David Harvey - Explanation in Geography-Edward Arnold (1986) PDF
David Harvey - Explanation in Geography-Edward Arnold (1986) PDF
'Dr Harvey has achieved a tour de force. He has written a book that is likely to
find lasting importance and, though he writes in the first place for practising
geographers, his book should be read by all who seek the leitmotif of modern
geographical research and teaching.' Geographical Magazine
: . V, ' ' ”, ’ ; · -^·'-· /a!
'What better example of frustration than to be told to review in 600 words or
less what may be the most significant volume to appear in one's discipline in
many years? Explanation in Geography could be just that—an impressive, en
during landmark in the intellectual evolution of my field.'
Professional Geographer
'Another World' by M. C. Escher is reproduced by kind permission of the artist and the
Escher Association. (Graphic work of M. C. Escher, Kon. Uitgeverij J. J. Tijl, 1967)
\l - - :: ;
Edward Arnold
\
. ·:: ~ - ^ '
Explanation in
Geography
David Harvey
To my father and mother
who sacrificed so much
for my early education
1
Si 3/ίΟ.οΐ
Η Α β .,
Explanation
in Geography
(D
Edward Arnold
DM'
And it was the philosophy of the scientific method which was implicit
in quantification.
Some people may flinch at the term ‘scientific method’, so let me
make it clear that I interpret this in a very broad sense to mean the
setting up and observing of decent intellectual standards for rational
argument. Now it is obvious that we can observe these standards
without indulging in quantification. Good geographers have always
observed them. But the curious thing was that it took quantification
to demonstrate to me how extraordinarily lax my own standards
were—hence all those unpublishable papers. I believe that the most
important effect of quantification has been to force us to think logi
cally and consistently where we had not done so before. This con
clusion led me to change the emphasis of my approach. Although it
was no accident that quantification was forcing us to up-grade our
standards of argument, we could, if we so wished, up-grade those
standards without any mention of quantification. The issue of quan
tification per se therefore faded into the background and I became
much more interested in the general issue of the standards and norms
of logical argument and inference which geographers ought to accept
in the course of research. These standards could not be divorced
from those of science as a whole. In short, I became interested in the
role of scientific method (however conceived) in geography. Now
there are many who have been nurtured so long in the ways of
science that they appear to need no formal instruction in its method.
To instruct such people seems like formalising what they already
intuitively know. But there are many geographers who need formal
instruction because, like me, they were not raised in the ways of
science. But even those geographers who possess a fair intuitive grasp
of scientific method cannot afford to ignore the formal analysis of it.
An intuitive grasp arises from teaching by precept and example.
Such a grasp is usually sufficient to handle routine work (and most
of science is routine). But it cannot always handle the new questions,
the problems for which there are no precedents. At this point it is
often necessary to understand the philosophical underpinnings of
scientific method as a whole.
Science provides us with very sharp tools. But as any craftsman
will tell you it is the sharp tools which can do most damage when
misapplied. The sharpest tools are those provided by mathematics
and statistics. The former provides us with a means for formulating
arguments rigorously and simply, while the latter provides us with
the tools for data analysis and hypothesis testing with respect to data.
I believe that these tools have often been misapplied or misunder
stood in geography. I certainly plead guilty in this respect. If we are
viii PREFACE
to control the use of these sharp tools in research we must understand
the philosophical and methodological assumptions upon which their
use necessarily rests. These assumptions are, of course, built up
explicitly through an analysis of scientific method. But we have to
ensure that the assumptions which we accept with respect to these
particular tools of science do not conflict with the broader assump
tions which we employ in setting up standards for rational argument
and inference. The problem of adequate method is therefore doubly
emphasised at the point where quantitative techniques and ordinary
rational argument and inference come together. Hence the impor
tance of quantification. We must, therefore, reconcile our assumptions
at all levels in geographical research. What started for me as a quest
to understand the nature of certain powerful tools of science, thus
ended as a quest for an understanding of the totality of the process
which leads to the acquisition and codification of geographical
understanding.
This book is therefore about the ways in which geographical
understanding and knowledge can be acquired and the standards of
rational argument and inference that are necessary to ensure that
this process is reasonable. I have sought to show that we can formu
late criteria to judge whether or not an argument is sound, a tech
nique properly used, or an explanation reasonable. I do not claim
to have specified these criteria correctly. Ignorance is a relative
thing. Compared to my situation five years ago I now feel much more
learned and wise, but relative to what I still have to learn I feel more
ignorant than ever. Indeed, since completing this manuscript in
June 1968 I have changed several opinions and I can already identify
errors and shortcomings in the analysis. This is therefore very much
an interim report—one person’s view at a particular point in time.
I do not wish it to form the basis for some new kind of orthodoxy and
I for one will certainly not defend it in those terms. My aim is to open
up the field of play rather than to close it off to future development.
In constructing this interim report, I have had a good deal of
help. I spent the year 1960-1 at the University of Uppsala on a
Leverhulme scholarship and I would like to acknowledge this finan
cial help for it gave me a year after the completion of my doctorate
in which to think about all kinds of things which I had not had time
for previously. During my stay in Uppsala, I formed a lasting friend
ship with Gunnar Olsson, and in the early stages we gave each other
much mutual support. In the summer of 1964 I received help from
the National Science Foundation to attend a conference on spatial
statistics at Northwestern University and I must say the experience
was a traumatic one. Michael Dacey, who provided considerable
PREFACE 1X
stimulus to my thought at that conference, has since provided me
with unpublished materials and I am grateful to him for allowing
me to quote from these. Waldo Tobler has similarly sent me unpub
lished materials, and I am also grateful to him for allowing me to
quote from them. I must also thank my colleagues in the University
of Bristol, particularly Allan Frey and Barry Garner, for providing
a convivial and stimulating atmosphere in which to think and work.
Peter Gould, during my year at Penn State, also provided me with
plenty to think about. Various people have nibbled at the manu
script and the ideas contained therein. Allan Frey, Art Getis, Les
King, Allen Scott, Roger Downs, Bob Colenutt, Rod White, Keith
Bassett, Conrad Strack, and several others, made suggestions some of
which I have incorporated. No acknowledgement would be complete
without a mention of the ‘terrible twins’ of British geography, Dick
Chorley and Peter Haggett. The former first introduced me to statis
tical methods just before I left Cambridge in i960, and since that
time has continued to throw ideas around in a- most stimulating
fashion. Peter Haggett has also been extremely kind to me and,
particularly since he became professor at Bristol, he has been a never-
ending source of advice and encouragement. I owe these two, and I
believe British geography owes them also, an enormous debt.
I also want to acknowledge those motley folk who helped to keep
me sane during the writing of this book. Marcia, Miles Davis,
John Coltrane, Dionne Warwick, the Beatles and Shostakovitch,
Titus, Phinneas T. Bluster and Jake, have all helped to maintain
my inner equilibrium when things looked bleak. I think Marcia
may help with the indexing too. To each and every one of this jolly
crew my heartfelt thanks.
D. H. Clifton, Bristol
March 1969
Acknowledgements
The author and publishers wish to thank the following for permission to use copyright
material: George Allen and Unwin Ltd. and Columbia University Press for table 2.3.1, p.
56, from Theory and methods of social research byjohan Galtung, 1967; the American Statistical
Association and the author for two maps, pp. 385-8, ‘Maps based on probabilities’ by M.
Choynowski, Journal of the American Statistical Association 54, 1959; the Association of Ameri
can Geographers and the author for map, p. 320, ‘The market as a factor in the localization
of industry in the United States’ by C. D. Harris, Annals 44, 1954; the Association of
American Geographers for extracts from ‘The spatial structure of agricultural activities’
by W. L. Garrison and D. F. Marble, Annals 47, 137-44, 1957, and from ‘The nature of
geography’ by R. Hartshorne, Annals 29, 1939; University of California Press for extracts
from The direction of time by H. Reichenbach, 1956, and from Land and life—A selection from
the writings of Carl Ortwin Sauer edited by John Leighly, 1963; Cambridge University Press
and the authors for figure 142, p. 294, On growth andform by W. D ’Arcy Thompson, 1961
(abridged edition), and for extracts from Scientific explanation by R. B. Braithwaite, 1953;
Chandler Publishing Company for extracts from The conduct of inquiry by A. Kaplan, 1964;
University of Chicago Press and the author for extracts from The structure of scientific evolutions
by T. S. Kuhn, 1962; Dover Publications Inc. for extracts from An introduction to symbolic
logic and its applications by R. Carnap, 1958 (English edition, translated by W. H. Meyer
and J. Wilkinson); The Free Press of Glencoe Inc. for extracts from Aspects of scientific
explanation and other essays in the philosophy of science by C. G. Hempel, 1965; C. W. K. Gleerup
Publishers and the authors for figure 38, p. 73, Innovationsforloppet ur korologisk synpunkt by
T. Hagerstrand, 1953, andforfigures 9.30 & 9.31, pp.278-g, ‘Theoreticalgeography’, byW.
Bunge, Lund Studies in Geography, Series C, x, 1966 (second edition); Charles Griffin & Co.
Ltd. and the authors for figure 22.1, Advanced theory of statistics, volume II, by M. G. Kendall
and A. Stuart, 1967 (second edition); P. Haggett for figure 10.10, p. 300, Locational analysis
in human geography, 1965 (Edward Arnold Publishers Ltd.); Harcourt, Brace and World Inc.
for extracts from The structure of science by E. Nagel, 1961; J. W. House for extracts from
‘Chance and landscape’ by L. Curry in Northern geographical essays edited by J. W. House,
Oriel Press Ltd., 1966; the Institute of British Geographers for figure 2, ‘Central Europe—
Mitteleuropa—Europe Centrale’, by K. A. Sinnhuber, Transactions 20, 1954; Methuen
Publishers for figure 3.1, ‘Models in geomorphology’ by R. Chorley in Models in geography
edited by R. Chorley and P. Haggett, 1967; the authors and Michigan Inter-University
Community of Mathematical Geographers for figure 5, p. 17, ‘A note on surfaces and paths
and applications to geographical problems’ by W. Warntz, Discussion Papers 6, 1965, and for
figure 14.6, p. 220, ‘Numerical map generalization’ by W. Tobler, Discussion Papers 8, 1966;
the University of North Carolina Press for extracts from Causal inferences in nonexperimental
research by Η. M. Blalock, 1964 edition; Aldine Publishing Co. and Routledge and Kegan
Paul Ltd. for extracts from Explanation in social science by R. B. Brown, 1963; John Wiley &
Sons Inc. for extracts from Scientific method: optimizing applied research decisions by R. L.
Ackoff, 1962, from Decision and value theory by P. C. Fishburn, 1964, and from Thefoundations
of statistics by L. J. Savage, 1954; and Yale University Press for extracts from The philosophy
of symbolic forms, Volume 2: The phenomenology of knowledge by E. Cassirer, 1957 (English
edition, translated by R. Manheim and C. W. Hendel).
x
ontents
page
Preface v
Acknowledgements x
List of figures X1X
Part I
PHILOSOPHY, METHODOLOGY AND
EXPLANATION
chapter
i PHILOSOPHY AND METHODOLOGY IN
GEOGRAPHY 3
2 THE MEANING OF EXPLANATION 9
I T h e m e a n in g o f e x p l a n a t i o n 10
A The need for an explanation ”
B Constructing an explanation 13
C The criteria forjudging whether an explanation
is satisfactory and reasonable 15
D On paradigms and world im ages 16
II E x p e r ie n c e , l a n g u a g e a n d e x p l a n a t io n 18
hi E x p l a n a t i o n a s a n a c t iv it y 23
Basic reading; Reading in geography 23
Part II
THE METHODOLOGICAL BACKGROUND AND
EXPLANATION IN GEOGRAPHY
3 GEOGRAPHY AND SCIENCE—THE METHO
DOLOGICAL SETTING 27
4 SCIENTIFIC EXPLANATION—THE MODEL
OF NATURAL SCIENCE 30
A The routes to scientific explanation 32
(7) Route 1 32
(2) Route 2 35
B Deductive and inductive form s o f reference 36
(1) The problem of verification 38
(2) The problem of inductive statements 4°
Basic reading; Further reading 43
XI
xn CONTENTS
chaPter page
5 PROBLEMS OF EXPLANATION IN THE SOCIAL
SCIENCES AND HISTORY 44
i T e c h n iq u e s o f in v e s t ig a t io n 45
ii The conceptual content o f e x p l a n a t io n s 4e
hi The l o g ic a l s t r u c t u r e o f e x p l a n a t io n 4s
iv V e r if ic a t io n — t h e p r o b l e m o f p r o v id i n g r e s p o n
sib l e SUPPORT FOR STATEMENTS IN SOCIAL SCIENCE AND
HISTORY 54
v E x p la n a t i o n in t h e s o c i a l s c ie n c e s a n d h is t o r y —a
CONCLUSION 59
Basic reading; Subsidiary references 61
6 EXPLANATION IN GEOGRAPHY—SOME
GENERAL PROBLEMS 62
i P h il o s o p h y o f s c ie n c e , g e o g r a p h ic m e t h o d o l o g y
AND EXPLANATION IN GEOGRAPHY— SOME BASIC RELA
TIONSHIPS 63
π S o m e m e t h o d o l o g ic a l c o n t r o v e r s ie s i n g e o g r a p h y 67
hi E x p l a n a t io n in g e o g r a p h y 78
Basic Reading 83
Part III
THE ROLE OF THEORIES, LAWS AND MODELS
IN EXPLANATION IN GEOGRAPHY
7 THEORIES 87
A The structure o f scientific theories 88
B The text o f scientific theories 91
C Incomplete theories—the problem o f partial
form alisation g6
8 HYPOTHESES AND LAWS IOO
A The universality o f law-statements 101
B The relationship between laws and theories 104
Basic reading on theories, hypotheses and laws 106
9 LAWS AND THEORIES IN GEOGRAPHY 107
1 Law s in g eography I07
ii T h e o r ie s i n g e o g r a p h y ii3
10 MODELS 141
I T h e f u n c t io n o f m o d e l s 141
II T h e d e f in it io n o f m o d e l s 144
h i L o g ic a l p r o b l e m s o f m o d e l u se 147
(1) Models of an # 147
(2) Analogue models 149
i v P r o c e d u r a l p r o b le m s o f m o d e l u s e 15i
(1) A posteriori models 151
(2) A priori models 152
v T ypes o f m od el i 55
vi T h e p r o b l e m o f m o d e l u se - 158
Basic reading 161
11 MODELS IN GEOGRAPHY 162
Basic reading 168
12 THEORIES, LAWS AND MODELS IN GEO
GRAPHIC EXPLANATION—A CONCLUDING
STATEMENT i69
(1) Purpose 172
(2) Form 173
(3) Strategy i75
Part IV
MODEL LANGUAGES FOR GEOGRAPHIC
EXPLANATION
13 MATHEMATICS—THE LANGUAGE OF SCIENCE ,79
A The structure of constructed language system s 179
B M athematical languages 181
C The application o f mathematical languages 184
Basic Reading i9o
14 GEOMETRY—THE LANGUAGE OF SPATIAL
FORM i 9i
1 C o n c e p ts o f sp a ce i 92
ii T h e f o r m a l r e p r e s e n t a t io n o f s p a t ia l c o n c e p t s 197
hi S p a t i a l c o n c e p t s a n d f o r m a l s p a t ia l l a n g u a g e in
GEOGRAPHY 206
XIV CONTENTS
chapter page
A The philosophy o f space in geography 207
B The measurement o f distance 210
C Formal spatial languages in geography 212
(1 ) Topology 217
(2) Projective geometry and transformations 219
(3) Euclidean geometry 223
(4) Space-time problems andMinkowskian geometry 226
I T h e m e a n in g o f p r o b a b i l it y 231
II T h e c a l c u l u s o f p r o b a b i l it y 243
hi T h e l a n g u a g e o f p r o b a b i l it y i n g e o g r a p h y 259
chapter page
Part V
MODELS FOR DESCRIPTION IN GEOGRAPHY
16 OBSERVATION 291
Basic reading 297
18 CLASSIFICATION 326
I T h e l o g ic o f c l a ss if ic a t io n 327
II T h e p u r p o s e o f c l a ss if ic a t io n 331
(1 )General o r ‘natural’ classifications 331
(2) Specific or ‘artificial’ classifications 332
hi T h e s e l e c t io n o f p r o p e r t ie s a n d t h e p r o c e d u r e f o r
CLASSIFYING 332
(1 )Logical division or ‘classification from above’ 335
(2) Grouping or ‘classification from below’ 337
i v Q u a n t i t a t i v e t e c h n iq u e s in c l a ss if ic a t io n 338
(1 )Mahalanobis’ generalised distance (D 2)statistic 342
(2) Principal components and factor analysis 342
(3) Grouping procedures 345
(4) Discriminant analysis 348
v C l a s s if ic a t io n — a c o n c l u d in g c o m m e n t 348
Basic reading; Reading in geography 349
XVI
CONTENTS
chapter page
19 DATA COLLECTION AND REPRESENTATION IN
GEOGRAPHY 35o
I T h e g e o g r a p h e r ’s d a t a m a t r i x 35o
II F il l in g o u t t h e d a t a m a t r ix — s a m p l in g 356
(1) Purposive or judgement sampling—the ‘case-
study’ approach 359
(2) Probability sampling 36i
(3) The sampling frame 567
(4) Sample data in geography 368
hi R e p r e s e n t in g d a t a — t h e m a p 369
i v R e p r e s e n t in g d a t a — t h e m a t h e m a t ic a l r e p r e s e n t a
t io n OF PATTERN 377
Basic reading 386
Part VI
MODELS FOR EXPLANATION IN GEOGRAPHY
20 CAUSE-AND-EFFECT MODELS 389
i The l o g ic a l s t r u c t u r e o f c a u s e - a n d - e f f e c t a n a l y s is 390
11 T h e a p p l ic a t io n o f c a u s e -a n d -e f f e c t m o d els 392
hi C a u s a l system s 396
iv C a u s e a n d e f f e c t a n d d e t e r m in is m i n g e o g r a p h y 400
v C a u s e - a n d - e f f e c t a n a l y s is i n g e o g r a p h ic r e s e a r c h 404
Basic reading; Reading in geography 406
Figure page
2.1 A diagrammatic representation of the relationship
between percept, concepts, and terms 19
2.2 A diagram to show how two rather different languages
may be developed within the same context of per
ceptual experience and conceptual development 20
4.1 A simplified hierarchical structure of scientific laws 32
4 .2 The ‘Baconian’ route to scientific explanation (Route
One) 34
4·3 An alternative route to scientific explanation (Route
Two) 34
1 0 .1 Chorley’s typology of models embedded within a map
of geomorphic activity 156
PART ONE
Philosophy, Methodology
and Explanation
λΑ
Chapter 1
Philosophy and Methodology in Geography
interesting phenomenon but one that did not really arouse his
interest. The questions that we ask, therefore, are partly conditioned
by our training. The explanations that we seek tend likewise to be
conditioned. Occasionally, however, this tradition is broken. A par
ticular generation may come to regard the dialogue of question and
answer that has taken place in the past as leading up blind alleys.
It is a charge that all disciplines at some time or other have been
liable to, that they have become enamoured of questions that have
no real interpretation in terms of concrete experience, or that they
have simply set up unrealistic questions and a neat mechanism for
providing seemingly satisfying, if equally unreal, answers. On other
occasions it seems as if a discipline has worked out a particular vein
of thought, and requires to shift the location of activity to some other
plane. At these points in the history of a discipline we are likely to
find an example of what Thomas Kuhn (1962) calls a scientific
revolution, the shift from one paradigm to another (a notion we will
consider later).
It is not, however, my concern to consider the history of geography
in this light. But it is vital to remember that the notions of conflict,
of surprise, of questioning, and of searching for some answer, demand
some preselection on the part of the individual, and that this pre
selection or filtering is very much conditioned by his training. At
any one time, therefore, a particular geographer is likely to be con
cerned with reducing a few seemingly unexpected results to expected
results. It would be impossible for him to be concerned with all
problems and all issues, even among those that come to his attention,
while the number of potential conflicts and questions seems to be
infinite.
B Constructing an explanation
The purpose of an explanation may be regarded as making an un
expected outcome an expected outcome, of making a curious event
seem natural or normal. In general there are three ways in which we
may do this. Some would argue that they are mutually exclusive,
some that they are complementary, some that they really amount
to the same thing in the long run. We shall, however, just briefly,
examine their characteristics.
(i) Probably the most important approach to constructing expla
nation is that generally known as the deductive-predictive
approach. It is developed in great detail by writers such as
Braithwaite {i960), Nagel (1961), and, above all, by Hempel
(1965), who has sought to extend an approach based on classical
I4 PHILOSOPHY, METHODOLOGY AND EXPLANATION
physics to explanation in all fields of enquiry. We will consider
this mode of explanation in some detail in later chapters. For the
moment it will be sufficient to note its main characteristics. The
objective is to establish statements or ‘laws’ and to show,
empirically, that these laws govern the behaviour of various
types of event. The law is then assumed to be a universally true
statement (that is, it applies independently of space or time). The
law functions in an explanation as follows: a set of initial con
ditions, which the law covers, is first stated and it is then shown
that these taken in conjunction with the law must necessarily
result in the event to be explained. It may be noted here that
prediction and explanation are symmetric in this case and that
there is no essential difference between them. The basic problem
which such a view of explanation poses is, of course, that of
providing adequate support for the supposedly universally true
statements.
(ii) An alternative view has been developed by a number of writers,
such as Hanson (1965), Toulmin (1960B), Wisdom· (1952), Bam-
brough (1964), and Ryle (1949). This view has been termed
the ‘relational’ view (Workman, 1964). In the relational view
explanation is regarded as a matter of relating the event to be
explained to other events which we have experienced, and
which, either through familiarity or analysis, we no longer find
surprising. The behaviour of planets, for example, may be
related to the familiar performance of apples falling from trees.
The essence of explanation, then, lies in providing a network of
connections between events. According to this view, laws are not
necessarily universally true statements, but simply a convenient
device for bringing information derived from particular instances
to bear on happenings in other instances (Workman, 1964;
Bambrough, 1964). The deductive form of explanation, there
fore, which seeks to explain the particular by reference to the
general, may be regarded as a particular form of relational
explanation. Relational explanation is, however, rather broader
in its approach than deductive explanation by itself.
(iii) Workman (1964) proposes a rather different form of explana
tion, which it might be convenient to term explanation by way
of analogy, or, as it is sometimes called, model explanation.
