Geographic Information Sciences

A Journal of the Association of Chinese Professionals in Geographic

Information Systems

ISSN: 1082-4006 (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tagi19

The Future of Spatial Analysis in the Social

Sciences

Luc Anselin

To cite this article: Luc Anselin (1999) The Future of Spatial Analysis in the Social Sciences,
Geographic Information Sciences, 5:2, 67-76, DOI: 10.1080/10824009909480516

To link to this article: https://doi.org/10.1080/10824009909480516

Published online: 12 Aug 2009.

Submit your article to this journal

Article views: 661

Citing articles: 12 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tagi20
GeoeraDhic Information Sciences Vol. 5. No. 2. December 1999 67

The Future of Spatial Analysis in the Social Sciences

Luc Anselin
Department of Agricultural and Consumer Economics
University of Illinois at Urbana-Champaign
Urbana, IL 61801-3608

Abstract
This paper presents a personal view on some emerging research directions at the interface of social science and
spatial analysis. Particular emphasis is placed on methodological challenges presented by developmentsin social
science theory, demands for data manipulation, and the need for education and dissemination.

I. INTRODUCTION

I n this paper, I formulate some general ideas of the art. To some extent, these research challenges
pertaining t o emerging research challenges and are qualitatively distinct from the impediments
promising directions for advances in spatial analysis encountered in the late 1980s in that the very
that are motivated by demands generatedby the social incorporation of GIS, spatial data and “spatial
sciences. These ideas are intended to paint a broad thinking” into the standard toolbox of the scientist
picture of a research agenda for the next decade, in has led to new questions. The solution to these
particular in terms of the contribution of the “next questions requires concepts, techniques and
generation”spatial analysis tools to the social sciences, implementations that go beyond the traditional
in the context of the development of “spatially paradigm that originated with the quantitative
integrated social science” [Goodchild et al. (2000), revolution in geography in the 1960s and 1970s.
Anselin (2000)l. While spatial analysis is often defined
as encompassing a wide range of spatial data In addition, a relatively recent phenomenon is the
manipulations, in this paper I will take a narrower renewed attention in the mainstream social sciences
view and focus on those techniques t h a t a r e to geography in general, and location and spatial
particularly relevant in the process of scientific interaction in particular. This, in turn, has created an
discovery, using the framework outlined in Anselin explosion in the demand for methods and tools that
and Getis (1992), In addition to limiting my remarks allow the explicit treatment of space in empirical
t o social science perspectives, I will also focus applications. Consequently, spatial analysis is playing
exclusively o n issues pertaining to spatial data a n increasingly central role in measurement,
analysis and spatial statistics, and deliberately not hypothesis development and validation of theoretical
address the broader functionality of spatial analysis, constructs, activities t h a t a r e crucial in the
which also includes, among others, spatial decision development of new scientific knowledge. The distinct
support systems, logistics and optimization.’ contribution of spatial analysis in this overall
framework is that it provides the means to explicitly
It has now been more than ten years since Goodchild recognize, assess and incorporate the importance of
(1987) argued for the importance of the spatial location and interaction within the methodological
analytical aspects of GIs to further the solution of toolbox of the social scientist.
generic spatial research questions [see also Goodchild
(1992)l. Since then, considerable progress has been Undoubtedly, spatial analysis and GIS have also
made, particularly from a technical viewpoint. revolutionized the manipulation of geographic
Familiar examples are the incorporation of spatial information in broader sections of the private and
analytical functionality within commercial GIS, the public sector, such as in urban planning, marketing
linkage of specialized statistical and other analytical and logistics. However, in this essay, these aspects w i l l
modules with GIS, and the discussion of not be further considered since the focus is on “science”
geocomputational issues associated with such in general and social sciencein particular as the driver
integration. However, as we enter the 21stcentury, a for new developments in spatial analysis. This role
number of new research challenges are emerging that for science is reflected in three important dimensions.
are not satisfactorily dealt with in the current state First, social science theory itself has generated a
demand for new developments in spatial analysis, in
Other recent assessments of research directions for spatial analysis that
are less focused on specific social science applications can be found in the
the sense that new concepts related to geography,
collectionof papers in Fotheringham andRogerson(1994),Fischer et al. (1996), location and interaction require an explicit treatment
Fischer and Getis (1997), as well as, among others, in Openshaw (1998),Miller
(1999), and Goodchild andhngley (1999). 1082-4006/99/0602-67$5.00
01999 The Association of Chinese Professionals in
Geographic Information Systems (Abroad)
68 Anselin: The Future of Spatial Analysis in the Social
of space. Similarly, the empirical verification of the social entities (context), and the empirical ver&%%%
new models (such as the “new” economic geography) of models for t h e s e i n t e r a c t i o n s r e q u i r e s t h e
requires appropriate spatial statistics that allow for combination of micro and macro data, as well as the
dependence and heterogeneity in the data. Second, combination of spatially aggregate data at various
there is a practical demand for sophisticated analysis scales (neighborhood, city, state, The methodological
to d e a l w i t h m e a s u r e m e n t issues a n d the framework for accomplishing d a t a integration is
manipulation of spatial (geocoded) data that a r e generically known as “small area estimation” in the
increasingly available to empirical workers. Third, this statistical literature [Ghosh and Rao (1994)], and as
growing demand also has a human capital component, spatial interpolation or areal interpolation in the G I s
in that there is currently a shortage of adequately literature [Goodchild et al. (1993),Mitas and Mitasova
trained spatial analysts. (1999)l. The application of GIS and spatial analysis
tools provides a means to obtain data for any scale,
In the remainder of the paper, I first formulate some but also raises important questions of accuracy and
general thoughts on the interface between spatial error propagation [Goodchild and Gopal (1989)], as
analysis and social science research. This is followed well as the fundamental concern about the “proper”
by three sections, each dealing with a particular scale of analysis [the ecological fallacy or modifiable
dimension of the challenges to spatial analysis that areal unit problem, see King (1997)l.
emanate from the social sciences, as suggested above
(theory, data, and disseminatiodeducation), In each A second major contribution is the application of
