Identifying local spatial association in flow data

Abstract.

In this paper we develop a spatial association statistic for flow data by generalizing the statistic of Getis-Ord, G _i (and G _i ^*). This local measure of spatial association, G _ij, is associated with each origin-destination pair. We define spatial weight matrices with different metrics in flow space. These spatial weight matrices focus on different aspects of local spatial association. We also define measures which control for generation or attraction nonstationarity. The measures are implemented to examine the spatial association of residuals from two different models. Using the permutation approach, significance bounds are computed for each statistic. In contrast to the G _i statistic, the normal approximation is often appropriate, but the statistics are still correlated. Small sample properties are also briefly discussed.