This article has been updated.

There are three numbers that are important to remember when thinking about the 2016 election.

There’s 137 million, the number of votes cast — an increase of 8 million over 2012 and the most votes cast in a presidential race in history.

There’s 2.87 million, the margin by which Hillary Clinton won the popular vote. Clinton received slightly fewer votes than did Barack Obama in 2012, 65.85 million to 65.92 million.

And there’s the most important number, 77,744. That’s the margin by which Donald Trump won Michigan, Pennsylvania and Wisconsin, giving him enough electoral votes to win the presidency. Or, really, we could talk about 33,452, the votes to just win Michigan and Wisconsin, enough to have given him 286 electoral votes and the White House. That’s 0.02 percent of all votes cast that made the difference — a tiny amount.

It’s interesting to consider a world where those 77,744 votes (or those 33,452 votes) flipped. If, in Michigan and Wisconsin, 16,728 voters who voted for Trump had instead voted for Clinton, his total would drop by that amount and hers increased — and she’d have won the states and the presidency. Or had she had the support of 33,453 more people in those two states, same result.

Given that, any conversations about what did or didn’t give Trump the presidency can necessarily take a shortcut to the end. The margin was so small that nearly anything might have or did make up the difference. Was it James Comey’s letter? Was it get-out-the-vote? Was it a rally? Sure, all three! Any and all three.

But that’s not what happened, so Clinton — surprisingly — lost, even though she got far more votes. And so we spend a lot of time dissecting why she lost and what might have contributed. And, so, we invite tech companies to Capitol Hill to testify about trolls and bots and Russians who tried to game the election on social media and — since basically everything could have been the difference-maker — might have made the difference in the result.

When considering the role of social media, though, we need to consider the alternative scenarios above. Did Clinton lose because more Trump voters turned out, energized by something they read in a Facebook post? Or did Clinton lose because Clinton voters stayed home, turned off by something they read about her on Twitter?

Shortly before the election, Bloomberg reported that this was an explicit strategy of the Trump campaign: Showing voters (especially liberals, young women and black voters) anti-Clinton ads aimed at discouraging them from turning out. We know that Facebook has proven its ability to increase turnout using encouragement tools among friends, but the Trump team, according to Bloomberg News, was hoping they could similarly tamp down the number of votes.

Is that what happened? It’s hard to say.

There are more than 3,100 counties in the United States. We took the number of votes in each county in 2012 and 2016 and compared the change in votes in each place with the change in the population in each over the same period. (Vote data was from the AP; population data from the Census Bureau.) Most of those counties were won by Trump, since Clinton’s vote was heavily centered in major metropolitan areas that usually make up only one county.

On average, counties that backed Trump saw the number of votes cast increase by 3.1 percent since 2012 and their populations increase by 0.27 percent. Counties that voted for Clinton saw the number of votes cast increase by 1.3 percent on average — and their populations increase by 2.7 percent.

In other words, Democratic counties saw more population growth on average than vote total increases. Among Republican counties, that was flipped.

We can look at that based on how strongly the county supported Trump or Clinton.

Counties that strongly backed Clinton saw bigger population increases than increases in the number of votes cast, on average. In Trump’s strongest counties, the opposite.

That’s not quite the entire picture, though, averaging counties regardless of population. (So, for example, a county of 200 people is on an even footing with one that is home to 2 million.) We can do similar calculations for Trump and Clinton counties overall. (Meaning that the vote totals and populations of all Trump and Clinton counties are tallied together — so: 2,000,200 people in the prior example.)

Overall, counties that voted for Clinton saw their collective population increase by 3.1 percent and the number of votes cast increase by 5.2 percent. In counties that voted for Trump, the number of votes cast increased by 5.3 percent — and the population went up by 2.7 percent.

More subtle — but still a shift that was to Trump’s benefit. Breaking it out by intensity of vote:

Update: A reader on Twitter made a good point. Population increases are often a function of more young people, who are less likely to be able to vote.

Again, it’s impossible to say why there may have been this difference in turnout. Was it because Hillary Clinton inspired less interest from Democrats than did Obama, particularly among black voters? Or is it that Trump inspired more interest from Republicans? Or is it because suppression efforts were effective at keeping at home voters who might have turned out for Clinton? Or was it any number of a thousand different things?

In a way, the inability to determine what the critical factor was means that more attention will be paid to all of the possible causes. Had Clinton won by 2.7 million votes and with over 300 electoral votes, there would be much less interest in dissecting the tiny ways in which those numbers might have been massaged. As it is, though, we will spend a lot of time looking at a number of different factors and learn that we may never know just how much one or another affected the outcome of the race.

Just that, combined, all of them might have.