May 19, 2005
Proportional Deduction

One of the least understood aspects of the election contest (at least misunderstood by the "real" journalists who are supposed to be explaining it to the public) is the "proportional deduction" method for attributing illegal votes. It's very simple. For every illegal vote in a precinct of 100 voters that gave 60 votes to candidate A and 40 to candidate B, you deduct 0.6 votes from A's total and 0.4 votes from B's total. It's not designed to predict how the illegal voter voted, and I'll be the first to admit that it's far from perfect, but it's the best you can do to correct for errors in light of the only legally admissible evidence that is available. It's also a standard procedure in many other jurisdictions.

I'm prompted to discuss this by George Howland's incompetent, factually incorrect article in the current Seattle Weakly. I'll dismantle more of his false assertions later, but first this one:

the Republicans are advocating a method of proportional deduction prepared by a respected statistician and political scientist, Jonathan Katz of the California Institute of Technology ... But three social scientists interviewed by Seattle Weekly, including one of the other leading experts on "ecological inference," say Katz is trying something very different in this case than has ever been tried before.
Nonsense. Had Howland actually read and understood Katz' report (a very simple 5-page paper, posted online at the Sec of State website) and spent a few minutes on google, he would have seen how standard his methodology is.

Here are some of the other places where the identical method of proportional deduction is considered standard (just search for proportional deduction on google and you will find them too) --

1. The United States House of Representatives. In this report on "DISMISSING THE ELECTION CONTEST AGAINST CHARLIE ROSE" (1996)

If a contestant is eventually successful in establishing convincing evidence of irregularities or fraud, the Committee could order remedies, including proportional deduction of improper ballots,\15\
which leads to Footnote 15:
The House's precedents allow for deletion of improper ballots by proportional deduction. This `general rule . . . in the House for deduction of illegal votes where it is impossible to determine for which candidate they were counted' requires reducing the total vote count in affected precincts in proportion to the percentage of votes received by each candidate in each precinct to eliminate the improper ballots from the vote count.
2. Arizona. In this AZ Court of Appeals ruling on a contested election in the City of Tempe
The City contends that when illegal votes are cast, the trial court must first apply the “proportional deduction rule” and should consider setting aside the election results only if the deductions would change the results of the election. See generally Huggins v. Superior Court, 163 Ariz. 348, 352, 788 P.2d 81, 85 (1990) (explaining the application of the proportional deduction rule).
I can't find the entire Huggins ruling online, but it is cited elsewhere, such as in

3. Florida: In this brief filed by Democrats in a FL Supreme Court Case in the 2000 election:

The same statistical evidence showed that 91% of the affected Martin County absentee ballots (all cast by registered Republicans) were probably cast for Governor Bush, and accordingly, that 91% of the ballots be deducted from the total Martin County votes for Governor Bush. (T-384). While this remedy suffers from being imprecise, it remains far preferable “to let[ting] illegal votes count.” Huggins v. Superior Court, County of Navajo, 788 P.2d 81, 85-86 (Ariz. 1990).
4. Illinois This IL Appeals Court ruling on the contested 2003 mayoral election in Calumet City cites an earlier "Frese" decision (can't find the original):
The Frese court then affirmed the trial court's apportionment of these votes between the candidates because the number of invalid votes could be ascertained, but it could not be determined who the votes were cast for and, as such, apportionment was the proper remedy. Frese, 76 Ill. App. 3d at 76.
Ironically, the later court ruled against proportional deduction in favor of total deduction, but Judge Bridges has already ruled that he won't allow that remedy.

Posted by Stefan Sharkansky at May 19, 2005 05:00 PM | Email This
Comments
1. A lazy piece of work by Howland, whom I'd previously thought of as a fair and diligent reporter.

Posted by: jsa on May 19, 2005 05:09 PM
2. Paul Berendt was just on KVI. He said there has never been a court ruling in favor of proportional deductions. What a moron.

