Radial KDE Visualization for Directed Flows
From some work I did recently for the Cincinnati Chamber:
These images summarize some fairly sketchy census migration data, showing the general direction from which or toward which people move relative to the Cincinnati region. So for example, a large red bump on the left may mean that many people are leaving Cincinnati and moving west. A blue bump to the northeast might mean some people are moving to Cincy from Cleveland or Columbus. Greens are balanced flows.
This visualization responds to a need to show some geographic dimension to data which, though detailed to the county level, has massive sampling error and a great many missing estimates. Estimating the number of migrants between, say, the Cincinnati MSA and the Los Angeles MSA is certainly possible with this data, but the estimated error are so high as to make a map of the estimates themselves nearly useless, especially with any degree of disaggration such as that seen in these images.
As long as I’ve understood margins of error myself (only a few years, honestly), I’ve understood that it should be basically hopeless to try to get lay people to understand the implications of MOE estimates.
Anyway, some interesting patterns emerge here. Note the big southward outflow of retirees for instance. Or note the high and relatively balanced interactions between Cincinnati and it’s sisters to the Northeast: Dayton, Columbus, Cleveland.
And here we see, at least, that people moving for military purposes are much less evenly distributed than people in other professions — surely the result of a small number of important military bases and a centralized cammand.
Now, the Chamber didn’t quite ask for these visualizations of course, but they’re what I produced because I didn’t want to feel responsible for any overconfident interpretations which would be inevitably arise if the data were simply mapped with census boundaries. With this presentation, the data looks like it can’t give you any real specifics, which is true.
The data comes from American Community Survey county-to-county migration estimates. Images were made with a combination of PostGIS, R and Inkscape. Some of the R code is available on GitHub.