Parallel developments in the quantitative analysis of political phenomena in Geography and Political Science indicate that the time is ripe for a stock-taking of the field and the development of a research agenda that will suggest future research directions. This proposed conference that we title "New Directions in Spatial Analysis for the Social Sciences" will bring together for the first time many leading practitioners of spatial analysis applied to political subjects and will build significantly on the existing reputation of the University of Colorado at Boulder as a center for training and research in this comparatively-new field. The best papers to be presented at the conference will be submitted as a special issue of Political Analysis, the leading methodological journal in the field of political science. A volume of conference proceedings will also be submitted to the methodology series at the University of Michigan Press. These outlets for the research results will certainly bring to the attention of the political science and related disciplines the issues, specific problems of geographic data and the specialized methods of analysis and display that have been developed by geographers. (See the call for proposals for graduate student funding to attend the conference and bio sketches and abstracts for conference papers.)
The stimulus for the conference lies in the tentative and sporadic attempts by geographers and political methodologists to engage each other in the common problem of how to analyze political phenomena that are spatially arranged, like electoral results on the basis of precinct counts or the geographic distribution of wars or coups d’état. As will be explained below, aggregate data that are geographically distributed must be analyzed with an array of special techniques, known as spatial analysis. The basic problem stems from spatial autocorrelation of the error terms in regression, the usual result of the clustering of similar values in the geographic configuration. One of the key regression assumptions is thus violated, the regression estimates are biased and inefficient, and recourse to special techniques to deal with this problem is necessary. Geographers, especially Luc Anselin (one of the participants) have developed a suite of methods strongly linked to the visual display of the data and the estimates in Geographic Information Systems (GIS), to cope with the special issues of two-dimensional data. Stated differently, most empirical analyses currently conducted assume that all the units analyzed are independent of one another. However, part of what defines the political world is the absence of true independence and the interrelationships among the actors of politics. It is extreme to assert that contemporary (or historical) politics are completely independent of one another; it is equally nonsensical to assert that electoral precincts (say Boulder and Denver) exist in a vacuum absent the other. Spatial techniques are being developed to correct this erroneous assumption and to allow insights to be constructed on the assumption of non-independence.
In political science, there is a growing acknowledgement of the need to examine aggregate geographic data with appropriate tools but this trend is hindered by the disciplinary barriers and statistical packages, such as SAS and SPSS, that are not appropriate for the analysis of aggregate spatial data. Most of the political science attendees at the conference have published preliminary works that try to modify existing political methodologies to cope with the unexpected challenges of geographic data, but the sophistication of the spatial analysis leaves a lot to be desired. There remains a significant gap between the well-developed political theories of how and why groups in geographic units should interact and influence each other and the analytical tools that are needed to examine these kinds of interactions. By contrast, geographers for the most part have developed high-level tools of analysis and display, but the applications have often been trivial and unenlightening. The purpose of the conference is to bring key members of the two communities together to find common ground and to elaborate an integrated set of inter-disciplinary techniques that can be diffused throughout both disciplinary communities.
To take one example of the special kinds of problems of spatial data, consider spatial correlation between neighboring cells, such as, the autocorrelation of votes for the Democrats across the precincts of a city. It is usually assumed that autocorrelation is the same across a whole data set and can be summarized by a clustering statistic like Moran’s I. Recent work by geographers (Getis, Ord and Anselin) have shown that dis-aggregation of the global statistic into its component parts (how much each unit contributes to the overall Morans I value based on the unit’s correlation with its specific neighbors) offers valuable insights into the structures and distributions of the problem under review (such as the distribution of conflict). Interactive graphical methods such as those available in ArcView offers a powerful methodology for exploratory spatial data analysis but too often the resulting cartograms or chloropleth maps turn out to be heavily dependent on the classification used and even the cleverest tricks of the cartographers cannot reproduce the complicated structures that are important. Integration of display and analysis in an interactive ESDA (exploratory spatial data analysis) highlights the structures of the data and suggests paths to more formal and complex analysis.
Empirical validation of the new "spatial" concepts and models requires an explicit spatial econometric methodology that tackles issues of spatial dependence (as a result of contiguity) and spatial heterogeneity (as a result of regional contexts), as well as their extensions to the space-time domain. Anselin and other geographers have pointed to the need for new developments in three important domains of spatial analysis:
a) Extending concepts of "space": Spatial analysis needs to go beyond dealing with physical geographical locations to include location in "social" space (social distance, economic distance). This will require further consideration and development of distance metrics for "social" space, for space-time dynamics and notions of "topology" in space-time (the counterpart of the "weights" matrix in spatial autocorrelation analysis). Promising avenues are current work on GIS data models, object-oriented GIS, and the like. An illustration of this problem is the choice of a weights metric for the measure of interaction between countries; should it be contiguity, distance, ratio of shared borders, an interaction statistic (like trade or phone calls), shared membership in international organizations, etc. The choice of the metric has profound implications for the results and so must be chosen carefully and with due attention to theoretical considerations.