Given the confusion that surrounds the term ‘model’, however,
we will avoid it here. Workman begins by noting how an
unexpected occurrence can be made less unexpected by develop
ing some ‘picture’ of events in such a way that the unfamiliar
becomes or seems more familiar. The explanation thus contains
a description of something that is not observed but which may,
for example, be obtained by analogy. This description provides
adequate predictions and can be shown to be true in the sense
that it does not lead to contradictions. In these circumstances
THE MEANING OF EXPLANATION I5
these issues which does not presuppose too much with respect to the
philosophy of perception or the philosophy of language. Since this
book is primarily concerned with rational explanation (with scienti
fic explanation interpreted in its broadest sense), we shall develop
this schema with respect to this particular style of explanation.
Caws (1965, 69) writes:
Starting from perception, which is regarded as a gross smoothing over of the
surface of reality (owing to the crudity of our sense organs), science proceeds to
the construction of conceptual schemes whose order reflects the order of per
ception, and links these with specialised languages for the purpose of making
predictions.
Fig. 2.2 A diagram to show how two rather different languages may be
developed within the same context of perceptual experience and concep
tual development. The languages have only a small area of overlap and
hence only a few terms can be translated from one language to the other
(1after Lenneberg, 1962, 107).
of what there is.’ Within each segment of the large circle, we may
distinguish an area of percepts which can be related to terms, an
area of concepts that may be related to terms, and so on. But there
are large areas of conceptual thought that cannot be represented by
terms, many terms that have no relation to percepts, and so on.
Any particular language will succeed in representing only a
limited number of concepts and percepts and it is quite likely that
different languages will represent rather different sets of concepts
and percepts. This kind of situation is shown in Figure 2.2. Different
languages therefore vary in their capacity to convey information:
English is thus sometimes regarded as the language of empiricism,
THE MEANING OF EXPLANATION 21
Basic Reading
Caws, P. (1965), chapter 10.
Boulding, K. E. (1956).
Kuhn, T. S. (1962).
Reading in Geography
Lowenthal, D. (1961).
ι --
.
PART TWO
The Methodological
Background and
Explanation in Geography
Chapter 3
Geography and Science—the
Methodological Setting
logy rely upon the intuitive reasonableness of their theories and not
upon empirical verification. By contrast, there are those, such as the
behavioural psychologists and the econometricians, who proceed by
observation and measurement and who rely upon direct empirical evi
dence for the confirmation of hypotheses. This latter group probably
have more in common with the physicist than they do with the
historian as regards their methodological thinking.
Given these contrasts even within disciplines, a complete dis
cussion of the methodological background to geographic thought
becomes impossible. Yet there are a number of common themes in
the arguments regarding an appropriate methodology. Smart’s
(1959) argument regarding biology, and Simpson’s (1963) argu
ments regarding geology, have a very similar ring to those of the
historian resisting the notion of laws in history. The idea of unique
ness is common to history, geology, geography, sociology, and so on.
These arguments need to be examined somehow. We shall therefore
begin by examining what might be called the standard model of scien
tific explanation. This is the methodological system which philoso
phers and logicians have derived from a study of explanation in the
natural sciences—particularly physics. We shall ignore the deviant
view in the natural sciences, but discuss the problems involved in the
application of this standard model to explanation in the social sciences
and history, for it is here that controversy has been most vigorous
and the issues most clearly stated. By such means it is hoped that the
central issues ofmethodological controversy in all sciences will become
clearer. It is of course against this background that geographers have
worked and developed their own methodological opinions. What
will become clear, however, is that there is little, if anything, in the
way of methodological controversy in geography that is not fully
covered in other disciplines. We can expect, therefore, to derive
a great deal from a look at methodological controversy elsewhere,
particularly since the arguments are often far sharper and far clearer
in other disciplines than they are in our own.
Chapter 4
Scientific Explanation—the Model
of Natural Science
Francis Bacon. It does not describe how the scientist in fact proceeds,
but it does describe one of the ways in which a scientist might
describe his actions so as to meet with the approval of other scientists.
The defect of this particular structure, however, is the assumption
that the processes of ordering and structuring data are somehow
independent of the theory ultimately constructed. In his critique of
this, the popular view of the scientific method, Churchman (1961,
SCIENTIFIC EXPLANATION 35
71) has argued that ‘facts, measurements and theories are methodo
logically the same5. Applying an a priori classification system to a
set of data may thus be regarded as an activity similar in kind to
postulating an a priori theory.
(2) Route 2
The second route (see Figure 4.3) whereby we may justify scientific
conclusions clearly recognises the a priori nature of much of scientific
knowledge. It firmly rests upon intuitive speculation regarding the
nature of the reality we seek to know. At its simplest, this involves
some kind of intuitive ‘picturing’ of how that reality is structured.
Such a priori pictures we shall later identify as a priori models. With
the aid of such pictures we may postulate a theory. That theory
should have a logical structure which ensures consistency and a set
of statements which connect the abstract notions contained in the
theory to sense-perception data. The theory will enable us to deduce
sets of hypotheses which, when given an empirical interpretation,
may be tested against sense-perception data. The more hypotheses
we can check in this fashion, the more confident we may feel in the
validity of the theory provided, of course, that the tests prove positive.
In the process of elaborating or seeking to test a theory we may resort
to another kind of model—an a posteriori model—which expresses the
notions contained in the theory in a different form—say in mathe
matical notation. In some circumstances model-building may here
amount to developing an experimental-design procedure, and a
primary function of this procedure is to lay down the rules whereby
we may define, classify, and measure the variables which are relevant
for testing the theory. By using such experimental designs we may
amass evidence to confirm the hypotheses contained in the theory.
But we can never prove an individual hypothesis in an absolute sense.
All we can do is to establish a certain degree of confidence in the
theory. Statements contained in a theory which commands con
siderable support we may call scientific laws. The difference between
a hypothesis and a scientific law may be regarded, thus, as a matter
of degree of confirmation or degree of confidence.
We may view scientific knowledge as a kind of controlled specula
tion. The control really amounts to ensuring that statements are
logically consistent and insisting that at least some of the statements
may be successfully related to sense-perception data. The procedures
involved in this second route to scientific explanation are extremely
complex. In later sections, therefore, we shall explicitly consider the
nature of theories, laws, and models, and examine some of the pro
cedures which a scientist calls into play in order to ensure the utility
36 METHODOLOGICAL BACKGROUND, EXPLANATION IN GEOGRAPHY
a solution within the context of these rules. The anomalies that arise
are the leaven for the rise of a new paradigm. But within the paradigm
the rules for confirmation are firmly laid down even if they can
be given no strict logical justification. Scientists could thus accept
Boyle’s law as proven even though there were only twenty-five
observations, simply because the manner in which those observations
were obtained conformed to the paradigm of scientific experiment.
Certain heuristic rules govern the use of induction and these rules
are established in accordance with the value system of a community
of scientists. It is, according to Kuhn, a hopeless task to try to provide
logical justification for such rules.
I TECHNIQUES OF INVESTIGATION
to u n c o v e r th e m e c h a n ic a l law s w h ic h g o v e rn e d h u m a n b e h a v io u r.
R e d u c in g e x p la n a tio n to som e v a ria tio n o n th e law s o f m ech an ics
p ro v o k e d a v ig o ro u s re a c tio n . H isto ria n s a n d sociologists su ch as
D ilth ey , W e b e r, a n d P a re to , re a c te d a g a in st th e p ositivism o f C o m te,
S p en ce r, M ill, a n d M a rx , w h ile F re u d b e g a n to u n c o v e r ev id en ce o f
m e n ta l processes t h a t c o u ld n o t easily b e b ro u g h t w ith in th e fra m e
w o rk o f m e c h a n istic e x p la n a tio n (H u g h es, 1959). T h e co n tro v ersy
o v er m ech a n istic e x p la n a tio n a t th e b e g in n in g o f th e tw e n tie th ce n
tu ry , p a ra lle le d in g e o g ra p h y b y th e d eterm in ist-p o ssib ilist a rg u
m e n t, c re a te d a n in te lle c tu a l leg ac y o f g re a t significance.
The evidence now suggests that mechanistic explanations con
tained in Newtonian physics and, less appropriately, in Spencer’s
sociology, can be applied only in limited circumstances. In the mid
nineteenth century statistical explanations were already develop
ing in physics, although it was not until Heisenberg enunciated the
‘uncertainty principle’ in quantum physics that chance variation
became a central concept in physics. The overthrow of the Newtonian
concept of a mechanistic universe in physics is but one example of a
radical change in the concepts employed by a particular discipline as
it grows and develops. Similarly, there are wide differences between
disciplines as regards the concepts employed. The problems of
biology are increasingly being approached by way of systems, and
systems analysis has become a powerful tool in social science. Acker
man (1963) has suggested that geographers should also turn to a
systems framework for formulating their problems. There are many
different types of conceptual framework possible in rendering ex
planations and it may be questioned whether it is reasonable to
presuppose that only one conceptual framework does or can domi
nate all science. Mechanistic, genetic, and systems explanations are
examples of alternative conceptual frameworks.
Different disciplines adopt different conceptual frameworks. To
some, mechanistic explanations may seem appropriate, to others the
more complex stochastic form may seem more suitable. It is not
possible to differentiate between natural and social science in terms
of the concepts adopted since there is clearly as much variation
within disciplines as there is between them. The demand for metho
dological differentiation between natural science and the social
sciences and history on the grounds of major conceptual differences
cannot be sustained.
This general conclusion that differentiation is not possible on the
grounds of either techniques or conceptual development does not
involve denying the difficulty of transferring techniques and concepts
from one discipline to another. It is obvious that the experimental
48 METHODOLOGICAL BACKGROUND, EXPLANATION IN GEOGRAPHY
(1 9 3 3 , 154) took the view, for example, that ‘the moment historical
facts are regarded as instances of general laws, history is dismissed.
Dray (1957, 45) portrays the argument as follows:
History is different in that it seeks to describe and explain w hat actually
happened in all its concrete detail. It therefore follows a priori that since laws govern
classes or types of things, and historical events are unique, it is not possible lor the
historian to explain his subject-matter by means of covering laws. 11 he is to
understand at all, it will have to be by some kind o f special insight into particular
connections.
This declaration of ‘methodological independence’ on the part
of history came in the latter part of the nineteenth century. Win e
band, Dilthey, Rickert, and other historians, chose to differentiate
between those subjects which they regarded as being susceptible to
the idiographic method (the exploration of particular connections) and
those which were concerned with establishing generalisations and
were nomothetic in character. The fundamental concern with unique
events was, to them, sufficient justification for separating history
from many other sciences and for regarding explanatory form itself
as being radically different. In the ensuing controversy over this
proposed dichotomy different disciplines were forced to take sides
It is perhaps a tribute to the strength of the German school of
historiography that the idiographic view has not only dominated the
practice of history until the present day, but has overlapped into
other areas of social and natural science. Thus the social scientist
has similarly been enjoined to accept the uniqueness thesis. Max
Weber (7949, 80), one of the father-figures of modern sociology, has
written:
Laws are important and valuable in the exact natural sciences, in the measure
that those sciences are universally valid. For the knowledge of historical phenomena
in their concreteness, the most general laws, because they are most devoid oi con
tent are also the least valuable. T he more comprehensive the validity— or scope
of a term, the more it leads us away from the richness of reality since in order to
include the common elements of the largest possible number of phenomena, it
must necessarily be as abstract as possible and hence devoid of content. In the
cultural sciences the knowledge of the universal or general is never valuable in
itself.
Simpson (1963) has similarly described the methodological attitude
of many geologists as conforming to the uniqueness thesis (above,
p. 28). The argument has not been without force in geography
either, and it is perhaps no surprise that German geographers, such
as Hettner, claimed that geography was an idiographic rather than
nomothetic science. Given the important influence of German
methodological thinking in Hartshorne’s work it is hardly surpns-
PROBLEMS OF EXPLANATION, SOCIAL SCIENCES AND HISTORY 51
does not involve an appeal to any set of laws, but involves showing
that a particular person or group of people had a rational reason
for acting in a particular way at a particular time. Once these
rational reasons have been discovered the historian has completed
his task. Further, it is quite sufficient for the historian to use ‘common
sense’ explanations of events in terms of the logic of the situation.
Such common-sense explanations are perfectly adequate, according
to Dray, and although there is no reason to deny the possibility of
developing laws of any significance in history, there is no need to
invoke the existence of such laws to provide perfectly adequate
explanations of historical events. This view of Dray’s has provoked
a strong counter-argument from Hempel (1965, 469-87), but at this
point the controversy becomes extraordinarily complicated.
Through this tangled web of argument and controversy in the
philosophy of history a number of conclusions may dimly be per
ceived. These conclusions extend to those disciplines which have a
strong relationship to history and look to the idiographic method:
(i) The uniqueness thesis cannot be sustained without substantial
modification.
(ii) There can be no explanation or rational description without the
employment of generalisations, classes, concepts, and principles.
Beyond this it is difficult to find any general agreement and it is only
possible to summarise some of the major areas of controversy:
(iii) Whether the generalisations which historians undoubtedly
employ in providing explanations qualify as laws depends upon
the view taken of the function of the generalisation in historical
explanation and the criteria set up to determine whether a
particular statement may be regarded as a law.
(iv) Although there is no doubt that explanation in history could
conform to the norms derived from natural science, there is
considerable disagreement as to whether such a procedure would
in any way ease the difficulty of providing explanations of
complex historical events.
The Weber-Winch thesis denies that the social sciences and history
can achieve a sufficient degree of objectivity to allow the independent
verification of hypotheses. This automatically restricts the develop
ment of theory in the usual scientific sense and restricts the kinds
of question that can be asked regarding social phenomena.
This view then poses the problem of finding some alternative
mode of validating statements. To Weber and Winch explanation
in the social sciences involves understanding a particular event by
empathy with the individual or individuals involved in the event.
Putting oneself in another person’s shoes an operation some
times called verstehen—provides us with the only kind of under
standing which is possible regarding human behaviour. This view is
common in social science and history. Collingwood (1946, 283)
writes:
For science, the event is discovered by perceiving it, and the further search for
its cause is conducted by assigning it to its class and determining the relation
between that class and others. For history, the object to be discovered is not the
mere event, but the thought expressed in it. T o discover that thought is already
to understand it.
as in the past, upon intuition and judgement alone. Clearly the ideal
is creative imaginative effort backed by the control which scientific
method gives us over the reasonableness and consistency of statements
which we make about reality. Such an ideal is surely not so unattain
able?
Basic Reading
Dray, W. (1964).
Rudner, R. S. (1966).
Subsidiary References
Brown, R. (1963).
Hempel, C. G. (1965).
Kaplan, A. (1964).
Nagel, E. (1961), chapters 13, 14, and 15.
Chapter 6
Explanation in Geography—Some
General Problems
The discussion of the previous two chapters has centred on the nature
of scientific method and on the difficulty of extending that method
to the social and historical sciences. The views discussed were those
of philosophers of science and the methodologists of the various dis
ciplines. The relationship between these views and the conduct of
empirical investigation by the practitioners of some discipline were
only incidentally considered. In examining explanation in geography
this relationship becomes of great importance. Methodological in
vestigation in geography should not be regarded as an end in itself.
It is, rather, concerned with elucidating the conduct of empirical
enquiry.
It has been suggested on occasion, however, that the work of
philosophers and logicians bears little relationship to the conduct of
empirical work. It is sometimes claimed that much of the philosophy
of science is too normative to be useful and that the standards set
up are far too rigorous for the practical purposes of most disciplines
(cf. Churchman, 1961, 341-3; Wilson, E. B., 1955). Zetterberg’s
comment, quoted at the end of the last chapter, is indicative of such
an attitude towards the rigorous employment of scientific method
in the social sciences. But philosophy of science has changed. The
practical problems of extending the scientific method to the social
and historical sciences (examined in some depth by Kaplan (1963),
Nagel (1961), and Hempel (1965), for example) and the growing
recognition of the behavioural aspects of scientific decision-making
(identified by Churchman (1961), and Kuhn (1962) for example)
have provoked a continuing revolution in the attitudes of philoso
phers of science. Explanation in physics is no longer their sole con
cern. It is still relevant therefore to examine the extent to which
philosophers of science have made general statements which are of
use to geographers in the conduct of empirical work.
The methodology of a particular academic discipline is not
determined by philosophers of science. It is partly evolved in ‘rule
62
EXPLANATION IN GEOGRAPHY 63
of thumb’ fashion by the practitioners of that discipline and partly
developed by methodologists trying to formalise procedure within that
discipline. To some extent both practitioners and methodologists may
be influenced by philosophy of science. But it is possible for, say, the
methodologist of geography to make statements which are at variance
with those of philosophers of science or to make statements which are
at variance with actual practice in most geographic research. The
purpose of this section, therefore, is to try to identify such contradic
tions where they exist, attempt to resolve certain methodological
arguments in geography, and finally to attempt to provide a broad
general statement of the explanatory frameworks appropriate for
geographic investigation. In this, a number of important general
questions need to be considered:
(i) What is the relationship between the methodological argument
as developed in geography and the methodological argument
regarding knowledge in general? Put another way, how far do
the views of the methodologists of geography tally with the views
of philosophers of science and, if there are differences, what
rational basis can be provided for such differences?
(ii) What is the relationship between the statements made by the
methodologists of geography and the practice of geographers as
revealed in their empirical work?
(iii) What is the relationship between the explanatory form accepted
by practising geographers and the explanatory form accepted
by the practitioners of other disciplines?
Most attention will be paid to the first question, mainly because the
relationships are easier to determine. Some comments will be made
on questions (ii) and (iii) here, but question (iii) will be examined
in some detail in later sections (see below, pp. 107-29).
the result of what Ackerman (1963, 430) has called a too close
association between geography and the disciplines of geology and
history—disciplines which, as we have already seen (above, pp.
28-9, 49-54) were dominated by the notion of uniqueness and
the idiographic method. The net result for geography was ad
herence to a particular view of the nature of geography coupled to a
particular view of explanation in geography. This association be
tween a particular set of objectives (description and interpretation
of unique areas) and a particular explanatory form (the idiographic
method) formed a powerful orthodoxy from which geographers
found great difficulty in breaking free.
This view is expressed in its most rigorous form in the argument
over exceptionalism in geography. This argument centres on a state
ment about the place of geography in the system of knowledge
which was made by Kant and which has subsequently become a basic
tenet of orthodox geography as expounded by Hettner and Harts-
horne. There is a very considerable literature on exceptionalism
in geography (Hartshorne, 1939; 1955; 1958; 1959; Schaefer, 1953;
Bunge, 1966, chapter 1; Blaut, 1962; Haggett, 1965A, 2-4; Lewis,
Ϊ965), and it will, therefore, be examined here.
Kant apparently characterised the position of both geography and
history in relation to the other sciences as follows:
W e m a y cla ssify o u r e m p ir ic a l k n o w le d g e in e ith e r o f tw o w a y s; e ith e r a c c o r d in g
to c o n c e p tio n s or a c c o r d in g to tim e a n d sp a c e in w h ic h th e y are a c tu a lly fo u n d . . . .
T h r o u g h th e fo rm er w e o b ta in a sy stem o f n a tu r e , su clq as t h a t o f L in n a eu s;
th r o u g h th e la tte r , a g e o g r a p h ic a l d e sc r ip tio n o f n a tu r e . . . G e o g r a p h y a n d
h isto r y fill u p th e e n tire c ir c u m fe r e n c e o f o u r p e r ce p tio n s: g e o g r a p h y th a t o f sp a ce ,
h isto r y th a t o f tim e . (H a rtsh o r n e , 1939, 1 3 4 -5 ·)
There is, therefore, less excuse at the present time for avoiding the
EXPLANATION IN GEOGRAPHY 77
construction of geographic theory. It might be claimed, however,
that geography, like history, is a consumer rather than a producer of
theories and laws. The covering laws which enter into geographic
explanation may thus be regarded as derivative of some other dis
cipline. If theory in geography is derivative then the geographer
should be aware of the wide range of theoretical constructs which
are available to him. Derivative theory, as we shall later see (below,
pp. 117-24), plays an important role in geography but, until
recently, there has been a tendency to avoid the full responsibilities
inherent in deriving theory from other disciplines. For the most part
geographer! have tended to imply either rather trivial observation!
about_ human behaviour a§ eovering laws in human geography, or
have implied erode deterministie (physieal environmental) law§=
laws whieh most geographers have elaimed to rejeet, An examina=
don of almost any regional textbook will serve to demonstrate how
strong the environmentalist tradition still is: This is not to say,
g that there is anything inherently wrong with the envirom
mentalist thesis, lo t there is something wrong with pretending to an
©bjeetive regional synthesis whieh implies a set of environmentalist
laws whieh have previously been diselaimed:
Derivative theory is undoubtedly more strongly developed in ge©=
graphy than indigenous theory* Mueh of the theorising now eurrent
in geography is of this type* The question therefore arises as to
whether indigmms theory ean be developed in geography as opposed
to derivative theory and, if so, what the relationships are between them,
This issue will be examined later, but for the moment it is worth
noting a preliminary eonelusionj that when geographers develop
their analytie thinking in terms of spaee=time languages (the spatial
framework for thought as proposed by ICant or later relativistie
spatial frameworks), then indigenous theory may be developed, but
when geographers resort to property languages the resultant theory
is elearly either aetually or potentially derivative of some other
diseipline, This important statement will not be justified here for it
is the subjeet of analysis in ehapter § (see also Harvey, tgi^A;
More generally we may eonelude that there is no logical reason
for supposing that theory eannot be developed in geography or that
the whole battery of methods employed in seientihe explanation
eannot be brought to bear on geographie problems. There are, it
must be admitted, some serious praetieal problems involved, lu t
these praetieal difheulties eertainly eannot be invoked to prove that
geographie thought is essentially different with respeet to explanatory
form from all other diseiplines exeept history and, perhaps, geology,
78 METHODOLOGICAL BACKGROUND, EXPLANATION IN GEOGRAPHY
The Kantian thesis is not totally irrelevant to geographic thinking—
indeed it contains some important insights into the structure of
geographic thinking. But it contains some premisses which cannot be
accepted at the present time (e.g. the premiss of an absolute space)
and it contains some conclusions which are not rationally derivable
from the premisses. The Kantian thesis, it must be concluded, needs
profound modification if it is to be satisfactory for the current needs
of geography as an independent discipline embedded in an over-all
structure of knowledge.
I l l EXPLANATION IN GEOGRAPHY
(a) Cognitive description. Under this heading are included the col
lection, ordering, and classification of data. No theory may be
explicitly involved in such procedures, but it is important to note
that a theory of some kind is implied. Thus classification involves
some a priori notions about structure and these notions really amount
to a primitive theory. In the early stages of a discipline’s development
such theoretical assumptions may be amorphous and ill-defined.
