Counting by date of death

It always looks like the pandemic is ending

Minimizers always prefer to display deaths by the date of death. They claim that displaying the date of death is correct, while “deaths reported” plots are “bad data.” Sometimes minimizers claim that reported deaths are used to “backfill” or “launder” deaths for unexplained shenanigans.

This is the graphical version of a common minimizer trope I’ve written about: “#actually, X people didn’t die yesterday; X people were first reported dead yesterday.” It’s a dumb trope because while it’s true X people did not actually die yesterday, newly-reported deaths are the freshest data we have access to. In fact, when deaths are rising (as they were until three weeks ago), daily reports likely underestimate deaths because health departments are still working though older deaths.

When minimizers insist on plotting deaths by date of death, they can paint a rosy picture even when the colors are bleak:

Justin Hart @justin_hart

A reminder that the United States has NOT averaged 1000 death / day since the week of August 15th. What you see are deaths "layered" into the past. Yes, deaths will accumulate but right now fatality rates are still dropping as our excellent folks in healthcare are managing C19

In fact, rates were not “still dropping” as Justin Hart claimed. Deaths were going through their steepest increase since March. Over 1,200 a day died from COVID-19 around November 11. By the end of the year, more than 3,000 a day died, easily exceeding the highest rate of death in spring.

Data source: CDC Provisional Death Counts

Deaths graphed by date of death always look like they’re dropping to zero.

That’s why minimizers prefer them. While they’re slightly more accurate for temporality locating deaths, they’re much less accurate in representing the present world as a result of very incomplete recent data.

A few seconds of reflection should explain why. Mortality reports get received by some jurisdictions quickly for some deaths. But other reports lag, sometimes months.

This can be seen in the colored bars on the plot Hart posted: a small percentage of deaths (maybe 15%) are tallied by the CDC within one week, and a larger percentage over the next two weeks (perhaps 30% per week) and the final reports trickle in very slowly over the course of months.

This means, unless deaths are increasing at an absurd rate—like tripling from one week to the next—deaths will always look like they’re declining, even if they’re going up. Here’s an animated plot illustrating this principle of reporting lag with CDC pneumonia deaths:

CDC pneumonia death figures over time, by Tyler Morgan

Another example is Sweden, where the COVID-minimizing government only reports deaths by date of death. Thus, every new report shows deaths racing to zero, even when they’re actually increasing:

David Steadson 🇦🇺🇸🇪🇪🇺🌍 @DavidSteadson

The effect is that that it *always* looks like deaths are declining in Sweden - it takes about 2 weeks before the actual true situation is known. Mortality has remained high, at about 90 deaths per day or higher from abt mid-Dec to mid-Jan.

Sweden data is frequently misunderstood or misrepresented because of this deceptive reporting gimmick—so much so that Our World In Data wrote a note about it: “Why Do COVID-19 deaths in Sweden’s official data always appear to decrease?“

Bottom line: whenever you see someone insisting that only graphs showing the date of death should be used, they’re probably trying to mislead you with reporting delays. The difference between date reported and date of death is not significant when reporting is complete—the “date reported” graph tends to just be shifted a bit to the right, which makes sense. We’re always reporting on the past while trying to make sense of the present.

Orange bars are date of death, blue bars date reported. By tlowden.

We don’t make sense of the present by ignoring the recent past.

Stay away from those who insist that old data is better.