Professional Documents
Culture Documents
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Society for American Archaeology is collaborating with JSTOR to digitize, preserve and extend access to
American Antiquity.
http://www.jstor.org
132
AMERICAN ANTIQUITY
sonality of Britain, 1943). Subjectivity in map interpretation was a general problem in political and economic
geography, as well as in a number of other fields but
one which was recognized and corrected at a comparatively early stage in the development of these disciplines (Harvey: Transactions of the Institute of British
Geographers 44:85-95). Consequently, the authors
argue that we have a methodological repertoire developed in other disciplines which can, to varying degrees, be adapted to spatial problems of concern to
archaeologists. A short introduction to statistical concepts germane to spatial analysis is appended to Chapter 1, for a quantified approach is deemed essential if
objectivity and replicability are major research objectives, and if large amounts of data are to be processed.
Chapter 2 delineates problems related to the interpretation of distribution maps of settlement systems
and of isolated artifact types. Distortion of original
configurations by erosion and site destruction, and a
differential in fieldwork intensity are identified as principle causes of sampling error in studies of regional
scope, regardless of whether site or artifact distributions are being evaluated. Reliability of map distributions is also seen to be influenced by the scale upon
which data are recorded; the effect that varying quadrat sizes has on statistical assessments of randomness
and/or aggregation is a well-known example. Once factors of distortion are taken into account, analysis can
proceed either inductively, through pattern search
procedures, or deductively, through the formulation of
hypotheses about what Schiffer (Behavioral Archaeology, 1976) has called "cultural formation processes."
The latter can be modeled mathematically, in many
cases, and some of these models are considered in detail in Part II.
Part II is the longest and certainly the most valuable
section of the book; it comprises three chapters which
treat point and block pattern analyses in general terms,
and then evaluate statistical methods for assessing the
departure from randomness in models for settlement
systems and for the regional distributions of single artifact types. Only by vitiating the randomness assumption can patterning which may be interpretable in behavioral terms be demonstrated to exist. Tests for randomness can be divided into quadrat methods and distance methods, and these are surveyed in Chapter 3.
Data recorded by quadrats (blocks) are analyzed by
comparing the observed frequency distributions of
point counts per unit with that of the Poisson function
adjusted for density using the variance/mean ratio
statistic, and the X2 distributed index of dispersion
(Greig-Smith, Quantitative Plant Ecology, 1964). An
approach which allows the search for structured pattern to be carried out at different scales (block sizes),
and which tests for goodness of fit at each scale is what
Whallon has called "dimensional analysis of variance"
(Pielou, An Introduction to Mathematical Ecology,
1969). If a graph of the variance between blocks for
each block size is drawn, block sizes which approximate a clustered pattern will appear as peaks in the
graph, which can then be tested by the variance/mean
ratio test, or interpreted subjectively. Applications
most familiar to American readers include Dacey's
analysis of the Sde Divshon lithic data (American Antiquity 38:320-28), Whallon's DANOVA paper (American Antiquity 38:266-78), and Thomas' simula-
REVIEWS
tion of Great Basin Shoshonean settlement subsistence
systems (Models in Archaeology, D. Clarke, ed., 671704, 1972).
The main advantage to quadrat based methods is
that much archaeological data is, and has been collected in the form of counts by grid units. However,
results are seriously affected by quadrat size (and it is
not always possible to vary quadrat size unless pointprovenienced data are available to begin with), the
variance/mean ratio statistic is unstable when the
mean is small, the X2 goodness of fit test cannot be
applied if the number of quadrats is small and if overall
density is low, and, finally, DANOVA is constrained
by the use of a rectangular grid of contiguous quadrats
in which the number of units along each side must be a
power of 2-a feature which limits its usefulness to
situations where the scale and shape of the excavated
(or collected) archaeological surface can be manipulated to form a grid of the required dimensions.
Because of these limitations, tests based upon distance measures are considered more appropriate for
archaeological research if point-provenienced data are
available. This means nearest neighbor analysis (Clark
and Evans: Ecology 35:445-53), or some variation on
it; the basic method and rationale are outlined and illustrated with an examination of Iron Age hillforts in
Wiltshire. Nearest neighbor analyses are perhaps more
familiar to archaeologists than are quadrat approaches.
On a regional scale they have been applied to surface
survey data from Pueblo sites on the Rio Puerco
(Washburn: American Antiquity 39:315-35), Peebles
on mound sites in Alabama (Moundville: The Organisation of a Prehistoric Community and Culture, 1974)
and by Hodder and his colleagues to analyze the distributions of Iron Age and Romano-British walled
towns in S. England. On the intrasite level, Whallon
(American Antiquity 39:16-34) has used the method to
describe the distributions of four major artifact
categories on a "Protomagdalenian' (Periogordian
VII) occupation floor at the Abri Pataud, in the Dordogne.
