The “On…” series is a collection of short blog posts relating to data visualization, economics, presentation skills, or data communication. In each, I discuss an issue, concept, or idea that I have not fully developed, a work in progress, or just some thoughts about a topic or issue I’d like to share.
Like every media organization right now, NPR has its own coronavirus tracker that includes different visualizations of COVID cases and deaths. The first visualization is a hex grid map that shows changes in daily cases for each state in the US. Last week, I asked my Twitter network whether they found the uncentered legend bothersome.
Does the off-centered legend bother anyone else or is it just me? (from https://t.co/xd545YgvfW) pic.twitter.com/g5vkuHKMEq
— Jon Schwabish (@jschwabish) August 7, 2020
I should clarify here: my question related not to the positioning of the legend over the map—though that now bugs me too (and for which I ruined some people’s weekends)—but to the fact that this diverging color palette is not centered over zero. Instead, zero is contained in the pale yellow segment (between -5% and +5%), the third block from the left.
I’m not sure that the asymmetry is wrong but it does bother me, primarily because it throws off my view of the extremes in the range. For example, if you look immediately at the extremes, your instinct is likely that those two segments should capture symmetric ranges, here, 100% or more and -100% or less.
In his wonderful book, Cartography, Kenneth Field doesn’t argue that diverging color palettes necessarily be symmetric:
Diverging schemes are appropriately used where the data has different extremes that might be best presented with different hues. A diverging scheme emphasizes the midpoint critical class with al light colour and then the two extremes with two diverging hues. This sort of scheme can also be used to represent ’no difference’ or ‘no change’ as well. Diverging schemes can also use a critical break approach.
I wonder, though, should they? Should a diverging color palette necessarily be symmetric? Even if no data exist in that category—and note, there are currently no states in the NPR map in the far left category—should a diverging palette necessarily be balanced so the reader can more easily compare on either side of the midpoint?
Carni Klirs (of the History of Fugazi infographic fame, among other things), made two suggested changes to the legend:
I see a couple options for improvements. Treat them like categories, rather than a diverging palette (in which case you’d expect the legend to be symmetrical), or make the uneven number breaks more obvious. That dark red category seems useful to denote *really* bad pic.twitter.com/Go5KaVHsO3
— carni_dc (@CarniDC) August 7, 2020
Carni’s second option makes the differences clearer, I think, but I might more simply just make the legend diverge around 0% and call it a day.
Another consideration is whether having unequal segments—that is, 5% for the inner two, 45% for the next pair, and so on—hides certain changes or not. Maybe a continuous diverging color palette in which we don’t have to make any of these arbitrary decisions at all would be a better choice.
and unless there’s something substantively important about 50 pt categories sitting along side 10 pt categories, a 50 pt category crazy HUGE: 6% gets same color as 49%, but 49% gets different color than 51%?? Justify THAT
— Rob Santos (@_Rob_Santos) August 8, 2020
At its heart, this is a binning decision and most binning decisions highlight an aggregation problem—one set of geographic units gets put in one group, say 0-25%, and another goes into the 25%-50% bin. Is there a big difference between 24% and 26%? Is there an important difference between 26% and 49%. In the case of the former, the two units get two different colors and in the latter they are the same color.
Last thing, not on the legend itself but on the choice of using the hex map. NPR, you might recall, is a fan of the hex grid map (see this previous blog post by David Napoli on how to build the hex map in Excel) but if you don’t like hex maps—as my Urban Institute colleague Claire Bowen noted on Twitter—it might be worth your time reading this blog post about why they can be useful. This, and other issues related to different map projections, cartograms, and bins are covered in more detail in my forthcoming book, Better Data Visualizations.
What also bothers me is the positioning of the states. I thought they forgot New Mexico (people genuinely think that it’s part of Mexico). Lots of other states are off too.
— Claire McKay Bowen (@ClaireMKBowen) August 8, 2020
So, what do you think? Should diverging palettes be symmetric on both sides? Should each unit contain the same interval? And while we’re at it, what about hex maps—like ‘em or hate ‘em?