Hello, good afternoon, let's do another mappy Twitter-torial, shall we? Today's topic is visualizing aggregate data. Mostly it's a critique of this map, which you've probably seen quite a bit. It has great data! But I take issue with the techniques used. arcgis.com/apps/opsdashbo…
We are all familiar with maps that visualize aggregate data. The most common one you've probably seen is a "choropleth" map -- a map of areas that are filled in with a color or pattern that tell you something about that area. Like this electoral map from the BBC: 
The colors on the map share information: which party won the district, and by what percentage. Choropleth maps have many of their own issues (making assumptions about large geographic areas), but are useful for showing data aggregated at an area level.
There are two other common ways to show aggregated data: a "dot density" map and a "graduated symbol" map. Dot density maps use one dot to represent some number of things you're mapping, and cluster the dots in geographic areas. Here's one cool example from Bill Rankin (2009)
Each dot represents 25 people in Chicago. That dot is not providing a specific geographic location of those 25 people, but a random location inside that geographic area (along with the other dots in the geographic area. Striking, no? It does a nice job of telling the story.
Finally, we have graduated symbol maps (aka "proportional symbols"). These use a symbol (often a circle) to represent a geographic area, and the size and/or color of the symbol is related to some aspect of the data. Example: the Johns Hopkins coronavirus map.
In the case of this map, each dot represents a state/province or country (depends on the region, in the US it's states) and the radius of the circle tells us how many confirmed cases are in that area. (Also fun: note that the scale isn't linear, but logarithmic!)
What is this map trying to tell us? It's trying to convey the total number of cases of COVID-19 across the world. Does it do a good job of that? Yes and no. Here are some of my critiques of this map.
1. Knowing the total number of cases in an area is useful. Knowing the number of cases within a political boundary is also useful... to a point. If I live on the border of Washington and Oregon (say, in Portland), that state border doesn't mean that much to me.
In terms of containment and state response, yes it feels like it's helpful to know how many cases are in a state. From an epidemiology perspective, viruses don't really care about state borders. This touches on a concept called the Modifiable Areal Unit Problem.
The Modifiable Areal Unit Problem is about how changing boundary areas can fully change the results of a spatial analysis. This leads to an ecological fallacy: the idea that these borders matter when it comes to this data. (For more, this Twitter-torial: )
So, in my humble opinion, number of cases per geographic area is useful, but only in terms of that geographic area's official response.
2. This is a slippy map, meaning I can pan around and zoom in and out. So useful! It also means that I'm going to zoom right into my hometown to see if there are any cases here. Right now I live in New York City, where I know there are cases. But what happens when I zoom in?
Nothing there! You know why? Because the graduated symbol denoting the number of cases in the state is placed on the geographical centroid of the state, triangulated roughly between Binghamton and Oneonta.
The centroid of an area is the mean position of all its points -- not quite the "center" of the area, but pretty close. If I was making a print map, this would make total sense. It's a lot easier to tell what's going on when you're zoomed out! But...
if I didn't know about graduated symbols, how these maps are made, what they're trying to tell me... I might assume that NYC is fine and all New York cases are centered around the town of Downsville, New York. That's where the marker is!
On top of that, the graduated symbols disappear once you zoom in past zoom level 7. So if I did believe that all New York cases were located there, I wouldn't even really know where "there" is because the symbol disappears before I can zoom in that far.
So. How do we make this map more clear? What steps can a cartographer take to tell this story better?
One option is to set zoom bounds. Don't let the user zoom in past a certain level. This signals that more fine-grained data doesn't exist and makes it clear that the graduated symbols represent an aggregate.
But honestly I'm also stuck on what this map is trying to show/tell. Showing confirmed cases aggregated by state or country doesn't seem that useful to me, as an individual. What about cases over time? How might that be visualized? What about population-adjusted stats?
I know this data is coming in hot and fast and we don't know how reliable it is, so I understand that it's important to get any information out there that helps the world understand what's going on. The problem is that so many maps cause active harm by telling the wrong story.
General advice: When you aggregate data, you're necessarily abstracting out details and nuance in that data. It is our responsibility as cartographers to make sure we do that responsibly and ethically, in particular by sharing our methodology and explaining our results.
Also, MAP LITERACY IS IMPORTANT AND WE SHOULD KEEP TEACHING IT TO PEOPLE, EVEN IN THE AGE OF GOOGLE MAPS. Maps are historical tools of oppression, and it's up to us to be critical of data visualization of all kinds rather than taking it at face value.
