Graphs and the abstraction of geography

A project related to voting patterns in NC USA. There are 100 counties in NC and each has a board of commissioners composed of a very small number (five to nine) of elected persons. Interest has arisen in some chapters of CodeForAmerica (such as (CodeForDurham) in describing the racial and gender representativeness of the boards compared to their counties. There are very substantial statistical issues stemming from the small board size and also some data quality challenges. Setting those aside for resolution for now, I would like to use a graph realization that resembles the abstraction of a GLMM, but is, of course, intrinsically visual. Having problems with interpreting GLMMs, I hope that an interactive visualization would contribute to understanding. My concept at this time is to declare counties as nodes, with properties such as population, racial percentages, selected socio-economic quantities, perhaps using adjacency or proximity (county-to-county), and other quantities that would fit into a GLMM. There would also be the node for the board associated with its county via an edge that would have one or more properties related to representation (entropy index, Euclidean distance, something else ...). The graph would have little if any resemblance to a map of NC. My feeling is that a robust visualization, easy to use, that is, to experiment with, would be essential. My personal competence is with R and conceptual issues. I am a very slow programmer - fiddle a bit here, jiggle a bit there, never quite satisfied. While I can, at least in principle, deal with D3 etc., the results would not be very interesting and would take almost forever to produce. I will be putting my work into a git repo soon; I intend that all the R and Rmd scripts, as well as the data, will be open to all. If anyone with a great deal of patience would like to participate, please let me know.

1 Like