Football Rankings Expert and Astrophysicist Nail 2004 Election Prediction!

Ultra-simple System Misses Only Hawaii, Confirms Polls

College football computer ranker Wes Colley (Colley Matrix and Old Dominion University [now at the Univeristy of Alabama in Huntsville]) and astrophysicist J. Richard Gott, III, (Princeton University) have a surprising conclusion from the 2004 election: trust the polls.

Colley and Gott had a very simple idea in September: take the median of the last month's polls in every state, and use the result to predict the election. Such a simple system is at odds with the much more common, much more elaborate technical and fundamental analyses applied to the polls, such as mathematical trend analysis, or adjustment for the "youth vote."

The median, as any high school student could remind you is just the middle item in a sorted list--the median of {3,4,1,9,17,1,6} is 4; sort the list and pick the middle one. The median has the advantage of throwing out statistically unlikely results (such as 17 in the list above), as opposed to the usual average (mean), which can be biased by a single bad entry.

Colley and Gott applied this method, without adjustment, as the polls came in each day. They began to notice that there was very little movement even over the course of a full month. As more polls rolled in, the state poll medians became more and more reliable. By the end, most swing states had at least 30 polls, and so the medians were very reliable indeed, but only if the polls themselves were unbiased, and there were no major up or down trends during the month, two big "ifs." (The second "if" required that no "October surprise" or major gaffe take place.)

While pundits and analysts scrutinized the polls, scouring the landscape for any emerging trend, Colley and Gott kept fast to their simple median method. With no obvious gaffes or surprises along the way, they had grown confident in what the medians were saying. By election day, they had a firm prediction of a Bush victory. As the nation watched and waited, one state after another met the Colley and Gott prediction, even the swing states, such as Pennsylvania and New Hampshire. At length, even Ohio fell into place. By the end, their prediction had missed only tiny Hawaii, where there had been only two polls in October.

With such a remarkable success, Colley and Gott had verified the two "ifs:"
  1. The pollsters were getting it right;
  2. The electorate moved very little over the course of a month.
A third conlusion follows immediately, which is that a more elaborate analysis of the polls was really unnecessary. The simple one-month median worked better than almost any other method. Colley, whose computer football ranking system is the simplest of those in the Bowl Championship Series Rankings is certainly pleased with the outcome, saying, "We did maybe the 2nd simplest thing you could think of, and what do you know, it worked."
Wesley N. Colley... Rankings Home