|
Abstract:
|
Many different problems in applied statistics can be viewed as special cases of a class of problems known as graph partitioning, i.e., the issue of dividing a graph into distinct sub-graphs while maintaining the original contiguity structure. These problems include, disease mapping and hotspot detection, spatially-constrained clustering, the construction of flow diagrams, modeling of eye movements. and community identification in social networks (Girvan and Newman 2002), among others. To illustrate our approach, we will apply our methodology to two recent marketing data sets. First, we will utilize our approach to perform spatial clustering on Netgrocer sales data in the United States to help understand the cross-zone correlated nature of the e-tailer's diffusion pattern. Next, we will apply our approach to construct flow diagrams of in-store movement using RFID data.
|
