Map Pairs are Bad at Visualizing Correlations

On Wednesday and Th, the Urban Institute published two posts on gun violence in America I co-authored with a colleague (post #1 and post #two). One of the data visualizations in the first post was inspired non past the outcome per se, but by another visualization I saw in the Washington Mail service that I thought did not fairly visualize the story they were trying to tell.

Importantly, on content: One of the points we made in our kickoff post is that of the more than 30,000 firearm-related fatalities in the U.South. in 2010, virtually one-tertiary (eleven,000) were classified as homicides while the residuum—more than 19,000 deaths—were classified equally suicides.

When I started looking into the literature on suicides, I came across this interesting Washington Mail service article in which the authors argue at that place is a strong correlation betwixt suicides and gun ownership rates. To visualize the signal, they present ii side-past-side tile grid maps.

As you may know, I'grand both a lover and hater of maps. I retrieve maps are terribly overused to visualize data—geographic size tends to distort the information (which is why I tend to have an analogousness for tile filigree maps) and mostly, the geographic patterns do non exist or are non completely obvious. More than chiefly, it's more than difficult to discern differences in quantities from colour than from other shapes such as confined or columns. That being said, people love maps—they are a familiar shape and can help brand the data more attainable and familiar.

In this example, however, the 2 maps do not assistance visualize the correlation betwixt the two serial, every bit the authors note in the paragraph that precedes the maps. I remade the two maps as a scatterplot (thanks to Ben Chartoff for pulling the WaPo data), sizing the circles past population and coloring them by region. (I made another plot in which I colored the circles by the Democrat-Republican vote in the 2012 Presidential Election; both versions are available in this Tableau workbook). In this version, the correlation becomes clear and the regional patterns are easier to see.

In sum, try to be careful with maps. Only because you have geographic data does not hateful yous should make a map. Or perhaps you should combine the map with another graph type that enables the reader to accurately compare the information values. Consider the story you are trying to tell and whether the map actually helps the reader.

What do you lot think? Is the original sufficient, or do y'all like the scatterplot or some other visualization blazon? The data are available here. Give it a try and post your visualizations in the comment section below or on Twitter using the hashtag #PolicyVizRemake.

schumacherdides1938.blogspot.com

Source: https://policyviz.com/2016/10/07/map-pairs-bad-visualizing-correlations/

Share this post