For those who long for the dulcimer sounds of my voice (Stefan describes me as “sweet”), tune in to The Kirby Wilbur Show, 570-KVI, Monday morning at around 6:30 AM. Kirby and I will be discussing tomorrow’s hearing in Chelan County, which has the potential to make or break Dino Rossi’s election contest.
There are two major issues at stake: whether the court will accept the GOP’s “proportional reduction” method of subtracting illegal votes from the candidates, and whether it will consider offsetting illegal votes uncovered by Democrats. As is his wont, Judge Bridges will likely issue a ruling from the bench, but I’m not entirely sure that it will be definitive. For example, he could leave the door open to some sort of a statistical analysis, but not necessarily accept the GOP’s precinct-level methodology.
In any case, this is an issue that will ultimately be decided by the Supremes, so no one should get their undies in a knot by tomorrow’s ruling, one way or the other.
UPDATE:
Well, it was short (and as Stefan would say, “sweet.”)
One point I don’t think I emphasized on the air was that really, the R’s need to prevail on both issues to stay alive in this contest. If Judge Bridges allows the D’s to put offsetting illegal votes into evidence, it’s all over, even with a precinct-level proportional analysis. By fighting the admission of offsetting illegal votes from pro-Rossi precincts, Rossi is essentially admitting that he lost.
UPDATE, UPDATE:
News is coming out of the Chelan County Courthouse, and I am reporting it here. So far, nothing earth shattering… Judge Bridges will consider a statistical analysis, but has not yet ruled it in or out, and votes by non-citizens will be excluded.
Goldy, Will you be wearing your Teddy Kennedy [D- Mary Jo] neck brace so you will look like “a victim”?
The Republicans have suggested that the respondents bear the burden of proof to show that votes on the Republican list are LEGAL. I predict that the judge will rule that both the Democrats and Republicans will bear the burden of proof to show that their lists of ILLEGAL votes are 100% accurate. The judge will allow a proportional analysis of the voting errors, but he will set a high mark for preponderance of the evidence. The judge will not rule on the complete mess argument, nor will he rule on proportional reduction of any vote totals; these issues will be decided during the trial phase. The Democrats will move for a Frey hearing to argue the qualifications of Dr. Katz and the relevance of his analysis, which the judge will grant. The Republicans will claim a sweeping victory, again, while the Democrats will not lose any sleep over the days events. Mark my words, when I’m wrong, Mr. Cynical will serve my crow with the speed of the Valkyries’, and a strong Wagnerian theme. …mmm, love that crow.
Shawn–
May tomorrow bring these beautiful words from my lips to you and that is, “Shawn, dinner is served!”
In any case, this is an issue that will ultimately be decided by the Supremes, so no one should get their undies in a knot by tomorrow’s ruling, one way or the other.
Yes. And the supremes may need to rule on the matter before the trial even starts. Thus, I would not be suprised to see the republicans ask to have the trial delayed while the interim appeal is heard.
Um… JCH… it’s the radio. Nobody sees what I look like. That’s why, as usual, I’ll be doing the show butt naked.
Goldy-
Butt naked? Are you going to wear one of those Tarzan loinclothes?
A scrawny little Jewish guy from Philly…
why don’t you try to cut loose with one of Tarzans famous yells:
AHHHHHHHHHHHHHH…AHHH..AHHHHHHHHH…AHHH…AH….AHHHH!
I’m afraid that with your genetically wretched, screechy voice the FCC would ban you permanently if you tried that Tarzan yell.
Goldy sans clothes – just say no Goldy… think of Kirby if nothing else!
Goldy…Please, No more images of a 110 pound Philly Jewish boy naked. Wear the neck brace. Hell, the neck brace helped Teddy get away with murder. Maybe, as a Democrat “victim”, you can wear it and find someone to sue. [That is the Democrat way.]
Goldy
6:30 AM? You expect me to get up at 6:30 AM just to hear you on radio? I don’t think so. I’m retired … I sleep til noon.