In the later stages classification procedures tend to become a part of
experimental design and, hence, determined by the particular theory
being examined, or the measurement and classification of data may
be derived directly from theory. Cognitive descriptions may thus
range in quality from simple primary observations through to sophis
ticated descriptive statements.
. . . nothing seems to me clearer than that geography has already suffered too
longfrom the disuse of imagination, invention, deduction, and the various other
mental faculties that contribute towards the attainment of a well-tested ex
planation. It is like walking on one foot, or looking with one eye, to exclude
from geography the ‘theoretical’ half of the brain power . . . Indeed, it is only
as a result of a misunderstanding that an antipathy is implied between theory
and practice; for in geography, as in all sound scientific work, the two advance
most amiably and effectively together.
W. M. D avis (1899)
Chapter 7
Theories
ig6j, chapter 3). This suggests at least one important criterion for
distinguishing a law. Unfortunately the interpretation to be put
upon universality is not entirely clear. We shall begin by taking the
notion of universality in its most rigid form.
The universality criterion requires that laws should not make
specific or tacit reference to proper names. Consider the proposition
that ‘towns of similar size and function are found at similar distances
apart.’ (This proposition may be taken from Thomas’s, 1962, work.)
The term ‘town’ can be defined only with reference to human social
organisation and it carries with it, therefore, an implicit reference to
the proper name ‘earth’. Within such a context the statement may
be true, but the universality criterion has undoubtedly been offen
ded. To get round this difficulty we may attempt to define ‘town’
in terms of a set of properties which we claim are possessed by towns
and only towns. In an infinite universe, however, there may well be
some phenomenon which possesses all the properties listed without
being a town. Again we are not justified in regarding the statement
as being a proper law.
J. J. Smart (1959) used the argument of the above paragraph to
show that laws could not be developed in biology. According to
Smart the only strict laws to be found in the whole of science are
those to be found in physics and, perhaps, chemistry. These, he
suggested, were truly universal in nature. This view automatically
excludes the development of laws in biology, zoology, geology,
physical geography, etc., except in so far as such disciplines can
reduce their statements to those of physics. The social sciences and
human geography are even more seriously affected.
But there are powerful arguments against interpreting univer
sality in such a strict manner. There are two ways in which we may
justify some relaxation of it. With ‘purely’ empirical propositions
it may prove useful to draw a distinction between philosophical and
methodological universality. Philosophical universality involves the
belief that universally true statements can be made. Such a belief
may be supported by reference to some set of metaphysical proposi
tions—such as the Platonic doctrine of universal essences—or else
it depends upon showing that a statement is in fact universally true.
The latter course is essentially an inductive step and, therefore, a
degree of uncertainty is involved. A proposition can never be shown
empirically to be universally true. This applies as much to the ‘strict’
laws of physics as it does to the ‘mere generalisations’ of biology and
economics. Philosophical universality implies methodological uni
versality, but the reverse relationship does not hold. We may regard
statements as f/they were universally true without necessarily believ-
HYPOTHESES AND LAWS 103
ing that they are or even assuming that they will ultimately be shown
to be so. This position we shall term methodological universality.
In such a case it becomes a matter of deciding whether it is useful
and reasonable to regard a statement as i f it were universally true,
and, hence, law-like.
It is interesting to note that Smart suggests that laws in the strict
sense cannot be developed to account for the behaviour of complex
systems. This seems to reflect the declining utility and reasonable
ness of regarding extremely complex organisations as being subject
to laws. But even such apparently intractable phenomena as wars
can be treated as i f they were subject to laws, with intriguing results.
Of course the laws which writers such as Richardson (1939) and
Rapoport (i960) have discussed cannot easily be regarded as
‘universally valid’ and certainly neither of those writers would claim
that status for their results. But then many of the so-called strict
laws of physics and chemistry are also questionable in this respect.
This brings us to the second point of interpretation of the universality
criterion which may lead to some relaxation in its application.
When considering the nature of scientific theories it was noted
that a theory requires the development of an abstract calculus. One
of the functions of this abstract system was to turn empirical relation
ships located in space and time into abstract sets of relata. It follows
from this that all deduced consequences of such a theoretical struc
ture must be stated as i f they were universally true propositions. If
we are dealing with fully formalised systems the statements are purely
analytic and have no empirical status. They are given empirical
status by means of the appropriate text. Invariably that text makes use
of proper names and this in itself makes a fiction of the supposedly
real universality of the statement. This applies as much to the
propositions of natural science as to those of social science. As
Brown (1963, 147-8) has pointed out:
T he qualification laid upon a scientific hypothesis by its ‘text’ includes the
definition of its terms and an accurate statement of the limits within which it is
supposed to hold. Neither of these need appear in the generalisation itself, and
however obvious each is, critics of the social sciences often speak as though . . .
hypotheses are deficient in not containing such qualifications on their face. It is
worth time, therefore, to remind ourselves that the generalisations of the natural
sciences do not differ on these points from those of the social sciences.
Given the exposition of the previous two sections we may now con
sider the nature of the statements we make in geographic research
and the form we use to link those statements together into a coherent
explanatory structure. The aim of this section, therefore, is to examine
the degree to which geographers have resorted to scientific explan
ations, and to investigate the potential of a more conscious use of ‘the
scientific method5 in geographic explanation. This general form of
explanation—and we are here referring to the very broad model of
scientific explanation—has been extremely efficient in both the
natural and social sciences. This is not to claim that all our out
standing substantive problems will be solved merely by the touch
of the glittering wand of scientific explanation. Far from it. Scientific
explanation, as all mature users of the method fully recognise, has its
limits. The efficiency of the method is, as Boulding (1956,73) has point
ed out, bought at the price o f‘a severe limitation of its field of enquiry
and a value system which is as ruthless in its own way in the censor
ing of messages as the value system of primitive man.5 But we shall
become conscious of these limits only when we have explored fully
what can be done within their confines. This, by and large, we have
failed to do.
I LAWS IN GEOGRAPHY
II THEORIES IN GEOGRAPHY
(d) The spatial distribution theme. The view has frequently been put
forward that the overriding aim of geography is to describe and
explain the distribution of phenomena over the earth’s surface.
Hartshorne regarded such a study as being an essential preliminary
to the study of areal differentiation, but to others this aim became in
itself a sufficient focus for geographic research. To a degree opinion
on this issue splits according to the regional-systematic dichotomy,
and we therefore find many of the systematic aspects of geographic
research (as in climatology and economic geography) developing
around this fundamental theme as their focal point of interest.
Locational analysis, at present an active area in geographic research,
may conveniently be regarded as a manifestation of interest in the
theme of spatial distribution.
n6 THE ROLE OF THEORIES
(e) The geometric theme. The ‘geometric’ tradition in geography is
an extremely old but relatively neglecte done (van Paassen, 1957;
Haggett, 1965A, 15-16). But since 1950 interest in this tradition has
increased markedly and with the statements by Bunge {1966) and
Haggett (1965 A) we must regard this theme as of major significance
in geographic research (see chapter 14).
The five major themes presented here are neither mutually exclusive
nor entirely inclusive of all geographic work. But each in its way is
capable of providing an operational definition of the ‘nature’ of geo
graphy and it is within the context of such operational definitions
that geographers have begun to formulate concepts and theories. In
some instances, as with locational analysis and geometric analysis,
theory has become more explicit and a formal approach has begun
to be adopted. In other instances the degree of formal theory con
struction has been negligible and the main aim descriptive. But
T he m om ent that a geographer begins to describe an area . . . he becomes
selective (for it is not possible to describe everything), and in the very act of
selection demonstrates a conscious or unconscious theory or hypothesis concerning
what is significant. (Burton, 1963, 156.)
Such concepts may ultimately act as the basic postulates for theory.
In most cases explicit theory has not been developed from them.
They thus function as either relative primitives or else as the far vaguer
postulates for Type 4 theories. In some instances particularly in
location theory, geometric analysis, and in analysis in physical
geography—the basic postulates have been more explicitly developed
and function more in accordance with standard scientific theory. But
in general we cannot identify the basic postulates upon which geo
graphic thinking is based nor does it seem desirable, given the dangers
attendant upon premature formalisation, that we should attempt to
cast geographic thought into such a formal framework. But it is useful
to set up a hypothetical argument regarding the basic postulates and
concepts for an adequate geographic theory, for this argument can
help us clarify certain important philosophical problems. We shall
begin by classifying the concepts used by geographers into those
which are indigenous and those that are derivative (Harvey, 1967B).
Devons and Gluckman (1964) in their book Closed Systems and Open
Minds have pleaded with their fellow-anthropologists for a deliber
ately cultivated intellectual naivety. This involves closing the domain
of their discipline and seeking explanations which can be provided
within that domain, while retaining a certain humility regarding the
effectiveness of the explanations offered. This plea for intellectual
naivety stems not so much from a wish to resist reductionism, as it
does from a recognition of the dangers of amateur interdisciplinary
dabbling. They conclude (p. 261) that
the different social and human sciences may be different realms, in whose border
lands trespass is dangerous save for the genius . . . a social or human scientist may
profit by studying disciplines other than his own. It is dangerous to practise them
without training and appropriate skills.
(a) In many cases concepts are developed to help explain and des
cribe geographical phenomena. When introduced into a rigorous
theory these concepts often require modification, and it frequently
occurs that the postulates are either actually or potentially derivable
from some other discipline. Concepts which turn out to be so deriva
tive usually relate to temporal processes. This observation fits the
generally accepted view of geography as being concerned with spatial
systems rather than with temporal processes per se. It is tempting to
conclude that all aspects of theory in geography which relate to
temporal process are actually or potentially derivative of some other
discipline.
The term model has been avoided as much as possible in the dis
cussion so far. Yet it has become very fashionable in geographic
research. In this respect geography is but a short distance behind
the social sciences in general. As Brodbeck (1959, 373) has pointed
out, model-building has a kind of ‘halo effect’ upon the research
worker since ‘Models are Good Things’. Yet there is no consensus
of opinion among philosophers of science as to what is meant by
a ‘model’, or as to its function in scientific research. The general
confusion surrounding the term is reflected in geographic research
where very different opinions can be found. The meaning and func
tion of the term model thus requires some careful methodological
investigation.
Relationship Function
In this case the model and the theory possess identical formal
structures and the physical properties of the phenomena described
in both model and theory are identical. But not all models are
theories nor are all theories models. In general it is held that the
model exhibits some underlying structure of the theory. A model
may thus be thought of as a kind of skeletal representation of the
theory, but still, it should be noted, the structural characteristics
148 THE ROLE OF THEORIES
The philosopher and logician may regard the problem of model use
as being essentially soluble by logical analysis. The scientist, on the
other hand, simply wishes to know how and under what circum
stances a particular research procedure is or is not justified. In fact
the dispute among philosophers of science and logicians reflects a
number of acute procedural difficulties in the use of ‘models’ (of
all types) in scientific investigation. Further, there is some suspicion
that many of the controversies in empirical science over seemingly
substantive issues are in fact generated by failure to agree as to the
correct procedure to be used in employing models in scientific in
vestigation. This procedural problem can be demonstrated in re
lation to the two ways in which a formal theory may arise in science
and the role which models play in such a process. In brief we may
distinguish between a posteriori and a priori models.
(1) A posteriori models
We have already described the two ways in which formal theory may
arise (above, pp. 32-6). The first route begins with empirical
observation from which a number of regularities of behaviour may
be extracted. To explain these regularities a theory is proposed which
may contain theoretical abstract concepts and eventually the theory
may be given axiomatic treatment and may be verified. This theory
may then be represented by some structural model, which can be used
to facilitate deductions and simplify calculations. In this case the
model is developed in order to represent the theory. By choice, we
can make the model either contain terms and structure which exist
152 T H E R O L E O F T H E O R IE S
in the theory (or can be defined with reference to the theory) or use an
analogue of some kind. In this case the function of the model is
simply to represent something which is already known, and the
only question which arises is that of the appropriateness of a model
for a given purpose and this, as we have already noted, can be fully
defined only if the appropriate general theory can be referred to.
When no general theory exists there must be some doubt as to the
appropriateness of the model representation of the theory.
If a general theory exists, and if the model contains terms and
structures referred to in the theory, then the conditions for Achin-
stein’s ‘model of an x’ are fully met. In such a situation the dangers
inherent in model-use are minimised. But the question arises as to
the effect upon model-use of varying sophistication in the theory
which the model is designed to represent. Model representations of
Type 2 or Type 3 theories which simply involve ‘harmless’ elliptical
argument need not be seriously affected. On the other hand, failure
to specify the theory fully must automatically mean a certain loss of
control of the model-theory relationship. Imprecise or partially form
alised theories may be represented by precisely specified models,
but the nature of the model-theory relationship must perforce remain
unspecified also.
One of the roles of a posteriori model-building is to allow the easier
manipulation of relationships and to facilitate testing procedures.
The less sophisticated the theory, the less control we possess over the
model-theory relationship, and the less we shall be able to tell if
(1) Model conclusions can be transferred to the theory or (ii) A
successful test of the model indicates a successful test of the theory
which the model represents.
With a posteriori models we are thus working along a continuum
from theoretical statements of great certainty represented by emin
ently controllable models, to extremely fuzzy theoretical statements
which may be given precise, but completely uncontrollable, model-
representation. At the latter end of this continuum it will be ex
tremely difficult to tell whether we are invoking ‘model’ or ‘analogue’
representations.
(2) A priori models
A particularly important form of the second route to theory construc
tion lies through giving an interpretation to a completely abstract
calculus (Brown, 1963, 174). This calculus may have arisen simply
from abstract analytic argument, or it may have arisen as a response
to a particular set of empirical problems in a different domain. But
for the moment we shall simply treat that calculus as having a
M ODELS 153
completely abstract formulation as, for example, a set theoretic
development of probability-theory, or an abstract development of
some geometric system. This abstract calculus is then found some
real-world interpretation by giving empirical meanings to the terms
contained in the calculus. By providing a set of interpretative sen
tences we ‘map’ some aspect of the real world into this ready-made
calculus. In other words we begin with the calculus and then seek to
identify a domain of objects and events to which it can be applied.
If it can be shown that this mapping is successful, then the calculus
may be accepted as a model-representation of a theory, and we may,
from the structure of the calculus, infer the structure of the theory.
The model is thus set up first and the theory is developed from the
model.
A priori models are probably more common than a posteriori models.
It is interesting to note that in Braithwaite’s (/96b, 89-90) view a
major difference between a model and a theory is simply that the
former is epistemologically prior to the latter. A model is thus an
a priori analytic construct which is applied to reality, whereas a theory
is a construct which grows out of experience of the real world. This
view of a model is particularly common in the social sciences. Thus
economists usually view a model as an a priori construct (usually
mathematical, but sometimes graphical) into which economic theory
is mapped for clarity of exposition or for set purposes of manipu
lation (Arrow, 1951; Beach, 1957; Koopmans, 1957).
A priori models have a number of different functions. In terms of
the context of discovery, ‘picture’ models of reality can be of great
importance and often play a key part in the psychological process of
theory formation. For the moment we will restrict consideration to
some of the logical properties involved in the use of a priori models.
Thus two of the main functions of such models are to suggest theory,
and to allow manipulations and conclusions to be drawn about
some set of phenomena even in the absence of full theory. In both
cases there are some difficult logical problems of inference and con
trol involved. In the first case it will not be clear whether the model
concepts are the same as those of the theory-to-be, or whether they
are in fact analogues. It is extremely dangerous to infer from a
successfully applied model that the theory which governs the be
haviour of the phenomena described in the model must necessarily
possess the same characteristics as the model. Similarly, predictions
made with the aid of an a priori model construct must be open to
doubt, since it is not possible to know in what respects the model
represents a theory and therefore the respects in which concepts and
structure are similar cannot be assessed.
154 T H E R O L E O F T H E O R IE S
V TYPES OF MODEL
In the same way that a model may have various functions and defin
itions, so it may perform its function through a multiplicity of media.
Again, certain academic disciplines tend to resort to certain kinds of
model. It is useful, therefore, to review some of the typologies of
models which have been developed.
Ackoff (1962), for example, differentiates between iconic models,
which use the same materials but involve changes in scale, analogue
models which also involve a change in the materials used in building
the model, and symbolic models which represent reality by some
symbolic system such as a system of mathematical equations. Each
type of model varies in its appropriateness for different functions,
but, again, there are no hard and fast rules. Ackoff’s classification
seems most closely geared, however, to model-functions which relate
directly to reality. Given the catholic definition of the term as used
by Ackoff, this is hardly surprising. As we have seen, however, a
theory may itself be defined ‘as a language for discussing the facts
the theory is said to explain’ (Ramsey, ig6o, 212; see also above, p. 88).
A theory may itself, therefore, be a symbolic system. Given Ackoff’s
typology it becomes difficult to discuss the crucial model-theory
relationship in a succinct way. But Ackoff’s point is nevertheless an
important one. By choosing to translate from one medium to another
(say, representing a symbolic system by an electric circuit) we may
facilitate manipulation of theory or the development of theory or
prediction etc. Ackoff’s views have been taken up in the geographic
literature by Berry (1963, 105-6).
Chorley (1964) regards all models as being analogues of some
kind, but suggests a classification of models into those which trans
late into analogous natural circumstances (this is similar to Ackoff’s
analogue model), models which involve experimental procedures
(which may either involve a change of scale or translation to ana
logous natural circumstances), and mathematical models (which
appear to correspond to Ackoff’s symbolic models). In a later
presentation, Chorley (1967) revises and extends this classification
system (Figure 10.1). Three main types of model are now envisaged
with a number of sub-types.
uoaes;|ejau96 Bujseajouj
Fig. ιο ί. Chorley’s typology of models embedded within a ‘map’ of geomorphic activity (from Chorley, 1967).
M ODELS 157
2. Physical System:
a. Hardware Model
(i) Scale
(ii) Analogue
b. Mathematical Model
(i) Deterministic
(ii) Stochastic
c. Experimental Design
3. General System
a. Synthetic
b. Partial
c. Black box
(b) Unidentified models. We may postulate a model and find that the
model gives excellent results, but that it is impossible to find any
firm theoretical interpretation. This kind of situation is characteristic
of some of Chorley’s ‘black-box’ models, and may also be character
istic of some statistical models. The rank-size rule, for example,
appears to have no firm theoretical interpretation, and the gravity
model is, to a lesser degree, in this category. In each case several
theories have been developed to explain these regularities, but the
model itself does not indicate in any way what the nature of that
theory should be.
(c) Identified models. A postulated model may give rise to one and only
one theory. This circumstance is clearly the most desirable—but it
is not very common in geographic research.
ι6 ο T H E R O L E O F T H E O R IE S
These two examples indicate the kind of problem which we face when
we use any a priori model in geographic research. Any logical schema
or mathematical system (and in the latter category we may include
geometry) may be regarded as an abstract calculus into which we
can map geographic problems. Such a mapping process provides us
with a powerful, if dangerous, device for generating theory or for
making predictions in the absence of theory. The question arises, of
course, why we should ever bother to use models when their use
poses such difficulty and danger. The short answer to this is that we
have no choice. With very weakly developed geographic theory and
a highly complex multivariate subject-matter, it is inevitable that
the model-concept should play a part in geographic explanation. In
the absence of firm geographic theory, a model can provide a ‘tem
porary’ explanation or an objective (if often inaccurate) prediction.
Such ‘temporary’ uses of the model concept are important, particu
larly in a world which demands some kind of answer to a whole range
of complex socio-economic problems. But in terms of basic research
the primary function of model-building in geography must be
168 T H E R O L E O F T H E O R IE S
so on. In this case we are translating the information on the map into
some other medium for purposes of manipulation—we are con
structing an ‘analogue’ model. In all of these cases we are, however,
making a posteriori use of the completed map or theory.
But the way in which we draw up the map, and the rules which
we follow in manipulating and using the map once it is completed,
contrast with the procedures used to create the map in the first
place. The method followed, the tools and implements used, the
assumptions made, and the activities followed, are very different
indeed. A scientific map or theory relates in some way to empirical
phenomena. In drawing up a map which will allow us to ‘find our
way around’ precisely we need to go out and survey the land—this
involves measuring, sketching, and recording. The cartographer
obeys one set of rules at his drawing-board and a different set of rules
when he is out surveying the land. In the same way that we cannot
judge the activities of the field surveyor by the criteria appropriate
to the drawing-board, so we should not judge the scientist in the
process of seeking for a theory by way of the criteria we use in drawing
up that theory in its final form. This important point has been
thoroughly emphasised by Gilbert Ryle (1949, 269-75). This is not
to say that the field surveyor’s activities are unrelated to the require
ments of the drawing-board. A field surveyor ignorant of the rules
of cartography will probably do a poor job; likewise, a scientist
ignorant of the logical requirements for the statement of scientific
theory will probably make a poor job of seeking out some theory.
The two activities are different but their requirements are closely
interrelated.
In the absence of detailed empirical knowledge the map-maker
may be forced to make a number of a priori assumptions in drawing
up his map. The medieval cartographers, for example, were forced
to assume that the world had some shape before they could record
anything. They may have assumed that the world was round, flat,
or, as did Cosmas Indicopleustes, assume it to be shaped like a
tabernacle. The point about such assumptions was that without
them and without the knowledge of the correct shape of the earth
no map could ever have been drawn. The assumed shapes thus func
tioned as a priori model-constructs. Predicting and explaining by
means of a priori models is, thus, rather like trying to navigate to
Jerusalem with the aid of the Hereford map.
The point of this extended analogy is to try and indicate simply
and clearly the rather different ways in which we should approach
the purpose of theory-construction, the rules which govern the
statement of theory, and the procedures we follow in discovering a
172 THE ROLE OF THEORIES
theory. By means of such an analogy it is also possible to make clear
the rather difficult notion of the model-theory relationship. But it
does not do to pursue the analogy too far. We shall therefore return
to the substantive problems of methodology at issue and try to
fashion some direct concluding statement on the purpose, form, and
strategy of investigation in geographic research.
