A two vote error would have changed the percentages somewhat.
Maybe Goldy can digest some of these statistics by 6:30 a.m. tomorrow morning. That way, he could tell Kirby Wilbur that Gregoire is likely to withstand Rossi’s election challenge if the proportional analysis is done at the county level.
(On the other hand, what about all the voterless ballots in King County? If some of these remain after considering all of the irregular provisionals and the various voter crediting errors, then this will help Rossi’s position. And Stefan believes that voterless ballots are more concentrated in the stronger Gregoire precincts in King County.)
Puddinhead @ 67
“Vitriol. Standard Democrapic playbook. Name calling. Standard Democrapic playbook.”
You must be new here. This is what we do at HorsesAss.org — call each other names. Which, by the way, is not limited to the liberal; your Republicon pals do it too. Read a random selection of posts by JCH, Mr. Cynical, chardonnay, et al. and you’ll see what I mean. Bloodsport is part of the HA culture, get the drift?
If the chest-thumping here is to include whacking each other over the head with degrees and credentials, then I took statistics in college (got an A — in fact, I had STRAIGHT A’s in my economics minor), I also went to graduate school (3.97 GPA), and I also graduated from law school and passed the bar exam on one try. You should understand, however, that I don’t put much stock in credentials — I’ve met too many educated fools to believe in paper qualifications anymore.
With the preliminaries out of the way, I’ll now address the substance of your post.
You said, “All you did was apply simple ratios and naturally by the laws of mathematics the ratios never change so your outcome never changes.” Well yeah. Because that was the premise of headless lucy’s statement. “If felons vote proportionately the same as other voters, then it makes no difference if felons vote,” remember? If you change the premise of the question, you’ll get a different answer, but who the hell authorized you to change the question? Is that what you did on your math exams? And you still got a passing grade?
Actually, before I let you get too sidetracked, the real issue is not a math calculation but your assertion that headless lucy’s objective recitation of a math calculation implies that headless lucy doesn’t care if felons vote. This is like saying, “Because you said 2 + 2 = 4, you don’t care if felons vote.” Stop and think about this for 30 seconds, and you should be able to see why I concluded you’re a dummy.
Okay, then you change the subject and raise a completely new issue by saying, “You should have tried a 70/30 ration for democrapic felons voting,” Why should I? In addition to being beyond the scope of the question on the table, what basis is there for assuming the illegal felon votes break 70/30 for Gregoire? A study by a college professor of how felons in Minnesota vote, who used demographic data from New Jersey, and doesn’t even know how felons vote in New Jersey? Oh, and never mind that Washington’s prison population has a much higher percentage of Caucasian males (who, as a demographic group, tend to vote Republican) than New Jersey or Minnesota. (I might not have the states exactly right, but who cares. Substitute Ohio or Indiana, if it makes you happy.)
Here’s why I won’t try a 70/30 ratio to guess how Washington felons voted. The best evidence we have right now is the statements 9 felons — 8 males and 1 female — made to news reporters. All 8 male ex-cons said they voted for Rossi. The female ex-con said she voted for Gregoire.
So my question is, why don’t YOU try an 89/11 ratio (89% for Rossi, 11% for Gregoire), since that’s what the evidence suggests? Or, if you want to get fancier, split the illegal felon votes into male and female groups, and assume 100% of the males voted for Rossi and 100% of the females voted for Gregoire, since that’s what the evidence suggests?
P.S., dumb one @ 67 — my vote was disenfranchised by a felon who voted for Rossi, so we’re even.
Richard @ 68
The problem I have with this bell-curve stuff is that Judge Bridges won’t approach this from the standpoint of “there’s an X% probability that Gregoire won.” Instead, he will subtract illegal votes from Rossi’s and Gregoire’s totals to see who won, because that’s how the RCW requires him to do it. Probabilty calculations are irrelevant and won’t be considered by the judge.
Whether anyone thinks “beyond a reasonable doubt that Gregoire was the true winner” may be relevant in a public opinion context, but it has no legal significance. If your intention was to predict the legal outcome, then you’re wrong about both the burden of proof and the standard of proof. The Democrats don’t have to prove beyond a reasonable doubt (the wrong standard of proof), or even by a preponderance (the correct standard of proof), that Gregoire was the true winner, because they don’t have the burden of proof; the Republicans have to prove by a preponderance of the evidence that Rossi was the true winner.