In Chapter 4, an effort is made to articulate theoretical models for interpreting observed spatial patterns
with the pattern search approaches outlined in Chapter
3. Aggregate random patterns are regarded as
anomalies seen to arise from random resource distributions, and more commonly from post-depositional factors, which may have distorted the original configuration. The point here is that the identification of a random pattern does not preclude the possibility that the
pattern may have been generated by randomizing disturbances of some other, original spatial pattern. Models for the horizontal and hierarchical organization of
settlement systems developed in classical locational
analysis in economic geography are reviewed and
found applicable in some instances to archaeological
data. Even as iconic models, they are theory-laden
formulations because ideal configurations are linked
with, or held to be directly attributable to explicit,
usually economic causal factors (e.g. resource distributions, population density gradients, subsistence
activities, marketing and productivity principles,
transport media and the like). The result is a package
of theoretical relationships which are derived from the
analysis of, and which have been tested against contemporary data from agrarian societies (cf. Smith: Re-
133
gional Analysis I, II, 1976), and which offer the potential for testing against the archaeological record. The
bulk of Chapter 4 consists of just such a series of tests
where hypotheses formulated on the basis of idealized
locational models are evaluated. The mechanics of
analyzing sites arranged into hierarchies on various
criteria are examined in the Midwest (Struever and
Houart: Anthropological Papers, University of Michigan 46: 47-79), in Mesoamerica (Hammond: Models in
Archaeology, D. Clarke, ed., 757-800, 1972) and in late
pre-Roman Iron Age Britain (Hodder: New Directions
in Archaeology, D. L. Clark, ed., 1975). The wellknown rank-size rule (Vining: Journal of the American
Statistical Association 18:44-64; Haggett: Locational
Analysis in Human Geography, pp. 101-107, 1965), an
alternative model for examining hierarchical situations
in which intralevel variability obscures interlevel distinctiveness, is also discussed at some length. Methods
for examining observed variation in the rank-size relationship are illustrated using British Iron Age data,
where centers are ranked using a path ordering algorithm. A potential difficulty is the necessity for tight
temporal control, and the remainder of Chapter 4 is an
attenuated discussion of the chronological significance
of site density and regularity relationships. Uniform
patterns are seen to develop through time as a consequence of the marketing and competition principles
first outlined by economic geographers. Clustered patterns emerge from resource localization and from the
process of contagion-the attraction of population to a
higher order center to take advantage of a broader
spectrum of services and activities than would be
available at a lower order center. Statistical models
using the probability distribution of the negative binomial are used to distinguish between "true contagion"
(where the distribution of points is assumed
homogeneous and the probability of the occurrence of
a point (settlement) directly increases the probability
of occurrence of other points nearby) and apparent or
"spurious contagion" (where random inhomogeneity
in population density is present and where the probability of the occurrence of a point is a function of a
differential in the availability of information). Observed frequency data based on quadrat counts from
Polish Bandkeramik, Lengyel, and TRB sites is tested
for goodness of fit against true and spurious contagion
models using maximum likelihood estimates. As these
data are ordered in time, a model for the pattern of
settlement development is outlined which entails an
initial stage of contagious growth, the expansion of
settlement clusters and finally the development of a
more dispersed pattern characterized by local variation in site density.
Regression analysis, the simulation of artifact dispersal patterns, trend surface analysis and spatial autocorrelation are considered in Chapter 5 with respect
to the distributions of single artifact types considered
in isolation. Regression analysis is discussed in general
and at some length; it is used to describe spatial trends
in the distributions of single artifact types from known
sources by analysis of residuals from fitted regression
lines (e.g. Iron Age/Roman pottery types from New
Forest, Savernake, Oxford kilns). Plots of positive and
negative residuals identify density fall-off patterns
which are interpreted subjectively-a drawback because sets of residuals patterned alike can result from a
134
AMERICAN ANTIQUITY
REVIEWS
The book ends with a brief synopsis of problems
with archaeological spatial analysis. Problems resulting from the nature of archaeological data, problems
arising from the constraints of the statistical methods
employed, and problems of the relationship between
spatial form and process are considered. Sampling
error due to a differential in site survival, fieldwork
intensity and a failure to control for temporal variation
are critical factors in the first case, whereas the statistical procedures surveyed here have sets of assumptions which at times limit their usefulness with recalcitrant archaeological data, in the second.
The major limitation at present is, of course, that
different spatial processes can produce the same spatial form. So, while it may be possible to describe and
to measure the strength of association between spatial
patterns using rigorous and replicable techniques, and
to test for goodness of fit with abstract theoretical
models, the fact remains that there is no 1:1 correlation
between spatial form and spatial process. Schiffer
(American Antiquity 37:156-65) has suggested that the
extraction of behavioral patterns from archaeological
remains is by no means the simple and straightforward
task which it is sometimes imagined to be since there is
no reason to assume that distributions recoverable
from the archaeological record are replications of distributions which are meaningful in behavioral terms.
From this perspective, archaeological residues are regarded as static representations of dynamic behavioral
and natural systems; consequently the distributional
aspects of archaeological remains must be carefully
scrutinized so that a host of cultural and natural processes which might influence the appearance of archaeological remains on the ground can be taken into
account. At present, we seem to be in a situation
where some fairly sophisticated theoretical statements
about spatial process have been made, usually in the
context of economic or locational geography. This
book presents some ideas about how these hypotheses
about spatial process can be assessed archaeologically, and that is its major contribution. Although archaeology (and other disciplines) clearly lack a comprehensive model which relates spatial process in general terms to spatial form, it is argued that the methods
outlined by Hodder and Orton are sorely needed if
explicit "systems for indirect observation of the past"
are ultimately to be developed (Fritz: Conternporarv
Archaeology, M. Leone, ed., pp. 135-37, 1972). Archaeological theory building is at a point when implementation of quantitative spatial measures is critical to the resolution of many, perhaps most of its
hypotheses. This book has been prepared with that
goal in mind. It is not a compendium of "answers' to
abstract questions about archaeological theory pertinent to spatial variables in general, but is instead a
necessary methodological bridge designed to span the
chasm between the purely abstract and the empirical
aspects of spatial behavior.
G. A. CLARK
Arizona State University
Tempe, AZ
135