By Michael F. Goodchild, Center for Spatially Integrated Social Science, University of California, Santa Barbara

This shows a GeoDa analysis of Jakarta, Indonesia, neighborhood consumption patterns combining dynamically linked maps and graphs, including an outlier map, a cluster map (local spatial autocorrelation), a scatter plot, and a conditional box plot. (Image courtesy of Luc Anselin.)

Space is what GIS is all about, and business knows the critical importance of the three Ls: location, location, location. But for many social scientists, location is just another attribute in a table and not a very important one at that. After all, the processes that lead to social deprivation, crime, or family dysfunction are more or less the same everywhere, and, in the minds of social scientists, many other variables, such as education, unemployment, or age, are far more interesting as explanatory factors of social phenomena than geographic location. Geographers have been almost alone among social scientists in their concern for space; to economists, sociologists, political scientists, demographers, and anthropologists, space has been a minor issue and one that these disciplines have often been happy to leave to geographers.

But that situation is changing, and many social scientists have begun to talk about a "spatial turn," a new interest in location, and a new "spatial social science" that crosses the traditional boundaries between disciplines. Interest is rising in GIS and in what GIS makes possible: mapping, spatial analysis, and spatial modeling. At the same time, new tools are becoming available that give GIS users access to some of the big ideas of social science.

Writing in Scientific American, economist Jeffrey Sachs and his colleagues Andrew Mellinger and John Gallup ask the basic question: "Why are some countries stupendously rich and others horrendously poor?" (Sachs, Jeffrey D., Andrew D. Mellinger, and John L. Gallup, "The Geography of Poverty and Wealth," Scientific American, March 2001). They go on to combine GIS analysis with the methods and equations of macroeconomics to show that location matters: where you were born globally has a lot to do with your chances in life. At a much more detailed spatial scale, geographer Danny Dorling and his colleagues have shown that location makes an increasing difference to your chances of early death in the United Kingdom (Yamey, G., "Study Shows Growing Inequalities in Health in Britain," British Medical Journal, 1999, Vol. 319, p. 1453).

One of the strongest arguments for looking at society through a spatial lens—through maps, GIS, and spatial analysis—is that it provides observations with context: processes can be examined in their geographic settings. A criminologist looking at community crime rates might otherwise miss the recent increase of policing in a neighboring community, which simply displaced the crime that had previously occurred there. This type of spillover process can be analyzed using a variety of methods that have evolved in the social sciences in the past decade or so under the general rubric of autoregressive models. Luc Anselin at the University of Illinois is a world leader in this area; working with the Center for Spatially Integrated Social Science (CSISS), he recently released GeoDa, a new suite of tools for this and other types of spatial analysis that is fully compatible with Esri products (downloadable at www.csiss.org/clearinghouse/GeoDa, along with extensive documentation and tutorials).

Distance Decay

Spillover is subject to the much more general principle of distance decay, which dictates that human interaction declines steadily and often predictably with distance. Despite the power of the Internet to link people across space, physical distance is still a major determinant of human interaction. Predicting is never easy or perfectly reliable for any social process, but market analysts and others are well aware that predictions can often be reliable enough to be useful. Spatial interaction models, for example, are widely used to predict retail shopping behavior, based on the principle that people balance distance with the attraction of shopping destinations in making choices.

This shows a GeoDa analysis of Seattle, Washington, house sales prices combining dynamically linked maps and graphs, including an outlier map, a cluster map (local spatial autocorrelation), a cartogram, parallel coordinate plot, Moran scatter plot (global spatial autocorrelation), and three-dimensional scatter plot. The points/locations highlighted in yellow are linked. Image courtesy of Luc Anselin.)

Geographic profiling, an important spatial tool in fighting crime, makes use of the principle of distance decay applied to the behavior of the offender and the locations of a series of crimes that show the same modus operandi. To put it crudely, a criminal prefers to work at some distance from home, but not too far. Distance decay surfaces are generated centered on each crime and superimposed to create a three-dimensional surface that likely peaks in the criminal's home area. The method has achieved some convincing successes (for a good summary overview and references to software, see Keith Harries' "Mapping Crime: Principle and Practice," Washington, D.C.: Department of Justice, 1999, available online at www.ncjrs.org/html/nij/mapping).

