This is a more conceptual than practical question and is about exploratory cluster identification. While there are very specific meanings in the geography canon as to what they are, in economics they seem to be somewhat loosely defined, often following Porter's lead:

Clusters are geographic concentrations of interconnected companies and institutions in a particular field. Clusters encompass an array of linked industries and other entities important to competition. Porter, M. E. (1998). Clusters and the New Economics of Competition. Harvard Business Review, (November-December 1998), 77–90.

However, I am interested in specifics and formal definitions. I work in city plannning, and there is plenty of talk about cluster x or cluster y being in existence at place z, but I would like to pin it down more and have a data-driven, rather than policy-driven inventory of any and all clusters.

Specifically, assuming a point data set representing business establishments, weighted by their employment, and I am interested in exploring, rather than confirming pre-conceived ideas of which clusters may or may not exist, which GIS approach is best suited for identifying which industries might exhibit clustering at a given scale? Is this a Ripley's K case where each industry grouping would be passed separately, or is there a more appropriate approach in the exploration stage when it is not 'known' which industry levels might cluster (depending on how one aggregates the classes, there could be several hundred distinct industries).


One type of clustering that provides inferences for whether the clusters would be found by chance are available for scan clustering techniques. These scan techniques typically simulate data according to the marginal totals over the spatial range and determine a reference distribution for the number of clusters and the density within the clusters that would be found by chance.

This is oppossed to say k-means (which just chops the data up into discrete groups) or hierarchical clustering (which you have to define how to agglomerate and cut the branches). Both of these have tautologies in saying that the clusters are real, as the routines by definition will return clusters according to the pre-specified criteria.

Types of clustering for spatial scan techniques can be found in CrimeStat (the STAC algorithm) or SatScan. For your particular application businesses would be indexed by their geographic locations and the intensity of the process would be coded as the number of employees. SatScan has the added functionality that you can adjust the Scan statistic for an underlying population.

| improve this answer | |
  • Thanks for the pointer. Will check out the STAC algorithm and see where it gets me. – ako Oct 24 '13 at 14:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.