b) Broadening the analytical toolbox : The toolbox of spatial econometrics and spatial analysis needs to be extended to deal with the challenges posed by the analysis of socio-economic (including political) space-time data. While much progress has been made, some unresolved issues are the estimation of space-time dynamics for limited dependent data (such as discrete choice data, duration data), modeling changing choice sets, distinguishing spatial dependence from spatial heterogeneity, and effective visualization of model fit. For many of these research questions analytical solutions are impossible or prohibitive, such that computational approaches must be followed (e.g., simulated moments, simulated likelihood, Markov Chain Monte Carlo). Indeed, some of the biggest advances have built upon these MCMC approaches to examine what are known as locally dependent random Markov fields, virtually unknown in political science and much ignored in geography. Several presentations at the conference will focus on this topic explicitly.
c) Technology transfer : Most of the current commercial GIS software comes in the form of (partially) open environments that allow the user to include customizations and extend the functionality. In a modern component oriented computing environment, there is therefore no longer a high priority to have commercial spatial analytical tools included in the "box", but rather to have the mechanisms to mix and match components to accomplish specific tasks. The commercial world will always be behind the curve when it comes to "state of the art" in terms of the statistical methodology it delivers. In contrast to the toolbox approach, shrink-wrapped commercial GIS software has tended to offer the lowest common denominator when it comes to spatial analytical (let alone statistical) methodology. This has had two major drawbacks. One is that uninitiated users identify "spatial analysis" with the (limited) set of techniques offered by a software vendor (e.g. some of the add-on features of the S-plus package). The other is that the geographic analysis is presented as being "easy" and underlying assumptions, algorithms and limitations are hidden from the user.
The theoretical questions posed in the political science literature offer an important challenge to the methodology of spatial analysis. However, this also constitutes a major opportunity for the spatial analytical perspective (as part of a geographic information science) to contribute to the theoretical debate in the core disciplines. The focus of this conference will be on how techniques for the analysis of spatial data can be effectively applied in a GIS environment, with a particular emphasis on the study of spatial patterns and spatial autocorrelation. In other words, special attention will be given to the "where" of the data. As an illustration, consider the recent article, based on research funded by the National Science Foundation’s Political Science Program. O’Loughlin, Ward and their students (Annals, Association of American Geographers Vol 88, 1998) used a spatial diffusion framework for the study because it allows emphasis on the inter-connections among temporal and spatial changes. Adding to valuable work on the mechanisms of democratization and democratic theory, an overarching view of trends and global patterns was identified with consistent evidence of the temporal cascading of democratic and autocratic trends, spatial clustering of regime types, and strong temporal-spatial autocorrelation; there is no single form of diffusion evident in the trends. This conference will follow up on this and related studies.
The format of the conference, to be held at the University of Colorado on the weekend of March 10-12, 2000, will feature research presentations and commentaries by CU faculty, statistical consultants and graduate students. Four of the presenters received their Ph.D.s from CU Geography or Political Science in the past 5 years and the conference will extend this reputation by engaging a new generation of graduate students in the spatial analysis of political phenomena. It will also contribute to the new Graduate School Certificate Program in Applied Behavioral Science by demonstrating the validity of inter-disciplinary analytical approaches to problems common to different disciplines.
The participants who have agreed to attend and make presentations at the conference:
(biosketches, titles of talks and abstracts)
John Agnew, Geography, UCLA
Luc Anselin, Econ and Geog., Illinois
Brady Baybeck, PolSc, Washington Univ.
Nathaniel Beck, PolSc, Cal. at San Diego
Paul Diehl, PolSc, Illinois
Colin Flint, Geography, Penn State U.
Kristian Gleditsch, Social Science, Glasgow
Robert Huckfeldt, PolSc, Indiana University
Peter Hugill, Geography, Texas A and M
Pat James, PolSc, Iowa State University
Carol Kohfeld, PolSci, Missouri at St. Louis
Jeffrey Kopstein, PolSc, Colorado
John O’Loughlin, Geog and IBS, Colorado
Mohan Penubarti, PolSc, UCLA
David Reilly, PolSc, Colorado
Michael Shin, Geog, U of Miami, Fl.
Daniel Sui, Geog. Texas A and M University
John Sprague, PolSc, Washington Univ.
Harvey Starr, PolSc., South Carolina
Douglas VanBelle, Tulane University
Michael Ward, PolSc, University of Washington, Seattle
Other Attendees – Paper Commentators:
Terri Givens, PolSc, University of Washington
Debra Javeline, Government, Harvard U
Jani Little, Statistical Consultant, IBS
Claudio Cioffi, PolSc, University of Colorado
Steve Chan, PolSc, University of Colorado
Keith Maskus, Econ/IBS, Colorado
Barbara Buttenfield, Geog., Colorado
Adrian Raftery, Statistics, Washington
John McIver, PolSc, University of Colorado
Edward Greenberg, IBS/PolSc, Colorado
Renske Doorenspleet, PolSc, Universitet Leiden, Netherlands