Analysis: Scientists go beyond the polls to forecast U.S. election

WASHINGTON Wed Aug 1, 2012 4:01pm IST

1 of 8. Emory professor Alan Abramowitz poses in front of a slide predicting President Barack Obama as winning the 2012 U.S. Presidential election at Emory University in Atlanta, Georgia, July 27, 2012.

Credit: Reuters/Tami Chappell

WASHINGTON (Reuters) - For anyone wondering how much of the vote President Barack Obama might lose in the November 6 election because he is black, a professor in Iowa has crunched polling data and come up with an estimate: 3.3 percent.

Another U.S. academic has penned a complex equation -- using data on the economy and presidential approval ratings -- that predicted who won the most votes in the last six presidential contests.

As the election draws closer, pundits and journalists are looking at all sorts of data, from persistently high jobless rates to the latest polls in the politically divided state of Ohio, for clues on Obama's chances of defeating Republican Mitt Romney.

But at a few U.S. universities, academics have boiled the art of prediction down to a dispassionate science. Some claim their forecasts in presidential elections -- typically issued months before Election Day -- have been more accurate than opinion polls taken the day before ballots are cast.

Plugging decades of data into spreadsheets, they calculate everything from how much a bad economy is hurting an incumbent to how the results of New Hampshire's presidential primaries, conducted 10 months before an election, can signal who the eventual winner will be in November.

"What this forecasting really amounts to is quantitative history," said James Campbell, a political scientist at the University at Buffalo, SUNY.

So far this year, forecasters in line with many current opinion polls see Obama squeaking out a victory over Romney.

In a Reuters poll of nine leading forecasters, the median prediction was for Obama to win 50.5 percent of the vote. Although under the complicated American system, that would not necessarily mean victory because the winner is determined by the state Electoral College results.

In 2008, the median forecast of the same group, which estimated that Obama would receive 52 percent of the vote compared with Republican John McCain's 48 percent, was about as close to the election results as Gallup's final poll from the last three days of the presidential campaign.

In the end, Obama received 52.9 percent of the vote to McCain's 45.7 percent, with other candidates receiving the rest.

PLACING THEIR BETS

As scientists, forecasters use their predictions to test theories about what drives election results.

But the endeavor also has a sporting side. Gamesmanship creeps into scholarly conferences, and some academics make small wagers over their predictive powers.

Alan Abramowitz, an Emory University professor and an accurate forecaster of the last half-dozen U.S. elections, won a bet with Campbell over the accuracy of their forecasts in the 2008 election. Campbell had to buy him a beer after Abramowitz's prediction was closer to Obama's margin of victory.

"I would say it's a macho thing, but if you saw what we look like, it's not a very macho-looking group," said Abramowitz, whose current forecast has Obama taking 50.5 percent of the vote.

Like most of his peers, Abramowitz is still gathering bits of data before issuing a final prediction by the end of the summer. More bad reports on the economy, such as the data released last Friday that showed U.S. economic growth slowed sharply between April and June, could shift forecasts to a Romney victory, he said.

A few models already suggest that the sour U.S. economy will be enough to boot Obama from the White House.

Three years after emerging from a deep recession, the economy has not recovered. The unemployment rate has been stuck above 8 percent for 41 months -- nearly Obama's entire time in office.

For his model, University of Iowa professor Michael Lewis-Beck looks at decades of data on job creation and presidential approval. His preliminary forecast, made with Charles Tien of the City University of New York, gives Obama 46.9 percent of the vote and Romney 53.1 percent.

Lewis-Beck also has tried to quantify how much racial bias could hurt Obama, the nation's first African-American president.

Views on race are famously difficult to measure. Data on race is easy to find: Roughly 78 percent of Americans identify themselves as white and 13 percent as black (Hispanics can identify themselves as either). But racist views typically aren't discussed in polite company, much less with pollsters.

With Tien and Richard Nadeau of the University of Montreal, Lewis-Beck studied an online poll from March that asked Americans which groups in society have been favored by Obama's presidency. Options included men, women, blacks and corporations, among others. Voters also were asked whether they would vote for Obama.

It turned out that people whose view of politics was affected by race -- that is, they believed the current government favors African-Americans -- were less likely to support Obama. Even people who thought the economy was getting better spurned Obama if they saw race as a factor in current politics.

The pattern allowed Lewis-Beck and his colleagues to put a number on what they called Obama's "racial cost" in the upcoming election: 3.3 percent of the total vote.

"There's going to be a double whammy -- the economic whammy plus the race whammy," Lewis-Beck said.

MEASURING THE ELECTORAL COLLEGE

In the last four presidential contests, more than 90 percent of leading forecasters picked the popular vote winner, says Thomas Holbrook of the University of Wisconsin-Milwaukee, who surveyed 40 forecasts over that period, including his own.

And therein lies a big caveat to their track record: Every major forecaster correctly predicted that Democrat Al Gore would win the popular vote over Republican George W. Bush in 2000.

Of course, Gore lost the election because he came up short in the Electoral College. Under U.S. rules for the presidential election, each state is worth a certain number of electoral votes, based on population. The first candidate to win enough states totaling 270 electoral votes wins the election, regardless of the popular vote.

"The 2000 election did a great job of reminding everyone about that," said Simon Jackman, a political scientist at Stanford University.

Some academics try to account for the Electoral College by using state-level data to predict the results of both the Electoral College and the popular vote.

It's a less traveled path because many professors find state-level economic data to be less reliable that nationwide data. Also, many think voters focus more on the national economy than on how their own states are faring.

Flouting this convention, Carl Klarner of Indiana State University uses a range of state and national data, from how much money people earned in Florida to presidential approval nationwide. A relative newcomer in the forecasting world, Klarner's prediction was among the most accurate in 2008 when he forecast that Obama would get 53 percent of the vote.

This year, Klarner sees Obama winning 51.3 percent of the vote. He gives the president a 57 percent chance of winning the Electoral College.

Several forecasters acknowledged that because a tight race is expected this year, they could come very close to estimating the vote tally for each candidate, but get the winner wrong.

"It's a lot like having a formula to try to predict football games," said Robert Erikson of Columbia University. "You can have a formula based on the past records of the teams, but then there will be intangibles, like the star quarterback breaks his leg."

(Editing by David Lindsey and Cynthia Osterman)