(1) Purpose
This book is mainly concerned with the way in which we make
statements and the strategy of developing general explanatory state
ments of considerable reliability. Yet these two issues are not in
dependent of purpose. It has thus been freely acknowledged that
speculation regarding the ‘nature of geography’ forms the main
spring for the creation of geographic theory. Without the motivation
and direction provided by such speculation, theory could not develop
(see above, pp. 114-16). Nothing in the preceding chapters can be
used to discredit the basic views which geographers have of the pur
pose of geographic investigation. In some instances it is hoped that
these traditional views have been extended and sharpened. Con
sider, for example, the following statement made by Hettner in 1923:
T h e th o u g h t o f a g e n e r a l e a rth s c ie n c e is im p o ssib le o f r ea lisa tio n ; g e o g r a p h y
c a n b e a n in d e p e n d e n t sc ie n c e o n ly as c h o r o lo g y ; th a t is, as k n o w le d g e o f th e
v a r y in g ex p r essio n o f th e d iffe r e n t p a rts o f th e e a r th ’s su rfa ce. I t is, in th e h ist
p la c e , th e stu d y o f la n d s; g e n e r a l g e o g r a p h y is n o t g e n e r a l e a r th sc ie n c e ; r a th er,
it p resu p p o ses th e g e n e r a l p r o p e r tie s a n d p ro cesses o f th e ea rth , or a c ce p ts th e m
fro m o th e r sc ien ce s; for its o w n p a r t it is o r ie n te d a b o u t th e ir v a r y in g a ie a l
e x p ressio n . ( Q u o ta tio n fro m S a u er, i g 6$ , 317·)
(2) F orm
T h e b asic c o n c e rn o f th e analysis so fa r h a s b e e n w ith th e fo rm o f
e x p la n a tio n in g e o g ra p h y . E x p la n a tio n in g e o g ra p h y h as, u n til
re cen tly , re m a in e d a process o f a p p ly in g in tu itiv e u n d e rs ta n d in g to
a la rg e n u m b e r o f in d iv id u a l cases. S cien tific th eo ries h a v e n o t, b y
a n d la rg e , b e e n ex p licitly d ev e lo p e d , law s h a v e c o n se q u e n tly n o t
b e e n fo rm u la te d , a n d th e u su a l re q u ire m e n ts o f scientific e x p la n
a tio n h a v e n o t b e e n m e t. S u c h a s itu a tio n c a n p a r tly b e a ttrib u te d
to th e m e th o d o lo g ic a l p o sitio n o f m a n y g e o g ra p h e rs (p a rtic u la rly
H a rts h o rn e ) w h ic h re ste d o n th e false in fe re n c e th a t b ecau se w e a re
essentially c o n c e rn e d w ith p a r tic u la r cases w e m u st necessarily seek
for o n ly p a r tic u la r e x p la n a tio n s. E a c h g a m e o f chess m a y be u n iq u e ,
b u t th e ru les o f th e g a m e a re ex tre m e ly sim p le a n d th e field o f p la y
v ery c o n stric te d . S u ch a m eth o d o lo g ic a l p o sitio n has led to th e fa ilu re
to in v e stig a te p h e n o m e n a w ith th e re q u ire m e n ts o f th e scientific
m e th o d in m in d — w e h a v e b e e n try in g to c re a te a m a p in ig n o ra n c e
o f th e ru les o f c a rto g ra p h y . B u t th e fa ilu re to ach iev e a n a d e q u a te
co rp u s o f re lia b le th e o ry m u st in p a r t b e a ttr ib u te d to th e ex tre m e
c o m p le x ity o f o u r su b je c t-m a tte r.
T h e r e is n o re a so n in p rin c ip le w h y law s sh o u ld n o t serve to ex
p la in g e o g ra p h ic a l p h e n o m e n a , o r th eo ries o f c o n sid erab le e x p la n a
to ry p o w e r b e c o n stru c te d . E x p la n a tio n s w h ic h co n fo rm to th e ru les
o f th e scien tific e x p la n a tio n as g e n e ra lly co nceived o f c a n , in
p rin c ip le , b e o ffered . T h is is o u r c e n tra l con clu sion.
T h e m a in d ifficulty com es w ith th e im p le m e n ta tio n o f this c o n
clusion . G iv e n o u r la c k o f u n d e r s ta n d in g a n d th e ex tre m e c o m
p lex ity o f m u c h o f o u r s u b je c t-m a tte r, it w ill b e a lo n g tim e before
w e possess re la tiv e ly c o m p le te th eo ries o f a n y g re a t e x p la n a to ry
p o w e r. E x p la n a tio n s w h ic h re st o n p a r tia l a n d in co m p lete a n d in
a d e q u a te ly specified th eo rie s a re b o u n d to b e rela tiv e ly w e ak a n d
inefficien t. I n su ch c irc u m sta n c e s it m a y w ell b e th a t e x p la n a tio n s
offered o n th e basis o f in tu itiv e in sig h t alo n e m a y be m o re effective.
I n th e lo n g te rm , h o w e v er, su ch a p ro c e d u re is inefficient. A n
in tu itiv e e x p la n a tio n im p lies a m o d e l o r a th e o ry o f som e sort;
174 THE ROLE OF THEORIES
otherwise it would not explain. Why not, therefore, attempt to develop
and expand such intuitive explanations with the requirements of
scientific explanation in mind? We may, after all, successfully navi
gate to Jerusalem using merely a sense of direction and travellers’
tales—but in the long run it is far more efficient to navigate by
map and it is a pity, therefore, that our successful intuitive voyage
was not recorded in detail for the purposes of map-construction. We
are not, as yet, in the position to create powerful geographic theory,
but it would be extremely useful if we proceeded in research with
the requirements of such a theory broadly in mind. We have not yet
achieved an adequate corpus of theory, nor does it seem that we will
do so in the very near future. To attempt to apply any too rigid form
of scientific explanation in the absence of adequate theory is to court
disaster. Our explanations must, therefore, continue to resort very
much to intuitive perceptions. But intuitive perceptions backed by a
slowly developing corpus of theory are a more comforting sight than
intuition grappling alone with complex problems.
It is pointless to judge current explanations or even the form of
those explanations by the standards of the scientific paradigm. Such
judgement will surely only discourage. If we cannot simply apply the
paradigm itself, then this requires that we develop our own paradigm
—one which is presumably weaker than, but not entirely unrelated
to, the scientific paradigm itself. The rules of such a paradigm need
to be flexible but firm. Perhaps the best way to approach such a
geographic paradigm is to note those elements in the scientific para
digm which can most easily be discarded without great danger.
Axiomatisation and formalisation of theory seem, for example,
unnecessary and may even prove stultifying. Similarly applying rigid
standards to the notion of a law appears pointless. Perhaps the central
rule of some geographic paradigm should be the willingness to regard
events as if they are subject to explanation by laws. This whole prob
lem of setting up our own standards for research will ultimately be
resolved by the geographer’s behaviour with respect to the sub
stantive issues he tackles. We may anticipate, therefore, that different
groups of geographers will lay down rather different rules of be
haviour. There will undoubtedly be conflicts between groups. But it
seems to me that all rules must concede that the overriding aim of the
geographer is to achieve a greater sophistication in the art of explain
ing the events he is concerned with. The strategy we adopt in seeking
for this greater sophistication is a matter for debate.
THEORIES, LAWS AND MODELS 175
(3) Strategy
W e w ish to n a v ig a te th e w o rld b u t w e h a v e n o a d e q u a te m a p . H o w ,
u n d e r su c h c irc u m sta n ces, sh o u ld w e p ro c eed ? W e c a n n o t p ro c eed
w ith o u t ta k in g a risk, m a k in g a n u m b e r o f assu m p tio n s, in sh o rt,
s ta rtin g w ith a m o d e l w h ic h w e re g a rd as re a l. S u ch a n a priori m o d el
allow s us to n a v ig a te , a n d , fu rth e r, it m a y allo w us e v e n tu a lly to c o n
s tru c t a m o re a c c u ra te m a p o f th e w o rld th a n w e c u rre n tly possess.
C h o rle y a n d H a g g e tt (1967, 3 3 -9 ) h a v e re c e n tly a rg u e d th a t
g e o g ra p h y sh o u ld a d o p t a m o d e l-b a se d p a ra d ig m o f a n e w a n d
sig n ifican tly d iffe re n t ty p e. O n e th a t is g e a re d less to th e a c tiv ity o f
re c o rd in g a n d classifying, a n d m o re to a n a ly sin g a n d a rtic u la tin g
th eo ry . T h is a m o u n ts to a m o re d ire c te d use o f a priori m o d el co n
ce p ts— d ire c te d , t h a t is, to th e o ry -c o n stru c tio n . G iv en th e p re se n t
situ a tio n in g e o g ra p h y th e re c a n b e n o d o u b t th a t th e key to stra te g y
is p ro v id e d b y th e n o tio n o f a priori m o d els. S u ch m odels h a v e a d u a l
a d v a n ta g e . I n th e first p la c e th e y allo w us to v e n tu re som e p re
d ictio n s (even if r a th e r su sp ect ones) in th e ab sen ce o f c o m p lete
th eo ry . S eco n d ly a priori m o d els c a n in d ic a te th e a p p ro p ria te th e o ry
o r a n e x ten sio n o r m o d ific a tio n o f som e existing b u t in c o m p le te
th e o ry .
T h e use o f a priori m o d e l-c o n stru c ts poses d a n g e rs— d an g e rs w h ic h
h a v e b e e n em p h a sise d in th e p re c e d in g c h a p te rs. A w areness o f th e
p ro b le m s o f id e n tific a tio n , in feren ce, a n d c o n tro l, is a n essential p r e
re q u isite fo r th e re sp o n sib le use o f a priori m o d el-c o n stru cts. S ailin g
th e h ig h seas w ith o u t a n a d e q u a te m a p is b o u n d to b e d a n g e ro u s,
b u t h o ld in g to a set co u rse w ith b re a k e rs s tra ig h t a h e a d is c o u rtin g
sh ip w reck .
G e o g ra p h y is p ro b a b ly in th e stag e o f d e v e lo p m e n t w h e n alm o st
all e x p la n a tio n sta te m e n ts m a k e som e use o f a priori m odels. S o m e
o f th ese m o d els w ill b e d e v e lo p e d as a specific resp o n se to th e g eo
g ra p h ic a l im a g in a tio n . T h e q u a lity a n d re q u ire m e n ts o f th ese c a n n o t
b e esta b lish e d in a d v a n c e . B u t m o re im p o rta n t, p ro b a b ly , w ill b e
th e c o n su m p tio n o f m o d els fro m o th e r are a s o f science w h ic h a re
m o re s o p h istic a te d in th e ir th e o re tic a l d e v e lo p m e n t. W e m a y a d a p t
th eo rie s a n d m o d els fro m som e o th e r a r e a o f e m p iric a l science, o r
fro m so m e a r e a o f logic a n d m a th e m a tic s , to fu n c tio n as a priori
m od els in g e o g ra p h ic re se a rc h . T h e se th eo ries a n d m odels c a n b e
e x a m in e d in a d v a n c e since th e assu m p tio n s c o n ta in e d in th e m c a n b e
sta te d , a n d th e ir a p p ro p ria te n e s s a n d a p p lic a b ility fo r c e rta in types
o f p ro b le m b e assessed in a d v a n c e . T h e r e a re v a st re a d y -m a d e calc u li
w h ic h c a n b e g iv en som e g e o g ra p h ic in te rp re ta tio n w ith g re a t
b en e fit. B u t in o rd e r to c o m p re h e n d th e resu lts w e m u st b e p re p a re d
176 THE ROLE OF THEORIES
to learn the language of the calculus we are proposing to use, under
stand its properties, modify it where necessary, and consciously mani
pulate it to suit our own needs. There is a world of difference between
mere unthinking model-consumption and model-use directed and
controlled in a creative way.
The current situation in geography appears to call for a strategy
of theory-creation through the interpretation of existing abstract
calculi in geographic terms. This, as we have seen (above, pp. 152-3),
corresponds to one of the major routes to theory-construction in
science. Each of the calculi open to us has a certain set of character
istics associated with it—it possesses certain properties and certain
limitations. The application of such calculi requires some prior
assessment of the possibilities and limitations inherent in them. Since
the general conclusion of the preceding chapters amounts to a plea
for the application of controlled a priori model-thinking to geographic
problems, it is appropriate that we should go on to consider in some
detail the nature, form, and limitations, of the more important
calculi which, on the face of it, appear to have great potential for
interpretation in a geographic context. This will be the subject of
the subsequent chapters.
PART FOUR
Model Languages for
Geographic Explanation
Chapter 13
Mathematics—The Language of Science
more than one S (i.e. we provide a calculus with more than one inter
pretation) and each S may be provided with more than one K (i.e.
several different calculi may be given the same semantical inter
pretation) . This simply amounts to the basic problem of the model-
theory relationship and demonstrates the problem of identification
which the construction of artificial languages may lead us into.
It is also instructive to note the more general features of the
relationship between semantical and syntactical systems. Firstly, as
Carnap (1958, 101) points out,
o n e w h o co n str u c ts a s y n ta c tic a l sy stem u s u a lly h a s in m in d fro m th e o u ts e t so m e
in te r p r e ta tio n o f th is sy stem . . . . W h ile th is in te n d e d in te r p r e ta tio n c a n r e c e iv e
n o e x p lic it in d ic a tio n in th e s y n ta c tic a l ru les— sin c e th e se ru les m u st b e str ic tly
fo r m a l— th e a u th o r ’s in te n tio n r e s p e c tin g in te r p r e ta tio n n a tu r a lly a ffec ts h is
c h o ic e o f th e fo r m a tio n a n d tr a n sfo r m a tio n ru les o f th e s y n ta c tic a l sy stem .
B M athematical languages
Symbolic logic provides us with the necessary tools to construct and
understand artificial language systems. Whitehead and Russell in
their Principia Mathematica (1908-11) succeeded in finally showing
that mathematical knowledge could be derived from logical prin
ciples and several mathematical systems have since been restated by
means of formalised languages. The publication of Principia Mathema
tica marked the end of an important stage in philosophical thinking
regarding the nature of mathematical knowledge. It did not, how
ever, mark the end of controversy. The nature of pure mathematics
l8a MODEL LANGUAGES FOR GEOGRAPHIC EXPLANATION
is still the subject of considerable argument. Korner (1960, 156)
states:
I t is lo g ic , says th e p u r e lo g ic ist; th e m a n ip u la tio n o f figures in c a lc u li, says
th e fo rm a list; c o n str u c tio n s in th e m e d iu m o f te m p o r a l in tu itio n , says th e in tu i-
tio n ist; sta te m e n ts w h ic h w e a b a n d o n m o re r e a d ily th a n so m e sta te m e n ts o f lo g ic
a n d m u c h less r e a d ily th a n e m p ir ic a l sta te m en ts, says th e lo g ic a l p r a g m a tist.
A n d th e r e are in te r m e d ia te p o sitio n s. T h e p rogress o f m a th e m a tic a l lo g ic sin c e
B o o le a n d F re g e h a s m a d e little d iffe re n c e to th e c o n tin u a tio n o f p h ilo s o p h ic a l
d isp u tes a b o u t th e n a tu r e o f m a th e m a tic s.
Second, what are the rules which govern the use of such abstract
mathematical languages in the context of sense-perception data?
As Carnap (above, p. 181) points out, a particular syntactical
system may be chosen with an interpretation in mind. In two later
chapters, for example, different axiomatic treatments of geometry
and probability theory will be examined. In each case examples may
be found where a particular interpretation of sense-perception experi
ence has led to a particular axiomatic formulation. Klein’s discussion
of the foundations of geometry (below, pp. 203-6) and Nagel’s dis
cussion of the foundation of probability theory (below, p. 230)
are perhaps the most eloquent examples. But the actual formulation
of the axioms is an analytic exercise—it amounts to constructing a
system with certain primitive terms, definitions, and rules. It is rather
MATHEMATICS—THE LANGUAGE OF SCIENCE 185
like inventing a new and elaborate game—a game played in sym
bolic terms.
The actual motivation of the mathematician in construction such
games—be it the need to solve some empirical problem or be it a
direct infatuation with constructing abstract symbolic systems—need
not concern us. But the relationship between this symbolic system and
real-world situations is of vital importance. Whatever Euclid’s moti
vation, it is certain that the system he constructed possessed the power
to systematise many of the empirical regularities which Babylonians,
Egyptians, and Greeks, had discovered with respect to the mathe
matical form of plane figures. Euclid thus managed to develop a
formal spatial language which could be so interpreted that known
empirical regularities, and new regularities, could be predicted. This
ability of Euclidean geometry undoubtedly did much to contribute
to the belief of many mathematicians and philosophers, from the
Greeks to Kant, that geometry was an empirical science. In so far as
the interpretation of geometric systems involved (and presupposed)
measurement, this view could be upheld. But later developments have
shown that Euclidean geometry need have no direct empirical refer
ence. The fact that the terms used in the calculus, K, can be given an
interpretation by means of a semantical system, S (i.e. points, lines,
angles etc., can be approximately represented), suggests that Eucli
dean, geometry functions as an interpreted mathematical language.
An infinite number of mathematical languages could be developed.
The invention and elaboration of a particular language out of this
infinite hypothetical set depends in part upon the inherent interest
of a particular language form to the pure mathematician and partly
upon the utility of that language for the scientist. Thus probability
theory remained a curiosity applied merely to games of chance until
problems in biology and physics necessitated the thorough exploita
tion of this particular mathematical language. As new applications
for a particular language are found, so new problems in the language
itself may emerge. The interaction between the extension of a mathe
matical language and the development of applications for that lan
guage is thus of some importance. But the application of the language
depends on being able to interpret mathematical symbols in terms of
sense-perception data and empirically based concepts.
This problem can best be approached by discussing the rules which
govern the mapping of empirical problems into some existing mathe
matical language. This amounts to determining the rules by which
we can tell when a semantic system, S, is a valid interpretation for
an abstract calculus, K. It is possible to state a set of logical rules.
Thus Carnap (1942) suggests that the semantical system, S, should
186 MODEL LANGUAGES FOR GEOGRAPHIC EXPLANATION
be isomorphic with the calculus, K. This of course is the character
istic view of the logician regarding the relationship between theory
(S) and model (K). But this view is not empirically very useful for it
amounts to claiming an isomorphism between real-world structure
and the mathematical language used to discuss that structure. It
necessarily assumes, therefore, that we already know all about real-
world structure—in other words K functions as an a posteriori model.
In practice a number of steps are involved in mapping real-world
problems into mathematical systems. These are described by Korner
(i960, 182) as follows:
(i) th e r e p la c e m e n t o f e m p ir ic a l c o n c e p ts a n d p r o p o sitio n s b y m a th e m a tic a l,
(ii) th e d e d u c tio n o f c o n se q u e n c e s fro m th e m a th e m a tic a l p r e m ises so p r o v id e d
a n d (iii) th e r e p la c e m e n t o f so m e o f th e d e d u c e d m a th e m a tic a l p r o p o sitio n s b y
e m p ir ic a l. O n e m ig h t a d d (iv) th e e x p e r im e n ta l c o n fir m a tio n o f th e la st-m e n tio n e d
p r o p o sitio n s— w h ic h , h o w e v e r , is th e task o f th e e x p e r im e n ta l scien tists ra th e r th a n
th e th e o r etica l.
The first step in this sequence is perhaps the most crucial. Empiri
cal concepts have to be extracted from an inexact real world and these
empirical concepts are then replaced by exact mathematical con
cepts and propositions. If the empirical concepts and propositions
are formulated in a vague and ambiguous way it becomes very diffi
cult to determine which mathematical language is appropriate, and
difficult to translate the empirical propositions and concepts into
mathematical ones. A prerequisite to the use of mathematical lan
guage, therefore, is precision in the formation of concepts and pro
positions. Failure in this respect is a common problem in many of
the social sciences. Coleman (1964, 3) thus suggests that in sociology
th e k in d s o f v e r b a l th e o r ie s a n d r esea r ch resu lts w h ic h h a v e b e e n se t forth are so
v a g u e ly s ta te d or so w e a k th a t it is d iffic u lt to tr a n sla te th e m to m a th e m a tic a l
la n g u a g e , a n d o n c e tr a n sla te d th e y o ften fa il to sh o w a n iso m o r p h ism w ith p o w e r
fu l p a rts o f m a th e m a tic s.
Basic Reading
A rro w , K . J . (1959).
Korner, S. (i960).
Chapter 14
Geometry—The Language of
Spatial Form
I CONCEPTS OF SPACE
I I T H E F O R M A L R E P R E S E N T A T I O N O F S P A T IA L
CONCEPTS
topology
affine
geometry
similarity equiareal
geometry geometry
of the corresponding point on the earth. Such a law is called the Projection on
which the map is drawn; and the equations of the projection are those which give
the relation between the terrestrial co-ordinates and those of the point on the map.
Solving such equation systems requires formal geometric methods.
Further, the properties of these equation systems can be analytically
investigated. Tissot (1881) was thus able to make a general study of
deformation involved in map projection. Tobler (1966A) writes:
I f the objective is to preserve some spherical property on a flat m ap, what
happens to the other properties when this is done? H ow are the angles distorted
on equal-area projections, and so on? Tissot’s indicatrix allows such questions to
be answered, in a local sense, with comparative ease.
GEOMETRY—THE LANGUAGE OF SPATIAL FORM 221
every point of the plane closer to that lattice point than to any other
lattice point’. The competitive process involved in central-place
theory then becomes a packing problem.
Dacey’s treatment of central-place theory demonstrates how
geographic problems can, once they are unambiguously conceptual
ised (and Dacey found some ambiguities in the usual presentation of
central-place theory), be represented in an appropriate mathematical
language. In this particular case the intention is to describe mathe
matically the nature of central-place systems as they are normally
described. There is no attempt to derive the geometry from process
postulates (e.g. the theory of supply and demand). Dacey’s stated
intention is limited to discussing the mathematical properties of the
diagrams usually used to demonstrate the spatial form of a system of
central places. He goes on to state that ‘a suitable mathematical
system is available, so this task only requires placing the appropriate
geometry in juxtaposition with central place concepts.’ The appro
priate geometry is provided by a special arithmetic and algebraic
development out of Euclid.
(4) Space-time problems and Minkowskian geometry
The examples discussed so far refer to situations in which formal
geometric concepts have been shown to be applicable to actual
geographic problems. This section considers a potential rather than
a realised development. For the most part formal geometry has been
applied in connection with one- two- or three-dimensional spatial
languages. Consider the following conceptual development of a four
dimensional problem in which time becomes an important co-ordin
ate; an individual person constantly moving about in a two-dimen
sional co-ordinate system (a:, y) with a varying amount of money to
spend (z) over time (t). Hagerstrand (1963) presents a simplified
version of this space-time conceptualisation in discussing the move
ment of individuals over space and in time (co-ordinate z is omitted).
