There are two momentous questions data nerds have been wrestling with since the race began. One: Can Hillary get Obama’s coalition of young adults and minority voters to turn out for her as strongly as they did for him? Doesn’t look like it, per the early-voting numbers for black voters — but Clinton might be able to make up the votes she’s losing from blacks thanks to more Latinos turning out this year (and her surprisingly durable advantage among college-educated whites, of course). A WaPo poll out today has her winning Latinos nationally 67/19, which puts her a bit ahead of Obama’s pace when he won that group 71/27 in 2012. If Latinos tilt more heavily towards her and more of them turn out, especially in swing states like Florida, Colorado, Nevada, and Arizona, Trump has a major problem. The Clinton coalition may end up looking different from the Obama coalition but “different” doesn’t mean “too weak to win.”

The other momentous question the nerds have been trying to answer has to do with the Trump coalition. His path to victory involves winning whites, especially working-class whites, overwhelmingly. Some polls have showed him underperforming Romney with those groups, a trend that would be lethal if it holds on Tuesday, but that might be changing now that more Republican voters have begun to come home over the last week or so. The deeper question is how many “missing” white votes are actually out there for Trump to harvest? If you look at the census data, the number is enormous. If Trump turned out even a modest fraction of irregular white voters and won them decisively, he’d waltz to a landslide win. On top of that, there are lingering suspicions that some voters who do plan to vote next week, especially upscale ones, aren’t being honest with pollsters about their secret support for Trump. A college grad who plans to vote Republican faces a lot of cultural pressure this year pushing in the other direction; there may be a kind of “Bradley effect” happening in which Trump’s actual support is several points higher than what people are willing to admit to pollsters. Put all of that together and there are two ways Trump could overperform next week. We could see working-class voters who didn’t vote in other recent elections turning out en masse for him this time, and/or we could discover that the “Bradley effect” among secret Trump supporters had led the polls to underestimate his electoral strength all along. If he pulls the upset next week, these are the first two lines of inquiry the nerds will pursue.

Some have already pursued them, though, as you know if you’re reading data sites like FiveThirtyEight or The Upshot or our own occasional digressions into this subject. Morning Consult published a study this morning that tried to gauge how much of a “Bradley effect” there might be in Trump’s numbers by conducting two polls simultaneously. One used a live pollster to call people up and ask them who they’re voting for; the other was conducted online, with no human interaction. The idea, of course, is that some people might lie to the human pollster about supporting Clinton when they secretly support Trump but they wouldn’t lie to an online widget. Compare the numbers in the two polls and, if the Trump numbers are lower in the live-interview poll than in the online poll, there’s your evidence of a “Bradley effect.” Result: They did find lower numbers for Trump in the live-interview poll. But the difference between the two was so small that it probably won’t matter next week.

Most of the change comes from exactly the demographics you’d expect — college grads and higher-earning voters, both of whom are considerably more likely to lie about not supporting Trump than other groups are. Given Clinton’s lead overall, though, the “Bradley effect” here is purely a matter of margins. A 52/47 win for Hillary is comfortable, along the lines of Obama’s 2012 victory. A 51/48 win is much tighter, along the lines of Bush’s 2004 victory. But both would almost certainly mean that Clinton hits 270 electoral votes and wins the presidency. The more the race tightens, though, obviously the more this matters. If the final polls next week show a 50/49 Clinton lead, clearly that could mean a 50/49 Trump win on Tuesday. But then, if it’s 50/49 in the polls, you don’t need a “Bradley effect” to explain the outcome. A race that close is a true toss-up; the polls would tell you little about who’s actually going to win. Also, it may be that as Trump’s poll surge this week makes news, the “Bradley effect” will be diminished: The more likely it seems that he’ll win the election, the more comfortable secret Trump voters may feel about admitting their preference. It’s also possible (although Morning Consult didn’t detect it) that there’ll be a sort of “Bradley effect” for Clinton among certain groups of Republicans, in which some righties who intend to vote for Hillary don’t admit that to pollsters. That group could offset some of the votes Trump will be receiving from his secret fans among college grads and high earners.

But forget the “Bradley effect.” What about the disaffected working-class white voters out there who are perfectly happy to tell you they support Trump but don’t have a habit of voting in national elections? If Trump can turn them out, the “Bradley effect” won’t matter. He’ll win. His whole pitch since the primaries has been that he’s able to bring people into the GOP who normally wouldn’t bother with the party — Jacksonian Democrats, blue-collar union members, and of course rural whites who either hadn’t bothered registering to vote before or who had registered long ago but then stopped voting at some point. Those people are his ticket to victory next week, if they show up. Nate Cohn of The Upshot looked at the trends in voter registration since 2012 to see if he could find evidence of a surge among Trump’s base. Should we expect lots of new white voters at the polls next week? Well, some — but not a blockbuster number, and not all of the ones who are voting will be voting Trump.

There has been no surge in registration among white voters since 2012, and the white voters who have joined the electorate are younger and likelier to support Mrs. Clinton than those who were already registered…

Mrs. Clinton has a considerable lead over Mr. Trump among newly registered voters in Pennsylvania, Florida and North Carolina combined, 47 percent to 31 percent…

Newly registered voters who aren’t affiliated with a major party lean to Mrs. Clinton over Mr. Trump by 42 percent to 21 percent; Gary Johnson runs a close third, with 20 percent. They may not improve the Democratic registration edge in these states, but they could contribute to Mrs. Clinton’s margin on Election Day.

What about voters who already registered in the past but haven’t bothered to vote lately (i.e. the “missing voters”)? As it turns out, “missing” white voters aren’t overwhelmingly pro-Trump; as noted in the excerpt, many of them are young and left-leaning. Trump does seem likely to win among “missing” white voters who say they’re certain to vote this year, 50/32 — but among the “missing” nonwhite voters who plan to vote this year, Hillary’s winning 61/20. Put all of that together and, among “missing” voters generally who claim they’re almost certain to vote next week, it’s Clinton who leads rather than Trump, 43/38. Which, really, isn’t all that surprising. With the country’s minority population growing, of course there’s going to be a sizable number of nonwhite “missing” voters to offset or even overtake the “missing” white ones. And she’s got a better GOTV operation than he does, so why shouldn’t she be more successful at picking the highest-hanging fruit on the electoral tree?

Put Cohn’s piece and the Morning Consult piece together and there’s some reason to believe that Clinton could actually outperform her polls next week. That’s not how you’d bet — the momentum is clearly with Trump right now — but if the “Bradley effect” is small and Clinton’s turnout operation is better than expected in turning out “missing” and new voters, especially Latino voters, some states that look like toss-ups now could be three-point Hillary wins. The early-voting returns in Nevada thus far point to “strong Hispanic support” for Clinton with the results that look a lot like 2012, when Obama won the state comfortably. Florida’s early voting is much tighter, with turnout up in Republican counties as well as Democratic ones. But note this:

Digging deeper into the voter file, Schale sees a surge of first-time voters in central Florida, where many newly-arrived Puerto Ricans are voting for the first time. In Orange County, Schale found, 29 percent of Hispanics who have voted are first-time voters and in Osceola it’s 31 percent. Schale says 55 percent of Orange County Hispanics have voted in no more than 1 of the last 3 elections, and the number in Osceola is 59 percent.

There’s a surge in first-time voters in Republican counties too. If Clinton wins the state, she may owe it to the fact that her “missing” voters showed up in enough numbers to hold down Trump’s advantage among his own “missing” contingent. We’ll know in five days.