Friday, February 25, 2005

"Carla's Apparent Paradox"...

Carla over at Preemptive Karma, alludes to a phenomenon well known to decision theorists, detection experts, ... IOW, guys like me...

Sharkansky cites a Seattle PI story quoting Democratic lawyer Jenny Durkin as saying the GOP will have to prove that the felons voted for Gregiore in sufficient enough numbers as to push the election to her, away from Rossi. Rossi's spokesperson Mary Lane, appears flustered in the piece, in my opinion. But Sharkansky of course spins it by namecalling Durkin and ignorning any actual substance.

Also cited in the Sound Politics piece is book review of John Fund's tome, Stealing Elections: How Vote Fraud Threatens Democracy. The review was written by Paul Froehlich, GOP Congressman from Illinois' 56th Congressional District. The review points out that 68% of the felons who voted in the 2000 election in Florida were registered as Democrats.

For Sharkansky, this is proof that Rossi won because felons registered as Democratic in Florida means they're voting Democratic.

Pssst....Stephan, which Presidential candidate won in Florida in 2000 despite Democratic registrations vastly outnumbering Republican?...

OK, maybe she didn't allude to it exactly, but here it is, and here's why it's important to realize that this is going to be the argument that puts the nail inthe coffin of Rossi's hopes:

1. In a decsion between any two hypotheses H0 and H1, if the data that is being used is of a statistical nature, and the statistics affect the decision, then there is always a probablity of error; that is pe =P(H0 decided but H1 was actually true) + P(H1 decided but H0 was actually true)

2. But, my conservative neighbors argue, we're not talking about error here! We're talkin' fraaauuuuud!

3. Well, yes and no: you're also speaking of the failure to detect fraud as well as the failure to admit a voter to vote who was rightfully entitled to vote.

Now that last item is analogous to the old English principle of jurisprudence about letting 100 guilty people go free to keep an innocent person from being imprisioned.

Traditionally, we Americans have never had zero tolerance for all guilty people getting away with something because that would mean that innocent people would likely be wrongly denied their rights.

In any election, there needs to be a tradeoff between making sure people who are entitled to vote can vote, and making sure that people who aren't entitled to vote don't or can't vote. Like any hypothesis test, setting the threshold in one area to minimize one kind of error tends to determine the error rate for the other kind of error.

Which brings us back to Gregoire and Rossi...

Note that the above is, as I said, based on a bias assuming people have rights to vote... not, I repeat not the inherent vote outcome of any particular election.

So, Carla's Apparent Paradox can now be stated:

1. Representative democratic systems, when they have elections, have error rates, and these error rates are based on ensuring that the rights of people to vote are not violated below a certain level.

2. Representative democratic systems, when they have elections, award the election to the person who had the most recorded votes, based on the premise that the one who had the most votes was who the most number of people wanted to get elected.

3. But- because of the inherent error rates of the system: Sometimes the right of people to vote, being more important than absolutely minimizing the number of wrong votes to zero means that sometimes the person who gets the most recorded votes is not the person who the most number of people wanted to get elected.

It's only an apparent paradox, though, because in a democratic system, we accept this tradeoff if the law is written - as it is in the case of Washington State- in such a way that the law is completely agnostic to vote numbers recorded by each candidate except where one candidate's percentage is equal to or greater than anothers.

Hmm... maybe I should apply to be an expert witness on this thing... not as a lawyer of course, but as an expert in statistical hypothesis testing...

Note however, that when it can be proved that the outcome would have been changed. , then the courts will indeed void the results of the election, and either seat the other person who really won or call for a new election.

That is the standard in Washington State. If it is "indeterminate," then it is alas for Rossi and the righties, Gregoire's.



What sort of evidence the Republicans will have to offer to win their case on the basis of illegal votes is something that Chelan County Superior Court Judge John Bridges, who is hearing the suit in Wenatchee, has yet to rule on.

Lawyers for the two political parties disagree about what that requires.

[Mary Lane, Rossi's spokeswoman] said the Republicans, despite Bridges' comments, still contend that the number of illegal votes should, on its face, require a new election because it makes it impossible to determine who really won.

"We believe that anything over 129 (illegal votes) should be sufficient to say that the judge can decide to go on a proportionality argument," Lane added, meaning that the Republicans should be allowed to argue that illegal votes from King County, for example, presumably favored Gregoire by the same proportion as lawful King County votes did.

Since 884 of the 1,108 felons voted illegally in King County, which Gregoire carried with 58 percent of the votes to Rossi's 40 percent, "that is not good news for Christine Gregoire," Lane said.

If Republicans try to make their case that way, [ Jenny Durkan, one of the Democratic Party's lawyers] argued, the Democrats can raise questions about "the demographics of felons in Washington state."...

The Rossi campaign said in an e-mail that the GOP was supplying the Democrats with the numbers of illegal voters later yesterday, as required by a court-imposed deadline, and would disclose the names of those voters today....

"If it's anything like the other lists they've produced, it's about 10 percent correct," Durkan said.

They actually have an even better argument: Chebyshev's Inequality. Take a sample of 1300 votes. What are the odds that the difference deviates from the general population? Well, actually there's a reasonable chance that any sample of 1300 could differ enough from such that a margin of, oh, say, 130 votes- either way is possible.

While of course the Repubs want to focus on King County alone, what about the rest of WA?

Poke around enough, and the Type I error (wrong detection) for either candidate becomes large enough...

No comments: