Spatial Autocorrelation

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

Spatial Autocorrelation

Dacey M 1965 A review of measures of contiguity for two and k- constitutes the basis for the evolution of spatial
color maps. In: Berry B, Marble D (eds.) Spatial Analysis: structures. In turn, the supply of facilities across space
A Reader in Statistical Geography. Englewood Cliffs, NJ, provides opportunities for individuals and households
pp. 479–95
to conduct their activities, but at the same time also
Dunn R, Harrison A 1993 Two dimensional systematic sampling
of land use. Applied Statistics 42: 585–601 sets the space-time constraints that restrict behavior.
Foster B 1980 Urban residential ground cover using Landsat Over the years, many different attempts at modeling
digital data. Photogrammetric Engineering and Remote Sensing spatial choice behavior have been reported in the
46: 547–58 literature. These models predict the probability that an
Geary R 1954 The contiguity ratio and statistical mapping. The individual will choose a choice alternative as a function
Incorporated Statistician 5: 115–45 of its locational and non-locational attributes. In this
Green P 1996 MCMC in image analysis. In: Gilks G, Richardson article, the background, theoretical underpinnings,
S, Spiegelhalter D (eds.) Marko Chain Monte Carlo in and basic methodology of these modeling approaches
Practice. Chapman & Hall, London, pp. 381–99
are examined. From the outset, it should be clear that
Haining R 1983 Anatomy of a price war. Nature 304: 679–80
Haining R 1984 Testing a spatial interacting-markets hypothesis. limited available space prevents us from discussing
Reiew of Economics and Statistics 66: 576–83 modeling approaches in considerable detail.
Haining R 1987 Small area aggregate income models: theory and
methods with an application to urban and rural income data
for Pennsylvania. Regional Studies 21: 519–30
Haining R 1990 Spatial Data Analysis in the Social and 1. Graity Models
Enironmental Sciences. Cambridge University Press, Cam- The modeling of spatial choice behavior gained con-
bridge, UK siderable momentum when Wilson (1971) introduced
Haining R, Wise S, Ma J 1998 Exploratory spatial data analysis
in a GIS environment. The Statistician 47: 457–69
a family of spatial interaction models, based on the
Isaaks E, Srivastava R 1989 An Introduction to Applied Geostatis- principle of gravity, to describe and predict social
tics. Oxford University Press, Oxford, UK phenomena. This principle states that the interaction
Krishna Iyer P 1949 The first and second moments of some between two bodies is proportional to their masses,
probability distributions arising from points on a lattice, and and inversely proportional to the square of distance.
their application. Biometrika 36: 135–41 Wilson elaborated this principle in a number of ways.
Marshall R 1991 A review of methods for the statistical analysis First, he suggested that any distance decay function
of spatial patterns of disease. Journal of the Royal Statistical that reflects the notion that, ceteris paribus, the
Society, Series A 154: 421–41 intensity of interaction decreases with increasing
Milne A 1959 The centric systematic area sample treated as a
random sample. Biometrics 15: 270–97
distance may be used to replace straight distance.
Mollie A 1996 Bayesian mapping of disease. In: Gilks G, Secondly, he argued that in spatial problems often the
Richardson S, Spiegelhalter D (eds.) Marko Chain Monte number of departures from particular origins and\or
Carlo in Practice. Chapman & Hall, London, pp. 359–79 the number of arrivals at particular destinations is
Moran P 1948 The interpretation of statistical maps. Journal of known, implying that so-called balancing factors have
the Royal Statistical Society, Series B 10: 243–51 to be introduced in the model to satisfy these con-
Ord J 1975 Estimation methods for models of spatial interaction. straints.
Journal of the American Statistical Association 70: 120–6 Wilson distinguished an unconstrained spatial in-
Richardson, S 1992 Statistical methods for geographical cor- teraction model (departures and arrivals are not
relation studies. In: Elliott P, Cuzick J, English D, Stern R
(eds.) Geographical and Enironmental Epidemiology: Methods
known), a production-constrained spatial interaction
for Small Area Studies. Oxford University Press, Oxford, UK model (number of departures per zone is known), an
pp. 181–204 attractions-constrained spatial interaction model
Ripley B 1981 Spatial Statistics. Wiley, New York (number of arrivals per zone is known), and the doubly
Whittle P 1954 On stationary processes in the plane. Biometrika constrained spatial interaction model (both number of
41: 434–49 departures and number of arrivals per zone are
Waldhor T 1996 The spatial autocorrelation coefficient Moran’s known). In the context of spatial choice behavior, the
I under heteroscedasticity. Statistics in Medicine 15: 887–92 production constrained spatial interaction model is
most interesting as it predicts which destination will be
R. P. Haining chosen. It can be expressed as:

pij l AiOiWj f (dij) (1)