This movement Hagerstrand terms a life-line:
every individual can be represented by a life-line, starting at the place of birth,
now and then making a jum p over to a new station, and finally ending at the
place of death. T he life-lines of a population in a block of time-space are twisted
together in a very complicated way.
probability theory are drawn from the theory of sets. These state
ments can be given an interpretation in the same way that the
primitive terms of geometry can be represented by dots and ruled
lines. This requires a general interpretation of probability theory,
however, and for the purposes of exposition a frequency representa
tion will be given.
The basic primitive notion in the mathematical theory of proba
bility is that of the sample space. Consider an experiment in which we
toss a coin. All possible tosses may then be characterised as forming
a sample space. The sample space contains what are called members of
the sample space or elementary events. In this example each elementary
event may be equated with the outcome of an experiment (the toss
of the coin). In set theoretic notation we can then define the sample
space as A = (a1} a2, a3, . . .). We may then consider any subset of
the sample space which we may call a compound event or sometimes just
an event. In our example an event might be all those tosses that are
recorded as heads. If there are only two events in the sample space
(e.g. heads and tails) then we have the condition that A — (A1} A 2).
It is important here to introduce the notion of mutually exclusive
events since this is also basic to the mathematical development of
probability theory. Two events may be defined as mutually exclusive
if the set formed by their intersection contains no members (or
A x n A 2 = φ, the empty set). It is easy to see in our example that
heads and tails are mutually exclusive events. All of the terms we have
used so far are primitive to probability theory and they can be put
in the language of set theory. All that remains is to define probability
as a setfunction defined on subsets of the sample space which translated into
simple English amounts to a rule for assigning values to the events
contained in the sample space. This rule then gives a number to the
probability of obtaining heads, say P(AX), and the probability of
obtaining tails, say P(A2). We can then state three basic axioms
which the numbers assigned must conform to.
(i) P(E) > o for every event E (i.e. no negative numbers).
(ii) P(S) — i for the certain event S.
(iii) P(E u F) = P(E) + P(F) if E n F = φ (the empty set), which
is the addition rule which states that the probability of the
union of two mutually exclusive events is the sum of their
probabilities.
The above statement provides foundation for the mathematical
theory of probability. There are, of course, a number of different
versions available but they are all very similar (Parzen, i960, 18;
McCord and Moroney, 1964, 20; Lindley, 1965, I, 6).
PROBABILI TY 'THEORY—THE LANGUAGE OF CHANCE a45
It is useful at this point to digress a little regarding the nature of
the sample space and in particular to note that the application of
probability theory to any empirical phenomena is dependent upon
finding an adequate operational definition (or interpretation) for
this abstract theoretic concept. In the example chosen above the
sample space contains an infinite number of members since we can
carry on tossing coins indefinitely. Suppose, however, that the space
is thought of as containing the 52 cards from an ordinary pack. In
this case the sample space contains a finite number of members
although we can, if we sample with replacement, generate an infinite
series from it. Suppose, in this case, we ask what suit a card is when
we draw it? There are four possible answers to this question and we
may think of the space as containing four mutually exclusive events.
If on the other hand, we ask what number or picture the card is,
there are 13 mutually exclusive events in the sample space. The
sample space can therefore be changed at will according to how we
wish to conceptualise it. Therefore the mathematical concept of the
sample space only has an interpretation by way of the question we are
asking. As Parzen (i960, 11) puts it
Insofar as probability theory is the study of mathematical models of random
phenomena, it cannot give rules for the construction of sample description spaces.
Rather, the sample description space o f a random phenomenon is one o f the
undefined concepts with which the mathematical theory begins. T he considera
tions by which one chooses the correct sample description space to describe a
random phenomenon are a part of the art o f applying the mathematical theory
of probability to the study of the real world.
Accept the
Correct Decision Type-II Mistake
null hypothesis
Geographical Mathematical
Concepts translation Language of
--------------------- ·>·
and (mapping) Probability-
Relationships Theory
Deduction
Y
Geographical translation Mathematical
<------------------- ;----
Conclusions (reverse mapping) Conclusions
fix) = - L - e- (* -n y 2 o ‘
2πσ2
(c) Specifying the sampling procedure also forms a vital part of the pro
cedure for making non-deductive inferences. The sampling problem
will be dealt with in detail later (chapter 19) but it is important to
note that statistical inference is essentially concerned with making
inferences from samples; and if these samples are to be used in relation
to substantive hypothesis, they must be collected in such a manner
that they are representative of the population of objects and numbers
being examined and relevant to the particular hypothesis being
examined.
(d) Specifying the appropriate test procedure is one of the most difficult
operations in non-deductive inference. Statistical theory provides
us with a battery of models, each with specific assumptions, and the
task we are required to undertake is to select one out of a whole
battery of available tests on the basis of the fact that it is most
compatible with the procedures involved in (i), (ii), and (iii), above.
Most conventional statistical tests specify measurement made at
least on the interval scale and specify also that the population of
numbers obtained should be normally distributed (thus the sample
distribution should be normal within the limits of sampling error)
and that the sample size should be reasonably large (conventionally
greater than 30). If these conditions are satisfied, then a whole range
of so-called parametric tests may be employed. These are character
istically the tests presented by Fisher {1936; 1956; 1966) and developed
also in the Neyman-Pearson theory. It is clear that the use of such
tests involves strong assumptions. Consider, for example, the assump
tions underlying a Y test:
(i) Each observation must be independent of any other observation
(i.e. obtaining any one observation must not prejudice the
probability of obtaining any other observation).
(ii) The observations must be normally distributed.
(iii) The populations must possess the same variance (this is the
condition of homoscedasticity).
(iv) The observations must be measured on the interval or ratio
scale.
This particular test is very powerful (in the statistical sense of power),
but the strong assumptions involved can rarely be fulfilled in geo
graphy. In some cases we shall be able to show that the necessary
conditions are met in the data, but in other cases we are forced to
assume their existence. Fortunately, the ct’ test has proved to be
fairly robust, working even when all the conditions are not strictly
met. Nevertheless, we can conclude with Siegel (1956, 19) that:
All decisions arrived at by the use of any statistical test must carry with them
PROBABILITY THEORY—THE LANGUAGE OF CHANCE 281
this qualification: ‘I f the model used was correct, and if the measurement required
was satisfied, then. . . .’
It is obvious that the fewer or weaker are the assumptions that define a par
ticular model, the less qualifying we need to do about our decision arrived at by
the statistical test associated with that model. T hat is, the fewer or weaker are the
assumptions, the more general are the conclusions.
Thus new types of test are being evolved which, although they may
be less powerful, are probably more applicable to geographical
situations. These tests are usually called non-parametric and distri
bution-free, although as Galtung (1967, 341) points out, these terms
are misnomers in the sense that the first really means ‘non-interval-
scale parametric’, while the second really means ‘non-normally
distributed’. The greater applicability of these tests by virtue of their
weaker assumptions is sometimes bought, however, by a general
decline in power. Here the geographer needs to be a rational
decision-maker, since he needs to select the most applicable and the
most powerful test and these objects are not always compatible. In
selecting a test, therefore, the geographer has to solve a particular case
of the general problem that dogs the use of any mathematical calcu
lus in empirical work, viz., to select the model that does least violence
to the empirical situation, and at the same time is simple to handle
and achieves the objectives that he has in mind. This is an evaluative
procedure in which we have a real need of guidance from the statis
tician. But statisticians have not always been as helpful as they might.
As the example of the ‘t’ test demonstrates only too well, the theore
tical assumptions necessary to derive a test are not always relevant
assumptions for the application of that test to a particular data set.
Statisticians have not always been willing to consider this and they
have not therefore provided adequate ‘rules of behaviour’ for the
practical application of the tests they derive. Tukey (1962) has
suggested that this will become a major area of statistical analysis in
the future. If so, the geographer is bound to benefit.
In the same way that ‘unlike other sciences geography can find great
use for principles that are applicable to relatively small portions of
the earth’s surface’ (above, p. n o ), so geography is forced to make
use of highly localised populations and highly localised samples
which, in many cases, are precisely the same as the population itself.
In these circumstances a number of different interpretations may
be put upon statistical inference in geography. Consider the following
five interpretations that might be put upon a statistical study of the
spacing of the 99 largest towns in Iowa:
(i) The population is the set of all Iowan towns and the 99 largest
towns are a sample. Inferences are thus made from this sample
(which is biased and almost comprises the population anyhow)
to the population.
(ii) The population is a set of all towns which are found in similar
environments to that of Iowa. In which case the sample consists
of the sub-population of Iowan towns and the assumption is that
Iowa is somehow representative of situations to be found in
other countries or states.
(iii) The population is the set of all possible configurations of Iowan
towns. In this case the 99 largest towns provide us with a sample
of the infinite number of ways in which the towns could have
been arranged.
(iv) The population is an imagined hypothetical set of all towns
everywhere and Iowa provides a reasonable sample to represent
this hypothetical universe of towns.
(v) The population is the set of all measures which could be made
on the distances between settlements, given a specified law
governing the distribution but a disturbance due to measure
ment ertor.
This last appears to be the construction that Dacey (1966A) puts on
his own studies of the pattern of town spacing in Iowa. He assumes
the pattern is generated by the central-place model, but that the
actual locations deviate from theoretical optimality by some error
term which may be conceived of as a kind of measurement error.
Given the central-place model, the total number of towns, and the
total area of Iowa, then the theoretical distance from a settlement to
any other settlement can be calculated. The actual measures made
PROBABILITY THEORY—THE LANGUAGE OF CHANCE 285
on distances between Iowan towns then amount to a repeated mea
sure made on this distance with a measurement-error term involved.
The actual measures should be approximately normally distributed
around the theoretical distance value if the model is correct. We have
already noted Dacey’s conclusions regarding the adequacy of the
model using a test of this form (above, p. 138).
It is possible to construct further variants on the interpretations
given above, but the main point is to demonstrate how confusions
might arise if the population is not specified in advance, and if the
statistical inference (for which rigid rules apply) is not kept apart
from the substantive inference.
Basic Reading
Ackerman, R. (1966).
Fishburn, P. C. (1964), chapter 4 contains an excellent summary of
probability theory.
Hacking, I. (1965).
Parzen, E. (i960).
Plackett, R. L. (1966).
Rescher, N. (1964).
Tukey, J. W. (1962).
Reading in Geography
Curry, L. (1964).
Curry, L. (1966A).
Curry, L. (1966B).
PROBABILITY THEORY—THE LANGUAGE OF CHANCE 287
Dacey, M. F. (1964A).
Dacey, M. F. (1966A).
Leopold, L. B., Wolman, M. G., and Miller, J. F. (1964).
Lukerman, F. (1965).
PART FIVE
Models for Description
in Geography
Chapter 16
Observation
The emphasis in this book has, so far, been heavily biased towards
the analytic and a priori aspects of geographic understanding. We have
thus dwelt upon the complexities of axiomatic systems, theoretical
and model constructs, mathematical languages, and the like. Such
an approach has not, probably, been to the taste of many geo
graphers accustomed to the more traditional mode of approach to
unravelling the mysteries that surround them. Indeed it may horrify
those who seek understanding by poring over maps, by wandering,
thoughtful and observant, down streets and over fields, by perusing
records buried in dusty archives, by rummaging among old news
paper reports, by digging soil-pits and diving deep into limestone
caverns, by watching and waiting, by sharpening their own per
ceptual experience, by training themselves to appreciate. Carl Sauer
(7963, 400) once wrote of the joys of field work:
Underlying what I am trying to say is the conviction that geography is first of
all knowledge gained by observation, that one orders by reflection and reinspec
tion the things one has been looking at, and that from what one has experienced
by intimate sight come comparison and synthesis. . . . T he important thing is . . .
to recognize kind and variation, position and extent, presence and absence,
function and derivation; in short to cultivate the sense of morphology.
To Sauer, the education of a geographer was essentially an educa
tion in how to experience; an education that relied upon interaction
between student and teacher,
engaging in a peripatetic form of Socratic dialogue about qualities of and in the
landscape. Locomotion should be slow, the slower the better, and be often inter
rupted by leisurely halts to sit on vantage points and stop at question marks.
Being afoot, sleeping out, sitting about camp in the evening, seeing the land in
all its seasons are proper ways to intensify the experience, of developing impression
into larger appreciation and judgem ent.
Such an approach to geographical understanding is deeply em
bedded in the geographical tradition. ‘We must ask the earth itself
for its laws,’ wrote Ritter (Hartshorne, 1939, 55)—a precept that has
governed geographical investigation in the minds of a long line of
distinguished geographers. The classic work of the early geologists/
geomorphologists such as Gilbert and Powell in the American Far
291
292 MODELS FOR DESCRIPTION IN GEOGRAPHY
definition. These circularities are not the only ones, however, for
it can be maintained that all definition is circular—cither that or it is
infinitely regressive. Ackoff (1962, 17°) comments on this circularity:
I n a sense this is tru e , b u t d e fin in g tak es p la c e in th re e d im en sio n s n o t tw o .
W h e n a full circle h a s b e e n m a d e , w e a re above o u r s ta rtin g p o in t a n d h a v e b ro u g h t
to it a ric h e r a n d m o re p re c ise m e a n in g th a n it h a d w h e n w e s ta rte d . I n effect,
w h e n w e define o n e c o n c e p t, th is d e fin itio n illu m in a te s th e co n cep ts o n w h ic h it
d e p e n d s as w ell as th e c o n c e p ts t h a t d e p e n d o n it.
I DEFINITION
water-gap: A low level valley across a ridge, through which flows a river; e.g. the
W ey (gap) near Guildford, the M ole (gap) near Dorking, through the N . D o w n s.. . .
This last definition is in part ostensive since it points to examples,
but the basic idea is to describe the meaning of a term by way of
other terms. The definiens thus describes a set of necessary and suffi
cient conditions for the application of the definiendum. A dictionary
approach to definition amounts, therefore, to a description of the
complex interrelationships between different terms and such a
description ensures consistency of use and serves to eliminate ambi
guity. But dictionary definitions are limited in a number of respects.
They assume, first, that explicit definitions can be provided (i.e.
reduction of the definiendum to the definiens can occur without any loss
of information content) and, second, that definitions can be pro
vided internally within a language system. In the first case we may
find it necessary to provide implicit definitions and these are parti
cularly important in the context of theoretical terms which cannot
easily (if at all) be reduced to other terms. Terms such as water-gap
can be explicitly defined, but there are other terms, such as culture
and ‘regional consciousness’, which cannot be defined in the same
way. These terms are the theoretical terms, the ideal types, the
abstract constructs, and the like, which are extremely important in
theory-construction. Here the role of definition changes from the
provision of synonyms to the provision of the correspondence rules
or textiov a given theory (above, pp. 91-6). The problem of defining
theoretical terms and idealisations is one that has taxed philosophers
of science for some time. Nagel (1961, 98) writes:
no infrangibly conclusive proof is available, and perhaps no such proof is possible,
that the theoretical notions employed in current science cannot be explicitly
defined in terms of experimental ideas. . . . It is pertinent to observe, however,
that no one has yet successfully constructed such definitions. M oreover there are
good reasons for believing that the rules o f correspondence in actual use do not
constitute explicit definitions for theoretical notions in terms of experimental
concepts.
Braithwaite (/p6 o, 77) has similarly concluded that the theoretical
terms of a scientific theory can only be implicitly defined by reference
to the postulates which form the basis of the theory itself. Indeed
OBSERVATION MODELS—DEFINITION AND MEASUREMENT 303
Braithwaite goes on to suggest that such theoretical terms ‘cannot
be explicitly defined by means of the interpretations of the terms
in . . . derived formulae without the theory thereby becoming in
capable of growth.5 The meaning of this is simply that derived
theoretical statements require translation into empirical statements,
but this translation does not in any way constitute an explicit defini
tion. Now the general importance of definition in this context has
already been examined (above, pp. 88-9), and it will be remem
bered that there is a key difference between theoretical terms which
can be related back to theoretical postulates, and many of the idealis
ations of social science which can be defined only intuitively (i.e.
their definition relates to images or conceptual idealisations and is
thus external to the scientific language system itself). We shall not
consider this problem in any further detail here, except to refer to
one important approach to definition which has attempted to solve
these complex problems by making certain assumptions. This ap
proach is termed operationalism, and it provides one solution to the
difficult philosophical problem of specifying meaning.
A theory or model takes on meaning only ‘when the symbols and
things they represent are defined5(Ackoff, 1962, 141), when adequate
correspondence rules are established (Nagel, 1961, 90-105), and
when the difficult task of somehow translating abstract theoretical
statements into tangible conclusions is accomplished (Braithwaite,
i960, 53). Operational definition tackles all these problems by
assuming that definition ‘consists simply in referring any concept for
its definition to the concrete operations by which knowledge of the
thing in question is had5 (Stevens, 1935, 323). The advantage of this
rule of definition is that the difference between implicit and explicit
definition disappears, and a rigid procedure is prescribed for ascer
taining the meaning of the term. Thus Ackoff (1962, 175) writes:
T he formal requirements for (operationally) defining properties are that we
specify what is to be observed, under what (changing and unchanging) conditions
the observations are to be made, what operations are to be performed, what
instruments and measures are to be used, and how the observations are to be made
and treated.
Kaplan (1964, 40) likewise writes:
T o each concept there corresponds a set o f operations involved in its scientific
use. T o know these operations is to understand the concept as fully as science
requires; without knowing them, we do not know what the scientific meaning of
the concept is, nor even whether it has a scientific meaning. Thus operationalism
provides, not just a criterion of meaningfulness, but a way o f discovering or
declaring what m eaning a particular concept h a s: w e need only specify the opera
tions that determine its application. Intelligence, in the famous dictum, is what
is measured by intelligence tests.
3o4 MODELS FOR DESCRIPTION IN GEOGRAPHY
Opinions vary as to the value of such an operational approach to
meaning. Ackoff (1962), Rapoport (1953) are strongly in favour,
Kaplan (1964) is sceptical. Certainly, there is a great deal to be said
in favour of making definitions operational wherever possible. We
may thus replace Monkhouse’s (1965, 257) dictionary-type definition
of a ‘region’ as ‘a unit-area of the earth’s surface differentiated by
its specific characteristics’ by an operational definition which specifies
the procedure whereby regions may be identified (a good example
would be Berry’s (1967B) operational approach to regionalisation
by way of multivariate analysis and grouping algorithms). Haggett
(1965A, 188-90), for example, discusses the problem of giving an
operational definition of an ‘urban settlement’ so as to ensure
standardisation and comparability on a world-wide scale. A more
operational approach to definition in geography, while in itself
desirable, is unlikely to solve all problems however. Kaplan (1964,
40-2) discusses some of the objections to operationalism, the most
important of which is the assumption that the conditions can be fully
specified and that relevant conditions and measures can be dis
tinguished from irrelevant. If we can believe Wallis’s (1965) experi
ences with computer programmes for factor analysis, for example,
we may anticipate that the outcome of a Berry-type approach to
regionalisation will depend very much on which kind of computer
programme is used. Operational definitions appear much more
consistent, but in practice it is difficult to ensure absolute consistency.
The question of consistency, rigour, and precision of definition,
however, is also of some interest. On the one hand the mathe-
matisation of geographical theory requires precisely defined un
ambiguous concepts (above, pp. 184-90). On the other hand there
are benefits to be had from retaining a certain ‘openness’, as
Kaplan (1964, 62-71) calls it, in the use and meaning of a term.
Thus:
T he demand for exactness of meaning and for precise definition o f terms can
easily have a pernicious effect, as I believe it often has had in behavioral science.
It results in w hat has been aptly named the premature closure of our ideas.
II MEASUREMENT
A Measurement m odels
Until recently, measurement models could be grouped into four
distinct categories (with one or two hybrid variants) all of which
provided simple scalar systems for measuring objects or their at-
so m e th in g a b o u t it; b u t w h e n y o u c a n n o t m e a su r e it, w h e n y o u c a n n o t exp ress
it in n u m b e rs, y o u r k n o w le d g e is o f a m e a g r e a n d u n sa tisfa c to ry k in d .’
OBSERVATION MODELS—DEFINITION AND MEASUREMENT 309
complete ordering are that reflexive relationships are allowed and the
condition of asymmetry has to be replaced by a condition of anti
symmetry (which amounts to specifying xt and x2 as identical only
when x1 < x2 and x2 < Χχ). A weak ordering does not specify, neces
sarily, a unique ordering among the xt and, characteristically, equi
valence classes are involved. For example, ranking people according
to socio-economic class will probably involve subsets of the xt
measures which are indeterminate with respect to order. Such a
form of ordinal scaling does not involve so many presuppositions
about the nature of the data, but it does convey less information.
(c) Partial ordering is basically the same as weak ordering except
that it is non-connected. We may weakly order a population accord
ing to socio-economic class, but there may be a group in the popu
lation for whom we do not possess any information and who, con
sequently, must lie off the scale altogether. Similarly, in asking
people to order preferences for places or objects, there may be some
places or objects which they have no information about and these
must, therefore, lie off the scale we are devising. We cannot expect
a complete or weak ordering of taste preferences among oranges,
bananas, pineapples, and ugli fruit if the people we are asking have
never savoured the ugli fruit.
Fig. 17.i The Measurement Scale Systems and Their Associated Methods (Mathematical and Statistical)
cCD —«1
'o
t 'co
<D ^
Om
OD 2c
> x e .92 xe 92 .2 3
(D I 'o
S e e pο co ® o COtr0)
E H_ co
in CD c
E— = co C
CO CD CO
D ie B co V .92 E
CO cr ~C .£ O
c Ό CO
XE o
3O
O
I—
<0 T£ O
o co 73
CD C
p
Ο I
_
co w
B m ^ X- *-> O
£ O .9 -C CX CD O Ο Έ
CD ex £ CD'o .® £2
ο o CD ex CO SZ X i Le. Q- CXTJ CD
Έ δ
CO \ p
C T3
— o <2 <S
«£ E 00 Ό
>-
_i O
< Z
oc LU
I—Q £o s^
Z 2
LULU sco E®
O h-
CD 2 o £ o >
2 O CO® CD c o ® O CO
S ' gCOc—
S- oc -C
® +C-O
»—c •■P ow
CO
—<
Q CC
£CCOD Of> I^> O CD O >. O> 3
C r~ CT '^ C c Cc CO
CD xe co uj CD y co X! co >
< CO O
<->> H
CO CO
<
DC
I
OBSERVATION MODELS—DEFINITION AND MEASUREMENT 313
Since 1958, however, a vast amount of work has been done on multi
dimensional-scaling procedures, particularly in psychology (Cattell,
1966; Kruskal, 1964), and the applicability of some of these to geo
graphic problems has also been considered (Downs, 1967). It is
clear, however, that the technical problems involved (particularly
on the geometric side) are precisely the same as those involved in
measuring geographic distance and that the same basic methodology
applies. Transformations, projections, and complex distance measures
(above, pp. 219-23) are just as applicable in the attempt to measure
distance on some complex attribute as they are in the attempt
to measure geographic distance. The generalisation of distance
measure from the special case of Euclidean geometry taken from
Adler (1966, 282-3; above, pp. 223-4) is thus a form of a general
equation for relating distances to projections given by Torgerson
( 1 9 5 8 , 293):
i* = [Σ(Ι - «,» D T ”
m
in which the distance between two stimuli j and k in an /--dimensional
space consisting of m = 1, 2, . . r, orthogonal axes is given by the
formula with c taking on a variable value. When c = 2, the simple
Euclidean measure of distance results; when c = 1, a rather special
distance measure devised by Attneave (1950), variously known as
the ‘city-block’ or ‘Manhattan metric’, results. Much larger values
ofr can be used and there is no reason for c to be an integer (Kruskal,
x964 )·
There is not space to go into multidimensional scaling methods
in any detail here. Those interested should refer to the psycho
logical literature (particularly Torgerson, 1958; Kruskal, 1964). But
it is interesting to note the basic similarity between the measurement
problem as it occurs in scaling complex attributes by way of various
stimuli, and a parallel situation in a more traditional geographic con
text. Thus (Tobler 1966A) has used the psychological techniques of
multidimensional scaling in constructing map projections, while
Gower (1967) has used an understanding of map-projection techniques
to transform complex data sets to more manageable dimensions.
