Predict Obama's odds in the 2012 election
Political scientists at George Washington, Yale and UCLA believe most elections can be predicted with just a few pieces of information. They created a formula that uses economic growth, presidential approval ratings in June and incumbency to forecast President Obama’s share of the two-party vote in the Nov. 6 election. Read more from Ezra Klein »
% change
Pessimistic:
'09 financial crisis
Optimistic:
Nixon landslide
Obama's low:
Aug. 2011
Obama's high:
Jan. 2009
Running election simulations
if the economic growth is -- percent and his approval rating is -- percent.
ELECTION SIMULATIONS
250
500
750
1000
See how former incumbents fare on the elections predictor
George W. Bush
1.4 percent change in GDP
46% approval rating
→
76% of election simulations
Bush won the 2004 election, just as the model predicted. The error was pretty small, too: It called Bush’s percentage of the two-party vote to within a percentage point.
Bill Clinton
2.6 percent change in GDP
55% approval rating
→
97% of election simulations
Amidst a growing economy, Clinton was easily reelected in 1996. Hard to lose when you’re popular and the tech boom is pouring rocket fuel on a recovery.
George H.W. Bush
2.1 percent change in GDP
38% approval rating
→
64% of election simulations
The Elder Bush’s reelection campaign confounded the model. It thought he would win easily. Instead, he lost big. Ross Perot’s third-party candidacy — not to mention Pat Buchanan’s primary challenge — were likely a big part of the reason. But 1992 was also an odd year: Even though the economy was recovering from the 1990-1991 recession, unemployment was rising.
Ronald Reagan
2.7 percent change in GDP
55% approval rating
→
97% of election simulations
Remember ‘Morning in America’? So does the model. But the Gipper won even bigger than the model predicted. Perhaps Walter Mondale’s strategy to run on tax hikes wasn’t such a good idea.
Jimmy Carter
-2.2 percent change in GDP
34% approval rating
→
5.7% of election simulations
Poor Carter. His loss was overdetermined: The economy was shrinking, inflation was high, and the hostages were trapped in Tehran. He lost, just as the model predicted. Interestingly, the polls were tight in that election till the very end, making it a good year for those who focused on fundamentals rather than the campaign.
Learn more about the methodology
To forecast the presidential outcome, we use ordinary least squares regression to predict incumbent presidential party vote share (in percent) of the two party vote from 1948 to 2008, (i.e. one hundred times the incumbent party votes divided by the sum of the incumbent party votes and the other major party votes, setting third party votes aside).
Our spartan model uses three explanatory variables: percentage change in gross domestic product per capita from quarter 1 of the election year through quarter 3 of the election year (the first nine months of the year); average presidential approval as measured by Gallup in June of the election year; and an indicator taking the value of one when the incumbent party's candidate is the sitting president, and zero otherwise.
From the regression results, we take 1000 draws from the sampling distribution of the model estimates, i.e. from a multivariate normal distribution with a mean of the vector of model coefficients and a variance matrix from the model's variance-covariance matrix. We also drew 1000 samples of the standard error of the regression, using a Student's t distribution (twelve degrees of freedom) as is appropriate in a forecasting exercise. With the 1000 vectors of coefficient and error draws, we construct 1000 predicted 2012 vote shares given hypothetical values of the three explanatory variables.
This reflects sampling uncertainty as well as fundamental uncertainty conditional on the model. Finally, we calculate what percentage of the 1000 predictions have the incumbent vote share greater than fifty percent, which is our estimated probability that Obama wins in 2012 given the hypothetical values of economic growth and June presidential approval.
SOURCES: Seth Hill, postdoctoral associate at Yale University; John Sides, associate professor at George Washington University; Lynn Vavreck, associate professor at UCLA; Gallup polling data. GRAPHIC: Jeremy Bowers, Emily Chow and Ezra Klein - The Washington Post. Published April 24, 2012.