Shawn @ 3
High mark for preponderance of evidence? What the heck does that mean? Preponderance is preponderance, Shawn. There aren’t different kinds of “preponderance.” Burden of proof is another thing altogether. There should be only one burden of proof, and it should be on the GOP; they should have the burden of proving that illegal votes changed the election result.
Mr. Cynical @ 3
Since you refused to eat when it was your turn, I don’t think you’re in a position to serve crow to anyone else.
Erik @ 4
I’ll be VERY surprised to see the Republicans file an interlocutory appeal. Here’s why. The GOP attorneys pushed for the May trial so the Supremes could make a decision by early July. Any later than that, and it’s too late to get on the November ballot. It makes sense for the Democrats to appeal Judge Bridges’ interim rulings, but it makes no sense for the Republicans to do so. However, one of the Democratic attorneys told me they won’t do that.
Erik @ 4
I don’t think that the Supreme Court is going to accept an appeal from any of Judge Bridges’ ruling until the trial is held and a final judgment entered. Especially if the ruling, whichever way it is, allows for each party to present its desired evidence at the trial.
We will have a trial and a final judgment within a few weeks. The Supreme Court could take a month or two (or more) to decide any appeal. Since hearing an appeal before final judgment is discretionary on the part of an appellate court, I seriously doubt that an interim appeal will be entertained.
Goldy is entirely correct that the proportional reduction method is a legal issue that will end up being resolved, one way or another, by the state Supreme Court.
And if proportional reduction analysis is used, I would also agree with Goldy that all illegal votes must be taken into account, and not just those illegal votes from precincts or counties that were carried by Gregoire.
And I will also predict that Judge Bridges will issue a less definitive ruling. He will probably allow any scientific statistical analysis into evidence at trial, and then try to frame his ruling after trial to be based on factual findings about the analysis method that he finds most acceptable (very hard to reverse on appeal), as opposed to legal conclusions that a particular statistical method is legally required (something that the Supreme Court would independently use its own view of the law on, and have a 50-50 chance or so of reversing).
At this rate, I will still be in bed at 6:30 a.m. this morning and miss Goldy on KVI :( Nothing unusual — I never listen to Kirby Wilbur anyway. He hasn’t had a kind word to say about me on the air for several years.
And I will also predict that Judge Bridges will issue a less definitive ruling. He will probably allow any scientific statistical analysis into evidence at trial,
If so, it certainly lessens the issue for appeal.
Since hearing an appeal before final judgment is discretionary on the part of an appellate court, I seriously doubt that an interim appeal will be entertained.
However, this is a very unusual case and the Supreme Court is going to much more willing than usual to get involved. They are going to be nice and close to this case.
The more general the ruling, the lower the chance for an interim appeal and review.
On the other hand I do not think either party will pass by making an interim appeal if they are told that they will have to deal with an unfavorable contest standard.
OK I can see this, the libs are already shooting for the last of three counts…….
the third one
Richard @ 9
“And I will also predict that Judge Bridges will issue a less definitive ruling. He will probably allow any scientific statistical analysis into evidence at trial, and then try to frame his ruling after trial to be based on factual findings about the analysis method that he finds most acceptable (very hard to reverse on appeal), as opposed to legal conclusions that a particular statistical method is legally required (something that the Supreme Court would independently use its own view of the law on, and have a 50-50 chance or so of reversing).”
Excellent analysis.
Chuck @ 11, 12
What are you trying to say?
Goldy, good job on the Kirby Wilbur show. I have no idea what he is talking about for the rest of his show this morning, since I tuned in just for you. (Well, I guess I might as well listen to Paul Berendt too, since he is following you.) Kirby did treat you very nice and was seemed pretty “fair and balanced” (in the true meaning of the term) in presenting the issue.
Kirby says that the Democrats’ evidence of illegal votes should be considered by the judge also. If Kirby doesn’t think that the GOP arguments for excluding the Democrats’ evidence make any sense or seem fair, they probably won’t make sense to the judge either. Look for the Democrats to win on that issue.