3 16 MODELS FOR DESCRIPTION IN GEOGRAPHY
(c) Error due to the environment results when conditions in the environ
ment change to affect the observer, the instrument, or the thing
observed. A change in temperature may affect a measure of length, a
change in credit restrictions may alter a person’s views on place
utility (by affecting his potential to move), a hot humid day may
affect the stress felt by the observer and the observed. It is one of the
functions of experimental design to control for such errors, but in the
non-experimental sciences such control is extremely difficult. Again,
it is necessary to have some estimate of the error involved.
(d) Error due to the observed may result either from the inherent varia
bility of what is being observed (e.g. people change their views on
place utility over time, and may even think differently in the evening
from the way they felt in the morning) or from the behaviour of the
observed being affected by the behaviour of the observer. The classic
case of the latter is the indeterminacy principle in physics formulated by
Heisenberg. In the social sciences the indeterminacy principle ap
pears to be very general, especially in questionnaire work. It is almost
impossible for an interviewer not to influence an interviewee, since
observer and observed are placed in a highly reactive situation.
Controls may be introduced but it is unlikely that error will be
eliminated or that the error will be non-systematic. An enthusiastic
interviewer is likely to get rather different results from a cynical
322 MODELS FOR DESCRIPTION IN GEOGRAPHY
D Measurement in geography
In the preceding sections we have concentrated on some of the
methodological problems inherent in the measurement process. In
many areas of geographic research these problems may appear
largely irrelevant not because they are unimportant but because
appropriate methods of dealing with them have been devised.
There is nothing very controversial about the way in which measure
ment problems in surveying are overcome. In most traditional
areas of geographic research, therefore, there is a well-developed
methodology of measurement. In human geography, on the other
hand, where there has been a very rapid change in emphasis over the
past few years, no such conventional methodology has yet been
evolved. Given the growing interest in the behavioural aspects of
location—studies in spatial perception (Gould, 1966; Lowenthal,
1966; Downs, 1968), locational behaviour (Pred, 1967)» environ
mental perception (Kates, 1962; Saarinen, 1966), and the like it
is important that the bases on which we measure be clearly under
stood. It cannot be denied that our methods of measurement in these
areas of research leave a great deal to be desired. This is in part due
to the inherent difficulty involved in measuring ‘images’, ‘values’,
‘satisfaction’, ‘utility’, and the like, but it also results from a failure
to grapple with basic methodological problems. It is therefore appro
priate to close this section on measurement by returning to the
problem of measuring place utility (above, pp. 317-18).
OBSERVATION MODELS—DEFINITION AND MEASUREMENT 323
A number of measurement procedures have been devised for
measuring place utility. Rather than survey all of these, or devise
new measures, however, it is convenient to set up just one model
and examine the validity of it. Consider, therefore, the following
model. We begin by choosing eight areas in a city which are rela
tively homogeneous with respect to their living conditions. Let us
assume that such areas can be identified and that we can put names
to them, e.g. ‘Whiteoak’, ‘Clifton’, ‘Chelsea’, and the like. We then
choose a sample from the population living in the city by some
specified method and ask them to rank-order the eight names they
are given in order of their preference for living there. What inform
ation does such a model yield us?
In direct terms the model simply yields us a response to given
stimuli. Torgerson (1958, 46) suggests three ways in which we can
conceive of such a model:
(i) The subject-centred, approach, in which the variation in response is
attributed to the individuals (i.e. we are measuring the utility
of each individual with respect to places).
(ii) The stimulus-centred or judgement approach, in which the variation
in response is attributed to variation in the stimuli (i.e. we are
measuring the utility of places as seen by individuals).
(iii) The response approach, in which variation in response is attributed
to both stimuli and individuals (i.e. we are measuring a mixture
of the utility of places and the utility scales of individuals).
The response approach is probably the least satisfactory because
it mixes two rather different things. But as our model stands, it
clearly belongs to this category. Now it is possible that the sampling
design can be used to transform the model into a judgement model.
Suppose we know, for example, that the key variables with respect
to the utility scale of individuals are education and income. By strati
fying our sample to hold education and income constant we may
control out the variation in the utility scale of individuals in large
measure and hence obtain a measure of the utility of places. Let us
now pursue this particular tack.
We are now certain that we are scaling the stimuli and not the
people. The question then arises what the stimuli actually represent
and how they are actually arranged by people. The stimuli consist
of a series of names, and the response relates directly to these names.
We can go on to make statements about place utility only provided
that certain assumptions can reasonably be made (here we are
involved in construct validation). Consider the following reasons
why people might arrange the names in a given order:
(i) Some names sound nicer than others—thus people might rate
324 MODELS FOR DESCRIPTION IN GEOGRAPHY
‘Clifton’ persistently higher than ‘Coalpit Heath’ simply because
it sounds nicer as a name.
(ii) People may have a variable amount of information about the
places they are asked to rank. Variation in the rank-ordering
may thus partly measure variation in the information which
people have. In situations where people possess zero information,
it may be inappropriate to use a completely ordered scale; a
partially ordered scale would be better.
(iii) People may possess different ‘images’ of places (because of
reputation, and other factors which are difficult to specify), and
they may order them solely in terms of these images.
(iv) They may order places with respect to the actual living condi
tions in the areas named and their preference for such living
conditions.
These four reasons are not the only ones that could affect the out
come, nor are they independent of each other. The net outcome,
therefore, is that we do not possess any very good control over the
stimuli we are presenting, nor are we sure what it really measures.
If we compound with such difficulties those associated with vari
ability in the population, we find ourselves in an extremely complex
situation in which it is difficult to judge the validity of the measure
proposed. The specification of error is likewise extraordinarily diffi
cult. It is true that many of these difficulties may be partly overcome
by a sound sampling procedure. Such procedures will, however, be
considered in chapter 19.
It may be argued from this example that, rather than pursue the
chimera of measurement, we should abandon the idea altogether.
There is indeed some justification for adopting such advice. But the
interesting thing is that in order to measure effectively we require
deep analytic understanding and considerable thought regarding
the controls necessary and the errors naturally incurred. Measure
ment may not be a satisfactory end in itself, but we can be sure that
in pursuing this end we will turn up problems and difficulties, the
solutions to which will provide major advances to our understanding.
This, however, depends entirely on a sound understanding of the
nature and principles of measurement. Without such an understand
ing we are plainly lost.
Basic Reading
Ackoff, R. L., et al. (1962), chapters 5 and 6.
Churchman, C. W., and Ratoosh,- P., (eds.) (1959).
Coombs, C. H. (1964).
Kaplan, A. (1964), chapter 2.
OBSERVATION MODELS—DEFINITION AND MEASUREMENT
It has already been made clear that the utility of a given system of
classification cannot be assessed independently of its purpose. The
many purposes of classification can, however, be grouped into two
types.
(1) General or ‘natural’ classifications
These are intended for general use by all scientists (Sokal and Sneath,
1963, 11-20). The notion of a ‘natural’ classification was developed
in detail by Mill (7950, 300-1):
T he ends of scientific classification are best answered when the objects are
formed into groups respecting which a greater number o f general propositions
can be made, and those propositions more important, than could be made respect
ing any other groups into which the same things could be distributed . . . a
classification system thus formed is properly scientific or philosophical and is
commonly called a natural, in contradiction to a technical or artificial classifica
tion or arrangement.
I--------------------------- 1
COMMERCIAL SUBSISTENCE
I------------- 1---------------- 1
OWNED RENTED SHARE-CROPPED
1------------1 1
OWNED RENTED SHARE-CROPPED
Fig. 18.2 D iag ram to illustrate two stages in th e logical division o f a
universal set o f farm s into m utually exclusive classes.
which are characterised by the property that ‘the possession of a
unique-Set of features is both sufficient and necessary for membership
in the group thus defined.’ The kind of classification which emerges,
however, is very much affected by the criterion selected at each step
and the order in which the criteria are employed. In developing
classifications by this route, therefore, we need to place the criteria
in order of significance and this, of necessity, assumes that we know
a great deal about the phenomena being classified; in other words,
we possess an adequate theory about structure and we can employ
that theory deductively to identify the classes. Such an approach has
its dangers. Sokal and Sneath thus go on to point out that
Any monothetic system will always carry the risk o f serious misclassification if
we wish to make natural phenetic groups. This is because an organism which
happens to be aberrant in the feature used in the primary division will inevitably
be removed to a category far from the required position, even if it is identical with
its natural congeners in every other feature. . . . T he advantage o f monothetic
groups is that keys and hierarchies are readily made.
336 M O D E L S F O R D E S C R IP T IO N I N G E O G R A P H Y
UNIVERSAL SET
Fig. 18.3 D iagram to dem onstrate grouping ‘tree’ for a set o f seven
counties.
IV QUANTITATIVE TECHNIQUES IN
CLASSIFICATION
Objects
>, k2 . . kj
Pm xml> . . . . . xmn
/ * p CLUSTER ?
< x x * .·
x) /
1 ; * χ
* x* ;
CLUSTER?
t - z, π 1 IQTF
CLUSTER?
100%
Fig. 18.4 The measure of distance for classifying in a two-dimensional
orthogonal space, showing : A : A simple case for grouping six observations
in the space made up of two variables; B : A complex case in which it is
difficult to discern groupings in the space.
this case the grouping is obvious, but in cases where it is not (Figure
18.4B) it is possible to search among all combinations of the objects
located in the space and select that particular combination that
maximises the squared distance between groups and minimises the
mean squared distance within groups.
In principle such a method may sound easy enough (even if rather
tedious) but there are a number of difficulties. The most serious
difficulty arises with respect to the geometry of the m-dimensional
space. If it is Euclidean then no problems arise, but this amounts to
saying that the correlation matrix between the attributes has no
entry significantly different from zero, i.e. all the attributes are
independent of each other. If the attributes are in any way correlated
(and they nearly always are) then the space is non-Euclidean and we
need to know the structure of the m-dimensional space in order to
calculate the distance between objects. This implies that we also need
GLASSIFICATION 341
to know what the interrelationships are among the attributes. Now
there are two ways we can know this. Given a sophisticated theory
about structure, we can predict what the interrelationships ought to
be among the attributes and hence define the m-dimensional space
theoretically. If we do not possess such a theory, we can seek to group
the attributes according to the way in which they co-vary over the
n objects. Such a procedure amounts to developing a general (if
temporary) theory about the structural interrelations among the
attributes and then using this information to define the structure of
the m-dimensional space. Grouping of the objects is thus contingent
upon some prior analysis of the interrelationships among attributes.
The net result is to draw attention to both aspects of our basic data
matrix—the attributes as well as the objects. The same modes of
analysis may, in fact, be employed to examine both aspects. Follow
ing conventional usage (Cattell, 1965) we may differentiate between
72-mode analysis (which examines the interrelationships among the
m attributes or variables) and (Tmode analysis (which examines the
interrelationships among the n objects or observations).
The procedural problem posed by the nature of the m-dimensional
space should not be regarded as a minor technical difficulty, to be
thrust aside at the first opportunity, to be assumed away without
much analytic effort. The tremendous importance of classification in
fact centres on this one basic problem, for its solution implies a con
siderable increment in our understanding of the phenomena in
question. To develop understanding of the interrelationships among
attributes is one of the basic aims of any investigation. The import
ance of classification as a search procedure rests on the fact that it
poses this very basic problem. Indeed, it may be argued that the solu
tion of this one procedural difficulty, implying as it does the con
struction of a theory of interrelationships, is far more important than
the end-product, the classification system itself.
Given that we possess measures of some kind on the m attributes
of the n objects, it is possible to estimate the nature of the inter
relationships among attributes or objects by using the various
measures of similarity, association, and correlation, that have been
devised. There are a large number of such measures available, but
which is appropriate depends on the nature of the measurement
system used (nominal, ordinal, etc., see Figure 17.1), on the size of
the sample (small-sample statistics usually being different from large-
sample statistics), and on the form of the distribution (non-normally
distributed data requiring non-parametric measures unless the data
can be transformed to a normal distribution, in which case the para-
metiic tests may be used). We may thus use a wide range of measures
342 MODELS FOR DESCRIPTION IN GEOGRAPHY
bode well as search procedures. They can lead us to new ideas, new
frameworks for analysis, and so on. They have tremendous potential
in geographic research. Whether that potential is realised depends on
whether mindful evaluation occurs rather than mindless application
by those using such techniques.
a farm may overlap into two or more parishes, a town may over
lap into two states, and so on. The hierarchy of areal units is
much looser than that which we might construct by the logical
division of some universal set (above, pp. 3 3 5 ~7 )· This loose hier
archical arrangement poses conceptual and inferential problems.
The same areal unit may be conceived of as singular or collect
ive. Consider, for example, two counties, A and B, which we wish
to compare with respect to agricultural activity. If we say that
A has 60% of its land under arable cultivation compared with 24%
in B, then the counties are being treated as singular areal units.
If we say that 60% of the farms in A are arable farms compared with
24% of the farms in B, then the counties are being treated as col
lective areal units. This difference is important for the two state
ments are completely different from each other in import. Further,
to say that 60% of the land in A is under arable compared with 24%
of the farms in B being arable, says nothing of any significance.
We are, in short, trying to compare completely different types of
areal individual. To establish a relationship between climatic con
ditions in a set of counties and the percentage of land under arable,
says nothing whatsoever about the relationship between climatic
conditions and the percentage of arable farms.
Similar problems of comparability and inference exist when differ
ent levels in the hierarchy are simultaneously studied. These diffi
culties are usually referred to as the scale problem. We shall begin by
considering the special case in which distinct steps in the hierarchy
of areal units can be recognised, i.e., areal units at one level can be
included in areal units at the next level. In such a ‘nested’ hierarchi
cal situation it should be observed that comparisons can only be
made between similar individuals (i.e. individuals at the same level
in the hierarchy) and that inferences made about relationships at
one level cannot be extended, without making strong assumptions,
to any other level (McCarty, et al., 1956; Haggett, 1965B; Duncan,
et al., 1961; Harvey, 1968B). This is not to say that conditions at
one level are irrelevant to conditions at another. It does indicate
that the nature of the analysis is contingent upon whether the
individuals being compared or analysed are at the same, or different,
levels. Three kinds of situation can be identified:
(i) Same-level analysis means that the individuals can be directly
compared since they are at the same level in the hierarcny.
(ii) High- to low-level analysis yields a contextual relationship (e.g.
price policy and price support at the national level forms the
context within which variation in farm production can be
analysed).
DATA COLLECTION IN GEOGRAPHY 353
(iii) Low- to high-level analysis yields an aggregative relationship
(e.g. national output is made up of the output of individual
firms).
All of these situations are of interest, but each requires its particular
mode of thought and, incidentally, a particular data-collection
procedure.
Such natural hierarchies in the phenomena being studied cannot
always be identified. With continuously distributed phenomena the
areal units are imposed rather than natural, and it is clear that areal
units of any size can equally well be chosen. Such arbitrary areal
units may be arranged into a hierarchy, but this design is imposed
not given. How, in such a situation, can appropriately sized areal
units be chosen? One possible method is to choose some arbitrary
co-ordinate system and identify uniformly sized areal units within
that co-ordinate system. We may collect data, for example, by way of
grid squares (Hagerstrand, 1967). It is also possible to use areal
individuals which are discrete in one respect as areal units for the
collection of other data. It is very tempting in such situations to
regard the discrete individual as being natural for the secondary
purpose for which it may be used. This is, of course, fallacious.
Geographers long ago discovered that national units were not ap
propriate for discussing climatic characteristics (Hartshorne, 1939,
44-7); and there is, similarly, nothing ‘natural’ about using counties
as units for discussing spatial variation in farming types. Administra
tive units, which from the point of view of administrative structure
may be regarded as singular areal individuals, are frequently used
to collect information regarding either continuously distributed or
small-scale discrete geographical phenomena. In such a situation
there is nothing natural in the way the data are aggregated and it
may well be that the administrative units are not ‘similar’ or com
parable from the point of view of the data aggregated within them.
Hence arises the necessity to adjust such areal data for the variable
area of the administrative unit itself—an adjustment that has not
always been made. Thus Chisholm (i960) criticises Dickinson’s
(19 5 7 ) discussion of commuting patterns in W. Germany and
Belgium on the grounds that Dickinson’s analysis ignores the effects
of varying sizes of administrative unit. Migration studies have been
particularly concerned with this type of problem (Hagerstrand, 1957;
Kulldorf, 1955), while Robinson (1956) and Thomas and Anderson
(1965) discuss the necessity of weighting data according to areal
size prior to correlation analysis. A general discussion on these and
similar problems is provided by Duncan et al. (1961) and Haggett
(1 965B )·
354 MODELS FOR DESCRIPTION IN GEOGRAPHY
The construction of the basic data matrix for areal individuals thus
faces a twofold difficulty. Firstly, the appropriate size of the data-
collection unit has to be determined. Secondly, it has to be ensured
that the units so devised are comparable with one another (or else
some weighting system has to be devised for the data recorded in
them). Both of these problems appear intractable without major
assumptions being made. There are, however, some glimmers of
hope for their solution emitted by the interaction between individuals
and attributes.
e f
Fig. 19.2 Types of spatial point sampling Design : A : simple random
sampling; B: areally stratified random sampling; Ci systematic random
sampling; D : systematic unaligned random sampling; E and F : two
versions of nested random sampling ( a fte r B e r r y a n d B a k e r , 1968; a n d
K r u m b e in , 1959).
DATA COLLECTION IN GEOGRAPHY 365
between terrain and farm size, and farm income. It is likely that
terrain affects farm size and there is, therefore, a multicollinearity in
the set of relationships. Blalock (1964, 89) writes:
One w ay out of the difficulty posed by multicollinearity is to make two in
dependent variables completely unrelated in a sample, even though they are highly
correlated in the population. This can readily be achieved through stratification,
which has a certain analogy with manipulations in experimental design.
I f f : a
a
c ac
be abc
b ab
b ab
The map has validity (with respect to the real world) only if the
conceptual schema which governs its construction itself has validity
with respect to that same real world. The map is, therefore, simply
a model of a theory about real-world structure. Constructing a map
without explicit theory amounts to stating an a priori model; with
explicit theory it amounts to stating an a posteriori model. The a
posteriori model, however, can be related only to the domain of the
theory which it represents. Using a particular a posteriori model to
examine phenomena outside the domain of the theory which it
represents amounts either to assuming the domain can be extended
to phenomena not initially covered by the theory, or to using the
a posteriori model as an a priori one. Consider the following example.
Maps represent, among other things, the relative location of
objects in space. Most maps are constructed with reference to a
Euclidean physical space and they are, therefore, a posteriori models
of a physical theory about real-world structure—a theory, incident
ally, that is generally accepted for the location of objects in physical
space and forms for the geographer a quite self-evident theory about
real-world structure. This ‘obvious’ theory and the a posteriori model
derived from it have subsequently been used to map complex socio
economic relationships. This use amounts to postulating a Euclidean
a priori model, or assuming that the relative location of objects in
socio-economic space can be adequately described by a Euclidean
theory of spatial structure. The nature of the spatial structure, how
ever, is not given a priori and the determination of it is essentially an
empirical problem. This problem has already been considered at
length (above chapter 14), but the important point for our under
standing of the map is simply this: the extent to which we can use the
map to discuss the location of objects and events in space depends
entirely upon the adequacy of the text linking map structure with
real-world structure. If we possess no text then no inferences can be
drawn about the real world. If we assume a text then we should be
clear as to the nature of the assumptions and not infer with respect
to the real world what we have already assumed in constructing a
text. If we possess a text we should be prepared to state it explicitly so
that we do not make inferences respecting domains not covered by
that text.
What this amounts to, as Dacey (unpublished) points out, is a
full discussion of the semantics of the map. Semantic issues are not
simply related to the selection of a geometry but also relate to the
symbols employed to represent phenomena. Here Dacey suggests
that ‘semiotics’ or the theory of signs, has much to contribute to our
understanding of the way in which the map can be used to convey
DATA COLLECTION IN GEOGRAPHY 373
different kinds of information. £By subsuming the map symbol
under the generic category of sign,’ writes Dacey, ‘the entire factual
and relational content of the highly general theory of signs is
made available as a basis for studying the map symbol.’ We can
thus apply formal semantics to understand the relationship between
the map and the reality it is designed to represent. Carnap (1942,
24-5) suggests that the final aim of any semantic system is to
establish rules of truth—that is, to establish a set of rules which
allow us to determine whether a particular ‘sentence’ derived from
a symbolic system, such as a map, is or is not true. Given such truth
rules, it is possible to classify statements made from the map as true
or false. Carnap points out that this procedure requires certain
preliminary operations. First, a classification of signs is required.
Second, rules of formation are required which state how and under
what circumstances new signs and symbols may be formed from this
initial classification (a simple example of such a formation rule might
be the interpolation of a new contour line between two existing
contour lines—the rules in this case state that interpolated contours
should not cross etc.). Third, designation rules are required which
relate the signs (map signs in this case) to some other sets of signs.
