Friday, November 26, 2004

Musings on the election process - WA's next governor

After the initial count, Rossi won by 261 out of 2.8 million.

After the machine recount, finished Wednesday, 11/24/2004, that lead was cut to 42. In one precinct, there was a box of ballots that had been left uncounted on election night, but counted during the recount. [That was in a precinct that went for Rossi, I

There is going to be a hand recount, which is good, otherwise I would complain that I'm deprived of the fun of watching the turmoil.

One thing we're missing is the statistical view - that the counting process is not a counting process but rather a sampling process with an inherent error rate. From that perspective, the result needs to be stated in statistical terms. E.g., "after today's recount result, there is a probability of 0.5000103 (or whatever the real number is, as a function of error rates) that Rossi got the most votes."

If we could bring that change of concept to voting, how could the population handle the concept? What does probability of victory rather than victory do to the social contract?

I keep thinking of the state (Nevada?) where in case of tie, the candidates draw cards and the high card wins. What should we do in case of a statistical tie? Can successful politicians talk about a mandate that came from the card drawing? (perhaps a mandate from God, who must have rigged the deck)


I posted that to a small list and got the reply: "Great questions, Carl. For a population that seems to recoil in the face of statistics, I would fear the worst if we moved in that direction."


To which I replied:

Even though I studied probability at MIT as a math major, it still took me quite a while to internalize that there is a reality (that is, someone really did get the most votes), but our knowledge of that reality is faulty and always will be. This took me perhaps a day of reading (of a statistics reference, several years ago) to get firmly ingrained.

So, how could we describe this to the man in the street?

As I was thinking about that, a Disney-like cartoon flashed in front of me: There's a secret ballot - people going into a curtained booth, bringing out a ballot and dropping it in a box. The eye point follows the ballot. [This is looking more like Schoolhouse Rock, as I watch it now in my head. :-) ]

The ballots get mixed up in the box. They're taken out and counted - maybe by optical scanner - but the scanner rejects some of the ballots as unreadable. It counts some of them incorrectly, because of misalignment. It might adjust the count, just a little bit, because the programmer of the counting machine wanted to affect the election if it was close enough. The machine gives us a count - totally up to something less than the total number of ballots - and with some error rate.

This is a secret ballot with a secret result. We get to ask questions about it, but the answers come back with error. There is a correct answer - only we don't know what it is - and if the election is large enough, we may never know. In a hand recount, if there are enough counters, there will be errors. [Of course, this is a computer system designer's dream case. We know how to design systems in which the error rate approaches 0. Throw enough redundancy at the problem and we can drive down the error rate - at a cost. In the old days, there were punch card operators who would send their decks to a neighboring operator who punched info from the same sheets onto the same cards - not punching, but verifying the punch. It cost money, but it reduced error rates.]

The process is different in each case, but vote counting is a statistical sampling process just like polls, exit polls, the US census, .... They have different error rates but they have error rates. Repeat the measurement and it comes out differently, especially when you do things a different way - e.g., a machine count, a machine recount, a hand recount. This error rate usually has nothing to do with fraud in counting - but that can happen too. It can happen when there are people doing the counting, but it's hard there because there are people of both parties watching. It's a whole lot easier if the fraudster can operate without being observed - e.g., when s/he is the programmer of the tabulating system. Remember the programmer who funneled fractions of a cent from each interest calculation off to his own account? It was ages before he was caught. The nice thing about fraud here is that if it's in the counting process, the measures we take to combat error should also combat fraud.

I think we can get better processes but I'll be surprised if we ever get flawless ones. So, if there's always error, there can always be an election that is undecidable - in which each recount produces a different result. There will always be a probability rather than a certainty that some candidate actually got the most votes.

How will we as a society handle that? Draw cards?


Blogger Pat said...

Still an optimist, eh Carl?
Given how numerically illiterate the US population is,
I don't hold much hope for explaining anything
about sampling to John Q Public.

I didn't even hear a rational comment during
the election about why we have an electoral college.
Or that the Founding Fathers fought for months
before deciding not to have popular elections.

8:26 PM  

Post a Comment

<< Home