This article has been updated.

When President Trump first unveiled his administration’s guidelines aimed at limiting the spread of the novel coronavirus, his team presented a graph showing how bad and how good — speaking relatively — the pandemic could be in the United States. Without any effort to limit the spread, the graph suggested, at least 1.5 million people might die and perhaps as many as 2.2 million. Practice social distancing and limit person-to-person contact? The number might be closer to 240,000 — or as few as 150,000 deaths.

Those estimates were based on models assessed by the White House coronavirus task force, including one from Imperial College and one from the University of Washington’s Institute for Health Metrics and Evaluation. Trump subsequently shifted the way he talked about those numbers, insisting for weeks that the actual toll might be in the range of 50,000 to 60,000. It won’t be, obviously: The United States has already recorded more than 67,000 confirmed deaths from covid-19, the disease caused by the virus.

On Sunday, Trump offered a new number. Maybe, he said during an interview on Fox News, the death toll would be somewhere between 75,000 and 100,000. That was obviously not going to be the case for a number of reasons, as we wrote on Monday morning. But now we know that Trump should have known that such a toll wasn’t likely, since his own administration says it won’t be.

The New York Times obtained a briefing document produced last week that showed, among other things, the government’s own projections for the number of people expected to die of the virus each day. The key graph looks like this.

There are a lot of unexpected aspects to that graph. One is that the actual reported deaths each day have been consistently not only above the mean projected value for the number of deaths but, in fact, beyond even the widest range of uncertainty in the model.

Put simply, if those blue dots fall outside the gray area, there is “some issue with the model,” as Washington Post data guru Lenny Bronner put it on Twitter. Put more explicitly, the model the government is using has been projecting less than half the daily deaths we’ve actually been seeing. The Post later reported that, in fact, the estimates do not constitute a complete model of what’s expected.

That’s important context for the other surprising aspect of the graph. It indicates that the number of daily cases will top 3,000 by the end of May. The way in which the data are presented makes the increase that the model expects seem more muted than it actually is. We took the graph and estimated the actual values represented by the red line, creating a visualization that uses a traditional y-axis to show what the government appears to think will happen.

The number of deaths per day would begin to climb quickly by the middle of the month. The cumulative death toll would then surge as it did.

(To determine the number of cumulative deaths under the government projection, we applied the daily increases from May 1 onto the cumulative number of deaths recorded by Johns Hopkins University on April 30 and thereafter incremented the total based on the model’s daily projections.)

Why the projected change in the middle of the month? It’s not clear. It’s possible that this includes an estimate of the effects of relaxing social distancing measures in the states. In 22 states that haven’t relaxed stay-at-home and other mandates (as here), the number of new daily cases in the past two weeks has increased by 21 percent on median. In the 24 that have partially reopened, the median increase in daily new cases has been 33 percent.

What’s interesting about the current number of recorded deaths is that it actually tracks fairly well with the IHME model used by the government in March and April to determine the results of social distancing measures. The IHME model has moved around a lot, but the number of cumulative deaths has tracked with what the model showed in early April.

It’s worth noting that, as of Monday morning, the IHME model estimated that there will be 72,433 deaths by Aug. 1 if trends were to hold — with a slowdown in daily cases over the first half of May. By Monday afternoon, news broke that the IHME estimate would be adjusted upward significantly, to nearly 135,000 deaths, in part due to the relaxing of distancing guidelines.

As Bronner noted, the model used for this estimate seems imperfect, to put it generously. But there are other factors we already know will push the total number of deaths higher. The government will eventually generate an estimate of the disease burden of covid-19 that includes consideration of apparent undercounts in deaths. The models being considered here also look only at the first wave of infections. If the virus is still spreading in the wild this fall and winter, we can expect a possible resurgence in cases that complicates health-care resources strained by the seasonal flu.

That graph Trump and his team presented in March, the one that figured that as few as 150,000 people might die, didn’t include a time frame. The chart released from the administration figures that we’ll get more than two-thirds of the way there about three months after the first recorded death in the United States.

clarification

Further Post reporting offered more insight into the genesis of the graph. We also learned that the IHME estimate would be revised after this article published.