The Washington PostDemocracy Dies in Darkness

The White House’s self-serving approach to estimating the deadliness of the coronavirus

President Trump speaks to reporters Tuesday at the White House. (Patrick Semansky/AP)

When a draft report prepared by the Centers for Disease Control and Prevention emerged on Monday, the White House was quick to disavow one of its most contentious determinations: that the number of deaths from covid-19, the disease caused by the novel coronavirus, would pass 3,000 a day by the end of May. Such a surge would mean more than 100,000 deaths overall by that point, prompting the administration of President Trump to quickly point out that the data “is not reflective of any of the modeling done by the task force” focused on battling the virus.

What Trump and the White House stand by, however, is disconcerting in another way.

Shortly after the CDC report leaked, a leading model from the University of Washington’s Institute for Health Metrics and Evaluation was updated to indicate that more than 134,000 people might die of the virus before the end of July. That was a significant increase from the IHME’s most recent prior version of its model, which showed fewer than 80,000 deaths in that same period.

Shortly before boarding Air Force One for an event in Arizona on Tuesday, Trump was asked about the shift in the IHME model. After all, his administration had at one point used the IHME data to make its case for social distancing measures aimed at limiting the virus’s spread. Trump was asked if he was willing to accept the new projections.

“Those projections are without mitigation,” Trump replied. “We’re doing a lot of mitigation.”

He then quickly transitioned.

“And we have to keep our country — our country wants to open,” he said. “The governors — it’s in their hands, but our country wants to open. … I created the greatest economy in history, the greatest employment numbers, the greatest success in history. And then one day, we had to close it down. We’re going to beat those numbers, and I’m going to beat them soon.”

Setting aside the inaccurate assertion about the “greatest economy,” Trump’s mid-reply reversal of emphasis is remarkable. His push to get the economy running again is, after all, directly opposed to strong social distancing efforts. To claim that the country is “doing a lot of mitigation” and also that the economy would be back at full strength “soon” is like claiming that your favorite dessert is ice cream and also that you’re severely lactose-intolerant. The two simply don’t overlap very well.

More importantly, though, Trump’s claim about the IHME model is misleading. The estimate that put the number of deaths under 80,000 assumed that full mitigation measures would be in effect through May. The revised estimate incorporates consideration of precisely what Trump is referring to: a relaxation of distancing measures aimed at restarting economic activity. The model has a nuanced consideration of where and how different measures are in place and has determined that an increase in new cases and deaths will result. Trump seemingly considers no such nuance in his presentation.

Again, the IHME model was one of the components of the White House coronavirus task force’s estimate of the effects of the pandemic. The model from late March and early April assumed a more rapid drop-off in the number of deaths each day from the virus than the most recent iteration. The recent iteration indicates a continued upward trend in early May, meaning tens of thousands more deaths. (We’ve included an important consideration below: the range of uncertainty included in each model.)

The IHME daily death toll data indicated by the solid orange line doesn’t match entirely with other data sets. The data compiled by Johns Hopkins University, for example, shows a higher spike in April (in part due to its incorporation of presumed deaths in New York) and a lower current daily tally.

There is some unevenness in when numbers are calculated and what’s included. There is also, as we noted, uncertainty in major models of what to expect, including the IHME’s and the one presented in that CDC report (which was created by a Johns Hopkins researcher). You can see that uncertainty illustrated in red and gray on the graph included in the report.

And then there’s this, from Trump’s Council of Economic Advisers.

This graph shows what The Washington Post first reported was being referred to as the “cubic model,” apparently deployed by CEA Chairman Kevin Hassett. You can see the “cubic fit” on the chart above; it shows the number of deaths each day dropping precipitously over the next few days and landing at zero by the middle of the month.

Even before the CEA acknowledged its “curve-fitting exercise” in a tweet on Tuesday, FiveThirtyEight’s Nate Silver speculated that the team had created a rough fit using a third-degree polynomial trend line in Excel. He included a graph showing that, sure enough, doing so would result in deaths hitting zero by mid-May.

If the preceding paragraph doesn’t make sense, the key part is the relative simplicity of it. The IHME’s presentation of what it includes in its most recent iteration of its model runs more than 5,300 words. Clicking four buttons in Excel is somewhat less refined.

It’s also subject to the assumptions one makes. For example, here’s a “cubic fit” using the seven-day average of daily deaths from Johns Hopkins, running back one month from May 4. The line hits zero deaths in a bit over a week.

But if we simply use different ranges of data, the line goes all over the place. In some iterations, the variations in the daily seven-day average yield quickly spike upward — presumably not the sort of result anyone would want to see, the CEA included.

Jason Furman, who held Hassett’s position under President Barack Obama, made a similar point on Twitter.

“The ‘cubic fit’ is based on an approach to epidemiology that has long been absent from any serious epidemiological discussions,” he said. “It made terrible predictions back in March and April. The functional form was chosen to get the result they wanted.”

That last point is clearly right. Perhaps something more sophisticated occurred than that someone working for the CEA fiddled with a graphing tool until the number of deaths each day reached zero. But by its own admission, they were using “curve fitting” to “improve data visualization” — not, it seems, to necessarily predict what might happen.

The IHME model isn’t perfect and has been continually revised since its launch to incorporate new information. That’s the goal, though, to better establish a tool to understand what’s likely to follow. When social distancing measures are relaxed, the tool thinks that more deaths will follow.

For Trump and his team, the goal is something else. It may simply be to push for a quick reversal of economic fortunes and to nudge unhelpful curves until they support making that reversal happen.

It’s an academic exercise, to some extent, but one in which errors are measured in lives.