As for the “cherry picking” of illegal felon votes, it may or may not be fair to accuse the Democrats of doing this also. The GOP seemed to do a very thorough job of finding illegal felon votes in pro-Gregoire and perhaps toss-up precincts. So the only cherries left on the tree for the Democrats to pick might have been pro-Rossi precincts.
Has the Audio been put on the WEB yet?
I was up but didn’t have a chance to listen.
Goldy.
Berendt was caught with his pants down on the Mike Siegel show when Vance said that they did consider the 432 “potential” felons. 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. I thought it was more important to hear Berendt vs. Vance.
Pudster
Richard @15,
Thanks for the kind words, and for your contribution to the threads here… you have certainly been very fair and balanced. I hope you understand that much of my vitriol is for entertainment purposes only; I always try to be fair myself (if not balanced.)
Kirby has always been friendly and noncombative, and so I respond in kind. It might not make for quite as fun radio, but it’s more informative. I do get the sense myself, listening to him, that he’s not expecting Rossi to prevail, and that the complaints about this election have been somewhat overblown due to the extreme closeness and scrutiny.
Though I agree with the comments here that the Judge will likely not rule out some kind of statistical analysis today, a kind of wish he would, so that we can just get this thing over with. I’m tired of election contest coverage, and would rather move on to more constructive issues.
Puddy @17,
Oh… so you are the person who listens to Mike Siegel. (Sorry Mike… just saw the Arbitron ratings.)
Well, I didn’t hear the Mike Siegel Show, but I should point out that Paul frequently stutters. I personally wouldn’t take at face value anything the party chairs say to the media… their job is to spin.
Alan @9,
You’re right, I should have said, a high mark for burden of proof.
Goldy, you answered little old humble me. Wow I am honored. It wasn’t about Berendt’s stuttering, it was about him being caught when saying the vetting wasn’t done on those felons. Vance said yes we did look at them butt we weren’t sure who they really are. Paul had nothing more credible to say.
Pudster
Does anyone know of a website that will have upto date announcements about the contest hearing today in Wenatchee?
I heard the Berendt this morning also. He seemed a bit calmer than last weeks combative and defensive tone. Sure sign of a scared loser.
It really sucks that the Kerry fund raiser only brought in $30-40k for the dems yesterday. Maybe it’s time to bring in the really big gun, Hillary. LOL
Kerry did re-read the New Testament and he seemed ready to barnstorm in defense of Democratic values.
Speaking of values, how many of you horsesass groupies went to Antioch Church yesterday wearing rainbow arm bands?
The failure of HB 1515 seemed to be a big dissappointment to crissy.
looks like another WIN for the GOP. Sorry guys.
Goldy, I will disagree with your first UPDATE. Or at least say you MIGHT be wrong on that point.
The county-level proportional analysis done by DJ would seem to indicate that Gregoire still has a slight lead over Rossi, after the Dems throw in another 432 illegal felon votes in mostly pro-Rossi territory. But it isn’t much of a lead — maybe 20 votes or less (DJ didn’t provide the mean results of his 100,000 Monte Carlo simulations — just that Gregoire prevailed slightly over 70% of the time.)
Precinct-level proportional analysis could be quite different. The illegal King County felon votes found by the Republicans tend to be from precincts much more pro-Gregoire than King County as a whole.
I will reserve judgment on this one, until the Dems illegal felons are further scrutinized, and the whole kit and kaboodle is subjected to precinct-level proportional analysis.
chardonnay, are you drinking before lunch again?
Richard @ 25
“The county-level proportional analysis done by DJ would seem to indicate that Gregoire still has a slight lead over Rossi, after the Dems throw in another 432 illegal felon votes in mostly pro-Rossi territory. But it isn’t much of a lead – maybe 20 votes or less (DJ didn’t provide the mean results of his 100,000 Monte Carlo simulations – just that Gregoire prevailed slightly over 70% of the time.)”
