The identification of spatial clusters is an important and critical task in many scientific fields. Areas which exhibit a raised incidence of some phenomenon (e.g. disease or crime) are often targeted for increased intervention efforts, such as additional public health safeguards, increased allocations of human resources, or modification to existing public policies to deter negative outcomes. However, the ability to precisely identify significant spatial clusters continues to be challenging. Problems associated with imperfections in spatial data, geographic scale, cluster shape and size, and temporal dynamics often co-mingle to create a somewhat chaotic environment for developing reliable and robust solution approaches. Therefore, while there is no single "best" spatial clustering approach for identifying areas of elevated risk, several techniques, including spatial scan statistics, remain popular and widely used in geography, epidemiology, and criminology for identifying hot spots. This project will develop cutting-edge mathematical and statistical approaches combined with exploratory spatial data analysis techniques to provide a more accurate and precise framework for identifying irregularly shaped spatial clusters for hot spot detection. Specifically, this research will develop more rigorous contiguity and relative contiguity-based spatial cluster detection approaches for identifying clusters with maximum statistical significance while quantitatively tracking their geographic structure. In addition, a suite of innovative diagnostics will be developed to better recognize errors of misidentification, such as missing high-risk units or including extra non-significant units in the detected clusters. The goal is to bring these developed methods to bear on the problem of identifying and assessing spatial clusters over a wide range of spatial scales and application areas.
Building upon preliminary research, this team is poised to develop the next generation of spatial clustering approaches and make major advancements to the STEM fields of applied mathematics, operations research, epidemiology, and geographic information science. Further, the substantive components of this project will generate new empirical evidence to help inform local and regional public policy and public health issues regarding alcohol outlets and their relationship to violence and morbidity. Results of this project also support vulnerable populations and places that are socially and economically disenfranchised in two major metropolitan areas (Cincinnati, OH and Philadelphia, PA). Published research and participation in major international conferences, in combination with websites, forums, and sponsored activities hosted by both Drexel and ASU will enable effective dissemination of project results to a wide audience.
This project examined and developed quantitative methods for identifying spatial clusters in geographic information. A fundamental problem with many cluster discovery approaches is their use of a predefined, geometric scanning window (e.g. circular or elliptical, etc.) for detecting clusters. While mathematically convenient, disease and/or crime clusters rarely conform to a convex shape in the real-world. Further, the use of these windows predisposes spatial scan statistics to include irrelevant, low-risk areas in detected hot spots. Not only does this diminish the overall character and quality of the cluster(s) identified, it prevents many cluster discovery approaches from identifying elongated and/or irregularly shaped hot spots. Although recent developments in cluster detection approaches have displayed an ability to better identify oddly-shaped clusters, many of the existing approaches utilize poorly articulated contiguity constraints (often shape-based proxies) for identifying regions exhibiting elevated risk. This research developed contiguity and relative contiguity-based spatial cluster detection approaches for identifying clusters with maximum statistical significance while quantitatively tracking their geographic structure. In addition, a suite of innovative diagnostics will be developed to better recognize errors of misidentification, such as missing high-risk units or including extra non-significant units in the detected clusters. The substantive goal of this project involved exploration of the connection between alcohol outlets and urban violence through spatial analysis and the developed cluster discovery approaches. Specifically, while alcohol has been implicated as a risk factor for violent outcomes, and there is overwhelming evidence that neighborhood-level ecological characteristics are strongly correlated with higher rates of crime, little is known about how clusters of alcohol outlets may influence outcomes like offending, victimization and injury. More notably, there is limited empirical evidence connecting agglomerations of alcohol outlets to clusters of urban violence. Alcohol outlets and violence specifically looked at communities in Philadelphia, PA and Seattle, WA. Products: A.T. Murray, T.H. Grubesic and R. Wei (2014). "Spatially significant cluster detection." Spatial Statistics 10, 103-116. T.H. Grubesic, R. Wei and A.T. Murray (2014). "Spatial clustering overview and comparison: precision, sensitivity and computational expense." Annals, Association of American Geographers 104, 1134-1156. A.T. Murray, R. Wei and T.H. Grubesic (2014). "An approach for examining alternatives attributable to locational uncertainty." Environment and Planning B: Planning and Design 41, 93-109. A.T. Murray and Grubesic, T.H. (2013). Exploring Spatial Patterns of Crime Using Non-hierarchical Cluster Analysis. In, Leitner, M. (ed)., Crime Modeling and Mapping Using Geospatial Technologies. Berlin: Springer. T.H. Grubesic, Murray, A.T., Pridemore, W.A., Tabb, L.P., Liu, Y. and R. Wei. (2012). Alcohol beverage control, privatization and the geographic distribution of alcohol outlets. BMC Public Health. 12: 1015. A.T. Murray, Grubesic, T.H., Rey, S.J. and L. Anselin. (2012). Spatial Data Uncertainty and Cluster Detection. Proceedings of the 2012 GIScience Meeting. Columbus, Ohio. Thesis / Dissertation Ran Wei. Addressing geographic uncertainty in spatial optimization. (2013). Arizona State University. Ph.D. dissertation. Stephanie Kleinschmidt. Positional uncertainty in spatial data and its effect on cluster detection. (2013). Arizona State University. Masterâ€™s thesis. Presentations Murray, A.T., T. Grubesic and R. Wei. Spatial cluster detection and interpretation. 6th International Conference of the ERCIM WG on Computational and Methodological Statistics (ERCIM 2013), London, UK, December 14-16, 2013. Grubesic, T.H., Wei, R. and A.T. Murray. Evaluating the Spatial Precision of Cluster Detection Approaches. North American Regional Science Conference. Atlanta, GA, November 13th – 16th, 2013. Grubesic, T., A.T. Murray and L. Tabb. "Evaluating spatial precision of approaches for irregularly shaped spatial cluster detection." 2013 Annual Meeting of the Association of American Geographers, Los Angeles, New California, USA, April 9-13, 2013. Pridemore, W.A., T. Grubesic, A.T. Murray, L.P. Tabb, Y. Liu and R. Wei. Using spatial optimization modeling to examine alcohol beverage control and to estimate future dispersion of alcohol outlets. 68th Annual Meeting of The American Society of Criminology, Chicago, Illinois, USA, November 14-17, 2012. Kleinschmidt, S., Murray, A.T., Rey, S.J. and T.H Grubesic. Spatial patterns and geographic data uncertainty. Annual Meeting of the Association of American Geographers, New York, New York, USA, February 24-28, 2012. Murray, A.T., Wei, R. and T.H. Grubesic. Spatial Contiguity in Cluster Detection. North American Regional Science Conference. Ottawa, Canada, November 7 -10, 2012. Murray, A.T., Grubesic, T.H., Rey, S. and L. Anselin. Spatial Data Uncertainty in Cluster Detection. GIScience 2012. Columbus, OH, September 18th – 21st, 2012. Grubesic, T.H. Bouve College of Health Sciences. Northeastern University. Alcohol, Violence and Control. April 2012.