From John Snow’s pioneering work on cholera in the 19th century until the present day, placing illness and disease within the context of a geographic framework has been an integral, if understated, part of the practice of public health. Indeed, geographical/spatial methods are an increasingly important tool in understanding public health issues. Spatial analysis addresses a seemingly obvious yet relatively misunderstood aspect of public health, namely, studying the dynamics of people in places. As advances in computer technology increase almost exponentially, computer intensive spatial methods (including mapping) have become an appealing way to understand the manner in which the individual relates to larger frameworks that compose the human community and the physical nature of human environments (streets with intersections, dense vs. sparse neighborhoods, high or low densities of liquor stores or restaurants, etc.). Spatial methods are extremely data intensive, often pulling together information from disparate sources that have been collected for other purposes such as research, business practice, governmental policy, and law enforcement. Although initially more demanding in regard to data manipulation compared to typical population level methods, the ability to compile and compare data in a spatial framework provides information about human populations that lies beyond typical survey or census research. We will discuss general methods of spatial analysis and mapping that will help to elucidate when and how spatial analysis might be used in a public health setting. This discussion will include a method for transforming arbitrary administrative units, such as zip codes, into a more useable uniform grid structure. In addition, a practical research example will be discussed focusing on the relationship between alcohol and violence. A relatively new Bayesian spatial method will be part of this example.
Background: Gis Capabilities And Prevention
A basic understanding of the capabilities of geographic information systems (GISs) is critical to the development of prevention activities because alcohol-related problems are not evenly distributed across space. GIS can be defined as a combination of computer hardware, software, spatial data (digital maps), and data with a geographic reference (e.g., alcohol outlets or crime locations) that facilitates spatial analysis. The key functions of GIS provide access to the broad spectrum of potential spatial analyses that can support the simple targeting of resources as well as the development of more complex models of spatial interactions. Both simple maps of problem rates or clusters and spatial interaction models may be useful for targeting traditional individual-based prevention programs or environmental interventions. Spatial interaction models, however, may be more appropriate for identifying the locations of events (e.g., assaults or crashes) that may be most amenable to environmental or regulatory prevention. In addition, GIS capabilities promote the development of a basic spatial/geographic epidemiology of alcohol use and related consequences, which is critical to the development of prevention programs (see Wieczorek, 2000, and Wieczorek and Hanson, 1997, for more details).
The key functions of GIS include geocoding, data overlays, reclassification functions, and distance/adjacency measures. Geocoding is a generic term used to describe the GIS function of providing a specific location to descriptive data. Geocoding applies to point data (e.g., alcohol outlet) as well as to area data (e.g., number of assaults in a census tract). Sometimes geocoding is known as address matching because the process of matching points to addresses is very common. The advent of the Census Bureau’s TIGER system has made geocoding a relatively low cost and widely available GIS function. However, professional geocoding services have developed to assist persons who are not comfortable in geocoding their own data or because of the high cost of updating digital maps based on TIGER in areas of changing population. Geocoding is the most basic of GIS functions because it transforms descriptive information into a format suitable for spatial analysis.
Key Terms in this Chapter
Uniform Grid Unit Transformation: A method for taking information from arbitrary administrative spatial units like zip codes and transforming the data into a uniform grid structure.
Geocoding: A generic term used to describe the GIS function of providing a specific location to descriptive data. Geocoding applies to point data (e.g., alcohol outlet) as well as to areal data (e.g., assaults in a census tract).
Kriging: A technique that can be used to develop contour maps (e.g., maps that show lines of equal value such as DWI rates) from a limited number of points or areas (which can be given a value at the centroid).
Spatial Decision Support Systems (SDSS): Decision algorithms and software developed for particular types of decision support that deal with geographic space. Developed with GIS technology, but without requiring programming skills or knowledge of GIS software, SDSS uses digital maps, tables, and charts that are intuitive and help reduce the amount of information processing needed to make complex decisions.
Geographical Information Systems (GIS): The geographic uses of data to develop maps and statistical relationships that help describe processes like the relationship between alcohol outlets and violence or vehicle crashes and alcohol outlets.
Spatial Clusters: A greater than expected geographically close group of occurrences or events (e.g., deaths, crashes, or alcohol outlets).
Bayesian Spatial Analysis: Spatial probabilities are often conditional and hierarchical. Bayes’s theorem provides a convenient way to compute such conditional probabilities. Because the computational impediments to computing posterior probabilities in a Bayesian setting are eliminated after the advent of Markov Chain Monte Carlo algorithms, this method is an efficient and flexible method for conducting spatial analysis.
Spatial Analysis: Using geographic data to mathematically model the relationship between measures such those mentioned previously, that is, alcohol outlets and violence.
Spatial Autocorrelation: The measure of similarity between values (for a given variable, for example, income) located in space. Similarity of values in spatial proximity may indicate some underlying mechanism that is spatial in nature and contributes to the spatial pattern of the predictor variable. Controlling for spatial autocorrelation reduces statistical bias in parametric modeling.