Spatial association detector (SPADE). Issue 10 (3rd October 2018)

• Record Type: Journal Article
• Authors: Cang, Xuezhi; Luo, Wei
• Is Part Of: International journal of geographical information science. Volume 32:Issue 10(2018)
• Page Start: 2055
• Page End: 2075
• Publication Date: 2018-10-03
• Subjects: Geographical detector -- spatial association -- multilevel discretization -- spatial variance; Geography -- Data processing -- Periodicals; Information storage and retrieval systems -- Periodicals; Géomatique -- Périodiques; Systèmes d'information -- Périodiques; 910.285
• DOI: 10.1080/13658816.2018.1476693
• Languages: English
• ISSNs: 1365-8816

Abstract

ABSTRACT: The geographical detector model can be applied to either spatial or non-spatial data for discovering associations between a dependent variable and potential discrete controlling factors. It can also be applied to continuous factors after they are discretized. However, the power of determinant (PD), measuring data association based on the variance of the dependent variable within zones of a potential controlling factor, does not explicitly consider the spatial characteristics of the data and is also influenced by the number of levels into which each continuous factor is discretized. Here, we propose an improved spatial data association estimator (termed as SPatial Association DEtector, SPADE) to measure the spatial data association by the power of spatial and multilevel discretization determinant (PSMD), which explicitly considers the spatial variance by assigning the weight of the influence based on spatial distribution and also minimizes the influence of the number of levels on PD values by using the multilevel discretization and considering information loss due to discretization. We illustrate our new method by applying it to simulated data with known benchmark association and to dissection density data in the United States to assess its potential controlling factors. Our results show that PSMD is a better measure of association between spatially distributed data than the original PD.