According to Postman’s article in the Seattle Times, Gill’s and Katz’s proportional apportionment at precinct level of 879 llegal votes alleged by Republicans would put Rossi ahead by 62 votes. But that doesn’t take into account 432 illegal votes turned up by Democrats in Rossi-leaning counties. These votes would more than offset the 62 votes if 57% or more of them are apportioned to Rossi, which seems likely as Rossi carried most of the eastern Washington counties by around 65%.
The Republicans will try to keep these illegal votes out of evidence, probably by arguing they didn’t have enough time to review this evidence because they didn’t get it until Friday. However, the proper remedy for that is to grant a time extension, and Judge Bridges would be inviting a Supreme Court reversal by excluding this evidence and basing a decision for Rossi on the GOP’s cherry-picked illegal votes.
So even if Bridges adopts Gill’s and Katz’s precinct-level proportional analysis, I think the odds of Rossi prevailing are slim to none. This methodology would narrow Gregoire’s lead but it’s unlikely to overcome it, once the illegal votes discovered by the Democrats are factored into their formula.
Any way you slice it, the math looks grim for Rossi and his supporters.
Of course, if the Repubs can’t prove all 879 votes were illegal, the numbers get even worse for Rossi.
Alan–
It will be mighty close if the Judge allows precindt level proportional analysis….but the Dems 432 have to be svrutinized and verified. I’ve heard that list is not perfect either.
In addition, there are other issues regarding how KingCo conducted the election (lacking required precindt reconciliations, certifying results with known problrms etc.)
It will be interesting to see what the Judge does on some of these other issues.
Airheaded Alan: I don’t live in Vitriol. Not my style. You can twist my moniker and that’s fine but to call me names it’s not how I deal in a conversation, regardless what Goldy allows. Congratulations on your intellegence quotient. My undergraduate is engineering and graduate is intercomputer communications. At least we are having an intelligent coversation.
I just heard Berendt on Mike Siegel show 6:42 AM, and his take was different on these “so called felons” now. Didn’t sound like he was now saying they were ignored by Republicans. In fact when put on the spot he stammered and studdered. Vance admitted on the air that they included those felons originally, but their names didn’t fully match so they were careful about the names. He used the example of a “potential” felon named Michael Siegel, but if he wrote Mike Siegel on the voting document, he didn’t include that name. Berendt had NO RESPONSE. Have you ever heard of dead air or pregnant pause for a few seconds. So Berendt put out names that were looked at and Berendt was called on his own words again.
What are most felons?: Poor and undereducated. This is for most blue collar felons in the US. I suppose you missed the Vancouver WA Columbian Newspaper article where they have documented Gregoire felon votes. You all focus on Berendt and what he says without doing any due diligence on your own. You never answer my question regarding Hilary Clinton and John Kerry wanting felons to vote, and you know why. You never answer the Sam Hunt question on his take on felons voting in WA State and you know why. They vote for your side. Who do the poor and undereducated vote for? Hilary and John Kerry knows more than you? You really want to admit that in a court of law? Wow. I stand behind my 70/30 otherwise you would not hear your wonderful Clinton/Kerry band wagon beating the “Let the Felons Vote” mantra.
Pudster
dj & Richard — thx for the nitty-gritty and reasoned colloquy. Unfortunately, we know that precincts divide more unevenly than counties, and we know a significant share of the questioned ballots are clustered by precinct … and unless I’m greatly mistaken, we don;’t yet know definitively which classes of questioned ballots Judge Bridges will eliminate, EVEN IF he does accept a statistical ballot elimination procedure.
So, good illustration as to methods, etc., and no necessary indication as to potential outcomes.
dj @ 65
You LEFTIST IDIOT!!!! It was a 129 vote margin, not 127! :-)
Thanks, Richard, for putting it so diplomatically:
“I hope you started with a 129 vote victory margin for Gregoire. Your post included a reference to a “127 vote victory margin”. A two vote error would have changed the percentages somewhat.”
I did screw up and use 127! What a dolt!
