Given Atlas’s route from conservative media to the West Wing, and given the argument that he was tapped to promote, we are therefore not surprised when we see tweets like this one:
It is a very Trump administration graph: goading political opponents, celebrating the president’s efforts — and relying on misrepresentation.
In this case, the problem is that the Y axis on the graph Atlas shared intentionally diminishes the number of deaths occurring. We understand his point: The number of deaths is low relative to the number of cases. But by presenting the data in this way, Atlas makes it look as though the number of deaths is insignificant. As intended.
It’s not that the data are wrong. It just makes it impossible to actually evaluate the number of deaths occurring. Consider this version of the same graph.
Notice anything? From July on, we actually cranked up the number of deaths by 25 percent to 50 percent. In other words, we added one death for every two that occurred — but you can’t tell, because the Y axis is so expansive. We could show, say, the number of auto deaths per day relative to the number of people who are driving, but that doesn’t really tell us much about auto deaths.
This is why we generally show the number of deaths recorded nationally on a graph like this one.
Here, we see an effect that Atlas would like us not to see: As cases go up, so do deaths.
Trump has argued that the number of covid-19 deaths has dropped precipitously — and he implies that it will continue to do so. But that’s not really the case.
There’s an understandable relationship between new cases and new deaths. Data from the COVID Tracking Project show that hospitalizations increase about a week after new cases surge. Deaths follow about a week after that.
To evaluate the relationship between cases and deaths, then, it’s more useful to see how that ratio has changed over time, not simply to compare deaths now with deaths in April, when the virus not as well understood and hospitals in New York were being overwhelmed.
Considering new deaths relative to new cases two weeks prior, the ratio has been essentially flat since early July. On a given day, the number of new deaths has been about 1.75 percent of the number of new cases 14 days earlier. That ratio varies a bit, but not a lot.
This has been consistent. On Oct. 15, when the country was seeing 680 deaths a day on average, I looked at the same metric.
“About two weeks after new cases are reported, we see new deaths that equate to between 1.5 and about 2.1 percent of the total,” I wrote then. “ … In other words, given the 51,000 new daily cases being added on average at the moment, we should expect to see between 770 and 1,080 deaths each day by the end of the month.”
The daily average on Oct. 31 was 810. On Oct. 28 and Oct. 29, the number of reported deaths each day exceeded 1,000.
I am a reporter for The Washington Post. I am not on the president’s coronavirus task force. Yet I can pick out trends suggesting how the number of new coronavirus cases necessarily leads to a predictable increase in deaths. But, then, I wasn’t tapped to work for the White House precisely because I could use my medical degree to make arguments diminishing the risk posed by the virus.
If the ratio holds, we can expect about 1,500 deaths a day by Nov. 18. If we do, remember that the best way to diminish the number is to compare it to some much, much larger number. The number of new cases each day at that point will probably serve nicely.