of these sections I outline some motivating examples exploratory spatial d a t a analysis (ESDA) a n d
and suggest a number of specific research challenges visualization in an inductive approach to discovering
to develop a research agenda for the next decade. p a t t e r n s , eliciting hypotheses a n d suggesting
Finally, some concluding remarks are formulated. associations. ESDA is a subset of exploratory data
analysis (EDA) [Tukey (1977)], but with an explicit
focus on t h e distinguishing characteristics of
11. SPATIAL ANALYSIS AND SOCIAL SCIENCE geographical data. It is a collection of techniques to
RESEARCH describe and visualize spatial distributions, identify
atypical locations or spatial outliers, discover patterns
The use of GIS techniques in general and mapping in of spatial association, clusters or hot spots, and suggest
particular has become increasingly common in social spatial regimes or other forms of spatial heterogeneity
science applications, in fields r a n g i n g from [Anselin (1999a)I. ESDAis particularly powerful when
anthropology [Aldenderfer and Maschner (199611, to no strong prior theoretical framework exists, as is
criminology [Weisburg a n d McEwen (1998)], often the case in interdisciplinary social science
epidemiology [Lawson et al. (1999a)], real estate analysis in fields such as criminology or human-
analysis [Can (1998)l and socio-economic analysis of environment interaction [see Messner et al. (1999)l.
tropical deforestation [Liverman et al. (1998)].2
Recently, the focus in these applications has moved A third major area of application pertains to contexts
from simple data manipulation and visualization to where a deductive approach is more appropriate, for
spatial data analysis, both exploratory as well as example when economic theory dictates the variables
confirmatory [e.g. Anselin (1998a)l. t o be considered as well as their functional
relationship. When the empirical work is based on
In general terms, one could arguably distinguish three spatial (cross-sectional) data or when the models
important ways in which spatial analysis contributes under consideration are “spatial” in nature (spatial
to the “toolbox”of the social scientist. First, it provides interaction), an application of t h e specialized
the basis for data integration, or the conversion of data methodology of spatial statistics a n d spatial
collected at one spatial scale (and time dimension) to econometrics is required [Cressie (1993), Anselin (1988,
other scales and dimensions. Specifically, this is needed 1999b)I. Broadly speaking, spatial econometrics covers
when geo-locational information must be manipulated the specification of spatial models, their estimation,
or when spatial data must be obtained for locations specification tests for spatial effects as well as spatial
or areal units for which they a r e not originally prediction. An important aspect of the methodology
recorded. T h i s is particularly r e l e v a n t in the is the “proper” incorporation of location and spatial .
combination of census data, remotely sensed images, arrangement, which are essential elements of a GIs.
maps and survey data towards the computation of
measures of access, distance and spatial linkages. Aside from these three areas in which spatial analysis
may contribute directly to the methodological toolbox
Many research questions in the social sciences pertain of the social scientist, the “geographical perspective”,
to the interaction between the individual and larger or thinking spatially also has a n important role to play
For a more general perspective, see also Martin (1996).
Geographic Information Sciences Vol. 5, No. 2, December 1999 69
in the refinement of the way in which space is feedback between t h e individual and the group
incorporated into social science theory itself. For (context). Concepts such as social norms, neighborhood
example, recent attention to precision farming has effects, peer group effects, social capital, strategic
forced agricultural economists to start considering interaction and copy-catting deal with interesting
how to incorporate the notion of spatial cost and output questions of how the individual interactions can lead
surfaces (rather than single numbers) into production t o emergent collective behavior a n d aggregate
theory [Weiss (1996)l. Similarly, emerging attention patterns. In these conceptualizations, the roles of
to social interaction in economics [Akerlof (1997)l location, space and spatial interaction are central.
requires the specification of models for the flows and Therefore, increasingly, the empirical verification of
intensity of spatial interaction. Unfortunately, this these models requires statistical and econometric
does not always build upon the wealth of insights and techniques that acknowledge and incorporate the
conceptual frameworks developed in economic spatial effects.
geography and regional science that deal specifically
with measures of accessibility, models to incorporate A few examples may help to illustrate this point. Some
friction of distance, complex interaction flows, and the recent work on models of social interaction and
like [e.g., Isard et al. (1998), Fotheringham and OKelly complex behavior in economics builds on principles
(1988), Sen and Smith (1995)l. An effective path developed in statisticalmechanics, such as interacting
towards spatially integrated social science would particle systems and random field models [Brock and
consist of recognizing and extending this existing body Durlauf (1995), Akerlof (1997), Durlauf (1997)l. The
of work and to avoid “reinventing the wheel”. basic underlying idea in these theories is a feedback
Finally, the interaction between spatial analysis and mechanism between the value of a phenomenon at a
social science research is not solely one-directional, given location and the magnitude at “neighboring”
in the sense of spatial analysis being simply a “tool” locations (where the notions of location a n d
to further social science research. In many respects, neighborhood are not necessarily in a geographical
the methods and conceptual frameworks of spatial sense). This leads to model specifications that are
analysis-as a part of geographic information science, formally equivalent to the spatial Markov fields
or GI Science-trace their origins to the natural and developed in spatial statistics [Cressie (1993)l. Similar
environmental sciences, and therefore may not always notions underlie some of the new macroeconomics of
be well suited to the demands of the social sciences. A Aoki (1996), where the interaction is in the form of a
major contribution of social science theory to GI “mean field” t e r m (some average effect of t h e
Science is therefore to provide superior social and aggregate upon the value at the micro level). Other
behavioral foundations for concepts of space and conceptualizations that explicitly incorporate
interaction and to suggest the basis for the formal interaction as part of the theoretical construct are
specification of spatial models. I n addition, demands models for evolving trading structures [Ioannides
from t h e social sciences raise t h e threshold of (1997,1999)],neighborhood spillover effects [Durlauf
sophistication required from spatial data analysis, in (1994), Borjas (1995), Glaeser et al. (1996)], yardstick
that they tend to focus on the manipulation of discrete competition [Besley and Case (1995), Bivand and
a n d categorical variables a n d stress space-time Symanski (1997)], and strategic interaction [Case et
dynamics. a1 (1993), Brueckner (1998)l.

