The last two polls, from FAU and the Florida Chamber of Commerce, had Trump ahead narrowly. This new one from Mason-Dixon has Hillary ahead by two, but I think the actual margin is less important right now than the fact that it does appear to be tight in Florida. Remember, last week Monmouth had Hillary ahead by nine and a few days ago Saint Leo’s had her up by 14(!!). There were inklings that she was pulling away in a state that’s must-win for Trump. Now, suddenly, it looks like she isn’t.

Many of the numbers here will look familiar:

Trump ahead comfortably with men and Hillary ahead more comfortably with women? Yep, we’ve seen that before. Trump polling miserably with the youngest voters while Gary Johnson does surprisingly well? Not the first time we’ve seen that either. Hillary polling a bit better with her own party than Trump is with his? That too, yeah. And of course there’s the familiar racial splits, with Trump up big among whites and Hillary up big among blacks and Latinos. Check the 2012 exit polls for Florida and you’ll see that the numbers above also line up very well with how Obama and Romney did in different demographics, which explains the two-point Clinton lead. Obama, after all, won the state by one point.

There are a few discrepancies worth noting, though. Although Trump is doing better with men than Romney did, Hillary’s doing significantly better than O with women. Obama won women in Florida by seven points; Hillary is doubling that margin. Likewise, Hillary’s doing better — or rather, Trump’s doing worse — with Latinos than Obama did. O won that group 60/39 in Florida but Trump’s stuck at 27 percent. The FAU poll a few days ago had Latinos splitting just 50/40 for Clinton, a major difference from what Mason-Dixon is seeing. It also had black voters splitting 68/20, which isn’t remotely close to the 91/5 lead above.

So why, if Hillary is improving on Obama among key groups, isn’t her overall lead bigger? It probably has to do with independents, the other notable divergence between 2012 and this poll. Obama won indies narrowly, 50/47. Trump, however, is blowing Clinton out here, 44/30, with Johnson cleaning up another 13 percent. That’s due partly, I think, to the sample: Independents were 33 percent of the Florida electorate in 2012 but they’re just 21 percent of the group Mason-Dixon polled. M-D may have decided to categorize most indies as “leaners” with only “true” independents left in a separate category. It’s not hard to believe that true independents would prefer Trump, the outsider, decisively to Clinton the establishment dinosaur.

It’s not clear how much Gary Johnson is pulling from each of the two major-party candidates here, but he tends to pull more from Hillary than from Trump in other polls. Compare her margins against Trump head to head in various polls (average lead of 6.0 points) and her margins in the four-way race (average lead of 4.5) and you’ll see. Johnson is helping to hold her lead over Trump down, in which case Harry Enten has some good news for Trump fans: Johnson’s polling has not declined over the past two months even though, historically, third-party candidates have tended to start fading by now. If Johnson’s voters will tilt towards Clinton if forced to choose between her and Trump then obviously Trump wants Johnson running strong the rest of the way. On the other hand, it could be that polling is systematically overestimating Johnson’s support. Pollster Mark Blumenthal wrote an interesting post a few days ago suggesting that some voters might be choosing Johnson (or Jill Stein) because they’re being given that option by pollsters as an alternative to Trump and Clinton. Are they really committed to their third-party choices, though?

Earlier this summer, SurveyMonkey conducted an experiment which generally confirms that prompting for Johnson and Stein overstates their support. In early June, we split our sample into random thirds, offering voters either the two-way or four-way vote questions, or a third alternative that asked voters to choose between Clinton, Trump or “another candidate.” Those that opted for “another” were prompted to “specify” their preference by typing it in.

Results for the two and four way questions were similar to current voter preferences. On the four-way vote, 9 percent chose Johnson and 5 percent selected Stein. When offered an unnamed alternative, 20 percent opted for “another” candidate, yet when we examined specific preferences just 2 percent had typed in Johnson or the Libertarian Party and just 1 percent typed Stein or the Green Party. More than twice as many (7 percent) typed in Bernie Sanders, who was still an active candidate at the time.

A very similar pattern emerges on the four-way vote when we ask voters a follow-up question about how certain they are to vote for their chosen candidate. Those who opt for Johnson and Stein are far less certain of their support than those who choose Clinton or Trump.

Even if many Johnson and Stein voters decide to simply stay home, that’s probably good news on balance for Trump in that it’ll likely cost Hillary more potential votes than it will him. Ideally Trump would prefer to convert those voters to supporting him, of course, but there’s not a lot of overlap between his nationalist platform and the libertarianism and socialism of the two third parties. Especially with Hillary making some disingenuous noises about protectionism to keep pace with Trump in pandering to Sanders and Stein voters. Stein is an asterisk candidate anyway.

One last point. Mason-Dixon’s sample consists of 625 registered voters who say they’re likely to vote in November. That strikes me as sub-optimal given the chatter out there about “undercover Trump voters” who are disengaged from politics and might not be registered yet but who are planning on registering and voting for him this fall. If you want to capture those people in your data, you’re better off sampling all adults, not just registered voters, and including some sort of screen to see how many unregistered people are planning to register so that they can vote. There’s still time left for those people to do that, although not a ton. And don’t forget, Trump doesn’t have a ground game to help the “undercover voters” navigate that process. The RNC does, I’m sure, but the Democratic turnout operation is certainly more robust. In fact, according to one study, a superior ground game can be worth seven to eight points in battleground areas where campaigns are targeting their GOTV efforts. There may be some new Trump voters at the polls this fall, but there are also bound to be some traditional Republican voters who don’t make it out to vote because they didn’t have a campaign hounding them to do so. How many “undercover Trump voters” will he need to stampede to the polls on their own in order to swamp the higher number of regular Democrats whom Hillary’s operation will be turning out?