Geographic profiling depends on getting the distance decay surfaces right, and as long as the method is applied to a single metropolitan area, there are good grounds for believing that this is possible (much of the development of the method occurred in the Vancouver, B.C., Canada, metropolitan area, led by researchers at Simon Fraser University). But there is no reason to suspect that the distance decay functions that work for Vancouver, with its low density and major highways, will also work for the comparatively cramped urban spaces of Amsterdam, the Netherlands, or Hong Kong. Despite decades of searching, social scientists are often frustrated by the evident lack of general, universal laws in social science—mathematical models that apply equally to human societies everywhere. But the combination of GIS and spatial thinking has produced a new and exciting option: the possibility that general principles might exist but that their expression in different areas might be substantially different. Many methods of so-called place-based analysis have been developed over the past two decades to exploit this potential. They rely on ready access to georeferenced data and on the kinds of computing power now available on the researcher's desktop.

Geographically Weighted Regression

One of the newest of these is Geographically Weighted Regression (GWR), developed by Stewart Fotheringham and his colleagues at the University of Newcastle Upon Tyne, United Kingdom. GWR looks for simple linear relationships between variables, just like ordinary regression, but allows the parameters of the relationship (the slope and intercept) to vary spatially. For example, one might be interested in the relationship between family income and expenditure or between age and voting behavior. In both cases a linear relationship is expected (higher income leads to more expenditure; older people are more likely to vote), but the details of the relationship are allowed to vary from one area to another. The result of the analysis is a series of maps that allows the user to assess how the characteristics of the relationship vary spatially—a radically different and useful spatial twist on an old nonspatial idea. A book is now available on GWR (Fotheringham, A. Stewart, Chris Brunsdon, and Martin Charlton, Geographically Weighted Regression: The Analysis of Spatially Varying Relationships, New York: Wiley, 2002), and software is available from www.ncl.ac.uk/~ngeog/GWR.

The Center for Spatially Integrated Social Science was established at the University of California, Santa Barbara, in 1999 to help social scientists learn about GIS and spatial analysis and to provide them with tools and other kinds of infrastructure support. CSISS runs seven programs:

• Conducting summer workshops for social scientists to introduce them to basic and advanced concepts in spatial social science
• Conducting specialist meetings that bring together people, working on major social issues, to discuss the importance of spatial methods
• Disseminating examples of best practice (Spatially Integrated Social Science, Oxford University Press, December 2003)
• Developing new tools that implement methods of spatial social science (directed by Luc Anselin at the University of Illinois, Urbana-Champaign)
• Providing an extensive set of resources on the CSISS Web site (www.csiss.org) including bibliographies and engines that search the Web for relevant information and tools
• Enhancing the ability of social scientists to search for data and other information by geographic location
• Providing a collection of online materials (e.g., syllabi, lecture notes, tutorials) to help social scientists learn about GIS and spatial analysis

CSISS was founded on the principle that space can be an integrating theme across the social sciences. Economists study economic processes, demographers study population, and criminologists study crime; to a large extent each social science exists in isolation from the others, studying its own piece of the social pie. Every GIS professional is familiar with the notion that location can integrate disparate layers of information. CSISS extends this argument to disparate social processes, arguing that it is at specific places and times that economic, demographic, and other social processes interact and combine and that GIS and spatial analysis therefore provide the key to interaction.

These are very early days in spatial social science. Only a fraction of 1 percent of the literature published in the social sciences takes a spatial perspective, so the potential for growth is still enormous. Very few university programs in the social science disciplines currently include GIS or spatial analysis, although interest is definitely growing. The National Science Foundation (NSF) recently announced a program to support research in spatial social science, and NSF also supports CSISS. In the coming years CSISS anticipates a rapid growth of interest—with GIS facilitating greater interaction among the social sciences and more productive connections outside the "ivory tower."

For more information, contact Michael Goodchild, University of California, Santa Barbara (e-mail: good@geog.ucsb.edu, tel.: 805-893-8049).