These designation rules are of great interest for, as Carnap points
out, there are no factual assertions in pure semantics, only con
ventions which relate one set of signs to another set. What this means
in the mapping context is that the map signs do not represent the
real world directly, but represent geographic concepts about the real
world. The factual statement which a map makes is at two stages
removed from the reality being described. Formal semantics simply
discusses the relationship between geographic concepts and the sym
bolic representation in map form. From this it follows that if the
geographic concepts are ambiguous and fuzzy, then the map state
ment, although it appears precise, will likewise be fuzzy or ambi
guous. Thus a map of the distribution of agglomerations is only as
good as our concept of an agglomeration. Here the process of defini
tion becomes of considerable importance (see above, p. 304). The
truth rules therefore allow us to identify necessary truths, given our
prior conceptualisation of reality. Formal semantics provides us with
a way of discussing the logical consistency of geographic statements
and map statements. The degree to which the logically true state
ments are empirically true can be assessed only by evaluating the
relationship between the geographic concepts and the reality these
concepts are concerned to discuss.
It is also possible to discuss the syntax of maps. Here the concern
is with the internal structure of the map statements and its basic
374 MODELS FOR DESCRIPTION IN GEOGRAPHY
tion. Maps have long fascinated people; and simply staring at them
and living with them has led to the formulation of many an interest
ing geographic hypothesis. Perhaps, therefore, there is a case for
designing maps to stimulate hypothesis formation—maybe even
stimulate our glands, as Sauer might say. Here the work of psycho
logists on the reaction of people to pictures has a great deal to tell
us about map construction. Simple psychophysical experiments on
the perception of shape, line, orientation, and so on (Ekman, et al.,
1963; Beck, 1967) here provide us with some rudimentary ideas, but
the perception of ‘information’ and ‘structure’ in multicomponent
situations has also been intensively studied by psychologists (see the
account by Garner, 1962). The evidence provided by such studies
shows how ambiguity and uncertainty, and the consequent ability
to discriminate structure, are dependent upon the stimuli and the
cultural background of the subject from whom a response is being
elicited. Segal et al. (1966) have demonstrated the considerable cross-
cultural variation that exists in the perception of shape and form.
Berlyne (1958) has also studied the preferences of subjects with
respect to pictures of varying levels of complexity and found that
in general people prefer to look longer at more complex pictures than
they do at simple pictures. But the greater the complexity the greater
the level of uncertainty and, hence, the greater ambiguity in inter
pretation. Too much complexity can saturate the ‘channel capacity*
of the human mind and merely succeeds, therefore, in eliciting a
negative response. Ambiguity and uncertainty are not in themselves
undesirable—a certain openness of meaning, as Kaplan calls it
(above, pp. 304-6) is necessary if our understanding is to progress.
It may even be possible to design maps deliberately that have the
right amount of ambiguity to stimulate our glands and not too much
to cause us to react entirely negatively. For most of us, Rembrandt
and Cezanne would have made better cartographers than Tackson
Pollock.
Maps have undoubtedly provided this kind of stimulus in the past.
Most maps contain a certain element of pure poetry, and hence con
tain all of Empson’s seven types of ambiguity, and more. But maps
are also designed to convey specific information and to do so in such
an unambiguous way that we are able to make specific decisions
relating to real-world activity on the basis of the map evidence.
The map functions as a communication system and it is relevant to
ask how much noise that system can contain. The map makes a firm
visual statement (colours change abruptly at boundaries, and so on)
and this can often be deceptive. If, for example, we possess poor
data, then how can we express notions about measurement error
376 MODELS FOR DESCRIPTION IN GEOGRAPHY
It has been shown that the ability of the map-user to discriminate and
evaluate the information contained in the map is not free from sub
jective elements and that the more the information contained in a
map the more ambiguity and uncertainty there is likely to be as
regards the interpretation to be put upon it. It is, however, possible
to measure some aspects of map information and therefore it is pos
sible to develop objective methods of map interpretation. These
objective ‘interpretations’ of map information in fact provide us with
higher-order information about the phenomena that have been
mapped. It is therefore possible to think of geographic information
as being provided at different levels in a hierarchy of generalisation.
In this hierarchy, as Bunge (1966, 39) suggests, maps lie in an inter
mediate zone between pre-maps (raw information conveyed in a
variety of ways) and mathematics (which provides a very general
statement regarding the structure of the spatial information).
Maps have traditionally formed the main data-storage system
which geographers possess. This use of the map as a locational
inventory or record has now been challenged by the use of the com
puter tape which stores a much greater amount of information much
more efficiently. Such a computer-storage system is still rather
expensive to operate, but many planning organisations are intro
ducing data banks to store this kind of information, and many
national censuses are now developing schemes of this kind (Tobler,
1964; Hagerstrand, 1967). From this ‘low-level’ information it is
possible to produce maps automatically by a computer-plotter
machine. There are a number of methodological problems that arise,
however, in the provision of automatic geographic information
systems. These problems relate back to basic issues, already discussed
in some detail, concerning geographic individuals and the develop
ment of space-time languages suitable for discussing geographical
distributions. There are, associated with these problems, a number
of technical difficulties of identification and map projection which
have been discussed by Tobler (1964) and Kao (1963; 1967). Dacey
and Marble (1965, 6) tackle the methodological problem, using
the basic linguistic concepts of map form already discussed. They
conclude
that maps, mathematical models, and other representations for spatial relations
use a ‘language’, but that this language differs in quite important respects from
everyday language, programming languages or the formal languages commonly
378 M O D E L S F O R D E S C R IP T IO N IN G E O G R A P H Y
1c 1c 1c * *
Twice Smoothed Map Z Thrice Smoothed Map Z
In this spirit Dacey (1964A; 1964B; 1966A; 1966B; etc.) has investi
gated the interrelationship between map pattern and hypothesised
spatial processes in various empirical situations. Other examples are
provided by Curry (1964) and Harvey (1966B).
This procedure of ‘searching’ map patterns with models con
structed from some hypothesised stochastic process raises some thorny
inferential problems. These problems are, for the most part, related
to what we have already identified as the scale problem (above,
(pp. 352-4). The various measures of pattern associated with
nearest-neighbour analysis and quadrat sampling are not independ
ent of the scale at which the point pattern is analysed. Different
quadrat sizes yield different frequency distributions, and therefore
provide different evidence for a hypothesised spatial process. Very
small quadrats (small relative to the density of the point pattern)
invariably yield a Poisson-like distribution and thus appear to sug
gest that the spatial process is random. Larger-size quadrats sampling
the same point-pattern may produce a negative binomial distribution
384 MODELS FOR DESCRIPTION IN GEOGRAPHY
The notion of cause and effect has been extremely important in the
history of scientific investigation. There is, however, probably no
other notion in the philosophy of science which has accumulated
around it so much contentious argument. The confusions are legion.
It is difficult, therefore, to cut through this mass of argument and dis
pute in order to show that cause and effect can provide us with a use
ful model for analysing geographic problems. There is no point in
attempting a full discussion of the history of cause-and-effect argu
ment or a full statement of the various nuances of meaning that are
currently attached to the notion. Those interested are referred to the
full discussion by M. Bunge {1963). But it is useful to clear away some
of the semantic difficulties associated with cause and effect at the very
beginning. M. Bunge (1963, 3-4) notes three principal meanings:
(i) Causation—the notion of a causal connection (sometimes called
the causal nexus) which associates a particular event (or set of
events) with a particular result (or set of results).
(ii) The Causal Principle—a law-like statement (i.e. a universal state
ment) cast in a cause-and-effect language.
(iii) Causal Determinism or Causality— a. doctrine asserting the universal
validity of the causal principle.
These three meanings are related. If we can establish the existence
of ‘causation’ in certain cases, then causal laws might also be estab
lished. The formulation of numerous causal laws (all of which are
demonstrably successful and in which we possess a good deal of con
fidence) may lead us to infer that causal determinism is the only
principle upon which we can obtain realistic understanding of the
world around us. We may also ask whether there are any other
forms of explanation than cause and effect which are acceptable to
us, and, further, raise the philosophical issue of whether the world of
phenomena is in fact governed by cause-and-effect laws exclusively.
Now it would be impossible to trace all the arguments and counter
arguments which have raged around these questions. But given the
views already set out regarding theories and models, as well as the
389
ggo M O D E L S F O R E X P L A N A T IO N IN G E O G R A P H Y
<
Such logical structures are, to say the least, enticing. They certainly
conform to many of our intuitive notions regarding what an explana
tory model ought to look like. Apart from this, the very power of such
structures suggests that they should play an important part in model
ling explanations and analytic investigations. But the outsize im
portance of such structures in geographic thinking at the present
time rests on the strong supposition that this logical form of analysis
accurately reflects real-world mechanisms and processes. Whether or
not this is true depends on the degree to which correspondence rules
may be established between the abstract logical structure and real-
world situations.
392 M O D E L S F O R E X P L A N A T IO N IN G EOG RAPH Y
The problem is, then, to identify the set of events A (defined by some
property or properties) which stands in such a relation to another
set of events B that we may assert that A causes B. The identification
of such sets is by no means easy.
It is implicit in the ‘common-sense’ version of the causal principle
that the set of events A should be somehow different from set i?. To
state, for example, that farmers grow hops because they are hop
growers scarcely conforms to our intuitive idea of the causal principle.
On the other hand, we are faced with the problem that as soon as
we connect two sets of events by way of the causal relation, we can,
if we search, insert intermediate sets of events between them. For
example, the sequence rainfall —> wheat yields, may be converted
into rainfall —> soil moisture-content —> wheat yields. In the first
statement rainfall is regarded as the direct cause, but in the second
it is treated as an indirect cause in a causal chain. Blalock (1964, 18)
thus points out:
it w ill u s u a lly b e p o s sib le to in se r t a v e r y la rg e n u m b e r o f a d d itio n a l v a r ia b le s
b e tw e e n a n y su p p o se d ly d ir e c tly r e la te d fa cto rs. W e m u st sto p s o m e w h e r e a n d
c o n sid e r th e th e o r e tic a l sy s te m clo sed .
1 Y ie ld p e r +*
a c re
2 F e rtilis e r
+*
in p u t
+* _ *
3 Labour
4 G ro s s f a r m g a te +* +* +*
re c e ip ts
T ra n s p o rt
5 co st to and _* +*
fr o m m a rk e t
In v e s tm e n t
+* _*
6 in m a c h in e r y
Fig. 2 0 . i (a )
Fig. 2 0 .1 (b)
F ig . 20.1 A : A tabulation of a set of hypothetical cause-and-eifect
relationships among a selected number of variables relevant to the study
of farm yields j B : A graphical portrayal of the same relationships.
CAUSE-AND-EFFECT MODELS 399
no unique causal ordering can be found. It is therefore simpler to
consider a particular form of this structural equation approach. The
following is condensed from Simon’s (1953, 58) and Blalock’s (.1964)
accounts.
Suppose we have the following causal system in which poor
growing-weather (Xx) small wheat crops (X2) —> increase in
wheat price (X3), and we assume the weather to depend only on a
parameter, the wheat crop on the weather (plus some random
shock), and the price on the wheat crop (plus some random shock).
Assuming the relationships to be linear we have the following system
of recursive structural equations:
Xx=ex
X2 — biXX x + e2
X3 = b32X2 T ^3*
This system requires, however, that X x has no effect on X3 except
through X2, and it may therefore be regarded as a special case of the
following system of equations in which b3X is set equal to zero:
Xx = ‘1
X2 — b21X x + e2
X3 = b3XXx + b32X 2 + e3.
If the parameters in this system are regarded as partial regression
coefficients then:
X3 — b31.2Xi + b32.xX 2 + e3.
Given the simple recursive model we are proposing, b3X.2 should be
equal to zero (apart from sampling variations). This means that
there is zero correlation between X x and X3when X2is held constant.
It should be possible, therefore, to evaluate such a causal model by
showing that b3X.2 is not in fact significantly different from zero. This
forms the basis of Blalock’s approach to the evaluation of causal
models. There are, of course, complications and some difficulties
with this approach and those concerned should read the rest of
Blalock’s analysis thoroughly.
Recursive causal systems are just one form of the structural
equation approach. Another form involves the analysis of reciprocal
causal systems. Here the model is designed to look at two-way inter
actions between variables but in order to do this it is necessary to
examine a system over a number of time periods. We may thus build
a special form of recursive system to deal with reciprocal interactions
of the form
400 MODELS FOR EXPLANATION IN GEOGRAPHY
Such a model seems peculiarly appropriate for examining Myrdal’s
notions of circular and cumulative causation—a notion which Pred
(1966; 1967) has followed up in some detail in the geographical
context.
It is perhaps worth while closing this section on the analysis of
causal systems, however, with a brief comment on the inclusion of
probability statements within such systems. When dealing with
aggregates of events, which can be treated statistically, it is
possible to construct probabilistic causal models. In some cases this
probabilistic element simply amounts to an error term or an inter
ference which enters into the closed causal system from outside.
Most econometric models are of this form. The terms can be par
titioned therefore into a group of variables whose interactions are
modelled deterministically and an error variable which is a catch-all
term for measurement error, external interference, and the like. In
other cases it is possible to construct causal systems in which the
variables within the system are modelled probabilistically. Consider
a causal-chain model in which the probability of event B following
event A, p(B | A), is known and the probability of event C following
event B, p(C | B) is also known, then by the multiplication theorem
the probability of event C following event A is given by the product
p(B I A) .p(C I B), provided there is no direct connection between A
and C. In this kind of model the deductive theorems of probability
can be used in combination with the logical structure of cause-and-
effect analysis to provide a convenient framework for analysis. Any
more substantial element of uncertainty is difficult to combine with
cause-and-effect logic, however, and non-deductive procedures
certainly cannot be incorporated. But the general ability to extend
cause-and-effect analysis to probabilistic situations has important
philosophical implications since it does provide some kind of chal
lenge to the metaphysical belief that cause and effect essentially
implies determinism. The sense in which this is so needs to be
explained.
V CAUSE-AND-EFECT ANALYSIS IN
GEOGRAPHIC RESEARCH
More complex forms of the simple causal model are also employed.
Causal chains, recursive systems, reciprocal recursive systems, and
so on, have all been used, often implicitly, in a geographical context.
Such models are particularly useful for examining the evolution of
spatial systems over time. Thus much of Pred’s (1966) work on
spatial dynamics has been cast in terms of cumulative causal inter
actions. Similarly, process-response models may be regarded as a
form of causal analysis which extends to reciprocal and recursive
systems. In all of these cases, of course, certain assumptions have to
be made, assumptions that concern causal ordering, identification
of the variables, and closure of the system. In each case we are im
posing something upon the empirical situation. There is nothing
wrong with imposing such assumptions. The mistake comes in draw
ing inferences which neglect to take account of the a priori assump
tions already made.
We may thus conclude that cause and effect provides a powerful
model for the analysis of geographical problems. We may use the
model to analyse individual occurrences, to indicate regular relation
ships, to examine dynamic systems, to construct theory, to state
laws, and so on. But the model has its limitations. There are critical
problems in mapping real-world problems into the model without
doing too much violence to the empirical situation. The one major
mistake to be avoided at all costs, is the false inference that because
the model may be applied in a number of situations with a con
siderable degree of success, that it is the only model open to us to use
in analysis and explanation and, further, that the real world is
necessarily governed exclusively by the operation of cause-and-effect
laws. To make such a false inference simply amounts to confusing an
a priori model which helps us to understand reality with a theory about
the nature of reality itself.
Basic Reading
Blalock, Η. M. {1964).
Bunge, M. (1963).
Reading in Geography
J o n e s , E . (1956).
M a r tin , A . F . (1951).
Montefiore, A. and Williams, W. M. (1955).
Chapter 21
Temporal Modes of Explanation
in Geography
while Wooldridge and East (1951, 82) come very close to committing
what is known as the ‘genetic fallacy’ when they write:
T h e b e st c la ssific a tio n s, in c lu d in g th o se o f la n d -fo r m s are g e n e tic , i .e ., b a sed o n
g e n e sis. T h e b e st w a y , in g e n e ra l, o f u n d e r s ta n d in g a n y th in g is to u n d e r s ta n d h o w
it h a s e v o lv e d or d e v e lo p e d .
Others have sought to preserve some balance between the genetic and
the functional mode of explanation. Thus Sorre {1962, 44) states:
W ith in th e g r o u p o f n a tu r a l a n d so c ia l sc ie n c e s— to w h ic h h u m a n g e o g r a p h y
b e lo n g s— w e u se tw o ty p e s o f e x p la n a tio n w h ic h a re n o t o p p o se d b u t c o m p le
m e n ta r y . W h a te v e r th e o b ser v e d p h e n o m e n o n m a y b e , it is r e c o r d e d in te m p o r a l
series, it is th e r esu lt o f a lo n g e v o lu tio n , a n d it is e x p la in e d b y a series o f a n terio r
sta tes. W h e n w e d e sc r ib e a p h e n o m e n o n in th e se term s w e g iv e it a g e n e tic
e x p la n a tio n ; le t u s s a y h isto r ica l e x p la n a tio n for it is b a s ic a lly th e sa m e th in g ,
h istoi'y in a la r g e sen se b e in g o n ly th e r esto r a tio n o f a su cc e ssio n . B u t th e sa m e
p h e n o m e n o n a p p ea r s a t th e sa m e tim e in a sp a tia l c o n te x t. I t m a in ta in s m u ltip le
c o n n e c tio n s w ith its e n v ir o n m e n t r a n g in g from sim p le ju x ta p o s itio n to c a u sa lity .
. . . C o n se q u e n tly , th e r e is a p la c e for a n e x p la n a tio n d r a w n fr o m th e r ela tio n s
o f a b e in g to its e n v ir o n m e n t— r e c ip ro c a l r ela tio n s b e c a u se w e are p r e se n te d w ith
a m a ss o f c o m p le x a c tio n s, r ea c tio n s, a n d in te r a c tio n s. T h is e x p la n a tio n , a c tu a lis t
b y d e fin itio n , is b a s ic a lly e c o lo g ic a l.
I TIME
(a) The substantive problem revolves around the elementary fact that
the geographer is dealing with the results of human decisions which
are partially dependent upon the perception of time by the individual
making that decision. There is often a considerable difference be
tween societies in which long-term investment is common (which
requires a progressive view of time) and societies in which short-term
T E M P O R A L M O D E S O F E X P L A N A T IO N I N G E O G R A P H Y 413
(b) The methodological problem is more basic, for it hinges upon finding
an acceptable interpretation for time in geography. It has to be
recognised that we, as individuals, are emotionally involved in
choosing a concept of time, and that we are not free of certain pre
dilections. How far emotion has affected geographical writing on
time is difficult to determine. Certainly, many have sought to support
the genetic viewpoint by referring to a particular concept of time-
flow. Thus Darby (1953, 6) writes:
C a n w e d r a w a lin e b e tw e e n g e o g r a p h y a n d h istory? T h e a n sw e r is ‘n o ’, for th e
p ro cess o f b e c o m in g is o n e p ro cess. A ll g e o g r a p h y is h isto r ic a l g e o g r a p h y , e ith e r
a c tu a l or p o te n tia l.
major swing in the nature of thought. Ritter in 1833 pointed out the
importance of the historical perspective in geography (Hartshorne,
1959, 83); and the historical perspective of geology had a consider
able influence upon physical geography. By the end of the century
Davis and Ratzel had both adopted evolutionary standpoints in their
studies of landscape form and diffusion respectively—an evolution
ary viewpoint that came to dominate geomorphology for half a
century and had a profound impact on human geography through
the work of Huntington, Griffith Taylor, Sauer, and others. In the
second place it is perhaps to be expected that some of the excesses
of the ‘temporal change’ school of thought should also be found in
geography. These excesses go under a variety of names such as the
genetic fallacy—the belief that the significance of something could
be evaluated only by reference to its origins (a point of view the
Victorians were particularly prone to) and historicism—the belief
that the nature of something could be entirely comprehended in its
development (Barraclough, 1955, 1). Such statements are not un
common in geography, and even respected writers such as Sauer
and Wooldridge and East (cf. the quotations on page 369) have been
guilty in this respect. The fallacy comes in supposing that the genetic
approach is the only one possible and that it produces entire and com
plete knowledge.
Perhaps the most important point about these ‘background’
features in developmental thinking, however, is that it is often
difficult to distinguish between subjective (or simply assumed) prop
erties of time and those properties that could be empirically dis
tinguished. In general, the temporal systems of history (cyclical,
linear, oscillatory, and so on) were assumed, and even the extensive
empirical studies of Spengler and Toynbee assume rather than
demonstrate the nature of historical time. In the natural sciences
the situation was less clear. Temporal change and evolution could
be established and in biology this appeared to take the form of
an irreversible branching process—but there was considerable un
certainty, largely because of the difficulty of distinguishing the
mechanism—a mechanism that was eventually provided by Mendel-
ian genetics and subsequent developments in genetic theory. Out of
this tangle of subjective and objective elements it was often difficult
to choose rationally a coherent view of evolution and developmental
change. Thus, as Stoddart (1966) is one of the rare articles on the
interaction between modes of geographic thinking and general
modes of thought, points out, evolutionary thinking in geography,
heavily influenced as it was by Darwin’s Origin took over the notion
of a necessary evolution but somehow lost the idea of the chance
T E M P O R A L M O D E S O F E X P L A N A T IO N IN G E O G R A P H Y 417
How far measures of time defined with respect to one process can
be applied to the examination of other processes is therefore an
empirical problem. There are conventionally established time scales
for locating events (calendars, and so on) but it is worth while
questioning whether these are appropriate for examining social pro
cesses which tend to vary greatly in terms of natural rhythm. It may
even prove useful to invent our own time scales (resource-exhaustion
time, amortisation time, erosion time, etc.) and to develop time trans
formations from one scale to another (Harvey, 1967A, 5 5 9 )· Most of
our notions regarding time are akin to a. priori models (they are
intuitive and subjective) as compared with the a posteriori models
which physics and astronomy provide. The use of such a priori models
requires the usual care and ‘eternal vigilance’.
A Process
The technical meaning of the term ‘process’ is rather different from
its ordinary everyday interpretation. In the ordinary sense we fre
quently use ‘process’ to refer to any sequence of events over time.
Such usage has little to recommend it in the context of explanation,
for it fails to differentiate between any sequence of events (which
cannot be regarded as explaining anything in particular) and a
sequence which is connected by some established mechanism (and
which can thus be regarded as explanatory). It is thus possible to
differentiate between sequences that exhibit necessary and sufficient
connections between the various stages in the sequence and those for
which no such connections can be found. A process, in the scientific
sense, can be given a more rigorous definition with the aid of some
of the conclusions drawn from our analysis of the logical structure of
cause-and-effect systems. A process law depends, therefore, on the
specification of:
(i) the system (a closed system) within which the process law
operates;
(ii) the relevant states of the system;
(iii) the relevant variables interacting within the system;
(iv) the parameters governing interaction among the variables and
the direction of the interactions.