Spatial Choice Models where, pij is the probability that an individual in zone i
will choose destination j; Ai is a balancing factor,
The interrelationship between spatial structure and guaranteeing that the number of predicted interactions
individual choice behavior has traditionally been at from zone i is equal to its number of observed
the forefront of geographic research. Geographers interactions; Oi is the number of observed interactions
have realized that spatially varying consumer demand from zone i; Wj is a vector of variables, representing

14768
Spatial Choice Models

the attractiveness of destination j; f (dij) is some purpose of the model was to derive rules of spatial
function of the distance between origin i and dest- choice behavior.
ination j. To that effect, choice alternatives were classified
Although much empirical work has been conducted into so-called locational types, which are a com-
on spatial interaction models since this seminal work, bination of an attractiveness category and a distance
most of the relevant literature has been concerned with category. Spatial choice behavior is assumed to reflect
operational problems. Theoretical progress has been a trade-off between attractiveness and distance sep-
limited. For example, different measures of dist- aration. Pairwise choice data are then used to calculate
ance (straight-line distance, Euclidean distance, city- the proportion of times a particular locational type is
block metric, and travel time) and attractiveness have chosen, given that both are present. Based on this
been suggested. Likewise, different functions for rep- data, the locational types are positioned on a uni-
resenting the distance decay effect have been proposed dimensional preference scale using a nonmetric multi-
(power, exponential, and more complex functions). dimensional scaling algorithm.
Finally, an exponent has been added to the spec- Over the years, this basic model has been elaborated
ification of attractiveness term to allow for the fact in a number of important ways. First, Rushton (1974)
that the larger destinations tend to have extra attrac- showed how graphical methods, trend surface analysis,
tion beyond their size because of the benefits of or conjoint analysis might be used to decompose the
economies of scale. preference scale into the contributions of the two basic
The estimation of spatial interaction models is variables. Second, Girt (1976) suggested linking the
typically based upon aggregate, zonal data of in- preference function to overt behavior by relating
teraction flows (shopping, recreation, migration, traf- distances on the preference scale to choice probabili-
fic, etc). To that effect, the study area is delineated ties. Third, Timmermans (1979) suggested deriving
carefully to avoid large external flows, and divided the choice sets of individuals from data on information
into a number of smaller homogeneous zones. The fields and consumer attitudes, and using these to scale
spatial interaction model is calibrated using observa- the alternatives. Although Rushton’s preference sca-
tions about interactions between these zones as input. ling model does have considerable appeal, it never
As a result, the parameters of the spatial interaction received any major following. One of the reasons
model are highly influenced by the geometry of the might be that the model is based on two explanatory
study area. This caused a search for improvements of variables only, while heterogeneity in preferences is
the model, and for alternative modeling approaches. not accounted for. The model has also been criticized
In this context, some authors suggested the devel- because of its conceptual basis. It has been suggested
opment of models that estimate the effects of attraction that the model may not represent preferences at
variables endogenously. Another way of dealing with all, but rather inconsistencies in individual choice
this problem was to incorporate some measure of behavior.
spatial structure into a spatial interaction model
(Fotheringham 1983). The simplest approach to the 3. Discrete Choice Models
specification of this additional term is the use of a
Hansen-type accessibility measure. This so-called The spatial interaction model has been criticized for its
competing destinations model may be expressed as: reliance on aggregate, zonal interaction flows. This
criticism has led to the development of disaggregate
pij l AiOiWjC δj d −ijβ (2) discrete choice models that are based on individual
choice behavior.
where Discrete choice models may be derived from at least
two formal theories: Luce’s strict utility theory and
Ai l
A
 WjC δj d −ijβ
C
−" (3) Thurstone’s random utility theory. In particular, Luce
B j D assumed that the probability of choosing a choice
and alternative is equal to the ratio of the utility associated
with that alternative to the sum of the utilities for all
the alternatives in the choice set. Luce thus assumed
Cj l  Wjd −jkγ (4) deterministic preference structures and postulated a
kj constant-ratio decision rule.
In contrast, random utility theory is based on
stochastic preferences in that an individual is assumed
to draw a utility function at random on each choice
2. The Reealed Preference Model occasion. An individual’s utility for a choice alterna-
tive is assumed to consist of a deterministic component
The fact that the spatial interaction model is heavily and a random utility component. Given the principle
influenced by the geometry of the study led Rushton of utility maximizing behavior, the probability of
(1969) to develop a preference-scaling model. The choosing a choice alternative is then equal to the

