Tuesday, November 6, 2012

The Person With the Most to Lose in the Election – NYT Mathematical Election Analyst Nate Silver

A Romney Win, Nate’s Next Gig is as a Junior Actuary at Metropolitan Life

In 2008 a little known statistician and election model builder Nate Silver had the most accurate prediction of the Presidential election.  As a result of his success Mr. Silver was elevated to the top Blog pages of the New York Times, and this year he is now on record on election day as having models which forecast about 315 electoral votes for Mr. Obama and a 91% chance that he will win the electoral college.  Obscure no more, Nate Silver now has the focus of the political pundit world right on his face.

Mr. Silver’s methodology is to use past relationships and current polling data to simulate the election, something called Monte Carlo simulation.  It is the same technique casino firms use to determine the various probabilities  of a customer winning a casino game, and so allows the casino to set the payouts.  If the technique is correct the casino will, day after day, always payout less than it takes in.  That’s why casino’s don’t call what they do gambling.

Essentially Mr. Silver simulates an election thousands of times, and in his final runs Mr. Obama wins 91% of the time.  The problem with relying on this is that unlike casino games, which are played tens of thousands of times in a day, the election will take place only once.  An even bigger problem is that an election is a sample of the voters, since not everyone will vote and sampling error is present.  In Virginia, for example, it may be that Mr. Obama is favored 49% to 47%, but if more Romney voters than Obama voters actually vote, the result could easily be in Mr. Romney’s favor.  This sampling error is critical when the race is close.

Even worse for the mathematical models is that they rely on polling, which is a sample of the sample.  So another sampling error is injected.  Mr. Silver counters that problem using a large number of polls.

Averaging polls together increases their sample size — making them much more powerful statistically than any one poll taken alone. 

But this only works if the polls themselves are independent, a problem Mr. Silver acknowledges.

 But the errors in the polls are sometimes correlated, meaning there are years when most of them miss in the same direction. Mr. Romney remains close enough to Mr. Obama that he could fairly easily win the popular vote if there is such an error in Mr. Obama’s favor this year.

So maybe the election will turn out the way Mr. Silver says it will, but until the votes are in nobody should expect that will be the case.  When the race is close sampling error can be a killer.  Good luck Nate, we are all pulling for you. 

No comments:

Post a Comment