Both methodological and theoretical considerations A second important strand of theoretical literature
suggested by recent developments in t h e social that emphasizes the importance of location, space and
sciences will require a rethinking of some of the central spatial interaction is the “new economic geography”
concepts in spatial analysis. This provides for a rich popularized in the work of Krugman (1991a, 1991b,
research agenda, which will be elaborated upon in the 1996,1999), Arthur (1989), Glaeser et al. (1992), and
following three sections. other^.^ The resulting models of increasing returns,
path dependence and imperfect competition induce
various forms of spatial externalities, agglomeration
1II.THEORY economies a n d spillovers, whose spatial imprint
requires a spatial econometric approach in empirical
An important motivation for the recent explosion in work [see Anselin et al. (1997)l.
the attention paid to GIS and spatial analysis in the
mainstream social sciences derives from a number of In sociology, a recent renewed attention to the
exciting theoretical developments in economics, ecological perspectives pioneered by the Chicago
sociology, and political science. These developments School in the early 1920s has yielded a growing
share a common interest in the interaction between
a For a recent review, see the collectionof papers in Pleskovic (1999), Fujita et
decision makers (as interacting agents) and t h e al(l999); also Martin (1999) for a more critical assessment.
70 Anselin: The Future of Spatial Analysis in the Social
iences
number of studies in which computerized mapping algorithms. In addition, new statistical methc%s may
and spatial analysis techniques have become central. be needed to ensure that proper inference is obtained
These new efforts follow directly from the theoretical when manipulating such “spatial” data. Promising
requirements that relate individual behavior to that directions are contained in new results obtained for
of the “context”and thus result in attempts to quantify t h e asymptotics of spatial econometric methods
notions such as social capital and neighborhood effects [Kelejian and Prucha (1999), Pinkse (2000)l on the
[see, e.g., Abbot (1997), Morenoff and Sampson (1997), statistical side, as well as in the work on object
Sampson et al. (2000)l. In political science as well, oriented GIs, participatory GIs and 3-D GIs. However,
especially in the study of international relations, a much remains to be done to address the sophisticated
spatial perspective is increasingly prominent in concepts of space suggested by social science theory.
theoretical as well as empirical approaches, suggesting
the formation of a new geopolitics [Starr (1991), Ward A second important challenge to spatial analysis
(1992), O’Loughlin et al. (1999)l. driven by theoretical concerns follows from the need
to provide a meaningful theoretical interpretation for
Challenges t h e role of “space” as it is incorporated in spatial
statistical and spatial econometric models. The effect
The resurgence in attention to space a n d spatial of neighbors (contiguous locations) as included in
interaction in social science theory provides a models for spatial dependence (e.g., through spatial
challenge to spatial analysis as we know it, in the sense weights), or the effect of location as expressed in
that many (most) of t h e currently available data models for spatial heterogeneity (e.g., in the form of
models and analysis methods are not particularly spatially varying coefficients or spatial regimes) does
geared to deal with these evolving theoretical concepts. not in and of itself provide an “explanation” of the
Three broad challenges in particular require some phenomenon under study. Instead, it may suggest the
further discussion. role of a spatial multiplier effect, the extent of that
effect and its strength, but it does not and cannot
First is t h e concept of “space” itself, how it is discover the actual socio-behavioral mechanisms that
incorporated in statistical models (regression models generate the effects. Similary, there is much recent
in particular) as well as stored in digital form in a excitement over methods to model local geographic
GIs. The standard approach is to treat space as a heterogeneity such as LISA Docal indicators of spatial
container for spatial objects or as a field by means of association, Anselin (1995)l and GWR [geographically
which spatial distributions are described, but several weighted regression, Fotheringham (1997)], although
of the theoretical frameworks suggest a notion of space such methods do not in themselves explain t h e
that is endogenously determined and changes a s a underlying heterogeneity. To many sceptics, these
function of t h e s t r e n g t h of i n t e r a c t i o n (e.g., more sophisticated spatial formulations are nothing
neighborhood sense, perceptual space). In addition, but models of geographic determinism in disguise. The
while the development of data models that incorporate real challenge to spatial analysis is not only to develop
space-time dynamics is an active area of research in new techniques of “local” spatial analysis, or more
GI Science, the extension of this to address individual- sophisticated models that formally express spatial
group interactions, perceptual space a n d other effects, but also to provide the means to discover and
cognitive aspects is still in its infancy [Talen (1999)l. understand t h e underlying social and behavioral
Similarly, several conceptualizations of spatial mechanisms that yield the revealed spatial patterns.
interaction [such as strategic interaction between This is further complicated by t h e observational
states in Case et al. (1993)l suggest the importance of equivalence between spatial dependence and spatial
non-Euclidean a n d non-geographical notions of heterogeneity in cross-sectional settings. In many
distance and distance decay. In terms of methodology, instances, this is a form of the “inverse problem”
this has direct implications for the specification of the encountered in the physical science^.^ Here, there is
so-called spatial weights in spatial regression models an important role for cross-fertilization between
[Anselin (1988)l. The full extent to which the standard theory and the tools of analysis.
estimators and specification tests also extend to more
general metrics is not yet fully understood and an area A third, related, b u t somewhat more technical
of active research in spatial econometrics. challenge to spatial analysis is to develop data models
and modeling techniques to handle spatial interaction
More generally, the incorporation of abstract spaces as well as space-time interaction. Especially in the
(such as policy space, attribute space) and distance context of theoretical models of diffusion a n d
metrics (such as economic distance, social distance, contagion, a proper metric for the distance in space-
political distance) within spatial analysis will require time (or speed of diffusion) is required. In addition,
a rethinking of some of the standard data models and most of the theoretical models of interaction a r e
For a recent discussion, see, e.g., ChilBs and Delfiner (1999, Ch. 8).
Geographic Information Sciences Vol. 5, No. 2, December 1999 71
developed at the micro-level and deal with discrete standard in the analysis of human-environmental
dependent variables [e.g., Brock and Durlauf (1995)], interaction, such as in investigations of changing land
whereas the current state of the art of spatial use and land cover and the assessment of tropical
econometric methodology pertains primarily to the deforestation [e.g., Bockstael(1996), Wood and Skole
standard (continuous dependent variable) regression (1998)l. Such studies a r e characterized by t h e
model. Some initial advances have been made, some availability of a wealth of spatial data, recorded at
based on analytical econometric approaches [e.g., various scales and with different resolutions. Often
Pinkse and Slade (1998)], others using Bayesian and they involve the combination of survey information
computation-intensive estimators [e.g., LeSage at the individual or household level with census data
(1997), Waller et al. (1997)], but this still constitutes at the administrative areal unit level, such as in
a vast area of research with important theoretical, spatial targeting, risk mapping and poverty mapping
methodological and computational ramifications. [e.g., Nelson and Gray (1997), Lawson et al. (1999b)l.

