Search

Most spatial inquiries seek to investigate causal questions about spatial processes, but many quantitative spatial methods are designed to identify correlations and spatial patterns. Studying the structure of associations that make up a spatial pattern can provide information about what the process that generated that pattern is likely to be, but it does not provide a means of testing any one explanation against alternative explanations. Causal inference provides a set of approaches to formally make comparisons between explanations. An opportunity exists to incorporate these techniques and spatialize the theory of cause in GIScience by building on recent advances in computer science and statistics. However, implementing causal inference in geography may require a shift in the design of geographic information systems.

Regionalization refers to the design of areal zones by spatially aggregating smaller units into larger clusters. Algorithms to conduct regionalization typically require the desired number of clusters to be specified a priori, though a reasonable number is not always clear. Therefore, a heuristic is proposed to endogenously determine the number of clusters in a supervised setting (i.e., model-driven) by balancing the fit of a spatial model and the average area of clusters used as input. The heuristic is applied in a spatial interaction modeling context and a workflow is presented for integrating regionalization algorithms into larger spatial analysis frameworks.