14769
Spatial Choice Models

probability that its utility exceeds that of all other their levels, the next step involves creating an ex-
choice alternatives in the choice set. perimental design to generate a set of hypothetical
The specification of the model depends on the choice alternatives. This design should be created such
assumptions regarding the distributions of the random that the necessary and sufficient conditions to estimate
utility components. The best-known model is the the model of interest are satisfied. Respondents are
multinomial logit (MNL) model. It can be derived by then required to evaluate the resulting attribute
assuming that the random utility components are profiles. Their responses are decomposed into the
independently, identically Type I extreme value distri- utility contribution of the attribute levels. Once part-
buted. It can be expressed as: worth utilities have been estimated, choice behavior
can be predicted by assuming that an individual will
exp [U(Xk, Si)] choose the alternative with the highest overall evalu-
pik l (5) ation. Alternatively, different probabilistic rules can
 exp [U(Xj, Si)] be postulated.
j To avoid such ad hoc rules, Louviere and Wood-
worth (1983) suggested estimating the preference
function and choice model simultaneously. To that
where U(Xk, Si) is the deterministic part of the utility of end, the attribute profiles should be placed into choice
choice alternative k of individual i with socioeconomic sets. Again, the composition of the choice sets depends
characteristics Si. on the properties of the choice model.
One of the most important criticisms that have been Over the years, the basic conjoint preference and
expressed against the multinomial logit model con- choice models have been elaborated in many different
cerned the fact that the utility of a choice alternative is ways. To avoid the problem of information overload
independent of the existence and the attributes of and reduce respondent demand, Louviere (1984)
other alternatives in the choice set (the Independence suggested the Hierarchical Information Integration
from Irrelevant Alternatives or IIA property). It method. It assumed that individuals first group the
predicts that the introduction of a new, similar choice attributes into higher-order decision constructs. Sep-
alternative will reduce market shares in direct pro- arate designs are constructed for each decision con-
portion to the utility of the existing alternatives, which struct, and a bridging experiment is designed for
is counterintuitive in the case of a high degree of measuring the tradeoff between the evaluations of the
similarity between particular alternatives. decision constructs. Oppewal et al. (1994) suggested
Therefore, various alternative models have been an improved methodology by using multiple-choice
developed to relax the IIA assumption (Timmermans experiments to test the implied hierarchical decision
and Golledge 1990). Similar work in this regard has structure.
been conducted by Williams (1977). The best-known Other recent advances include the development of
model that circumvents the IIA-property is the nested group preference models, the development of models
logit model. Choice alternatives that are assumed to be of portfolio choice, the inclusion of contextual effects
correlated are grouped into the same nest. Each nest is and constraints, and the combined use of experimental
represented by an aggregate alternative with a com- design and revealed preference data, to name a few.
posite utility, consisting of the so-called inclusive
value, and a parameter to be estimated. To be
consistent with utility-maximizing behavior, the in-
clusive values should lie in the range between 0 and 1, 5. Complex Spatial Choice Behaior
and the values of the parameters should change The previous models are all typically based on single-
consistently from lower levels to higher levels of the purpose, single-stop behavior. Over the years, how-
hierarchy. ever, it became clear that an increasing proportion of
trips involved multistop behavior. Moreover, Ha$ ger-
strand’s time geography had convincingly argued that
4. Conjoint Preference and Choice Models behavior does not reflect preferences only, but also
constraints. Thus, various attempts have been made to
Unlike discrete choice models, the parameters of the develop models of trip chaining and activity-travel
conjoint preference and choice models are not derived patterns.
from real-world data, but from experimental design Originally, most models relied on semi-Markov
data. Conjoint models involve the following steps process models or Monte Carlo simulations. More
when applied to spatial choice problems (e.g., Tim- recently, utility-maximizing models have dominated
mermans 1984). the field. An important contribution in this regard was
First, each choice alternative is described by its made by Kitamura (1984), who introduced the concept
position in a set of attributes, considered relevant for of prospective utility. It states that the utility of a
the problem at hand. Next, a series of levels for each destination is not only a function of its inherent
attribute is chosen. Having defined the attributes and attributes and the distance to that destination, but also