Overall, while certainly much progress has been Much of the spatial analysis needs driven by this
made since the late 198Os, the existing tools of spatial explosion in the availability of spatial information can
analysis, spatial econometrics and spatial statistics be met by today’s technology. However, the very size
are still lacking and much remains to be done to of the data and the multitude of Games for collecting
develop a flexible dynamic modeling toolbox that is them raise a number of issues that are currently not
able to reflect the complexity of spatial components adequately m e t a n d require f u r t h e r research,
of the new theoretical frameworks that are emerging specifically with respect to spatial scale, the size of
in the social sciences. the data sets and spatial sampling.

Challenges
IV.DATA
A first challenge is to address the issue of spatial scale
Arguably, the most commonly cited reason for the in light of the increased data availability and growing
increased interest i n spatial analysis by social analytical power of GIs. The choice of the “proper”
scientists is the explosion in the availability of geo- scale of analysis has become a n essential part of the
coded socio-economic data sets (i.e., data sets that design of scientific inquiry in the spatial sciences.
also contain the location of the observational units). Today, all kinds of geo-coded data sets have become
The existence of an extensive infrastructure of easily accessible, with information collected Gom the
spatially referenced road networks (e.g., the Tiger individual to the global level. Moreover, powerful GIS
files of the U.S. Census bureau) as well as digital base tools allow one to move from one scale of analysis to
maps for a wide range of administrative units, and another a s well as to integrate data collected at
the affordability and availability of Global Positioning different scales. Clearly, observations for one level of
Systems (GPS) has made the explicit recording of analysis (e.g., at an aggregate level) do not necessarily
location a routine matter. Where privacy laws permit, provide useful information about lower levels of
data on a wide range of socio-economic variables, from analysis (such as individual behavior), especially when
employment to crime, public h e a l t h a n d t h e spatial heterogeneity is present. Also, as observations
environment are distributed i n formats that are are re-arranged into “zones”,several statistics change
amenable to geographical analysis. Increasingly, in value, such as correlation coefficients and measures
place-based search is becoming implemented in of spatial autocorrelation. This is an old and familiar
digital libraries and web-based initiatives to facilitate methodological problem, known to sociologists at the
spatial data sharing and dissemination [Goodchild ecologicalfallacy [for a recent review, see King (1997)],
e t al. (2000)l. Furthermore, since many federal to geographers as the modifiable areal unit problem,
regulations are spatially explicit (e.g., the U.S. or NIAUP [Openshaw and Taylor (1979)l and to earth
Community Reinvestment Act to ensure equitable scientists as the upscaling or change of support
access to mortgage lending in “neighborhoods”) they problem [Chilhs and Delfiner (1999)l.
carry reporting requirements on a wide range of
transactions in a n explicit geographic framework One perspective on the ecological inference problem
[e.g., Thrall (1998)l. This has also spawned an is that it is impossible to solve, since the properties of
explosion of activity in the private sector in the form any predicted value however constructed remain
of value added reselling of public d a t a , unverifiable.6 Alternatively, one could argue that
geodemographic analyses and target marketing
The typical context for an ecologicalinference is when data are not avail-
[Birkin and Clarke (1998), Birkin et al. (1999)l. able at some lower level of aggregation (such as a household) but inference
Empirical research in a range of fields has begun to for that level is based on observations at a higher level (such as census tracts).
If the micro-data were available, they should be used, and there would not
take advantage of this plethora of spatial information. be an ecologicalfallacy problem. When they are not available,the accuracy
For example, a spatial perspective is increasingly the of any predicted value cannot be verified.
72 Anselin: The Future of Spatial Analysis in the Social
‘ciences
studies should be based on scale-invariant concepts principle applicable, their implementation ?n very
or scale-invariant variables, such as densities and large data sets can easily constitute a computational
surfaces. While t h i s may have a n intuitive (permutation approaches for each observation) as well
attractiveness in the physical sciences, many processes as conceptual (multiple comparisons) challenge. The
in social science are discrete in nature, and modeling role of geocomputation has become more important
frameworks to deal with this characteristic are still t h a n ever. Since many s p a t i a l problems a r e
i n their infancy.6 The problem remains how t o intrinsically of order N2,they cannot be effectively
construct or estimate the relevant surfaces, and while tackled in current computational environments unless
methods such as indicator kriging from geostatistics special purpose algorithms are developed to handle
offer considerable promise [e.g. Goovaerts (1997), memory management, efficient searching, sorting and
Chilhs and Delfiner (1999)], their applicability to socio- data manipulation [Anselin (1998b)l.
economic phenomena remains largely untested.
Similarly, the exploratory (graphical) and simulation Another issue related to the large size of geospatial
tools for “ecological inference” (EI) proposed by King data sets is the choice of inferential paradigm.
(1997) are a start, but so far they do not take into Classical asymptotic theory in spatial statistics and
account spatial autocorrelation and other spatial s p a t i a l econometrics has been developed t o
aspects of the problem, and this remains an active approximate the properties of estimators and tests
area of research (and controversy). statistics in finite samples, but they are not as
meaningful when the sample a t hand actually does
A related issue is the integration of multiple scales of approach infinity.8 Alternative paradigms, based on
analysis, as in hierarchical modeling. While such Bayesian notions or purely computational (simulation
