By the numbers: How Donald Trump won

Democrats stayed home. Democrats have no one to blame but themselves for losing this race. Take a look at Wisconsin (which Trump won) and Minnesota (which Trump barely lost). Trump got an almost identical number of votes as Romney did four years earlier. There was no white surge in either state, or if one did exist, it was cancelled out by Trump losing other Romney supporters. The problem for Clinton was a massive drop-off in Democratic turnout. In Wisconsin, she received 200,000 fewer votes than Obama. In Minnesota, she lost 180,000 votes. The fact that she received 40,000 fewer votes than Obama in predominately African-American Milwaukee was particularly problematic. She lost the state by 27,000 votes.

There’s a similar dynamic in Michigan. Trump improved over Romney by more than 160,000 votes. Clinton lost close to 300,000 votes. More than a third of those missing votes came in Wayne County and Genesee County, home to Detroit and Flint, respectively. Considering she lost the state by 12,000 votes, the diminished turnout in these two cities alone was decisive. As a result, there’s a strong case to be made that Clinton’s decision not to spend more time campaigning in Michigan and Wisconsin was disastrous.

But even in states where Clinton paid lavish attention, her support cratered. In Ohio, Trump improved over Romney’s 2012 results by 110,000 votes. Clinton lost 500,000 votes relative to Obama. Clearly, some of this was a case of Obama voters switching to Trump, but that doesn’t tell the complete story. In union households, Clinton lost by 12 points — Obama won them in 2012 by 23. The percentage of the electorate that was Republican jumped from 31 percent to 36 percent. In relatively affluent communities like Franklin County (Columbus), Clinton did better than Obama, but lost votes in heavily African-American Cuyahoga county and did far worse in white working-class communities like Trumbull, Ashtabula, and Mahoning. But in Ohio, the story seems to be that a ton of voters, particularly white working class voters, who supported Obama in 2012 simply stayed home in 2016.

The missing white voters emerged. No state offered a more shocking and unexpected outcome than Pennsylvania. For Democrats, there’s a rule of thumb in Pennsylvania politics: Get out a massive vote in Philadelphia and you’ll win. Clinton did that. In fact, she got more votes in Philadelphia County than Obama in 2012. In the blue Philly suburbs she got 56,000 more votes than Obama. She even won in Chester County, where Obama lost in 2012. She mobilized the so-called Obama coalition of affluent, college-educated whites and nonwhite voters. And yet, she lost the state by 68,000 votes. Obama won it in 2012 by 320,000 votes.

Trump reversed the Democrat’s advantage in Pennsylvania by doing what almost no political observer thought he could do and no polling showed possible: mobilizing a huge number of white rural voters. There was a 32,000-vote shift in Luzerne County; a 24,000-vote shift in Lackawanna County; a 22,000-vote shift in Erie. Clinton should have been able to handle those losses. It was the additionally higher turnout among whites in red countries that doomed her. In red county after red county, Trump did better than Romney did in 2012. Some of that was clearly white rural voters switching from Obama to Trump, but take a place like Northampton County, just north of slightly blue Bucks County. Clinton matched Obama’s vote total there. Trump found 12,000 votes Romney didn’t get.

A similar dynamic is evident in Florida, where Clinton got 250,000 more votes than Obama. She netted 80,000 more votes in Dade County; 40,000 more in Broward; 22,000 in Palm Beach; and 55,000 more in Orange. Those are all communities with large numbers of nonwhite voters. Overall, except for Pinellas, St. Lucie, and Jefferson, she won every county in Florida that Obama won, and often by larger margins. For Clinton in Florida, at least in nonwhite communities, the Obama coalition came through.

Her problem was that Trump got 440,000 more than Romney. How did he do it? Not in the same way that he did it in Pennsylvania. Instead, in suburban and exurban counties like Volusia, Brevard, Polk, Lee, Pasco, and Sarasota, he ran up much bigger margins than Romney had four years earlier. Unlike Pennsylvania, these aren’t necessarily working class of rural voters. They are suburban dwellers and retirees — a very different demographic from those who turned out for Trump in Pennsylvania.

The Obama coalition isn’t completely dead. The failure of Clinton to win in Pennsylvania and Florida would seem to suggest that the Democratic strategy of mobilizing college-educated whites and minority voters is doomed. The one corrective to that argument is Virginia. Clinton’s popular vote margin there was smaller than Obama’s, but she still won by almost 5 points. In the affluent Northern Virginia suburbs of Loudoun and Fairfax, she got 115,000 more votes than Obama and maintained his advantage with black and Hispanic voters. Elsewhere in the state, she lost white rural voters, but in Virginia, overall, the Obama coalition held.

The other two places that offer some fleeting hope for the Obama coalition are Arizona and Texas. Clinton lost Texas by 9 points, but she received close to 600,000 more votes than Obama did four years ago. Trump received only 112,000 more votes than Romney. Texas may still be a cycle or two away from being competitive, but Democrats have some reason to believe they can challenge Republicans there.

Arizona is a more interesting story. Since there are still outstanding votes in the state (and, as of this writing, Clinton still has a chance to win it), it’s hard to draw concrete conclusions. But four years ago, Obama lost the state by 10 points. This year Clinton lost it by 4. The exit polling on Arizona is a bit hard to take seriously, since it shows her dramatically over-performing among white voters, compared to 2012, and underperforming among Hispanic voters. In general, most of the exit polling on Latino voters is highly suspect. Latino Decisions, a polling firm that did pre-election exit polling in both English and Spanish, had Clinton winning Hispanics by an 79-18 percent margin. Nonetheless, Clinton was clearly more competitive in Arizona’s Maricopa County, and that probably can’t be explained by Hispanic turnout alone. Arizona could be a true battleground as soon as 2020. One other source of hope for Democrats is that in places like Arizona, Florida, Texas, Virginia and perhaps Nevada there does appear to have been an Hispanic surge for Democrats

Voter suppression wasn’t necessarily a factor. The diminished turnout in black communities may lead one to conclude that voter suppression efforts, particularly in Wisconsin, might have been a factor. Considering that Clinton lost the state by 28,000 votes, it seems fair to argue that the state’s strict new voter ID law blocked enough voters for Clinton to win.

But another state where voter suppression efforts have been a huge issue, North Carolina, suggests otherwise. Clinton did about as well as Obama did in 2012. Trump improved by 70,000 votes over Romney, who won the state. But in counties with large numbers of African-American voters like Wake, Mecklenburg and Guilford, Clinton over-performed on Obama’s numbers. Considering Clinton lost by 180,000 votes in North Carolina, it’s hard to make the case that voter suppression was the problem for her.

We still need to look at much more data to understand what happened and how Trump won, but if the initial download is any indication, there isn’t one clear answer. Considering that Clinton won the popular vote, perhaps even by more than a million votes, the saddest explanation is that she lost just enough votes in just enough crucial swing states to hand the presidency to Trump. Dumb luck is an inadequate and unsatisfying explanation, but that might be the answer we’ve got right now.

Michael A. Cohen’s column appears regularly in the Globe. Follow him on Twitter @speechboy71.