This article has been updated.

President Trump’s still-relatively new campaign manager, Bill Stepien, stepped into the job at a pretty opportune time. His new role was announced by presidential tweet on the evening of July 15, when Trump trailed former vice president Joe Biden by about 9 percentage points in RealClearPolitics’ average of national polls. Since then, the gap has narrowed to a bit over 6 points.

Stepien, clearly understanding how to keep his job, last week gave the credit to Trump and, specifically, Trump’s resumption of regular briefings on the coronavirus pandemic.

“His instincts are strong and there’s no one better at the podium than him,” Stepien told Bloomberg News. “That is a net positive every day of the week, polling shows it.”

This was the theory behind renewing the briefings, in fact: Trump polled better when he was holding them regularly, so getting back into the habit should help his poll numbers. It was a theory based on a flawed premise, but, to hear Stepien tell it, it’s working out.

Trump himself made a related claim during the briefing on Monday evening: that Biden’s policies were hurting the Democrat’s candidacy.

“I think it’s one of the reasons why if you look at polls — which I’m not a big believer in polls, I wouldn’t if I was, I guess I wouldn’t be standing here right now,” Trump said. “And by the way, our poll numbers are going up very rapidly, as you know, and Joe’s are going down very rapidly.”

On the surface, the RealClearPolitics numbers seem to reinforce the idea that a shift began once Trump resumed his briefings. The first one last month was on July 21, when Biden was polling at about 50 percent on average and Trump was at 41 percent. In the most recent RCP average, Biden’s at 49 and Trump is at 43, narrowing the margin by 3 percentage points.

But as a number of people, including the New York Times’s Nate Cohn, have pointed out, this is in part a function of the polls that are being included in the average.

Many major pollsters who rely on live telephone calls for the bulk of their contacts — a more expensive methodology — haven’t released new polls in a while, preferring to wait until closer to the party conventions to set a pre-convention baseline of support. The last purely live-caller national poll included in FiveThirtyEight’s aggregation of polls, for example, was conducted almost a month ago by the Kaiser Family Foundation.

Why does this matter? Because Trump’s numbers have consistently been a bit better in polls conducted online than in live-caller polls. He and his team insist that this is a function of respondents being less willing to offer honest assessments of their views to real people than they are when asked by a computer survey — a convenient rationale for claiming that there’s an unmeasured groundswell of support but a rationale that wasn’t bolstered by the results in 2016.

But the upshot is that pollsters which primarily use online surveys poll more often, since it’s less expensive, and most of the recent surveys included in the RCP average are ones conducted entirely or partly online, where Trump does better. Maybe his position has improved in recent weeks, and that will be reflected in live-caller polls, as well. Or maybe the shift in the RCP average is a function of the polls being included.

What’s important to note about Stepien’s assertions about his candidate’s position is that the modest improvement seen in that average still has Trump in much weaker position than he was four years ago. This is in part a function of how much more positively Biden is viewed than Hillary Clinton was four years ago.

Over the course of 2016, Clinton rarely earned 50 percent support in the RCP average. She never did after she clinched the Democratic nomination, in fact. Over the course of 2016, she earned between 43 and 50 percent support in the RCP average, overlapping with Trump’s range of 38 to 46 percent. So far this year, there has been no overlap between Trump’s and Biden’s support, with Biden ranging from 47 to 51 percent and Trump from 40 to 46. On more than half the days for which RCP has a value, Biden’s support has averaged 50 percent or higher.

The same pattern holds when considering the margin between the two candidates now and four years ago. Then, the margin between Trump and Clinton varied widely, with Trump finding himself within five points of Clinton on 3 of 5 days that year. He has been within five points of Biden on 3 of 10 days so far in 2020 — and never closer than four points behind.

There are a flurry of applicable caveats here, of course. That national polling doesn’t map cleanly onto the electoral vote is the most important, of course — though in 2016 it was an accurate measure of the national popular vote. But if Trump doesn’t narrow the current national gap, it’s hard to see how he could win in November.

Let’s do a rudimentary assessment of how states follow the national margin. If we assume that the margin in each state holds steady relative to the national margin as it did in 2016, a 6.4-point national lead for Biden would mean he flips six states: Arizona (Biden +1), Florida (Biden +3), Michigan (Biden +4), North Carolina (Biden +1), Pennsylvania (Biden +4) and Wisconsin (Biden +4). (As you might expect, given the national margin, the RCP averages in each of those states have Biden winning by four to six points.) Biden would win the electoral vote by more than 100 votes.

Even a four-point national lead — the lowest Biden has seen — drops only Arizona and North Carolina out of his win column, using this same calculus. He still wins the electoral vote by more than 70 votes.

This doesn’t account for variations in the results stemming from the differences between Biden and Clinton, relative to Trump. There are indications, for example, that Biden is outperforming Clinton with older White voters but faring about as well with Hispanics, meaning that the variation in states like Wisconsin would be larger than in states like Arizona.

The natural response from Trump and Stepien, of course, would be that Trump upended expectations set by pollsters in 2016. And that’s true. State polling suggested that Clinton had a high likelihood of winning Wisconsin, for example, though she ended up losing it narrowly. One factor was that many state polls failed to weight their results for respondent education levels, meaning that college-educated voters were often overrepresented. Since they tended to prefer Clinton, those responses tended to skew the results.

Another factor, though, was that Trump did well among undecided voters. We could (and did) see this happening in the closing days of the campaign in the RCP average. It was a function, at least in part, of his beating Clinton among voters who disliked both candidates. He won nose-holders by 17 points, as people who didn’t like their options ended up more heavily voting for him.

This year, the dynamic is different. Polling has repeatedly shown that it’s Biden, not Trump, who is the preferred candidate of prospective voters who view both candidates unfavorably. Whether those people vote is an open question, but, if they do, it’s worth noting that Trump’s advantage with those struggling to make up their minds has eroded.

All of which is to say that Stepien’s assessment of Trump’s having regained control of the race remains more optimistic than realistic. A central lesson from 2016 was to acknowledge that weird things can happen and, therefore, that the contest is far from decided. But it’s also fair to note that Trump’s position now is weaker than it was then, in a variety of ways, and that, if that weakness holds, the odds of his winning would be significantly lower than the 2-in-7 shot he had Nov. 8, 2016.