analysis is now well established in social science estimators, resampling methods, permutation
methodology [for example, hierarchical linear model approaches) hold considerable promise, but their
popularized by Bryk and Raudebush (1992), and its implementation in very large spatial data sets is still
Bayesian counterparts], its extension to dealing with far from trivial.
spatial data remains to be further developed [initial
approaches can be found in Langford et a1.(1999)]. This A final issue is related to spatial sampling, a topic
is particularly crucial in the analysis of categorical typically ignored in the design and application of
and discrete variables (such as counts of events), survey research in the social sciences. Many micro data
where there are no acceptable analytical solutions that sets are currently available to social scientists and
incorporate spatial dependence. Methods based on are being used to study various implications related
simulation estimators, Markov Chain Monte Carlo to geographic notions such as spatial interaction, sense
(MCMC) and Gibbs sampling are extremely promising of community, social capital and related concepts.
[Gilks e t al. (1996), LeSage (ZOOO), Beron and However, the stratification of the surveys on which
Vijverberg (ZOOO)], but a number of important they are based typically has ignored the role of spatial
methodological and computational issues remain to effects and may therefore be totally inappropriate for
be addressed, especially t o implement these the purposes of spatial analysis. Three issues are
techniques in realistic large sample settings. important here. A first situation is encountered when
the interest lies in designing or stratifying a survey
The sheer size of available geo-spatial databases in order to correct for any possible presence of spatial
constitutes a second challenge to spatial analysis. Most effects, such a s spatial autocorrelation. A basic
“classical” techniques of spatial data analysis were principle underlying spatial sampling in this context
initially developed for situations where the data sets is to assume a form of distance decay for t h e
contained less than a hundred o b ~ e r v a t i o n s I. ~
n autocorrelation. As a result, observations that are “far
contrast, the current norm is easily several orders of enough apart” can be considered to be spatially
magnitude greater, such as in the analysis of real uncorrelated and thus can be treated in the usual
estate transactions [Pace and Barry (199711. There are fashion [e.g., the principle underlying the so-called
several implications of this larger size. One is that DUST-dependent areal units sequential t e c h n i q u e
exploratory spatial data analysis, under the guise of sampling procedure in Arbia (1993)l.A second instance
“spatial data mining” has become crucial in the process is when the survey is designed to capture and estimate
of looking for patterns, clusters, associations and other the extent and strength of spatial interaction itself.
meaningful non-randomness. While many of the In this instance, some form of cluster sampling design
currently available techniques such as LISA are in is required, where the degree of clustering should
match the range of the spatial interaction process of
Early formulations of theoretical frameworks can be found in Isard and
Liossatos (1979),but these are near impossible to implement in an empiri- interest. I n both cases, it is essential to understand
cal setting. * This is often ignored in practice, but the essence of the problem is that in
’For example, the classic Irish county data set used in Cliff and Ord (1973) the limit most estimators and test statistics converge to a fxed constant,
to illustrate spatial autocorrelation tests and spatial autoregressive models with zero variance and a degenerate distribution collapsed onto the fixed
contained only about 25 observations. constant.
Geographic Information Sciences Vol. 5, No. 2, December 1999 73
and measure the underlying spatial autocorrelation, training in GI sciences is still the exception rather
since it forms the basis for respectively separating or than the rule.
grouping sampling units. Finally, the granularity of
t h e sampling design is crucial for understanding In light of these circumstances, three main research
processes characterized by spatial heterogeneity. challenges suggest themselves: t h e development of a
generic spatial analytical toolbox, the role of spatial
These problems are familiar in the physical sciences, analysis in social science methods curricula, and the
where a growing literature deals with spatial sampling ultimate contribution of spatial analysis to social
in the context of environmental monitoring a n d science education and research.
resource exploration [e.g., Arbia and Lafratta (1997),
Muller (1998)l.However, their extension to the domain Challenges
of social science survey design remains largely
unexplored. The search for a generic spatial analysis software
toolbox has generated considerable debate. The
research challenge associated with this is to find a
V. DISSEMINATION AND EDUCATION compromise between the commercial requirements for
sufficient market size and the sophisticated needs of
A third important set of research challenges follows a n ever-changing methodological state of the art. On
from the need to educate and train sufficient numbers the one hand is the “black-box” approach often favored
of spatial analysts to support the current and future in commercial environments, with limited demands
demand for spatial analytical expertise generated by on the sophistication of the user through an easy to
t h e social sciences. There are two aspects to this understand user interface, but typically insufficiently
question, one pertaining to the dissemination of tools advanced for a research environment. On the other
and techniques, the other to the provision of sufficient hand are the “programming” approaches offered in a
human capital. number of advanced computing packages, which put
most of the burden on the user. A compromise may lie
In t h e late 1980s, t h e lack of software tools in in the development of suites of components that can
commercial GIS environments to carry out spatial be mixed a n d matched by sophisticated users or
analysis in general a n d spatial d a t a analysis in wrapped in a shell and interface for less computer
particular was often cited as a main reason for the savvy users. Such development would be greatly
