WPRI’s False Death Chart and Games with Models, 5/29/20

RI-COVIDdeaths-perday-7dayaverage-052920-featured

Over the past decade or so, data visualizations have almost become part of the WPRI news department’s brand, with Ted Nesi taking the lead.  The idea of visualizations is to help the public have clarity on complicated and/or emotionally challenging public policy issues.

That’s why it’s especially perplexing to see the reporters using data visualization with false information about COVID-19.  Consider this tweet from Nesi earlier today (emphasis added):

A few stats that stand out to me in today’s @RIHealth data

• Test positivity rate down to a new low (3.7%) on a day with 3,000+ tests reported

• 20 new hospital admissions – bears watching

• 9th straight day with a double-digit death toll

That last bullet just isn’t true.  The “death toll” of a day is the number of deaths on that day.  What WPRI is reporting is the number of deaths being reported each day, and that includes backwards revisions.  From yesterday to today, the state revised death numbers back to May 7, which is three weeks ago.  Nine of the 16 newly reported deaths actually happened two or more weeks ago.

What difference does this make?  Well, let’s visualize some data.  Here’s the WPRI chart that Nesi is using to substantiate his claim.  Note again that they’re charting the deaths on the day that they’re reported by the state, not the day that they occurred:

WPRI-EliSherman-COVIDdeaths-052920
This chart makes it look like deaths were decreasing until about a week ago and are now on the increase… with the 7-day average almost back to its early-May peak.  But that’s simply not the case.  Here is the chart as it appears if we assign the deaths to the day that they actually occurred.

RI-COVIDdeaths-perday-7dayaverage-052920

What is actually happening is that deaths, like hospitalizations, are decreasing.  In fact, the numbers are back to early-to-mid April. Additionally, contrary to WPRI, no single day has seen more deaths than 22, with only two over 20, not the six shown by WPRI’s chart.

To be sure, we should have some skepticism that the numbers will be revised again, but as Rhode Islanders look at the news reports to develop their own “reopening” comfort levels and evaluate the governor’s policies, charts that give the false impression that deaths are on the rise are not helpful.

Turning to the daily update of how well (or poorly!) my predictive model is performing against reported numbers, here’s the chart for today.

COVID19-hospitalizationsandprojections-052920

(See here for my original methodology and here for a subsequent modification I made.)

Projections versus actuals (date of report).  Remember that the projected numbers were based on what was reported as of yesterday, so revisions to earlier reports would make the gap bigger.

  • Cases:
    • Projection for 5/29: 14,586
    • Actual for 5/29: 14,635
    • Projection for 5/30: 14,724
  • Hospitalizations:
    • Projection for 5/29: 212
    • Actual for 5/29: 219
    • Projection for 5/30: 208
  • Deaths:
    • Projection for 5/29: 681
    • Actual for 5/29: 693
    • Projection for 5/30: 697