Electoral Scoreboard Explained

Electoral Scoreboard: ... All polls in the last month

Disclaimer: I have no political affiliation with either party or candidate. All data have been gathered from www.electoral-vote.com, whose author has generously provided on the Internet the results from state polls he has admirably and laboriously compiled. That website's author is a Democrat, but he maintains that his results are strictly non-partisan; namely, he presents all poll data he is able to gather, without regard to the polls' results.

Statistcal Method
Colorado, Maine and Nebraska's Peculiarities

The Statistical Method

The statistical method is very simple, but also very powerful. We simply take the median margin of victory by either candidate in all available polls after a particular date, and present that figure. We also list wins-losses-ties for the candidate winning in the median in each state.

The median is formed by sorting the list of all polls, then taking the middle value. For example, if we look at Florida during September, we have, from most to least recent:

Bush +5, Bush +3, Bush +5, Bush +3, Kerry +1, Kerry +1, Bush +1, Kerry +1, Bush +2, Bush +8, Kerry +1, Tie, Bush +6, Bush +1, Bush +1

Kerry +1, Kerry +1, Kerry +1, Kerry +1, Tie, Bush +1, Bush +1, Bush +1, Bush +2, Bush +3, Bush +3, Bush +5, Bush +5, Bush +6, Bush +8

There are 15 entries in the list, so the middle entry is the 8^th, which is Bush +1, so the median result is Bush +1, which is what we would report for the month of September.

We also report Wins-Losses-Ties. One may look at the 1-point median lead for Bush in Florida in September, and say that the state was essentially a toss-up in September. However, Bush's record in September was 10-4-1 against Kerry, which is a healthy winning record.

The median is powerful, because it throws away discrepant points automatically, without bias. Though it is slightly slower to converge to the true value than is the sample mean (by a factor of sqrt[2*pi] for a normal distribution), it avoids the sample mean's high susceptibility to outliers. As such, a particular poll that may be biased, have systematic errors, or occur during a temporary "bump" for either candidate, introduces far less error in the median than in the sample mean.

An illustrative example of the power of the median may be found in this manuscript in which the median is applied to estimate the long and hotly contested value of the Hubble Constant (H₀):

Gott, J. R., Vogeley, M. S., Podariu, S., Ratra, B., 2002, Median Statistics, H₀, and the Accelerating Universe, Astrophysical Journal, 549: 1

The authors of this paper note an amusing, but high relevant quote by one of the greatest luminaries in astrophysics:

... Ya. B. Zeldovich commented on [the median]. He noted that [in his home country] some watches are not made very well, so when three friends meet they compare the times on their watches "one says `it's 1 o'clock,' the second says, `it's 5 minutes after 1,' the third says `it's 5 o'clock.' Take the median!" Perhaps no one has ever stated the benefits of the median over the mean better than Zeldovich.

Colorado, Nebraska and Maine

Colorado has introduced an intriguing ballot initiative to be considered by voters in this election. If passed, the initiative would award to each candidate electoral votes in a fraction (of Colorado's 9) that matches as closely as possible the candidate's fraction of the state's total popular vote. We have found 6 polls on the initiative: 51% for vs. 31% against (on 9/18); 47-35 (9/22); 45-44 (10/8); 39-53 (10/17); 35-49 (10/20); 38-55 (10/22); 35-44 (unknown date). The medians of these polls, overall, and in the last month are 9% against and by 14% against, respectively. We therefore regard the initiative as unlikely to pass, and present Colorado's electoral votes as winner-take-all.

If the initiative did pass, Ralph Nader could be a factor if he receives at least 1/27^th of the vote. Currently, his median vote fraction in Colorado is currently at 2%, which is below the 1/27^th = 3.7% threshold required.
Maine and Nebraska allocate electoral votes one per Congressional district, and give the remaining two to the overall state winner. To deal with this, we have gathered the 2000 election results from each Congressional district in those states and computed the margin of victory (for, say, Gore). We then compare Gore's total margin of victory in that state to our current median state poll. The difference in our current state poll and the 2000 state result is added to the 2000 Congressional district results, and the adjusted 2004 winner for each district is estimated.

Acknowledgement: I would like to thank J. Richard Gott, III, my thesis advisor at Princeton, not only for suggesting this project to me, but for his many suggestions and our many lively conversations on the topic of the election and its electoral nuances.

Wesley N. Colley... Rankings Home