slow dissemination of these techniques to empirical enhanced by the existence of an open and virtual
practice [e.g., Haining (1989)l. However, to a large community of scholars whose contributions would be
extent, this impediment has been removed. Not only made available. The creation of such a community is
has there been a slew of sofware tools developed in one of the cornerstones of the new Center for Spatially
academic e n v i r o n m e n t s t o a u g m e n t existing Integrated Social Science [Goodchild et al. (2000)l.
commercial GIS with spatial data analysis capability
[e.g., Anselin (1992), Anselin and Bao (1997), Zhang A second challenge pertains to the integration of
and Griffith (1997), Symanzik et al. (1999)], but also spatial analysis in the methodological curricula of
the commercial vendors themselves have entered this social scientists. Geographical information science has
arena. Examples are the spatial analysis modules t r a d i t i o n a l l y b e e n t h e “turf” of geography
provided with ESRI’s ArcView GIS and the S+ArcView departments, but some recent developments may
link of MathSoft [for a recent review, see Bao et al. bring this into question. For example, in a number of
(1999)l. However, there remains considerable tension places GI Science programs and degrees are offered
between the “lowest common denominator” approach in a multi-disciplinary a n d i n t e r -d i s c i p l i n a r y
taken in the COTS (commercial off-the-shelf) products environment outside the traditional departmental
and the advanced to esoteric methods incorporated boundaries (e.g., at the University of Texas at Dallas
in the academic ventures. Also, the incorporation of and the University of Utah). Some of these initiatives
spatial data analytical techniques into the market- a r e promoted by industry, using state of t h e art
leading statistical and econometric software packages technology in distance education a n d web-based
is still limited to mapping and some basic geostatistical learning (e.g., ESRI’s virtual university). The question
techniques. remains whether a “reinvented” Geography will claim
this terrain or whether this constitutes a threat to its
G r e a t s t r i d e s have been m a d e in t e r m s of t h e traditional role, in the sense that new “degrees for
incorporation of geographic information sciences into the 2lStcentury” will increasingly take on this task.
the mainstream curriculum of geography departments These issues will require considerable debate and
[e.g., the model curricula developed by NCGIA and seem far from resolved at this point in time. In addition
UCGIS], b u t outside geography, education a n d to the matter of who will provide the education, there
74 Anselin: The Future of Spatial Analysis in the Social
Sciences
remains much work to be done to integrate even basic Models, Boston: Kluwer Academic.
notions of spatial analysis into mainstream social [5] Anselin, L., 1992, Spacestat, a softwarepackage for the
science methods curricula. analysis of spatial data, National Center for Geographic
Information and Analysis, University of California,
Santa Barbara.
Finally, an underlying theme of this essay was that [6] Anselin, L. , 1995, Local indicators of spatial association
indeed “space does matter,” although this is by far not -LISA, Geographical Analysis, 27:93-115.
a widely accepted notion in the mainstream of the [7] Anselin, L., 1998a, GIS research infrastructure for
social sciences. An important challenge to the spatial spatial analysis of real estate markets, Journal of
analysis community consists of demonstrating in Housing Research, 9:113-133.
unequivocal terms that much is to be gained by a [8] Anselin, L., 1998b, Exploratory spatial data analysis
careful a n d explicit incorporation of t h e spatial in a geocomputational environment, In P. Longley, S.
element in social science research. Making the sales Brooks, R. McDonnell, a n d B. Macmillan (eds.),
pitch credible is a formidable task and much remains Geocomputation, A Primer, pp. 77-94, London: John
Wiley.
to be done.
[9] Anselin, L., 1999a, Interactive techniques a n d
exploratory spatial data analysis. In P. Longley, M.
Goodchild, D. Maguire and D. Rhind (eds.), Geographical
VI. CONCLUSION Information Systems, 2ndEdition, pp. 253-266, New
York: John Wiley.
A n u m b e r of developments in t h e t h e o r y a n d [lo] Anselin, L., 1999b, Spatial econometrics, In B. Baltagi
empiricism of the social sciences form the foundation (ed.), Companion to Theoretical Econometrics, Oxford:
for a growing importance of spatial analysis. We are Basil Blackwell (forthcoming).
at an exciting frontier and it is likely that the next 111 Anselin, L., 2000, GIs, spatial econometrics and social
science research, Journal of Geographical Systems 2
decade will bring tremendous gains in the technical (forthcoming).
and theoretical prowess associated with GI science. 121 Anselin, L. and S. Bao, 1997, Exploratory spatial data
In order to accomplish t h e s e advances, a tight analysis linking Spacestat and ArcView, In M. Fischer
interaction between theory, d a t a analysis a n d and A. Getis (eds.), Recent Developments in Spatial
computation will be necessary. In addition, these Analysis, pp. 35-59, Berlin: Springer-Verlag.
a d v a n c e s a r e more likely t o h a p p e n in a n 131 Anselin, L. and A. Getis , 1992, Spatial statistical
interdisciplinary and multidisciplinary environment analysis and geographic information systems, Annals
where traditional boundaries are brought down. It is of Regional Science 26:19-33.
a challenge to the GI science community to seize this [14] Anselin, L., A. Varga, Z. Acs, 1997, Local geographic
spillovers between university research a n d high
opportunity a n d provide the means to develop a technology innovations, Journal of Urban Economics,
“spatially integrated” social science. 42~422-448.
[15] Aoki, M., 1996, New Approaches to Macroeconomic
Modeling, Cambridge: Cambridge University Press.
ACKNOWLEDGMENTS [16] Arbia, G., 1993, The use of GIS in spatial statistical
surveys, International Statistical Review, 61:339-359.
Earlier versions of this paper were presented a t the Varenius [17]Arbia, G. and G. Lafratta , 1997, Evaluating and
Workshop on Status and Trends in Spatial Analysis, Santa updating the sample design in repeated environmental
Barbara, CA, Dec. 1&12,1998, and a s the opening address surveys: monitoring air quality in Padua, Journal of
to Geoinformatics ’99, Ann Arbor, MI, June 19-21, 1999. Agricultural, Biological a n d Environmental Statistics,