Ok, here is the analysis again, including statistics on the final election total. Your intuition serves you well, Richard :) :
D_win: 74299, R_win: 25701, tie: 0
%D_win: 74.3, %R_wins 25.7
mean: 22.3, standard deviation: 38.2
median: 22, Central 95% probability mass: (-59, 105)
Note: negative scores are wins for Rossi.
“Precinct-level proportional analysis could be quite different. The illegal King County felon votes found by the Republicans tend to be from precincts much more pro-Gregoire than King County as a whole.”
Yes, indeed! But, this really depends on how much of a statistical argument is allowed by Judge Bridges. If such arguments are accepted, he might demand some level of certainty like 95% or even 99% that Rossi appears to win after removing the invalid votes.
With levels of certainty like that, the Republicans would almost certainly have to have the 432+ invalid votes identified by the Democrats thrown out. As I showed here, a Rossi win is only 96.1% certain using only the Republican invalid vote list (but, admittedly done at the county level and missing a few pieces of data). At a very high level of certainty (99.9), a precinct level refinement may not be enough!
BTW: I’ll be happy to do this same analysis at the precinct level, even if it would appear to favor the Republicans. Any takers on tracking down the precinct-level votes and invalid ballot data?
DJ @ 27
How do you get a standard deviation of 38.2? I thought that standard deviation was the square root of the variance, which would make it 1459.24 — a rather high variance for eliminating just over 1600 ballots. Normally, the maximum expected variance on an individual ballot elimination should be .25 — at least if this was a binomial distribution with the formula of p(1-p)
Although a standard deviation of 38.2 seems to make sense with the percentage of victories you have. You get a Z-Score of -0.5759 with a mean of 22.3, which would result in Rossi being expected to win in 28.2% of the simulations.
When I was thinking 20 votes or less, I was also thinking of a smaller standard deviation — more like 20, instead of the 38.2 that you derived.
Do you think Professor Katz is off when he comes up with a 15.89 standard deviation using 1183 ballot eliminations (even when considering that he used a straight D-R binomial analysis)?
alan @ 26 Chardonnay does not eat lunch…has to watch her ‘working weight’, so naturally, she is always drinking before lunch
Richard @ 28
”How do you get a standard deviation of 38.2? I thought that standard deviation was the square root of the variance, which would make it 1459.24 – a rather high variance for eliminating just over 1600 ballots. Normally, the maximum expected variance on an individual ballot elimination should be .25 – at least if this was a binomial distribution with the formula of p(1-p)”
Wow, Richard, this is an excellent point. I had not really taken a close look at the numbers. It took me a couple of hours to get my head fully wrapped around this. Here is what is going on. There are several variances involved here. There is one for the votes to be subtracted from the D-column, one variance for the R-column. Then, when we are done, we compute the final election tally as: final = 129 – (numb_dem – numb_rep).
Statistically, numb_dem and numb_rep are each Binomial random variables (call them D and R, respectively). Therefore, we must compute the variance FINAL as
VAR(FINAL) = VAR[129 – (D – R)]
= VAR(D) – VAR(R) – 2COV(D, R)
(See rule 2 and rule 4 here to see why: http://www.kaspercpa.com/statisticalreview.htm).
When adding independent variables, the covariance term drops out. In this case, however, D and R are highly negatively correlated. This is because most of the time, if we don’t subtract a ballot from the D-column we probably are subtracting it from the R-column (the exception is for the “other” ballots if we do this as a trinomial outcome). The practical outcome is that the variance of final is nearly 4 times the variance of either D or R.
I re-ran the numbers again, but also print out the variances and covariance for 10,000 runs (I am too impatient to wait for 100,000 just now :) ). Note: E(X) means the average of random variable R.
D_win: 7518, R_win: 2386, tie: 96, %D_win: 75.9, %R_wins 24.1,
E(FINAL): 22.3, VAR(final): 1489.38,
E(D): 853.2, VAR(D): 378.25
E(R): 746.6, VAR(R: 379.93
COV(D, R): -365.57
”When I was thinking 20 votes or less, I was also thinking of a smaller standard deviation – more like 20, instead of the 38.2 that you derived.