In any case, here are a couple of analyses using the correct numbers. I did these analyses to point out some additional flaws of Professor Katz’s analysis:
1. No 2nd order, ignore ties, All. D_win: 74356, R_win: 24691, tie: 953, %D_win: 75.1, %R_wins 24.9
2. w/ 2nd order, ignore ties, All. D_win: 74544, R_win: 24597, tie: 859, %D_win: 75.2, %R_wins 24.8
3. No 2nd order, random ties, All. D_win: 74450, R_win: 25550, tie: 0, %D_win: 74.4, %R_wins 25.5
4. w/ 2nd order, random ties, All. D_win: 74972, R_win: 25028, tie: 0, %D_win: 75.0, %R_wins 25.0
5. w/ 2nd order, random ties, D&R. D_win: 71361, R_win: 28639, tie: 0, %D_win: 71.4, %R_wins 28.6
BTW, I did not change the number of bad ballots for this analysis. Richard mentions 8 invalid ballots in Benton County, but I also owe the Republicans 17 invalid ballots outside of King county—I do not yet know which county they belong to.
The first analysis ignores ties and is, essentially, the same as before but using the 129 margin victory. Gregoire wins 75.1% of the elections.
The second analysis is like the first, but I made a “second order” improvement. Each time an illegal ballot was “removed,” I reduced by one the total number of remaining ballots in the pool. To statisticians, this is the difference between “sampling with replacement” and sampling without replacement.” The latter is more realistic. If we really pulled actual ballots randomly, we would not throw that ballot back into the mix after recording the vote. Rather we would set aside the ballot and have a smaller pool of ballots for the next random draw.
As you can see, it makes a little difference (0.1%). But, the small effect is because I am working with county-level data. Using precinct-level data, the effect would be much greater, because the proportions of ballots for each candidate will change much more after a single ballot is removed. Professor Katz’s statistical method, in essence, assumes sampling with replacement, which when done at the precinct level biases the results.
The second pair of results (3 and 4) shows what happens if ties are randomly allocated to each candidate with 50:50 chances. This favors Rossi a bit. Interestingly, the effect of the second order correction is much greater now (0.6%).
Analysis 5 shows what happens when ballots from 3rd party, independents, and under-votes are not considered. Rossi gets a 3.6% boost. This assumption is bad because illegal voters may well have cast ballots for other candidates or left their ballots as under-votes for the Governors race.
Okay, Richard, we had quite a dialog about “tails of distributions” and approximations. Here are my results using only the Republican numbers. I think you suggested earlier that every simulated election should go for Rossi based on Professor Katz’s results (but perhaps you were discussing the precinct-level analysis). Here they are:
6. No 2nd order, random ties, D&R. D_win: 3207, R_win: 96793, tie: 0, %D_win: 3.2 , %R_wins 96.8
7. No 2nd order, random ties, All. D_win: 4048, R_win: 95952, tie: 0, %D_win: 4.0 , %R_wins 96.0
8. w/ 2nd order, random ties, All. D_win: 3857, R_win: 96143, tie: 0, %D_win: 3.9 , %R_wins 96.1
First notice that there are some wins of Gregoire in all three analyses. Analysis 6 drops the “other” ballots from the experiment (like Professor Katz does), and biases the results in favor of Rossi. Analysis 7 includes the “others” and % Dem wins goes up from 3.2% to 4%. Removing invalid ballots before continuing (analysis 8) drops that down to 3.9%. So if even a few ballots in the list are found to be valid, the probability would drop down below 95% for the Republicans (the level that Professor Katz was inclined to use for “certainty”).
Richard @ 68 and RonK @ 76 point out that the results will change using a precinct-level analysis. Indeed! I think this exercise is useful for pointing out flaws in Professor Katz’s analysis, including:
1. He used a binomial distribution (i.e. assumed ballots will be removed from Rossi’s and Gregoire’s pools only). This biases the results in favor of Rossi.
2. He uses an approximate method for determining statistical significance. The bias of doing so seems to favor Rossi.
3. His statistical method assumes “sampling with replacement,” (i.e. improper ballots are thrown back into the mix before each “draw”), which slightly biases results in favor of whatever party is losing. This bias could be much larger for small precincts.
Pudwhacker: Pull that Vlasic dill out of your anus immediately! It’s beginning to affect your vision.