The Primary Model predicted that Barack Obama would win a narrow victory with 50.1% of the major-party vote. Obama surpassed that forecast by just a little more than one standard error (2.5). How come the model came as close as it did with a forecast issued as early as January? What prevented it from coming even closer? And what might be done to improve the model?

To recapitulate, the key predictor of the model is the performance of the candidates in the New Hampshire primary. Obama's showing, by historical standards, was respectable for a candidate of the opposition party, even though he came in second. John McCain's showing, in turn, was below par for a candidate of the White House party, even though he won there. Granted, Obama surpassed his New Hampshire performance in other nomination contests. Had Obama done as well in New Hampshire as he did in the Iowa caucuses, for example, the model would have awarded him a vote share (51.1%) closer to the outcome in November. So it might be wise to use a broader measure of primary performance to predict the vote in November.

The model's other predictor exploits a cyclical dynamic of presidential elections. When a party has controlled the White House for two terms, as Republicans did the last eight years, the opposition party often has a better than even chance to win the next election. For 2008, however, the cyclical component on its own would have predicted a Republican victory. How so? One, the operation of the cycle was muted due to the closeness of the 2000 and 2004 elections. What is more, this dynamic was disrupted in 2000, when the winner of the popular vote failed to prevail in the Electoral College. But that is the kind of rare event that must be left to random error.

Even so, and without considering an unpopular president, a faltering economy, and a meltdown on Wall Street during the election campaign, the Primary Model was able to pick the winner of the 2008 election months ahead of time and well within two standard errors.