14770
Spatial Cognition

of the utility of continuing the trip from that des- Kitamura R 1984 Incorporating trip chaining into analysis of
tination. Dellaert et al. (1998) generalized Kitamura’s destination choice. Transportation Research 18B: 67–81
approach to account for multipurpose aspects of the Lenntorp B 1976 Paths in space-time environments. Lund
Studies in Geography, Series B, no. 44
trip chain.
Louviere J J 1984 Hierarchical information integration: a new
Trip chaining is only one aspect of multiday method for the design and analysis of complex multiattribute
activity\travel patterns. Several models have recently judgment problems. In: Kinnear T C (ed.) Adances in
been suggested to predict more comprehensive activity Consumer Research 11, ACRA, Provo, UT, pp. 148–55
patterns. These models can be differentiated into the Louviere J J, Woodworth G 1983 Design and analysis of
older constraints-based models (e.g., Lenntorp 1976), simulated consumer choice or allocation experiments: An
utility-maximizing models (e.g., Recker et al. 1986), approach based on aggregate data. Journal of Marketing
rule-based or computational process models (e.g., Research 20: 350–67
Golledge et al. 1994, ALBATROSS – Arentze and Oppewal H, Louviere J J, Timmermans H J P 1994 Modeling
hierarchical information integration processes with integrated
Timmermans 2000).
conjoint choice experiments. Journal of Marketing Research
31: 92–105
Recker W W, McNally M G, Root G S 1986 A model of
6. Conclusions complex travel behavior: part I – theoretical development.
Transportation Research, A 20A: 307–318
This article has given a very brief overview of spatial Rushton G 1969 Analysis of spatial behavior by revealed space
choice modeling. Due to the limited available space, preference. Annals of the Association of American Geographers
we have focused on some of the key modeling 59: 391–400
approaches, disregarding alternatives (e.g., dynamic Rushton G 1974 Decomposition of space preference structures.
models) altogether. In: Golledge R G, Rushton G (eds.) Spatial Choice and Spatial
If we examine the accomplishments, one cannot Behaior. Ohio State University Press, Columbus, OH, pp.
119–33
escape the conclusion that the models developed over Timmermans H J P 1979 A spatial preference model of regional
the years have been increasing in complexity. The shopping behaviour. Tijdschrift oor Economische en Sociale
theoretical underpinnings of the recent activity-based Geografie 70: 45–8
models are much richer than the Markov or gravity Timmermans H J P 1984 Decompositional multiattribute pre-
models of the 1960s. Moreover, the brief overview ference models in spatial choice analysis: a review of some
indicates that geographers have been using alternative recent developments. Progress in Human Geography 8: 189–
data, such as experimental design data and activity 221
diaries, in addition to the more conventional survey Timmermans H J P, Golledge R G 1990 Applications of behav-
data. ioural research on spatial choice problems II: preference and
choice. Progress in Human Geography 14: 311–54
See also: Spatial Analysis in Geography; Spatial Williams H C W L 1977 On the formation of travel demand
models and economic evaluation measures of user benefit.
Decision Support Systems; Spatial Interaction; Spatial Enironment and Planning, A 9: 285–34
Interaction Models; Spatial Optimization Models; Wilson A G 1971 A family of spatial interaction models and
Spatial Pattern, Analysis of; Spatial Statistical associated developments. Enironment and Planning, A 3: 1–32
Methods; Spatial Thinking in the Social Sciences,
History of H. Timmermans

Copyright # 2001 Elsevier Science Ltd.


Bibliography All rights reserved.
Arentze T A, Timmermans H J P 2000 ALBATROSS: a
Learning-based Transportation-oriented Simulation System.
Spatial Cognition
EIRASS, Eindhoven University of Technology, The Nether-
lands Spatial cognition concerns the study of knowledge and
Dellaert B, Arentze T A, Borgers A W J, Timmermans H J P, beliefs about spatial properties of objects and events in
Bierlaire M 1998 A conjoint based multi-purpose multi-stop the world. Cognition is about knowledge: its ac-
model of consumer shopping centre choice. Journal of quisition, storage and retrieval, manipulation, and use
Marketing Research 35: 177–88 by humans, nonhuman animals, and intelligent
Ettema D F, Timmermans H J P 1997 Actiity-based Approaches machines. Broadly construed, cognitive systems in-
to Trael Analysis. Elsevier Science, Oxford, UK clude sensation and perception, thinking, imagery,
Fotheringham A S 1983 A new set of spatial interaction models: memory, learning, language, reasoning, and problem-
The theory of competing destinations. Enironment and
solving. In humans, cognitive structures and processes
Planning, A 15: 1121–32
Girt J L 1976 Some extensions to Rushton’s spatial preference are part of the mind, which emerges from a brain and
scaling model. Geographical Analysis 8: 137–56 nervous system inside of a body that exists in a social
Golledge R G, Kwan M P, Ga$ rling T 1994 Computational and physical world. Spatial properties include lo-
process modeling of travel decisions using a geographical cation, size, distance, direction, separation and con-
information system. Papers in Regional Science 73: 99–117 nection, shape, pattern, and movement.

14771

International Encyclopedia of the Social & Behavioral Sciences ISBN: 0-08-043076-7

You might also like