stepleton
Segmentsylvania
Output of a Bayesian graph segmentation algorithm on the "high resolution" 2000 US Census commuter dataset for Pennsylvania. The algorithm clusters economically related municipalities by joining together places with lots of commuter traffic between them.
Without prior knowledge of the number of clusters, the algorithm does a decent job of identifying economically linked regions around Philadelphia, Pittsburgh, Erie, Harrisburg, Wilks-Barre, and Scranton, among others. In sparser areas, the algorithm's partitions of the state match county lines---proposing in effect that people who live in rural counties tend to work in those counties. This despite the fact that input data contains no information about counties: it's simply a matrix of how many people commute to work between municipality A and municipality B.
Segmentsylvania
Output of a Bayesian graph segmentation algorithm on the "high resolution" 2000 US Census commuter dataset for Pennsylvania. The algorithm clusters economically related municipalities by joining together places with lots of commuter traffic between them.
Without prior knowledge of the number of clusters, the algorithm does a decent job of identifying economically linked regions around Philadelphia, Pittsburgh, Erie, Harrisburg, Wilks-Barre, and Scranton, among others. In sparser areas, the algorithm's partitions of the state match county lines---proposing in effect that people who live in rural counties tend to work in those counties. This despite the fact that input data contains no information about counties: it's simply a matrix of how many people commute to work between municipality A and municipality B.