The Tech's Daily Electoral College Monte Carlo Simulation

By Keith J. Winstein Based on www.electoral-vote.com. Data from: November 1, 2004 Probability that President Bush is re-elected: 48% Probability that Senator Kerry is elected: 50% Probability that Ralph Nader is elected: 0.0% Probability that the electoral vote is an exact tie: 1.7%

How to read these results: "We cannot say with confidence that either candidate would win the election if it were held today."

Data from: November 1, 2004

Candidate	Percentage	Two-sigma Margin of Error
President Bush	51%	0.9%
Senator Kerry	49%	0.9%
Ralph Nader	0.3%	0.06%

President Bush has a 99% chance of winning the popular vote, based on these results. This does not match the actual national polls, which have much larger margins of error. It is probably not reasonable to use the exact voter turnouts from 2000 as a proxy (especially without error bars) for the 2004 turnout.

Using this questionable method, we arrive at a 55% probability that the winner of the popular vote is not the winner of the electoral college.

Thanks to Bill Sommerfeld for the computation of popular-vote results.

I performed a Monte Carlo simulation of the American 2004 electoral college, using the polling data available at www.electoral-vote.com. I used the first-listed poll (generally, the most recent poll) for each state.

Here is the source code, in Perl.

I made the following assumptions:

• Each poll represents a sample of 600 people per state, producing a 4-point two-standard-deviation margin-of-error on each major candidate's polling percentage. [Changed from assuming a 4-point two-standard-deviation normal distribution on Oct. 26.]
• Maine and Nebraska award electoral votes by giving two electoral votes "winner-take-two" to the state's winner. Also, each congressional district in Maine (2) and Nebraska (3) controls one electoral vote. I run a mini-election in each district, using the state-wide poll results, but assuming that only 250 people were sampled from the district.
• In each state (or district of Maine and Nebraska), the number of ballots cast for each of President Bush, Senator Kerry, and Ralph Nader is drawn from the distribution that results from sampling, 600 times, a random variable that produces the results "Bush," "Kerry," "Nader," and "no response" with probability equal with the poll results. [Changed from assuming independent normal distributions on each candidate on Oct. 26.]

I ran this simulation of the national election 100,000 times and counted how many times the president was re-elected, how many times Senator Kerry won election, how many times the result was an exact tie in the electoral college, etc.

Here's another statistical analysis of this year's polls. Thanks to Nathan Collins for the link.

Here's a 2000 presidential election analysis that used what looks like a similar method and predicted that Vice President Gore would win the electoral college (in the event of a popular-vote tie) with 84.7% probability. Thanks to Roger Ford for the link.