Complex Diagrams

1

A friend asked me what I thought about the NYT Mapping Migration visualization.

It’s interesting that they note this is a new (experimental) kind of visualization; frankly I think it doesn’t work very well.

To review, you (well, I) do visualization like this (one tweet version; hours long version).

In this case, it feels more like they had a structure they wanted to use (well, two structures, treemap and geographic map, they wanted to mash up and use). So that’s a fundamental design failure because structure drives meaning so strongly, you really ought to pick structure in response to your purpose, not structure first. But this seems structure-driven, not purpose-driven.

If I were to communicate this same information I’d say the purpose is to show the relative proportions of origins of each state’s population. Now a geographic map is a bad way to talk about population in general, because geographic size has nothing to do with population (NJ has about 10x the population and 1,000x the population density of Alaska), and so you get all kinds of accidental distortion. However, geographic maps are really good at showing things like regional trends, so there may be some value there.

So how to show proportion per state and also maintain regional relevance? For proportion a classic tree map (subdivided rectangular area, not Voronoi), or even pie graph, per state could work; they almost got that right. Instead of using the geographical shape of each state, each state could be represented with a size proportional to its population with a cartogram, similar to how the electoral vote results maps work.

The result would be the largest square for California, smaller squares for other states. Each state square would be subdivided into regional areas, each area proportional to population origin and colored as they have them here. If it was me, I’d use consistent placement for the colors, yellow/west always on the left, red/east always on the right, etc.

If you wanted to do something other than a tree map, bar graphs (either stacked or side by side) per state would work perfectly well too, but that’s a little harder to implement and keep the geographical relevance.