Thursday, June 11, 2020

Mapping COVID

Please, not another COVID map!  I get it.  Newspapers, universities, basement analysts and the like have all been producing COVID maps ranging from reasonably informative to downright awful.  But none is, how shall we say, exactly perfect.

Rather than create any more, I'm just going to try to improve some of those that I think are pretty good to begin with. 

Here's the NYT map tracking changes in new cases - each county gets a color based on the difference in new cases per capita from today compared with two weeks ago. Blue is falling, red is rising fast.


So what's good:

  • It's pretty!
  • You can see some clear national trends: the hard hit northeast is almost universally declining; huge swathes of the mountain west are barely hit at all; the west and parts of the southeast are rising
But the main problem is that coloring a county based on per capita effects means that county with a large area and a small population can draw the viewer's attention more than warranted.  
    • Madison (15.7K) and Washington (203K) counties in Arkansas create a giant red block in the middle of the map.  That's 3 more cases in Madison and 50 in Washington.  
    • Dodge (88.8K) county in Wisconsin accounts for 18 more cases in the most recent period compares with the previous two weeks.  Compare that with the tiny blue dot of Milwaukee (947.7K) representing a decline of 47 cases.  
    • The pacific northwest looks like a hot spot, but Lincoln (47.7K) county, Oregon accounts for only 19 more cases.
Sparsely populated counties are therefore driving what the viewer sees, but it's generally tiny dense counties that account for most of the changes.  As such, you can't really tell what's going on the the southeast or midwest or where the real hotspots are.  

So while the map looks pretty, it's really presenting your brain with a whole lot of noise.  Let's try fix the major issue by giving each increase or decrease the same amount of ink (or pixels).  Each county is filled with the number of blue or red dots corresponding to the total decrease or increase in cases:

OK, so their map is prettier, fine.  Let's look at the major trends.  We see the same sharply declining cases in the northeast; rising cases in the southwest; overall clear downward trend in the midwest despite a few patches of red;  NC, TN, and FL are clearly increasing.

Let's compare side by side to highlight how the NYT map misleads us a bit:
  • The pacific northwest looks like a hotspot in the NYT map, but is almost irrelevant in ours.  There simply isn't much change there and it shouldn't attract our attention
  • IA, WI, MN, and IL are clearly trending downward in ours, whereas it looks at best a mixed bag in the NYT map
  • FL is increasing in both, but in our you can see it's driven by Miami and the very southern part 
  • AK is increasing in ours, but not the attention grabber it is in the NYT map
  • The GA counties that grab our attention in the NYT map aren't actually the story - it's really big decreases in Atlanta combined with gradual increases around the state
  • The southwest story is pretty much the same, but we can see pretty clearly in ours that it's LA and the south driving the increase
  • Ours makes it clear the NM is a big nothing burger and CO is is declining, things that you wouldn't divine from the NYT map
If you want wall art, the NYT map is the better one, but it comes at the expenses of your brain learning the wrong visual messages.  I'd love to see the high quality maps shown on the NYT take the issue under consideration.