The traditional phrasing of presidential election polling asks that respondents assume the election is happening that day: If the election were today, how would you vote? In 2020, more than in any prior year, that question might be taken literally. On each day of the past two weeks, an average of 3.7 million additional votes have been cast by early voters or received by county election boards. Voters are voting today — and, according to polls and as can be inferred from partisan splits, largely against President Trump.

But that’s simply one pool of voters. There’s still a week in which voters can cast ballots, and there’s still Nov. 3, Election Day itself, when many Trump supporters have indicated they plan to cast a ballot in person.

Lingering over all of this is the specter of 2016, when robust leads for Hillary Clinton evaporated in the way that a Hemingway character goes bankrupt: gradually and then suddenly. Over the last few weeks of that contest, Clinton’s leads in key states narrowed and narrowed, a function of undecided voters breaking for Trump. On Election Day itself, an unexpected level of support for Trump shifted the final tallies in several states significantly to the Republican’s advantage, enough to win him the presidency.

Trump himself elevates concerns about the polling specifically because most polls show him trailing. On Tuesday, he claimed he held a national lead based on a poll from Rasmussen Reports, a pollster that offers results consistently far more favorable to Trump and Republicans than other pollsters. In 2018, for example, Rasmussen’s final measure of support in the midterm elections found Republicans had a slight edge. Democratic House candidates won more votes by an 8.6-point margin.

Nonetheless, it’s worth considering where the polls stand at the moment and how they compare with polling in the same states four years ago. For this, we’re using FiveThirtyEight’s helpful averages of polls from both years, current as of Tuesday morning. If you want regularly updated tallies of the same numbers, we created an interactive graphic providing just that.

Looking at both national numbers and 11 swing states, Trump trails in seven states he won last time around, states where 117 electoral votes are in play. If we assume the same difference between the final averages in 2016 and the actual result, Trump would still lose six of those seven states — as well as the national vote and three other states that he lost in 2016.

If those assumptions hold, in other words, Trump will lose.

Those assumptions may not hold. In several states, the race has tightened during the past week; it is safe to assume the race might tighten further. It’s also unclear what to expect in terms of actual shifts in votes. Have pollsters better incorporated support for Trump, reducing the margins by which the polls are off? Or can we still expect similar errors?

While the polls have narrowed in several states, they haven’t done so uniformly or, generally, as much as they did during the same period four years ago. Only in Ohio, which former vice president Joe Biden doesn’t necessarily need to win, has the shift to Trump been larger this year than in 2016. In every other state we measured, the shift has either been smaller or been to Biden.

This should be considered in the context of Biden’s having generally already held larger leads than did Clinton. In Wisconsin, for example, Clinton was up 6.6 points 14 days before the election while Biden was up 7.3. Over the next seven days, Clinton’s lead narrowed by 0.9 points. Biden’s has fallen by 0.2. The gap between the final polling average in Wisconsin (five points) and the final vote (Trump up by 0.7) was 5.7 points. If the race in Wisconsin narrows this year by the same amount as it did in the last week of 2016 (a 0.7-point shift to Trump), that still leaves Biden with a 6.4-point lead — enough to weather a 5.7-point polling shift.

All of this is extremely loosely drawn, based on a range of assumptions that will probably not fall into place. The point is simply that Biden is in better shape in Wisconsin than was Clinton four years ago, even given what we know about what happened subsequently.

Trump appears to be increasingly pinning his hopes on Pennsylvania. On Tuesday, he touted a poll showing him up by three points … but it’s not clear what poll that might be. (Even a recent Rasmussen poll had Biden up by that margin.) It is the case, though, that Pennsylvania appears to be following the path of 2016 more closely than other states. Biden’s lead there has tracked closely to Clinton’s as the days have passed. If the polls shift 2 points to Trump during the last week (as they did then), Biden’s lead falls to 3.3 points — less than the 3.8-point swing between the final 2016 polling average and the actual result.

But there are two other factors that exist today that didn’t then. The first is that voters are much more likely to have made up their minds. At this point in 2016, Clinton led Trump by a margin of 47.1 percent to 42 percent, meaning nearly 11 percent of Pennsylvanians were undecided or leaning to third-party candidates. Now, Biden leads by a 50.2 percent to 44.9 percent margin, indicating only about 5 percent of the electorate is still wavering. Trump necessarily needs to peel support away from Biden, not just build up his own support elsewhere.

Then there’s the fact that 1.7 million people in Pennsylvania have already voted, more than a quarter of the total vote four years ago. Despite Trump’s efforts to suggest people want to change their votes, those are almost certainly votes that aren’t going to move no matter what happens in the polls during the next week. Every hour that passes, the final results in every state are becoming more firmly set in stone.

There are an enormous number of complicating factors this year that make hard-and-fast predictions questionable. Among them is the extent to which absentee votes will be counted, given the Trump campaign’s very public effort to discount them. Those efforts might backfire, however. Relying on Election Day voting for the bulk of his support puts Trump at a disadvantage should there be problems with voting systems or should there be bad weather in key areas. The Supreme Court’s decision on Monday to halt changes allowing for late-arriving ballots to be counted might hurt his own voters more than he expects: Nationally, about 53 percent of ballots requested by Democrats have already been returned, compared with about 44 percent of Republican ballots.

It’s lazy to simply throw up one’s hands and say we don’t know what will happen, so feel free to call me lazy. We don’t. But, as was the case seven days ago, Trump needs something pretty dramatic to happen in order for it to seem like he’s the favorite in this contest.

Luckily for him, the odds that something pretty dramatic might happen in any seven-day period of 2020 are awfully high.