Posted by: Mark Beyer on May 19, 2005 05:29 PM
3. I think I'd prefer a rewrite of the elections law to make an election invalid if the 'margin of error' exceeds the 'margin of victory', where 'margin of error' is the _absolute_value_ of all substantiated claims. (If 50 felons vote in a 100% Rossi county, and 50 felons vote in a 100% Gregoire county, that contributes to the 'margin of error' in the amount of _100_ votes.)

Posted by: Al on May 19, 2005 05:38 PM
4. In all honesty, I think the system is crap too, legal or not. It's much better then the idiotic solution the Dems wanted, but I still don't like it. A vote is a sovergn thing, a personal thing. It shouldn't be a statistic, just like the census shouldn't be either.

It SHOULD be, if there is more screwed up votes then the margin of victory, revote. It's that simple. It's how they did it in North Carolina, it's how they should do it here.

Posted by: Cliff on May 19, 2005 05:52 PM
5. Katz is well-published in statistical analysis of elections so there's some weight behind his conclusions. However, the critics cited by Howland have a point, there really is a problem with inferring cell frequencies because of the "ecological inference" fallacy. Katz certainly knows about the problem (he has co-authored papers with Gary King who has a book on the it) so it would be interesting to hear from him on this. Statisticians familiar with ecological inference would dispute Table 2 in Katz's article. There's no way that his conclusion is the final word on the effect of felon voting on the election outcome.

By the way, it is my opinion that there are few felons who would vote for a former attorney general so I think this whole discussion is bogus.

Posted by: Ron Schoenberg on May 19, 2005 05:54 PM
6. Al:

Wouldn't your suggestion cause you some heartburn if Rossi was ahead and the illegal ballots all came from Gregoire counties? Even though you could convincingly argue the illegal votes did not change the outcome, your method would still require invalidation.

I still think the major problem with the GOP's proportional method is not the method but the data they apply it to. They obviously cherry-picked illegal votes primarily from Gregoire precincts. No matter how appropriate the method is, the result is not valid if the data you apply it to is not reliable. The proportional method is only reliable for assigning illegal votes if the illegal votes are representative of the whole state. If you take only illegal votes from King County for instance, the resulting adjustment is not sufficiently reliable for the Court to find it "appears" Rossi got more votes after removing illegal ones.

Posted by: Wayne on May 19, 2005 05:54 PM
7. Ron
Without getting way OT. Who would you vote for if you were a felon? Soft on crime with reduced penalties for felons or hard on crime with strict penalties.

Posted by: Mark Beyer on May 19, 2005 06:01 PM
8. While proportional deduction is the middle ground. Can anyone tell me how accurate proportional analysis/deduction is? Is it 100% accurate? I don't think so, maybe between 5-80% depending on the population and the parameters. The more accurate the parameters the closer the accuracy of the analysis. Unfortunately, the accuracy could always be 0% if the guess is wrong. Statistical analysis is nothing more then an educated guess. What if the guess is wrong?

Posted by: Mark Beyer on May 19, 2005 06:09 PM
9. Why do you presume the main concern of felons is who is tougher on crime? I suspect, to the extent that felons are interested in voting, most of them vote based on the same considerations affecting anyone else in their position. White male felons from Black Diamond are more likely to vote like other white males from Black Diamond and minority felons from Seattle are more likely to vote like other minority voters from Seattle.

I do think if the voting felons are predominantly males, using proportional analysis from a group that is gender balanced introduces a certain amount of uncertainty into the results.

Posted by: Wayne on May 19, 2005 06:12 PM
10. Mark,

I guess you're saying that a felon would vote against Republicans because they're hard on crime versus Gregoire who put them in jail. I think that Republican vs Democrat on crime penalties is a little abstract compared to the very concrete and personal impact of Gregoire throwing them in jail.

Posted by: Ron Schoenberg on May 19, 2005 06:16 PM
11. Do you buy a car based on fuel economy, safety, looks, performance, or emisions? Businesses and politicians play to their customers and constituents. Victims want hard time for criminals, criminals want reduced sentences. That's a pretty easy answer. Want to make it easier, who is courting the felon vote? The democrats are proud to show they are courting the criminal vote, why? I like the republicans advertising campaign, it targets helping the victims who are in the majority versus the criminals who are in the minority. Not only is it the right thing to do, its more effective in gaining the votes the candidate is looking for.