Indeed. Take the square root and you will see that the standard deviation of D and R are both about 20. It is when we compute FINAL that the variance gets much larger.
Do you think Professor Katz is off when he comes up with a 15.89 standard deviation using 1183 ballot eliminations (even when considering that he used a straight D-R binomial analysis)?
Yes, I believe you have uncovered a fundamental flaw in Professor Katz’s analysis (Sorry, Richard, that is what you get for being smart and curious!). Since he is using a straight binomial, his variance is underestimated by a factor of 4. If the Democrats put a good statistician on the stand, they may be able to show significant flaws in Professor Katz’s analysis. In fairness to Professor Katz, however, I have not examined his longer report—perhaps he does deal with this issue. But, if not, this means, that estimates of certainty are way overestimated in Professor Katz’s analysis. In defense of Professor Katz, I made the same mistake in my “quick & dirty” analysis last month. Of course, I wasn’t being paid for my efforts, either. :)
BTW: to get a better estimate of the standard deviation from his analysis, square the standard deviation, multiply it by 4, take the square root.
This is a beautiful example of the power of doing this kind of thing by simulation! As long as you code it correctly, it doesn’t lie.
Richard @ 28
The formatting on the last post sucked, and I noticed a couple of typos in equations, so here is a fixed version of my last post (yes, Goldy, I must say, a preview function would be sweet).
”How do you get a standard deviation of 38.2? I thought that standard deviation was the square root of the variance, which would make it 1459.24 – a rather high variance for eliminating just over 1600 ballots. Normally, the maximum expected variance on an individual ballot elimination should be .25 – at least if this was a binomial distribution with the formula of p(1-p)”
Wow, Richard, this is an excellent point. I had not really taken a close look at the numbers. It took me a couple of hours to get my head fully wrapped around this. Here is what is going on. There are several variances involved here. There is one for the votes to be subtracted from the D-column, one variance for the R-column. Then, when we are done, we compute the final election tally as: final = 129 – (numb_dem – numb_rep).
Statistically, numb_dem and numb_rep are each Binomial random variables (call them D and R, respectively). Therefore, we must compute the variance FINAL as
VAR(FINAL) = VAR[129 – (D – R)]
= VAR(D) + VAR(R) – 2COV(D, R)
(See rule 2 and rule 4 here to see why: http://www.kaspercpa.com/statisticalreview.htm).
When adding independent variables, the covariance term drops out. In this case, however, D and R are highly negatively correlated. This is because most of the time, if we don’t subtract a ballot from the D-column we probably are subtracting it from the R-column (the exception is for the “other” ballots if we do this as a trinomial outcome). The practical outcome is that the variance of final is nearly 4 times the variance of either D or R.
I re-ran the numbers again, but also print out the variances and covariance for 10,000 runs (I am too impatient to wait for 100,000 just now :) ). Note: E(X) means the average of random variable X.
D_win: 7518, R_win: 2386, tie: 96, %D_win: 75.9, %R_wins 24.1,
E(FINAL): 22.3, VAR(final): 1489.38,
E(D): 853.2, VAR(D): 378.25
E(R): 746.6, VAR(R: 379.93
COV(D, R): -365.57
”When I was thinking 20 votes or less, I was also thinking of a smaller standard deviation – more like 20, instead of the 38.2 that you derived.
Indeed. Take the square root and you will see that the standard deviation of D and R are both about 20. It is when we compute FINAL that the variance gets much larger.
”Do you think Professor Katz is off when he comes up with a 15.89 standard deviation using 1183 ballot eliminations (even when considering that he used a straight D-R binomial analysis)?”
