University professor’s statistical analysis predicts Obama’s reelection highly likely
September 12, 2012
Filed under Uncategorized
When Sheldon Jacobson launched his Bracket Odds website for the last March Madness, college basketball and statistics enthusiasts were tickled with his thorough analysis of seed distributions. For every possible bracket, gamblers had a precise measure of their bracket occurring — their odds of winning.
But March Madness is a game. The 2012 U.S. presidential elections are not, and Sheldon Jacobson is not a gambling man. He’s a computer science professor who works in operations research, a field that uses sophisticated statistics and mathematics to wrangle information to inform better decisions.
For the past four presidential elections, Jacobson has turned his field of expertise into a sharp-edged tool to analyze polling information and come up with a precise estimator of the victor each time voters head to the polls.
“If you look at the national numbers, Obama and Romney are almost equal,” Jacobson said. “But if you look at the actual forecasts for the elections, Obama is way ahead.”
This sounds like a contradiction, but it’s the result of the winner-take-all system 48 states use to apply their votes to the Electoral College. Most states are either clearly red or blue, but for some states, getting voters to turn either way is vital to winning the election.
“If you look at the approximately 10 battleground states, Obama’s ahead in almost all of them. So he has captured the states that are up for grabs better than Romney so far,” Jacobson said.
As of early September, most polling aggregates favored Obama to win, though not within margins that fall within a degree scientific acceptability. “270towin.com”:http://www.270towin.com/ estimated Obama’s victory as 84 percent likely. The New York Times’ “Fivethirtyeight.com”:http://fivethirtyeight.blogs.nytimes.com/ gave Obama a 75 percent chance of winning a second term at press time.
But according to Jacobson’s “Election Analytics website”:http://electionanalytics.cs.illinois.edu/, Romney’s victory is so unlikely that the conditions required to secure his win have only a 1.4 percent chance of occurring. Jacobson takes a two-pronged attack in processing polling data into the numbers he presents. The group starts by using Bayesian estimators, which, by mathematically inferring from prior observations, help fill in blanks in the polling data, blanks such as undecided voters. This Bayesian analysis translates into a state’s likelihood of voting in a certain candidate’s direction and includes varying “swing scenarios,” which determine how undecided voters are most likely to cast his ballot.
After that, information is processed into a probability of victory by using dynamic programming, which breaks the problem into smaller packets. Other groups use computer simulations to generate that final probability, but that method inherently incurs even more error. With the method Jacobson’s team employs, the errors have but a single source: the polling data itself.
“We’re bringing another angle to it, which is just a more sophisticated technical contribution,” Jacobson said. “Does that mean it’s going to be better or worse? Well, it’s going to be different.”
Jacobson’s work in elections analysis began before Nov. 7, 2000, as the American electorate watched in tepid excitement as ballot results came in from precincts across the nation. Going off of the raw number of votes, it looked as though Gore had clinched the election. But when it came down to several recounts in Florida and a U.S. Supreme Court case, George W. Bush came into office with less than half the vote.
But Jacobson wasn’t surprised; if you had been watching the Electoral College closely, you would have noticed Bush’s chances of winning were much stronger than Gore’s, even though it did come down to Florida’s 25 electoral votes.
The last time that happened was in 1888, when now-rarely-mentioned President Benjamin Harrison won. At that time, the earliest computers were run by a series of gears, wires and primitive punch cards. But with today’s machinery, analysts can run simulations much, much faster. And with computations that were on the order of hours in years past, technology has come a long way in assisting elections analysts.
“Four years ago, we took our computations; it took around an hour. And simply because we’re doing it better now and we have faster computers, it takes about five minutes,” Jacobson said.
As the Republican National Convention was wrapping up in Tampa, Fla., Jacobson and his lab met to discuss the most intense portion of the election. With only 70 days left to the election and the final stretch of the campaign in full swing, the team has to stay on top of daily updates to reflect the growing volume of polls in the crucial states that could be wooed by savvy politicians.
The group of undergraduate students, led by graduate student Jason Sauppe, are now in the daily process of taking each afternoon’s numbers and running them through the algorithms that will generate that day’s projections. In addition to the presidential race, the group is also turning to the U.S. Senate races, with the forecast of who will control the congressional chamber based on the same process used for the White House.
But the sheer uncertainty of elections limits September forecasts. Pollsters ask respondents whom they would vote for were the election held the day of the survey, rather than on Nov. 7. And even with high probabilities that forecast Obama to win, anything could happen.
“We know, theoretically through simple math, and empirically from recent elections, that 0.5 percent is far and away more than enough votes to swing the election in the Electoral College if they hail from particular geographic locations,” he said.
The entire election could hang on just 10,000 voters, Jacobson said. But exactly which 10,000 voters is unknown until after the ballots are in.
In 2008, Indiana was particularly hard to forecast. Going with Obama, the state stumped pundits and analysts who expected Indiana to be a narrow victory for Republicans. Jacobson said it was likely that the error came out of absentee votes that weren’t considered in the final days of polls.
But Jacobson and other analysts are expecting a closer election in 2012. And maybe this year, his analysis may be able to shed a more precise, focused light on how this election will pan out.
_Nathaniel can be reached at [email protected]_