By now, you’ve heard about the great American divide that ushered Donald Trump into office. It’s probably been pitched as a matter of money and wealth— prosperous city dwellers against the rural poor, or the white working class versus everyone else.

But that’s the wrong place to look. Education mattered more than anything else, it appears, even when controlling for economic factors.

States submitting their final election returns have made it possible to dig deeper into local ballots, combining Census demographics and county-level turnout to make conclusions with statistical heft. The chart below tracks 15 demographic factors and the relative strength they held in this election, as modeled through linear regression (and controlling for total votes and Mitt Romney’s 2012 turnout, which strips away some predictable partisan patterns):

Economic discontent defined this election, and a populist won it. But bare economics do not appear to have played a leading role in how voters cast their ballots. The proportion of people who held a bachelor’s degree or higher was the primary correlate in how a county voted, far more than how much money the average townsperson made, or how many had lost a job.

Education is tricky, because it’s tied up with so many other things. College graduates are generally more liberal; they’re also more insulated against unemployment and outsourcing than your average blue-collar factory worker (though that certainly could change). And, of course,  there’s the ever-widening income gap between college graduates and everyone else. What looks like a strong correlation between higher education and opposition to Trump might just be a proxy for a hidden economic variable.

But if that’s the case, it’s not a simple connection. The education gap persists even when controlling for a county’s median income, its industrial base, and whether it has lost local manufacturing jobs. All those factors predict support for Trump, but not to the degree that education does. Neither does population density, a decent indictor for whether voters live in a rural or urban area.

Counties with well-educated residents consistently broke against Trump even when they weren’t particularly wealthy, as Nate Silver recently noted was the case with his hometown in Michigan. They voted Democratic even when surrounded by relatively red-leaning neighbors (see Harrisonburg, Virginia, or Asheville, North Carolina). Some of the counties hosted college towns, distorting their overall demographics, but not all of them. Something about living in a better-educated area made people favor Clinton, and it didn’t have as much to do with money as widely suggested.

The exception to that might be the rate of male unemployment, which rivals a college degree for predictive power. That’s not surprising: white working-class men have long polled as the base of Trump’s support, and the past two decades haven’t been kind to them. But while the connection seemed strong, our analysis indicated it was also more likely than the other factors to be a product of random chance.

So what is it? Trump voters aren’t stupid. I have a hard time believing that simply obtaining a college degree would convince a voter to view Trump as a charlatan and not a champion. I’ve previously theorized that a “hope gap” might explain Donald Trump’s appeal, with people in communities where most residents have fewer educational credentials having a dimmer view of the future. But it’s also possible that a region’s education level is simply the primary ingredient in a larger medley of cosmopolitanism, an indication of its integration with a global world.

The Republican nominee did worse in communities with strong minority populations (including a disappointing turnout among Asian voters, who are typically sympathetic to the GOP). Conversely, his support shot up in counties where very few residents have left their home state. Polling before the election suggested Trump voters were more likely to have remained in the communities where they were born; his electoral returns largely held that up. He also fared better in regions that have seen a recent influx of Hispanics, where there were few before.

Read one way, these figures support the white anxiety theory—that white Americans saw their way of life changing along racial lines and decided to try their own hand at identity politics. But consider that Great Britain saw nearly the same voting patterns earlier this year in the Brexit referendum, with education playing the strongest role in whether a region voted leave or remain in the European Union. The United Kingdom has racial politics, too, but they’re different than America’s. Indeed, that contest centered on how closely the U.K. wanted to knit itself into the larger world, a question the U.S. is grappling with right now as well.

That presents a less simplistic explanation: Trump won not solely because of economics, race, or globalization, but by some subtle interplay between the three. Analysts haven’t yet hit on the Grand Unified Theory of Trump, the right set of variables (total international flights? Hours of Netflix watched per person?) to shove education out of the picture and show the true bonds that connect conservative voters. That equation might never balance. But there’s now enough evidence to say, emphatically, that money and race alone didn’t decide this election.