And, how many criminals did CG actually put in jail personally? Under a lower standard, how many was she instrumental in putting in jail in a way it would be held against her?

Posted by: Mark Beyer on May 19, 2005 06:26 PM
12. Wayne
In a small population the support overwhelmingly comes from what benefits the population. In the military soldiers support what supports the soldiers. Why do soldiers vote so consistently? because they talk about what affects them constantly. In prison there is consistent talking about this law and that law, jailhouse lawyers are fact, not fiction. When time weighs heavy and long there is ample time to talk about what benefits you the most whether its a soldier on guard duty or a criminal in the pen. Soldiers could care less about Stephanopolous but they didn't liek Clinton. Criminals could care less about Gregoire, they would dispise Locke though. What they would support would be what gets them out now and what would keep them out later. Funny, the republicans are againt getting or keeping them out without repaying society fully.

Posted by: Mark Beyer on May 19, 2005 06:34 PM
13. 1. As Attorney General, Gregoire was incompetent and under her lack of leadership numerous felons were aquitted who would have been convicted. Of course the felons would vote for her.
2. The systematic fraud in King County was conducted well enough to cover up the direct evidence so statistical inferrence is the only way to show that Gregoire probably did not win.
3. It can easily be proved that due to the fraudulent actions in King County there is no way to prove who got the most legitimate votes.

Posted by: Gary on May 19, 2005 06:34 PM
14. Berendt is still in denial about proportional deduction. Berendt foolishly sat back with his comrades and foolishly contended that the R's MUST identify who each illegal vote was for....even though they were already comingled with legal votes. They chanted that mantra over & over again....while they should have been identifying illegal votes. Then, all of a sudden, the Dems scream HOLY SH**!!! we better get going on this too. But they found out how complicated it is to Identify and validate these felons. Their list is terrible. They are in trouble. If they fall short and lose this...it's all Berendt's fault and he is going, going GONE!!

Wayne has repeatedly put forth the argument that the R's "cherry-picked" therefore invalidated any proportional analysis argument. That's a reasonable argument...but will be shot down because the R's will contend the D's have done this too...so the Judge will rule on what evidence is presented. PERIOD! Wayne's argument is also weakened because cherry-picking is precisely what the Dems did when Bawling Pauly sent out the LEFTIST BRIGADE to round up provisional ballot affadavits. Remember that Wayne??? So what is different about this?? The Dems have made an effort to cover Rossi precindts. If they did a lousy, inadequate job....who's fault is that????

Wayne--you are in big trouble....and I think you know it. Do you think this is a slamdunk Wayne??
I think Rossi is currently a 2-1 favorite. What say you?

Posted by: Mr. Cynical on May 19, 2005 06:42 PM
15. Ron - AGs don't put people in jail. They represent the state in various court and other legal matters. Also, an AG can give an opinion if a legislator requests it.

Who puts criminals in jail? County prosecutors, county district attorneys. Don't you think criminals know this? You apparently are not a criminal!

Wayne - you said: "I suspect, to the extent that felons are interested in voting, most of them vote based on the same considerations affecting anyone else in their position. White male felons from Black Diamond are more likely to vote like other white males from Black Diamond and minority felons from Seattle are more likely to vote like other minority voters from Seattle."

That is exactly why proportional analysis is a valid method of determining a probable vote distribution among felon voters on a precinct by precinct basis.

As other posters on this discussion group have noted on the topic of "cherry picking" - the Dems could easily have examined voter rolls in other counties looking for the rare Republican felons voting. Did they? I don't know.

Mark B. - statistical analysis is a well-established, respected academic discipline. The judge will know that after the academic expert's credentials have been introduced at trial.

Posted by: Mac on May 19, 2005 07:10 PM
16. Wayne,
I'd rather have an extra election or two thrown out _immediately_ than any elections going to court.

I'd also put provisions in any election to put pressure on the individual counties to clean up their own act. Like:
What if the county with the highest per capita 'error rate' has to pay for the revote? That would be KC. And a 10 million dollar revote would cause a leetle bit more interest in figuring out how to count.

