COVID-19 Doubling Times

(Just want the graphs? Scroll past the text.)

I see a lot of graphs of cumulative COVID-19 cases over time. While they definitely play well on TV with their dramatic upward curve, what I want from data visualization is to improve my situational awareness: are things getting better? are they getting worse? is that light up ahead the end of a tunnel, or a train? Graphing cumulative counts doesn't seem to help much with that, because the counts can keep growing for a long time even while we're bringing the situation under control.

What felt more useful to me was the rate of change, rather than the absolute number. We humans aren't great at looking at a raw-count graph and knowing the rate, though, especially as the numbers get large. So I assembled these graphs centered around the rates, rather than the counts:

• Red lines show the doubling time of deaths from COVID-19.
• Blue lines show the doubling time of confirmed COVID-19 cases.
• For Massachusetts, the orange line shows the doubling time of COVID-19 hospitalizations.
• The grey background shows the count of deaths, for context.

Doubling time is calculated over the seven-day window that ended on the date being plotted. The higher the value, the faster the number is doubling, so lower values are better. Once the number starts actually decreasing over time, the graph will show halving times (and, again, lower is better).

Two final cautions: first, I'm not an epidemiologist, I just think we're not good at intuiting about exponential growth rates. Second, when the numbers are small, the rates are very likely to be much less stable and thus much less useful for reasoning about. Use with caution. If you have questions or other feedback, you can reach me via email at <covid@kodachi.com>.

The fine print:

• No data are graphed for times during which a given location reported no deaths or cases.
• No point will be plotted if the number was constant over the whole seven-day window.
• The primary data source is the Johns Hopkins CSSE data, which is updated daily.
• Massachusetts hospitalization data is from ma.gov.
• Massachusetts and Johns Hopkins seem to have a stable data feed (again?) as of about 1 May 2020. Whew.
• Before 13 April 2020, Massachusetts reported only confirmed cases in the hospitalization counts. On 13 April 2020, they started including presumed positive COVID patients as well. New York City did the same on 16 April 2020. These changes manifest as significant jumps in the graphs.
• Massachusetts quietly restated their hospitalization counts when they released their new dashboard on 20 April 2020. The numbers graphed for Massachusetts reflect the restating.
• Massachusetts started reporting hospitalizations by county as of 14 April 2020. Data quality for the first week was pretty weird -- in particular, Middlesex and Suffolk data for 17 and 18 April are missing entirely.
• The sharp drop in Massachusetts hospitalizations on 20 April is surreal, but so far looks real based on other information I have from Partners Health -- anecdotally, their hospitalizations slowed a lot over the weekend of 18-19 April. I'm still keeping an eye on that.
• I would love hospitalization data for other regions, but I don't have it.