Saturday, January 15, 2005

My Letter to the Editor on the WA Vote

This wasn't published, AFAIK, but I still like it, so here goes. [Written 12/31/04]

(thread link)

=======================

Dino Rossi's claim that the election was a total mess is false. The election was a statistical tie, not because of any flaws in the system but because the voters happened to split nearly evenly. So, an overlooked ballot here and a machine mis-read there yields enough of a difference to change the winner.

Our system doesn't understand what to do with a statistical tie. If it were an exact tie, some states allow for drawing cards or flipping a coin, but if it's a statistical tie, we have no procedure other than the one we just went through.

We might want to rethink those procedures for the future. The answer isn't to keep voting until the candidate with the most strident voice finally wins or until people get tired of voting.

The answer is to learn to detect a statistical tie and devise methods for breaking that tie. We might start by asking a statistician to: 1) measure the probability of error in each counting process and 2) use that measurement to compute the probability that the one with the largest count actually had the most votes. We might then set some threshold for that probability - say 75% - below which we call it a tie and draw cards (for example).

9 Comments:

Blogger Bob said...

Hear, hear!

You're exactly right, Carl. This is what happened in Florida in 2000 also. People want to believe that "the other guy stole the election" or "my guy won fair and square".

They don't want to hear that what really happened was that there was no result, and the Supreme Court had to intervene to make a decision - because people generally don't like complexity or ambiguity.

Your proposal is right too; we should have a tiebreaker mechanism which doesn't threaten a constitutional crisis every time this happens.

1:31 PM  
Blogger cme said...

Now that we have such strong agreement - what do we do to make it come to pass?

3:53 PM  
Blogger cme said...

Meanwhile, I take issue with Bob's claim that the US Supreme Court had to intervene in Florida in 2000. Here in WA, we went through a recount - by normal procedures - and that decided it. :-)

What I really object to is two things, in the 2000 vote:

1) that a conservative court decided it by a 5-4 vote (making this at least as unacceptable as any recount result - easily contestible)

2) that it was allowed to be decided as if the voters had spoken. We needed a coin toss or the drawing of cards, so that the victor had no way to claim a voter mandate for his policies. In case of tie, the victor is responsible for carrying out compromise policies.

5:10 PM  
Blogger cme said...

This letter to the editor showed up today 1/17/05 in the Seattle PI:

==============================

Florida's 2000 votes could've been legitimized

Here is the difference between the 2000 presidential election and the 2004 Washington gubernatorial election: A recount that could have established the legitimacy of Florida's votes did not occur in 2000.

The difference is huge.

Nick Cullen
Seattle

12:44 AM  
Blogger Bob said...

Actually I think we still agree - mostly - about Florida. I agree that the recount procedures in Florida were a mess, and that what happened in fact would not have been able to determine the outcome if the outcome had been determinate.

However, I think the fact of the matter is that even if a proper recount had occurred, the result in Florida would have been found to be a statistical tie, and I believe that the investigations by regional and national newspapers after the fact bear this out.

Regarding what we can do to introduce the notion of statistical indistinguishability into elections... that's a hard question. Whenever I think about presenting this to the public, I come up against an impenetrable barrier of perception and political self-interest.

Recall that the Census bureau wasn't able to introduce statistical sampling in a case in which it clearly would have made the count more accurate; I suspect that any proposal to introduce a "too close to call" result and a random resolution processs would run up against all sorts of arguments to the effect that unless there is an actual numerical tie, resolving elections by any mechanism other than counting amounts to thwarting the will of the people. Look at the Kerry campaign's rhetoric this election cycle: "every vote should count, and every vote should be counted".

You got any ideas?

10:46 AM  
Blogger Bob said...

The two things I do disagree with you about in your analysis of Florida are:

1. The objection based on the ideological perspective of the court: this is simply poorly founded. Every court has an ideological orientation, but must nevertheless decide cases. It's simply pointless to object to the resolution of a particular issue on the basis of the judges' perspective - one must make an argument based on the reasoning of the opinion.

2. The suggestion that winners be required to pursue compromise policies is unecessary and, in my view, inadmissible. Once a winner is declared, the winner is accountable to his own conscience and to the voters. He or she must be free to make decisions the same way as other public servants - without added restrictions. If the voters are unhappy, they may vote the winner out, either in the normal course of the election cycle or via recall.

10:50 AM  
Blogger cme said...

In reply to your first comment, as I was reading it, I had visions of a long TV program - maybe an infomercial style public service ad - watching human vote counters go through recount after recount, while observers catch them making mistakes - until finally the recounters go postal. That might get the idea across to the non-mathematical public.

Of course, we could take this as a challenge to ourselves in the technical community: how can we run an election with 1 billion voters with an expected error of less than 0.01 vote? I've done fault tolerant system design. I assume you've looked at it too, at some point in your life. We kept claiming that we could achieve an arbitrarily small error rate, with enough redundancy. Of course, this would mean that the voter hirself would have to do redundant things (vote multiple times, probably on different design ballots) to verify correctness of the human-input part of the process. I can imagine how to do this with pure electronic voting (dealing with audit, fraud, etc. as well, of course) - but I'd need to think a little to apply this to absentee ballots.

4:53 PM  
Blogger cme said...

As to your last comment, of course you're right on both points.

That doesn't mean I have to like it when some TX village idiot ends up with the reins of power in his demented grasp.

4:54 PM  
Blogger cme said...

You're right. If there were to be another vote without the same tie, you have to change the rules - and H1-Bs might do it - unless they're tied, too. I wouldn't put it past the fates to make that happen.

6:25 AM  

Post a Comment

<< Home