Oakland computer ace finds election equation
________________________________________
Program gives Kerry an 81.8 percent chance of winning as of Friday
By Josh Richman
STAFF WRITER
oaklandtribune.com
Why read one presidential election poll when you can read 1,800 all at once?
That's what Lawrence Allen thought, and he decided to do something about it. The 56-year-old unemployed computer programmer from the Oakland hills wasn't seeking a national spotlight -- he just wanted a better survey.
"A nationwide sample with a 4 percent margin of error showing a 'statistical tie' tells you absolutely nothing," he said. "It seemed a self-imposed ignorance, a branch of 'science' that hasn't advanced in the last four decades."
As an astronomy buff, Allen was interested in statistical methods stargazers use to weed out sampling errors and deal with wide error margins. He'd talked to his brother, also a programmer, about composite polls summarizing results from multi ple individual surveys. And then he found Princeton University biology professor Sam Wang's Web site, offering a "meta-analysis" of state polls to produce likely Electoral College outcomes.
Allen said he'd always wanted to learn the Matlab statistical programming language anyway, and now he saw his chance. He created a program that combines a large number of state polls into a composite, reaching margins of error in heavily polled states of around 1 percent.
The states are then ordered based on how close they are to a 50-50 probability, and the 15 closest states are subjected to a "likelihood analysis" -- going through all 32,768 possible outcomes of those 15 states taken together, with each outcome assigned a probability based on multiplying individual states' probabilities.
Each outcome also has an associated Electoral College vote count; it's assumed that the 35 states plus the District of Columbia not included in the likelihood analysis will vote as their own composite polls suggest.
The outcome probabilities get sorted in bins according to what electoral count it produces, and then are added up in order; when the cumulative probability reaches 50 percent, that "bin" represents the likeliest Electoral College result.
This can differ from the Electoral College vote of the likeliest individual outcome, which is what happens if all states vote as suggested by the composite polls. For example, suppose Bush has a 60 percent chance of winning in each of Nevada, New Mexico and West Virginia -- the likeliest individual outcome result would have Bush picking up those states' 15 electoral votes.
But Bush has only a 21.6 percent chance of winning all three, giving Kerry a 78.4 percent chance of picking up the five electoral votes from at least one of those states.
Clear as mud to most of us; that's why the world needs people like Allen. "So I got the program written the way I wanted it, and then thought: 'How hard could it be to put this up on the Web?'" he said.
Not hard -- see it at www.arrowheadengineering.com. Friday afternoon, Allen's Web site gave Kerry an 81.8 percent chance of winning, versus Bush's 18.2 percent chance.
Earlier this week, Allen was featured in a front-page Wall Street Journal story on "an elite cadre of political amateurs unleashing the tools of statistics and mathematics on an extraordinarily close presidential race."
Also featured in the article is Matthew Hubbard, a math lecturer at California State University, Hayward, who created his own model that calculates the odds of victory in the Electoral College.
Hubbard's Web site, at binomial.csuhayward.edu, predicts that if the election were held today, Kerry would have a 73.9 percent chance of receiving 270 electoral votes -- the number needed for victory -- compared with a 24.6 percent chance for Bush.
Allen modestly calls this his "15 minutes of fame." His last electoral work "was as a precinct worker for Dukakis," the 1988 Democratic presidential candidate, "so I'm not politically active," he said. "But I'm interested in politics. I've got friends on the far right and on the far left."
He describes himself as somewhere to the left of most Democrats and says he voted for U.S. Rep. Dennis Kucinich, D-Ohio, in the Democratic presidential primary. So he and his liberal friends were dismayed as months slipped by and his analyses continued to show President Bush in the lead. But they're happier now that Kerry seems to be pulling ahead. |
