Two weeks ago, we unveiled our model of presidential elections based on data from the past ten election cycles, which currently predicts that President Obama has a fragile advantage over his eventual opponent. We currently project that the president will win 303 electoral votes in November. But if the economic or job approval numbers slide backward even a few ticks, he loses the election by a hair.
Today we’d like to examine one rather glaring error in the model and explain why we’re not going to fix it: Massachusetts stands at 74.4 percent likely to go to Obama, while all external signs dictate that it is a much safer bet for the Democrats. This is because we currently assume Mitt Romney will win the nomination, as the prediction markets suggest.
There is overwhelming empirical evidence that presidential candidates get abnormal returns in their home state. In 1984, for example, Walter Mondale still won Minnesota even though the other 49 states all went to Ronald Reagan. (Mondale also held down Washington, D.C.) So Yahoo! Labs’ Patrick Hummel and I tested this theory with data from the last ten election cycles. We determined the size of that abnormal return, calibrated on those past races. This boost shifts Massachusetts from a Democratic lock to a Democratic-leaning state. It currently flips for Romney if Obama’s approval rating falls to 41 percent, well before similar states.
Not only has Massachusetts voted Democratic in the last six presidential elections, but it has also done so overwhelmingly. The state’s House delegation is all Democratic. Scott Brown, who won a special election for Ted Kennedy’s former Senate seat, is the only Republican representing the state on the federal level, and Elizabeth Warren is 68.6 percent likely to take that seat off his hands this fall. George McGovern even won the state in 1972, another year when the Republican incumbent (Richard Nixon) won 49 states.
Furthermore, Romney has widely distanced himself from Massachusetts’ liberal image, especially when health care comes up. The health care law that Romney signed as governor, which bears many similarities to Obama’s national initiative, is widely popular in Massachusetts. But Romney has declared that his first act as president would be to toss out the so-called “Obamacare” legislation.
Here we have an empirical study of the data clashing with anecdotal political wisdom, and it’s tempting, of course, to work in some fudge factor, or arbitrarily decide that Romney’s real home state is Utah, the nexus of his Mormon faith. But our model will be judged by academia on its statistical significance, not on what it predicts in 2012, with a twofold goal. First, we want to create the most accurate forecast possible several months before there is enough accuracy in polls and prediction markets to use that data for predictions. Second, we want to provide a data-driven look at the real correlation between all the data points–economic growth, incumbency, regional affiliation, and so forth–that are constantly debated in the press and academia.
As the election season progresses, we will incorporate more polling and prediction market data into our state-by-state presidential predictions. These data points will help correct any 2012 anomalies that cannot be observed by historical data. At present, historical data is the most accurate predictor of the election–regardless of any temptation to tamper with its findings when it conflicts with our human instincts.