Sometimes the Best Map is not a Map

Forgive the click-bait title, but I recently ran into a dataset that reminded me that the best way to visualize geographic data isn’t always a map. There are many different ways to graph data, each suited to a different purpose.

I’ve been working me through the (excellent) online ESRI Cartography course, which features a selection of interesting datasets, including results from the 2015 UK general election. One course exercise – plotting turnout in the election – highlighted different ways of categorizing continuous data such as natural breaks, equal interval, or geometric. What struck me about the map (below, left) was that larger constituencies appeared to have higher turnout. (Protestant Northern Ireland is an obvious exception to this trend.) But it was hard to know for sure from the map by itself.

The first variation I tried was to map population density, or more specifically “electorate“ density, since the data had the number of eligible voters, not total population (above, right). All that map did was confirm that small constituencies had high population density. Which isn’t a surprise since all the constituencies are designed to have a similar total population. Eyeballing the two maps next to each other, it looks like the sparse constituencies had, on average, higher turnout than the dense ones. But was that real? Or simply an artifact of the large, low-density constituencies visually dominating the maps?

Which is where displaying spatial data as a graph works better than keeping the data on a map. Here’s two scatterplots showing turnout compared to the area of each constituency. The first graph uses a linear scale, which bunches up the majority of constituencies along the left axis. Which shows how even this simple scatterplot is complicated by the huge range in the size of constituencies — from 8 or 9 to over 10,000 square kilometers. The second graph switches to plotting area on a logarithmic scale. Each tick on the scale is ten times larger than the previous one. This makes the relationship more clear (greater turnout in larger constituencies), although it’s still not a huge effect.

Exploring data in one format (a map) piqued my curiosity, but I needed to switch to a different visualization method (a scatterplot) to better understand the data. And even with a scatterplot I needed to transform the data from a linear to a logarithmic scale to better see the relationship between turnout and area. I needed to experiment with a variety of formats to better see the relationships in the data.

If you’re curious about cartography or want to learn ArcGIS I think the ESRI Cartography course is well worth your time. Kenneth Field and John Nelson are the headliners, but Edith Punt steals the show. It’s inspirational while covering the nuts and bolts of making maps in ArcGIS Pro.

Links

ESRI Cartography MOOC: https://community.esri.com/t5/education-blog/register-now-free-esri-cartography-course/ba-p/1401181

Categorizing data (John Nelson): https://www.dhirubhai.net/pulse/truthful-mapping-john-nelson/

Log scales (Lisa Charlotte Muth): https://blog.datawrapper.de/weeklychart-logscale/