Yes, I believe you have uncovered a fundamental flaw in Professor Katz’s analysis (Sorry, Richard, that is what you get for being smart and curious!). Since he is using a straight binomial, his variance is underestimated by a factor of 4. If the Democrats put a good statistician on the stand, they may be able to show significant flaws in Professor Katz’s analysis. In fairness to Professor Katz, however, I have not examined his longer report—perhaps he does deal with this issue. But, if not, this means, that estimates of certainty are way overestimated in Professor Katz’s analysis. In defense of Professor Katz, I made the same mistake in my “quick & dirty” analysis last month. Of course, I wasn’t being paid for my efforts, either. :)
BTW: to get a better estimate of the standard deviation from his analysis, square the standard deviation, multiply it by 4, take the square root.
This is a beautiful example of the power of doing this kind of thing by simulation! As long as you code it correctly, it doesn’t lie.
DJ @ 30 & 31
Holy shit! I was going to blame Professor Katz, but I am the one who has screwed up this analysis. Katz has 15.89 as the standard deviation on both the change in D votes and on the change in R votes. Both have to be exactly the same in a pure binomial distribution, since p(1-p) always equals (1-p)p.
However, the standard deviation of (D-R) in a purely binomial distribution will always be twice the standard deviation of either D or R. And Katz does recognize this in constructing his 95% confidence interval — which ranges from -70.98 to -195.60 for Gregoire’s “victory” margin. (I am being facetious, since a negative margin would be a Gregoire loss.) This basically represents 1.96 standard deviations either side of the mean, with these standard deviations being about 31.78, instead of 15.89.
On the other hand, Katz did screw up by using a binomial distribution, instead of a trinomial distribution, and allocating all of the illegal ballots to either D or R. Overstating the effect by 14 votes might end up making the difference between winning or losing, and clearly effects the level of statistical certainty.
I had figure that Katz’s analysis (using precinct level and only the GOP illegal ballots) would require 7.49 standard deviations from the mean (even after subtracting 14 votes for the error of ignoring Bennett/undervotes) in order to have Gregoire victory outcomes. Those chances would have been over a trillion to one.
However, since the standard deviation of (D-R) is more like 31.78, instead of 15.89 (actually, it would be a little different in the trinomial analysis, but probably not by much), Gregoire victory outcomes would only be 3.74 standard deviations from the mean.
Such outcomes can be achieved, on average, one out of every 10,868 attempts. So, if you had the precinct level data on the 1,183 GOP illegal ballots, and ran the Monte Carlo simulation, you would probably have a Gregoire victory maybe 10 times out of every 100,000 runs.
That also explains why about 4% to 5% of your county-level analyses on Katz’s data showed Gregoire wins, when I was expecting maybe 1 out of 10,000. I was wrongly assuming the standard deviation to be one-half of its actual value, and computing the probabilities based on twice the actual Z-score.
Again, Katz does use the correct standard deviation for (D-R) in computing his confidence interval. But he wrongly ignores Bennett/undervotes, thereby overstating the effect on net Gregoire vote reduction. I am sure Katz can easily correct this flaw in his analysis with relatively little effort on his part.
Richard @ 32,
I looked more closely at his second document, and I agree. He did compute a proper confidence interval. I agree with your last paragraph, although I point out a few other problems at the end of my comments on a previous thread ( my-comments-popup.php?p=622&c=1#comment-27699 ).
I’d love to empirically test your statistical argument that Dems would win 1 of 11,000 elections using the original GOP list in a precinct-level analysis. I have a hunch the simulation result would be a little less conservative than would the statistical approximation.
RDC,
Due to some events in my neck of the woods, I am less likely to post on the world according to marks. I have enjoyed the opinion exchange. My (final?) observations are very simple:
1. The Democrats are doing great, so long as there are more people like you in their fold.
2. The Republicans are in trouble if number 1 is true.
I will be checking in on occasion, and if I have time I will make comments on the topic du jour. Cheers!
Commenting on number 1 would require that I be immodest or braggardly, neither of which fits well. My still lit hope is that your number 2 will come true.
I’ve enjoyed the exchange. Good luck on your endeavors, wherever they may take you and whatever they may be.