In my last post I demonstrated an algorithm that splits areas of the United States (lower 48) into subregions based on US Census population and boundary data. A main feature of that is using k-medoids to cluster adjacent census geographies, where adjacency was determined by a Delaunay triangulation of polygon centroids rather than by physical adjacency. Additionally, edge weights for the k-medoid calculation were re-weighted so that edge weights increased proportionally with real-world distance, but decreased quadratically as the populations of the linked nodes increase. This weighting is designed to capture the real-world effect of lower distance increasing the likelihood of real-world connections. Though the previous post focused on the counties of Long Island, this one will focus on actually separating the whole US into regions. ...
Redrawing the Boundaries of Long Island's Counties
I have created an application, available on GitHub, that allows me to split areas of the United States into subregions based on US Census population and boundary data. One of my first uses of this app was to replace the four counties of Long Island into four subregions based on my app’s algorithm rather than by the historical accidents that created the four extant counties. Background Long Island has four counties: Kings, Queens, Nassau, and Suffolk*. Portions of the boundaries of these counties are demarcated by physical features, such as the Newtown Creek between Kings County (the borough of Brooklyn) and Queens, or Cold Spring Harbor between Nassau and Suffolk. But mostly the counties are separated by political boundaries which were drawn over 350 years ago to establish county and town-level governing institutions in the colony, and later the state, of New York. ...