Nobody saw it coming. Not the media. Certainly not Hillary Clinton. Not even Donald Trump’s team of data scientists, holed up in their San Antonio headquarters 1,800 miles from Trump Tower, were predicting this outcome. But the scientists picked up disturbances—like falling pressure before a hurricane—that others weren’t seeing. It was the beginning of the storm that would deliver Trump to the White House.
Flash back three weeks, to Oct. 18. The Trump campaign’s internal election simulator, the “Battleground Optimizer Path to Victory,” showed Trump with a 7.8 percent chance of winning. That’s because his own model had him trailing in most of the states that would decide the election, including the pivotal state of Florida—but only by a small margin. And in some states, such as Virginia, he was winning, even though no public poll agreed.