Comments from Shuming Bao, referees and participants 2:451-466.
a t the workshops are gratefully acknowledged. Part of the [18]Arthur, W. B., 1989, Increasing returns, competing
research reported on in this paper was supported by NSF technologies and lock-in by historical small events: the
grant BCS-9978058 to the Center for Spatially Integrated dynamics of allocation under increasing returns to
Social Sciences (CSISS). scale, Economic Journal, 99:116-131.
[19] Bao, S., L. Anselin, D. Martin and D. Stralberg, 1999,
Seamless integration of spatial statistics and GIs: the
REFERENCES S-Plus for ArcView and the S+Grassland links, Journal
of Geographical Systems (forthcoming).
[20] Beron, K.J. and W.P.M. Vijverberg, 2000, Probit i n a
[l] Abbot, A., 1997, Of time and space: the contemporary spatial context: a Monte Carlo analysis, In L. Anselin
relevance of the Chicago School, Social Forces, 75:1149- and R. Florax (eds.), Advances in Spatial Econometrics,
1182. Heidelberg: Springer-Verlag (forthcoming).
[2] Akerlof, G., 1997, Social distance and social decisions, [21] Besley, T. and A. Case, 1995, Incumbent behavior: vote-
Econometrica, 65:1005-1027. seeking, tax-setting a n d yardstick competition,
[3] Aldenderfer, M. and H. Maschner, 1996, Anthropology, American Economic Review, 85:25-45.
space, and Geographic Information Systems, New York: [22] Birkin, M. and G. Clarke, 1998, GIs, geodemographics
Oxford University Press. and spatial modeling i n the UK financial service
[4] Anselin, L., 1988, Spatial Econometrics: Methods and
Geographic Information Sciences Vol. 5, No. 2, December 1999 75
industry, Journal of Housing Research, 9:87-111. Hall.
[23]Birkin, M., G. Clarke and M. Clarke, 1999, GIS for [45] Glaeser, E., H. Kallal, J. Scheinkman,A. Schleifer ,1992,
business and service planning. I n P. Longley, M. Growth i n cities, Journal of Political Economy,
Goodchild, D. Maguire a n d D. Rhind (eds.), 100:1126-1152.
GeographicalInformation Systems, 2ndEdition, pp. 709- [46] Glaeser, E., B. Sacerdote and J. Scheinkman 1996, Crime
722. New York John Wiley. a n d social interactions. Quarterly J o u r n a l o f
[24] Bivand, R. and S. Szymanski, 1997, Spatial dependence Economics, 111:507-548.
through local yardstick competition: theory and testing, [47] Goodchild, M., 1987, A spatial analytical perspective
Economics Letters ,55:257-265. on geographic information systems, International
[25] Bockstael, N., 1996, Modeling economics and ecology: Journal of Geographical Information Systems,1:327-
the importance of a spatial perspective, American 334.
Journal of Agricultural Economics, 78:1168-1180. [48] Goodchild, M., 1992, Geographical information science,
[26] Borjas, G., 1995, Ethnicity, neighborhoods, and human- International Journal of Geographical Information
capital externalities, American Economic Review, 85: Systems,6:31-45.
365-390. [49] Goodchild, M. nd S. Gopal, 1989, Accuracy of Spatial
[27] Brock, W. and S. Durlauf, 1995, Discrete choice with Databases. London: Taylor and Francis.
social interactions k Theory. NBER Working Paper No. [50] Goodchild, M. and P. Longley, 1999, The future of GIS
W5291. Cambridge, MA: National Bureau of Economic and spatial analysis, In P. Longley, M. Goodchild, D.
Research. Maguire and D. Rhind (eds.), Geographical Information
[28] Brueckner, J., 1998, Testing for strategic interaction Systems, 2ndEdition, pp. 567-580. New York: John Wiley.
among local governments: the case of growth controls. [51] Goodchild,M., L. Anselin, R. Appelbaum, B. Harthorn,
Journal of Urban Economics, 44:438-467. 2000, Towards spatially integrated social science,
[29] Bryk, A. and S. Raudenbush, 1992, Hierarchical Linear International Regional Science Review,23: 139-159.
Models, Newbury Park: Sage Publications. [52] Goodchild, M., L. Anselin, U. Deichmann, 1993, A
[30] Can, A, Ed., 1998, Geographic information systems in framework for the areal interpolation of socio-economic
housing and mortgage Finance, Journal of Housing data, Environment and Planning ,A, 25:383-397.
Research 9 (special issue). [53] Goovaerts, P., 1997, Geostatistics for Natural Resources
[31] Case, A., H. S. Rosen and J. R. Hines, 1993, Budget Evaluation. New York Oxford University Press.
spillovers and fiscal policy interdependence: evidence [54]Haining, R., 1989, Geography and spatial statistics:
from the States, Journal of Public Economics, 52~285- current positions, future developments. In B. Macmillan
307. (ed.), Remodeling Geography, pp. 191-203. Oxford:
[32] ChilBs, J-P. and P. Delfiner, 1999, Geostatistics:Modeling Blackwell Publishes.
Spatial Uncertainty. New York: John Wiley. [55] Isard, W. and P. Liossatos, 1979, Spatial Dynamics and
[33]Cliff, A. and J.K. Ord, 1973, Spatial Autocorrelation. Optimal Space-Time Development. New York: North
London: Pion. Holland.
[34] Cressie, N., 1993, Statistics for Spatial Data. New York [56] Isard, W., I. h i s , M. Drennan, R. Miller, S. Saltzman, E.
John Wiley. Thorbecke, 1998, Methods of Interregional and Regional
[35]Durlauf, S., 1994, Spillovers, stratification a n d Analysis. Aldershot: Ashgate.
inequality. European Economic Review, 385336-845. [57] Ioannides, Y. M., 1997, Evolution of trading structures.
[36] Durlauf, S., 1997, Statistical mechanics approaches to In W. B. Arthur, S. Durlauf, D. Lane (eds), The Economy
socioeconomicbehavior. In W. B. Arthur, S. Durlauf, D. as a n Evolving Complex System 11,pp. 129-167. Reading,
Lane (eds), The Economy as a n Evolving Complex MA: Addison-Wesley.
System 11,pp. 81-104. Reading, MA: Addison-Wesley. [58] Ioannides, Y.M., 1999, Economic geography and the
[37] Fischer, M. M. and A. Getis, 1997, Recent Developments spatial evolution of wages in the United States. In L.
in Spatial Analysis. Berlin: Springer-Verlag. Anselin and R. Florax (eds), Advances in Spatial
[38] Fischer, M. M., H. Scholten and D. Unwin, 1996, Spatial Econometrics.Heide1berg:Springer-Verlag
Analytical Perspectives on GIS. London: Taylor & (forthcoming).
Francis. [59] Kelejian, H. and I. Prucha, 1999, On the asymptotic
[39]Fotheringham, A. S., 1997, Trends in quantitative distribution of the Moran I test statistic, w i t h
methods I: stressing the local, Progress in Human applications. Working Paper, Department of Economics,
Geography, 21:88-96. University of Maryland, College Park, MD.
[40]Fotheringham, A.S. and M. O'Kelly, 1988, Spatial [60] King, G., 1997, A Solution to the Ecological Inference
Interaction Models: Foundations and Applications. Problem. Princeton: Princeton University Press.