Posted by: Al on May 19, 2005 07:22 PM
17. In regard to whom the felons voted for: Did the felons have any idea what CG prior occupation was? Did they care? Did they know what the AG's duties were?

Not many people would know, if you polled the felons, it is not a priority in their lives.

Posted by: Chris on May 19, 2005 08:51 PM
18. "statistical analysis is a well-established, respected academic discipline. The judge will know that after the academic expert's credentials have been introduced at trial." - Mac

That may well be, but Katz's findings can be blown out of the water because he didn't take the "ecological" problem into account.

Posted by: Ron Schoenberg on May 19, 2005 09:43 PM
19. Katz may not have taken the ecological fallacy into account (I would disagree--his initial footnote suggests he modified his formulas to account for at least a portion of the uncertainty the ecological effect would add to a statistical sample from a known population), but even if he had ignored it completely this doesn't blow his analysis "out of the water." The ecological fallacy describes a vulnerability to misrepresentation but does not prove misrepresentation, which would be required for something to be "blown out of the water." Instead, it suggests possible issues, but rather than being some iron-clad statistical law, the fallacy only says one thing: in aggregating a statistical sample and then trying to disaggregate it again, you *may* hide variances that would otherwise appear if you were going in the other direction with a smaller sample size.

Posted by: Marc on May 19, 2005 11:04 PM
20. Mr. C:

The so-called cherry-picking by democrats during the vote count has nothing to do with the scientific validity of proportional analysis when dealing with illegal votes. The GOP's analysis has to stand on its own as a reliable method of determining who got the most votes. All the democrats have to do is show the GOP overlooked a significant number of illegal votes and the votes overlooked were primarily in GOP precincts. That means the GOP's data is not representative and the GOP's proportional analysis results are not reliable. You seem to feel the Court will just add in the illegal votes found by democrats and the advantage goes to whichever party was able to find the most illegal votes. But this is an election contest, not a basketball game.

I predict Rossi is a 3 to 1 underdog. He has even less chance on appeal.

Posted by: Wayne on May 19, 2005 11:13 PM
21. Can someone tell me why they can't use the proportional deduction arguement to argue that the votes should be proportioned based on studies of how felons vote (mostly democrat) instead of basing the proportionality for a particular candidate based ont how that district votes. Did that make any sense to anyone out there?

Posted by: mavrik34 on May 19, 2005 11:18 PM
22. Howland who?

I must be out of the loop.....(and I'd like to stay that way.....heh)

Posted by: Deborah on May 20, 2005 12:14 AM
23. Marc:

The problem is that these statistical issues are likely to be ill-understood by the judge and the ecological issue could be used to cast doubt on Katz's results leaving them tainted. And I think any taint no matter how small could have a critical effect on a judge who won't have much understanding of the statistical issues.

In any event, Sharkansky's claim about Howland's article being "incompetent, factually incorrect" is overstated. I'd be interested in knowing who those three social scientists are that Howland interviewed. UW is the equal to any institution in the world with regard to statisticians in the social sciences.

Posted by: Ron Schoenberg on May 20, 2005 07:43 AM
24. Wayne
You had better hope Rossi has better chances then that. Otherwise you will have to wait until the voting rosters are cleaned up and the procedures fixed to "Know" beyond a shadow of any doubt that your vote was counted and not disenfranchised or altered in any way. Right now, I have no faith that my vote went to the candidate I selected. What you should do is list the benefits of having Rossi win versus the benfits of having Gregoire continue the mess she is pushing not forgeting the junk she has pushed so far. Also, make a list of facts, figures, and errors for each side of the election contest. Very little proof on the side of the Democrats. They can't even use much of the election laws as our debates here have shown. The democrat side is on very very weak ground.

Ron
Statistics: a means of mathematically representing the same information three different ways by three different people, trying to say the same thing.

Posted by: Mark Beyer on May 20, 2005 08:04 AM
25. Ron
Forgot to mention, none of the three agree with their results.

Posted by: Mark Beyer on May 20, 2005 10:58 AM
Post a comment
Name:


Email Address:


URL:


Comments:


Remember info?