Could Trump still pull off an upset?

Fifteen days to go.

Former Vice President Joe Biden has maintained his polling lead over President Donald Trump; the FiveThirtyEight polling average has Biden at 52.4 percent with Trump at 41.9 percent. That's a whopping huge lead with only two weeks to go and more than 28 million votes already tabulated. Still, it's not hard to see where Trump could make up enough ground to win, even though it's by now an unlikely outcome.

So let's go through one more time all the ways that Trump could do better than the current polls suggest — or worse. The numbers here aren't meant to be taken too precisely; they're just rough estimates to show the range of realistic expectations. 

Biden's recent gains fade (Trump potential gain: 0 to 4 percentage points). As recently as Sept. 29, Biden's lead was only 7 percentage points. That was on the low end of where polling averages have been, but overall — until the news about Trump's taxes, the first debate and Trump's coronavirus diagnosis — the typical Biden lead had fluctuated between 7 and 9 percentage points, only occasionally going a bit lower or higher. It's possible that as those events fade so will Biden's surge. 

Late events shift the contest (-5 to +5 percentage points). Nate Silver has a chart comparing polling leads 15 days out to the final numbers in the last 12 presidential elections, and the news isn't good for Trump. In nine out of the 12 contests, the lead shifted by fewer than 2 percentage points. Still, the polls shifted significantly in 1992, with Bill Clinton losing half of a 14-point lead; Trump gained 3.1 points in 2016, and Bob Dole gained 2.1 points in 1996. Let's say that the plausible maximum shift here is five points, in either direction, but it's much more likely that any such shift will be small.

Polling error (-4 to +4 points). National polls may not get things exactly correct, but they're usually pretty good, and large errors are rare. In 2016, Trump only exceeded the final national polls by 1 percentage point, although he did better in a handful of states. As with late events, this could go in either direction. I can imagine some reasons that polls could be underestimating Trump's support, but I could do the same for Biden. I don't see any reason to think one is likelier than the other.

Electoral College bias (2 to 4 points in Trump's favor). It's likely that Trump wins the Electoral College if the national vote is tied. The question is how big his advantage might be. Right now, it looks pretty large — the "tipping point" state, the one that would give Trump the election if each state shifts the same amount in his favor, is Pennsylvania, where the president is losing by only 6.7 percentage points. That may be an artifact of which outfits have conducted recent polls. Still, the forecast models from FiveThirtyEight and from the Economist project about a 3-point tilt. 

If you add up the best-case scenario for Trump from each of those possibilities, you can see how he could win. But remember: There's no particular reason to think that there will be any shift at all from late events, or that the shift would be toward Trump rather than away from him. Same for any potential polling error. 

The best news for Trump in October: Biden has been unable, so far, to push any of the other states that Trump won in 2016 into the same range as Wisconsin and Pennsylvania. What that means is that Biden probably has to win both of those (and Michigan, where he's opened up a somewhat bigger lead). If Biden's lead in Arizona or Florida or North Carolina were comparable to those two states, he'd be safer. As long as Biden's lead is large it won't matter, but if it gets under 5 percentage points, then Electoral College considerations start kicking in.

The best news for Biden in the last week: His post-debate surge doesn't appear to be fading. If anything, his national lead seems to be a bit larger than it was a week ago. That strongly suggests that the top category above is going to zero out. Which is good news indeed for Biden, given that the next two categories, late events and polling error, are as likely to help him as hurt him. 

Of course, we could toss all of that out and go back to basics: A president at 42.8% approval and 54.2% disapproval two weeks before the election, and who has been underwater virtually his entire presidency, is simply very unlikely to win re-election.

Jonathan Bernstein is a Bloomberg Opinion columnist covering politics and policy. He taught political science at the University of Texas at San Antonio and DePauw University and wrote A Plain Blog About Politics.

Disclaimer: This article first appeared on bloomberg.com, and is published by special syndication arrangement.