Dordrecht: Kluwer Academic. [61] Krugman, P., 1991a, Increasing returns and economic
[41]Fotheringham, A. S. and P. Rogerson, 1994, Spatial geography, Journal of Political Economy 99~438-499.
Analysis and GIS, London: Taylor & Francis. [62] Krugman, P., 1991b, Geography and Trade. Cambridge,
[42] Fujita, M., P. Krugman, A. Venables, 1999, !l'he Spatial MA: MIT Press.
Economy: Cities, Regions and International Trade. [63] Krugman, P., 1996, Urban concentration: the role of
Cambridge, MA: MIT Press. increasing returns and transport costs, International
[43]Ghosh,M. and J. N. K. Rao, 1994, Small area estimation: Regional Science Review 19, 5-30.
a n appraisal, Statistical Sciences, 9:55-93. [64]Krugman, P., 1999, The role of geography i n
[44] Gilks, W., S.Richardson, D. Spiegelhalter, 1996,Markov development, International Regional Science Review 22:
Chain Monte Carlo in Practice. London: Chapman and 142-161.
76 Anselin: The Future of Spatial Analysis in the Social
Sciences
[65] Langford, I., A. Leyland, J. Rasbash and H. Goldstein, London: Pion.
1999, Multilevel modeling of geographical distributions [82]Pace, R.K. and R. Barry, 1997, Quick computation of
of diseases. Applied Statistics, 48:253-268. spatial autoregressive estimators, Geographical
[66] Lawson, A., A. Biggeri, D. Bohning, E. Les&e, J-FViel, Analysis, 29:232-246.
R. Bertollini, Eds. 1999a, Disease Mapping a n d Risk [83]Pinkse, J., 2000, Moran-flavored tests with nuisance
Assessment for Public Health. Chichester: John Wiley. parameters, In L. Anselin andR. F l o r a (eds.),Advances
[67]Lawson, A., D. Bohning, A. Biggeri, E. Les&e, J.-F. Vie1 in Spatial Econometrics. Heidelberg: Springer-Verlag
1999b, Disease mapping and its uses, in Lawson, A., A. (forthcoming).
Biggeri, D. Bohning, E. Lesaffre, J-FViel, R. Bertollini [84] Pinkse, J. andM. Slade, 1998, Contracting in space: an
(eds.), Disease Mapping and Risk Assessment for Public application of spatial statistics to discrete-choice
Health, pp. 3-13. Chichester: John Wiley. models, Journal of Econometrics, 85:125-154.
[68]LeSage, J., 1997, Bayesian estimation of spatial [85] Pleskovic, B., 1999, Guest editorial: challenges for the
eautoregressive models. International Regional Science new economic geography in the twenty-first century.
Review 20:113-129. International Regional Science Review, 22:139-141.
[69]LeSage, J., 2000, Bayesian estimation of limited [86] Sampson, R. J.,J. Morenoff and F. Earls, 2000, Beyond
dependent variable spatial autoregressive models, social capital: neighborhood mechanism and structural
Geographical Analysis 32 (forthcoming). sources of collective efficacy for children, American
[70] Liverman, D., E. Moran, R. Rindfuss, P. Stern, Eds., Sociological Review (forthcoming).
1998, People and Pixels, Linking Remote Sensing and [87l Sen, A. and T. E. Smith, 1995, Gravity Models of Spatial
Social Science. Washington, D.C.: National Academy Interaction Behavior. New York Springer-Verlag.
Press. [88]Starr, H., 1991, Democratic dominoes: diffusion
[71] Martin, D., 1996, Geographic Information Systems: approaches to the spread of democracy, Journal of
Socioeconomic Applications. London: Routledge. Conflict Resolution, 35: 6-31.
[72] Martin, R., 1999, The new geographical t u r n i n [89] Symanzik, J.,T. Kotter, S. Schmelzer, S. Klinke, D. Cook,
economics: some critical reflections, Cambridge Journal D. Swayne, 1999, S p a t i a l d a t a analysis in t h e
of Economics, 23:65-91. dynamically linked ArcView/Xgobi/XploRe
[73] Messner, S.,L. Anselin, R. Baller, D. Hawkins, G. Deane, environment, Computing Science a n d Statistics
S. Tolnay, 1999, The spatial patterning of county (forthcoming).
homicide rates: a n application of exploratory spatial [go] Talen, E., 1999, Constructing neighborhoods fkom the
data analysis, Journal of Quantitative Criminology, 15: bottom up: the case for resident-generated G I s .
423-450. Environment a n d Planning, B, 26x533-554.
[74] Miller, H., 1999, Potential contributions of spatial [SlIThrall, G., 1998, G I s applications in real estate and
analysis to geographic information systems for related industries, Journal of Housing Research, 9:33-
transportation(G1S-T), Geographical Analysis 31:373- 59.
399. [92] Tukey, J., 1977, Exploratory Data Analysis. Reading:
[75] Mitas, L. and H. Mitasova ,1999, Spatial interpolation. Addison-Wesley.
In P. Longley, M. Goodchild, D. Maguire and D. Rhind [93] Waller, L., B. Carlin andA. Gelfand, 1997, Hierarchical
(eds.), Geographical Information Systems, Znd Edition, spatio-temporal mapping of disease rates, Journal of
pp. 481-492. New York: John Wiley. the American Statistical Association, 92:607-617.
[76] Morenoff, J. and R. J. Sampson, 1997, Violent crime [94] Ward, M. D., 1992, The New Geopolitics. London: Gordon
and the spatial dynamics of neighborhood transition: and Breach.
Chicago 1970-1990, Social Forces, 76:31-64. [95]Weisburd, D. andT. McEwen, Eds. ,1998, CrimeMapping
[77] Miiller, W., 1998, Collecting Spatial Data: Optimum a n d Crime Prevention. New York: Criminal Justice
Design of Experiments for Random Fields. Heidelberg: Press.
Physica-Verlag. [96] Weiss, M., 1996, Precision farming and spatial economic
[78] Nelson, G. andD. Gray, 1997, The use of spatial analysis analysis: research challenges and opportunities,
a t the World Bank. Environmental Economics Series American Journal ofAgricultura1 Economics, 78:1275-
Paper No. 56. Washington, DC: The World Bank. 1280.
[79] OLoughlin, J.,M.D. Ward, C.L. Lofdahl, J.S. Cohen,D.S. [97] Wood, C. andD. Skole, 1998, Linking satellite,census,
Brown, D. Reilly, K.S. Gleditsch and M. Shin 1999, The and survey data to study deforestation in the Brazilian
diffusion of democracy, 1946-1994. Annals of the Amazon, In D. Liverman, E. Moran, R. Rindfuss, and P.
Association of American Geographers, 88:545-574. Stern (eds.), People and Pixels: Linking Remote Sensing
[80]0penshaw, S., 1998, Some t r e n d s a n d f u t u r e and Social Science,pp. 70-93. Washington, DC: National
perspectives for spatial analysis i n GIS,Geographic Academies Press.
Information Sciences, 4:5-13. [98] Zhang, Z. and D. GrifEth, 1997,Developinguser-friendly
[8l]Openshaw, S. and P. Taylor, 1979, A million or so spatial statistical analysis modules for GIs, a n example
correlation coefficients: three experiments on the using ArcView, Computers, Environment and Urban
modifiable areal unit problem, I n N. Wrigley (ed.), Systems, 21:5-29.
Statistical Methods in the Spatial Sciences, pp. 127-144.

-- c
!'