CMStatistics 2017: Start Registration
View Submission - CMStatistics
Title: Use of the two-dimensional Kolmogorov-Smirnov test to measure spatial concentration in geospatial data Authors:  Takaaki Ohnishi - The University of Tokyo (Japan) [presenting]
Takayuki Mizuno - National Institute of Informatics (Japan)
Tsutomu Watanabe - University of Tokyo (Japan)
Abstract: Since micro-geographic data is becoming available, it is necessary to develop robust methods applicable to such data. Industry concentration is an universal property observed in most countries and at various spatial scales. The spatial concentration of industries has been traditionally measured by using cluster-based methods which is defined on a discrete definition of space. In these methods, space is divided into subunits, so the position of the boundaries and level of observation have an impact resulting in the modifiable areal unit problem. In order to avoid this problem, recent studies have applied distance-based methods which consider space as continuous. Although these methods allow an exact and unbiased analysis of the spatial concentration at all scales simultaneously, it is necessary to define the values of some parameters and the shape of the kernel function. To overcome these disadvantages, we present the use of the two-dimensional Kolmogorov-Smirnov two sample test to measure spatial concentration. The Kolmogorov-Smirnov test computes the maximum difference between two cumulative distribution functions yields a P-value. The spatial concentration can be characterized by P-value. We illustrate the proposal by analyzing Japanese telephone directory data which contains about 7 million establishments with latitude, longitude, and industrial sector information.