Process laws succeed in explaining only in so far as these features can
be specified. Bergmann (1958, 93, 117) therefore puts the scientific
process model in the following schematic form. Let C — (cl3 c2, ... , cm)
be a description of the system and St = (x^, x2\ . . ., x j) be the
420 M O D E L S F O R E X P L A N A T IO N I N G E O G R A P H Y
description of the state of the system at time t. Then given C and Sh,
it is possible to predict Sta, from the process law, but the reverse
prediction can also be made (i.e. knowing Sta we can predict what
the state of the system was at Sh). Thus ‘if the system is known then
any two of its states can by means of the process law be inferred from
each other.’ A process law thus defines what we shall later call the
‘trajectory of a dynamic system’ (below, p. 461).
This logical schema illustrates the meaning of process in its strict
scientific sense. This meaning is typified by a markov process in
which states of the system are identified, an initial state given, and
a matrix of transition probabilities provides the necessary process
law (see Harvey, 1967A for an elementary account of such structures
in a geographic context). In empirical work, this view of process
implies that process laws succeed in ‘explaining’ only in so far as the
empirical situation can be regarded as isomorphic with, and hence
mappable into, a logical schema that possesses these characteristics.
This translation poses difficulties. It is thus not easy to identify the
limits of the system being considered (i.e. to close it), identify the
relevant states and variables, and to know everything there is to
know about the interaction among the relevant variables. In control
lable situations (e.g. under experimental conditions) these difficulties
are not insuperable, but in most areas of geographical enquiry such
a mapping requires some powerful assumptions. Bergmann (1958,
127) suggests that process laws, in the strict sense, cannot yet be
identified in social science because the conditions necessary for such
laws to operate have not been identified. Nevertheless, giving a strict
specification of process has a number of advantages:
(1) Narrative
It is not usually thought that plain historical narratives are con
cerned with explanation. The aim is usually to describe the occur-
422 M O D E L S F O R E X P L A N A T IO N I N G E O G R A P H Y
Basic Reading
Toulmin, S., and Goodfield, J. (1965).
Meyerhoff, H. (i960).
Nagel, E. (1961), chapter 16.
Reichenbach, H. (1956).
432 MODELS FOR EXPLANATION IN GEOGRAPHY
Reading in Geography
Hartshorne, R. (1959), chapter 8.
Harvey, D. (1967A).
Sauer, C. (/o6Y), chapter 17.
Smith, C. T. (1965).
Stoddart, D. (1966).
Chapter 22
Functional Explanation
for otherwise the economy will not function. Hempel suggests that
the appeal of functional analysis is ‘partly due to the benefit of
hindsight’, in that we already know that markets do exist. Functional
analysis provides a convenient method for identifying some of the
necessary conditions for the existence of central places, but it cannot,
apparently, provide necessary and sufficient conditions for explaining
a given trait, such as the occurrence of central places.
This analysis of the logical deficiencies of functional explanation
does not in any way vitiate the use of functional statements in explan
ation. Taken in conjunction with certain other specific conditions,
function statements may indeed be used in the course of offering an
explanation. These conditions have already been hinted at in the use
of phrases like ‘recurrent activity’, ‘proper working order’, and so on.
Brown (1963, no ) suggests that functional relations are a sub-class
of causal ones:
T he difference between the two sorts of relations is that functional relations
hold only between traits within a specified system of a certain type— a self-
persisting one—while the class of causal relations is much larger. It includes these
and others as well. . . . Functional relations, then, are certain causal ones which
operate within self-persisting systems.
which the cause lies in the future or in some end that may be future
or present. Functional explanation as we have considered it so far is
a particular form of this. It is possible to argue in favour of teleo
logical explanations in a number of ways. We can maintain that:
(i) There are ‘final causes’ which are God-given or nature-given.
This is a metaphysical assertion which is incapable of empirical
test and which has no relevance to empirical understanding.
(ii) There is some degree of purposive behaviour on the part of
individuals or objects. My staying in Bristol may thus be
associated with an intention on my part to finish writing this
book. Such purposive behaviour may seem a reasonable asser
tion about individuals, but many would maintain that it is
needlessly anthropomorphic to assert that cultures have inten
tions, economies have intentions, and so on.
(iii) Teleological explanations are in principle (as both Nagel and
Braithwaite show) reducible to ordinary causal explanations (a
conclusion which Hempel’s analysis also indicates).
Out of these we must necessarily reject (i) as unscientific (which
is not to say it is untrue), while (ii) provides us with a rather weak
form of explanation (see above, p. 57)· The problem with (iii)
is that the circumstances for such a reduction are not in general
known. Braithwaite (ig6o, 3 3 4 ~5 ) t^lus concludes that
in general irreducible teleological explanations are no less worthy of credence
than ordinary causal explanations. . . . It seems ridiculous to deny the title of
explanation to a statement w hich performs both of the functions characteristic of
scientific explanations— o f enabling us to appreciate connexions and to predict
the future.
Braithwaite’s conclusion is not at variance with the conclusions
drawn by other analysts, except that it shows a greater preparedness
to accept teleological explanations as first-stage approximations to
some more rigorous scientific explanation form. The general con
sensus being that teleological and functional explanations cannot
be regarded as different in kind from other varieties of explanation,
it seems that the only important decision we need to make regarding
such explanatory forms is how long and under what circumstances
we are willing to put up with first-stage approximations of this sort.
It may well be that rough functional and teleological explanations
are the best we can achieve in the given state of our understanding.
But in principle there seems no reason why we should not resort to
more complete explanatory forms. Even with these, functional and
teleological explanations may still have a role to play since they can,
as Lehman (1965) concludes, often provide a quick and efficient
way of answering a particular kind of question.
FUNCTIONAL EXPLANATION 439
C ‘Wholes’
Hartshorne, in the passage on functional regions quoted above,
suggests that a functional unit constitutes a whole which is in some
way more than the sum of its parts. In doing so he refers to a general
444 MODELS FOR EXPLANATION IN GEOGRAPHY
philosophical problem that has been a source of major controversy
in organic biology and in psychology (where gestalt psychology pro
vides a whole school of thought stemming from the belief that there
are certain indivisible units). Functionalism is thus sometimes associ
ated with a doctrine of ‘wholeness’. In geographic thinking such a
doctrine has not been a central one (Hartshorne, 1939, 260-80),
although it has had some impact by way of the cruder applications
of the organismic analogy to, for example, regional and political
wholes. It has also been used in connection with gestalt concepts
to describe the reaction or relation between man and environment
(Kirk, 1951; Hartshorne, 1939, 276). More recently the concept of
organic or functional wholes has been raised in a new and rather
interesting form through the application of systems thinking to geo
graphy. It is perhaps useful to take a closer look at such concepts.
Nagel (1961, 380-97) shows that the exact meaning of the state
ment that ‘the whole is more than the sum of its parts cannot be
determined, simply because the terms ‘whole’, ‘p a rt, and sum , are
variously interpreted and hence ambiguous (he lists eight possible
meanings given to the term ‘whole’ for example). Given this am
biguity, it is impossible to prove or deny that objects or systems
actually exist which are indeed more than the sum of their parts.
Such an assertion is in part a matter of a priori belief and thus belongs
more to the realm of metaphysics than to that of logic. Nevertheless
there are numerous situations in which it is common to talk of such
functional wholes. The distinctive feature of such situations is that
the behaviour of some system is not determined by the individual
elements within it, but rather that the behaviour of the individual
elements is determined by the intrinsic nature of the system itself
(i.e. the whole). The individual elements may thus show a high
degree of mutual interdependence, but to demonstrate this empiri
cally is not to demonstrate that the whole determines the part. Some
writers have suggested however, that any system that exhibits a
high degree of internal organisation should be deemed as a functional
whole. Nagel (1961, 393) thus concludes that
although the occurrence of systems possessing distinctive structures of inter
dependent parts is undeniable, no general criterion has yet been proposed which
makes it possible to identify in an absolute way systems that are genuinely
functional’ as distinct from systems that are ‘merely summative’.
Now systems that can be regarded as the sum of their parts are
analysable in terms of those parts. This Nagel terms an additive
form of analysis. If functional wholes exist, some other ‘non-additive
form of analysis would be necessary. Nagel thus goes on to examine
the suitability of additive forms^of analysis and contrasts the particle
F U N C T IO N A L E X P L A N A T IO N 445
physics of classical mechanics with the field approach of electro
dynamics. This example is of considerable interest, for thinking
in terms of fields of influence, hinterlands, and so on, is very common
in geography, while more recently the field concepts of psychology
(which are closely related to gestalt psychology in spirit) have been
explicitly introduced into geographic theory (Wolpert, 1965; Berry,
1966; Harvey, 1967B). The question whether such field phenomena
should be regarded as functional wholes incapable of additive analy
sis is thus of some significance. Nagel does not provide a firm answer
to this problem. He points out that it may be convenient to treat
field phenomena as functional wholes and to develop non-additive
methods of analysis to deal with them. This does not mean that field
phenomena must necessarily be so regarded; and indeed in certain
ways they may be treated additively without any loss of information.
The question of functional wholes, indivisible units, and the like,
cannot therefore be solved by logical analysis. Yet that analysis
yields sufficient understanding for Nagel (1961, 397) to assert with
some confidence that the issue ‘cannot be solved, as so much of the
extant literature on it assumes, in a wholesale and a priori fashion’.
The position we adopt with respect to such functional wholes may be
incapable of full empirical support, but it cannot, either, ignore the
results of empirical and logical analysis. These suggest that it may
be safest to construe the differences between units conceived of
as functional whole and units conceived of as summative units, as
differences of degree rather than of kind.
D Methodological functionalism
Whatever may be said about the logic of functional explanation or
functionalism as an a priori assumption, there can be no doubt of the
very substantial achievements and insights that have been gained
through adopting functionalism as a working hypothesis. To attack
functionalism as a philosophy is not, therefore, to attack it as a
methodology. Indeed, it has been claimed that the heuristic rules
that form the basis of functionalist methodology are extremely useful
and need not be associated either with functionalist philosophy or
functional forms of explanation (Jarvie, 1965). It is perhaps in this
light that we should accept Philbrick’s (1957) plea for a greater use
of the functional approach in geography.
The methodological strength of functionalism really lies in its
emphasis upon interrelatedness, interaction, feedback, and so on, in
complex organisational structures or systems. This kind of approach
to problems has been extremely rewarding where it has been freed
44g MODELS FOR EXPLANATION IN GEOGRAPHY
from metaphysical connotations. It thus seems very valuable to
enquire how central places function in an economy, simply because
it opens up a whole range of questions which, if we can provide
reasonable answers, will yield us deeper understanding and greater
control over the phenomena we are examining. The world is mani
festly a complicated place, and the heuristic strength of functionalism
is that it directs our attention to this complexity. As a philosophy it
assumes that this complexity is essentially unanalysable into consti
tuent parts—a point of view that seems unscientific since it provides
licence for the employment of intuitionism in (who can tell?) reveal
ing or specious form. As a methodology it provides us with a series
of excellent working assumptions about interactions within complex
systems. But as the analyses in the previous two sections have shown,
functionalism cannot suffice either as philosophy or as a form of
logical analysis. In a discipline still heavily dependent upon first-
stage approximations’ a strong case might be made for a fuller em
ployment of methodological functionalism as a heuristic device, and
even for explanations in terms of function. But first-stage approxima
tions must presumably give way at some stage to full-blown theory.
The danger we must here avoid at all costs is that mortal inferential
sin of erecting an a prion functional model into full theory without
knowing it and without the necessary confirmatory evidence.
Basic Reading
Hempel, C. G. (1959). This is an essential reference.
Brown, R. (1963), chapter 9.
Nagel, E. (1961), pp. 380-446 and 520-35.
Reading in Geography
Philbrick, A. K. (1957)·
Chapter 23
Systems
In the last chapter it was suggested that the valid use of function
statements was entirely dependent upon the objective specification
of some system (above, p. 435). This conclusion is of considerable
significance, for it indicates that the rendering of an explanation of
any type may be contingent upon the specification of a system, and
that therefore our understanding of the process of explanation itself
may be contingent upon our understanding of the concept of system.
Gouldner (1959, 241) writing in a sociological context puts it this
way:
T he intellectual fundam ent of functional theory in sociology is the concept
o f a ‘system’. Functionalism is nothing if it is not the analysis o f social patterns
as parts o f larger systems of behavior and belief. U ltim ately, therefore, an under
standing o f functionalism in sociology requires an understanding of the resources
o f the concept of ‘system’. Here, as in other embryo disciplines, the fundamental
concepts are rich in ambiguity.
I SYSTEMS ANALYSIS
The concept of a system is not in any way new. Newton wrote on the
solar system, economists have written on economic systems, biolo
gists on living systems, plant and human ecologists have used system
concepts, and geographers have certainly made considerable use of
the notion of a system ever since the discipline originated. Although
systems concepts are very old they have tended to remain on the
fringe of scientific interest—acting as contraints almost—rather than
MODELS FOR EXPLANATION IN GEOGRAPHY
being the subject of intensive investigation. The modern emphasis
on systems as an explicit (and indeed central) item for analysis may
be seen as part of a general change in emphasis from the study of
very simple situations in which the interactions are few, to situa
tions in which there are interactions between very large numbers
of variables. It has been pointed out (von Bertalanffy, 1962, 2;
Ashby, 1963, 165-6) that classical science was concerned almost
entirely with simple linear causal chains or with unorganised com
plexity (such as that characterised in the second law of thermodyna
mics). Indeed, Ashby (1963, 165) suggests that the success of science
for the last two hundred years or so must be attributed to its exploit
ation o f‘the many interesting systems in which interaction is small’.
We have already noted how, in the behavioural and biological
sciences, the complexity of interaction poses special problems for the
application of classical methods—indeed, the functionalists were
adamant that the classical concepts of science were inappropriate
for the study of complex systems. The interest in these complex
systems has grown rapidly in the twentieth century, but the problems
they posed seemed scarcely tractable without some conceptual and
technical breakthroughs. The necessary technical apparatus ^ to
handle systems has been steadily built up out of the mathematical
development of communications engineering, cybernetics, inform
ation theory, operations research, and the like. Thus:
Since 1940 . . . a serious attempt has been made, aided by the new techniques,
to grapple with the problems of the dynamic system that is both large and richly
connected internally, so that the effects of interaction are no longer to be ignored,
but are, in fact, often the focus of interest. . . . So has arisen systems theory the
attempt to develop scientific principles to aid us in our struggles with dynamic
systems with highly interacting parts (Ashby, 1963, 166).
The intimate connection between systems analysis and the analy
sis of complex structures makes this approach very attractive to those
disciplines dealing with phenomena that are highly interconnected.
Given the multivariate nature of most geographical problems, it is
hardly surprising that systems analysis provides an appealing frame
work for discussing these problems.
(a) The element is the basic unit of the system. From the mathematical
point of view an element is a primitive term that has no definition
(rather like the concept of point in geometry). Mathematical sys
tems analysis can thus proceed without further consideration of the
nature of elements. But the use of the mathematical theory of systems
to tackle substantive problems depends entirely upon our ability to
conceptualise phenomena in such a way that we may treat them as
elements in a mathematical system. Put the other way around, it
depends upon our ability to find a substantive interpretation for the
mathematical element. It is not always easy to hnd such interpre
tations which we can agree upon as reasonable and unambiguous.
There are two basic problems here. First the scale problem has to be
faced. The substantive interpretation of an element is not independ
ent of the scale at which we envisage the system’s operating. The
international monetary system, for example, may be conceptualised
as containing countries as elements, an economy may be thought of
as being made up of firms and organisations, organisations them
selves may be thought of as systems made up of departments, a
department may be viewed as a system made of individual people,
each person may be regarded as a biological system, and so on. The
definition of an element thus depends on the scale at which we
conceive of the system, or as Klir and Valach (-1967, 35) call it, the
resolution level:
Every element is characterised by forming, from the point of view of the cor
responding resolution level (at which the system S is defined), an indivisible unit
whose structure we either cannot or do not want to resolve. However, if we
increase the resolution level in a suitable manner . . . the structuie of the element
can be distinguished. In consequence, the original element loses its meaning and
becomes the source of new elements of a relatively different system, i.e. of a
system defined at a higher resolution level.
In substandve terms, therefore, we face the problem that systems
may be embedded in systems, and that what we choose to regard
as an element at one level of analysis may itself constitute a system
at a lower level of analysis. A car may be an element in the traffic
system, but it may also be regarded as constituting a system. This
feature of systems analysis carries with it some problems. Blalock
and Blalock (1959) thus point out that there are two ways in which
we can conceive of an element at some higher level in the hierarchy
of systems. It can be regarded as an indivisible unit that acts as a
unit (e.g. we may think of a firm deciding or responding) or it may be
regarded as some loose configuration of lower order elements (e.g.
individuals within an organisation interact with individuals in other
organisations). These two interpretations are portrayed in Figure
SYSTEMS 453
23.1. Blalock and Blalock (1959) point out the considerable con
fusion that may result from a failure to distinguish between these
two points of view and suggest that major controversy in sociology
regarding the nature of ‘the social system’ can be attributed almost
entirely to semantic difficulties surrounding these two different
(b) The relationships or links between the elements provide the other
component in the structure of a system. Three basic forms of
relationship can be defined (Figure 23.2). A series relation is the
simplest and is characteristic of elements connected by an irreversible
link. Thus —> «3· forms a series relation and it may be observed
that this is the characteristic cause-and-effect relation with which
traditional science has dealt. A parallel relation is similar to multiple-
effect structures in that both a{ and β;· are effected by some other
element ak. A feedback relation is the kind of link that has been newly
introduced into analytic structures (mainly through the develop
ment of cybernetic analysis by Wiener, ig6i, and others). It describes
a situation in which one element influences itself. Thus the value
assigned to an attribute of an object is affected by that value itself.
It is possible to combine these relationships in a number of ways (see
Figure 23.2), so that two elements may be connected in various
different ways simultaneously. The links thus form a kind of ‘wiring
system’ connecting the elements in various ways. This emphasis on
connectivity between elements can be discussed in terms of topo
logical relations and it is not surprising to find, therefore, that graph
theory is an important descriptive device in the analysis of the struc
ture of systems.
(2) The behaviour of a system
When we speak of the behaviour of some system we are simply referring
to what goes on within the ‘wiring system’ that makes up its struc
ture. Behaviour has to do therefore, with flows, stimuli and responses,
456 MODELS FOR EXPLANATION IN GEOGRAPHY
inputs and outputs, and the like. We can examine both the internal
behaviour of some system or its transactions with the environment.
A study of the former amounts to a study of the functional ‘laws’ that
connect behaviour in various parts of the system as discussed in the
last chapter. Most analyses of behaviour tend to concentrate on
the latter aspect, however, and it will be examined in this context
here. Consider a system that has one or more of its elements related
to some aspect of the environment. Suppose the environment under
goes change. Then at least one element in the system is affected and
effects are transmitted throughout the system until all connected
elements in the system are affected. This constitutes a simple stimulus-
response, or input-output system without feedback to the environ
ment:
stimulus response
system
(input) (output)
The mathematical meaning of behaviour can easily be defined. We
can define a vector of inputs X = (xx, x2, · . xn) which function as par
tial stimuli to the system, and a vector of outputs T — (y x, y 2, - . ·, y n)
which refer to the response of some system. The behaviour of a
system can then be expressed in general as the transformation T of
the vector X into the vector T:
r = T(X)
In other words the behaviour is described by the equations (deter
ministic or probabilistic) that connect the input vector with the
output vector (Klir and Valach, 1967, 31-2). The simplest example
of this is provided by input-output analyses of economies, in which
a vector of final demands (e.g. derived from exports, home con
sumption, or however) is related to a vector of final outputs in various
sectors in the economy. In this case the system is composed of all
inter-industry links within the economy and the effects are traced
throughout the economy to yield final outputs. The system is
represented in this case by the matrix of technological coefficients.
The mathematical aspects of the behaviour of systems will not be
considered further here, but it is useful to mention a practical con
straint imposed by the feasibility of mathematical manipulation.
In general it is easiest to handle linear systems. Much of the liter
ature on systems behaviour, therefore, is concerned with linear
systems, and in practice it proves necessary to conceptualise (or
measure) real-world relationships as if they were linear even if they
are not. This constraint is not one imposed in principle, it is simply
related to the relative difficulty of handling large systems of non-
SYSTEMS 457
linear equations. In practice we may be able to observe the trans
formation by recording inputs and outputs in tabular form, yet fail
to identify any mathematical function which fits that transformation.
Ashby (1966, 21) suggests that simple mathematical functions are
rarely found in practice. It is useful to differentiate, therefore, be
tween situations in which we can handle the behaviour analytically
and situations in which we possess observations of input and output
only.
(3) The boundaries of a system
It is possible to investigate the structure and behaviour of some system
only if the boundaries of that system are first identified. Mathematically
this is no problem, since the boundaries are given by defining certain
elements as being in the system and other (relevant) elements as
belonging to the environment. In order to use the mathematical
properties of systems analysis, however, we require some operational
method of defining boundaries. This is no easy problem. In some
cases the boundaries are fairly self-evident (particularly when we are
dealing with systems that are discrete and have well-defined con
nections with their environment, e.g. a firm within an economy).
In other cases we are forced to impose boundaries in some fashion,
and in doing so employ our own judgement as to where the system
begins and ends. Forrester (1961, 117-18) points out that such judge
ment decisions do not mean that the choice lacks £a foundation of
fact or contact with reality’. We may evaluate the choice of system
boundaries by referring that choice to the objectives of an investi
gation together with our experience of such systems. The choice of
boundaries, however, may have a considerable impact upon the
results obtained from systems analysis. A good example is provided
by input-output analysis of an economy, in which the final demand
(the stimulus) can be conceptualised in a variety of ways. If it is
treated solely as exports, for example, then the household sector and
labour inputs become a part of the network of interactions within
the economic system. In some studies, however, household demand
is treated as a component of final demand and therefore lies outside
the system (Isard, i960, chapter 8). The predictions of final output
will vary according to which of these choices is made. Both ap
proaches are reasonable; and which is chosen depends mainly on
what is feasible and what we wish to investigate.
(4) The environment of a system
In general terms the environment of some system may be thought of as
everything there is. But it is useful to develop a much more restricted
45g MODELS FOR EXPLANATION IN GEOGRAPHY
definition of an environment as a higher-order system of which the
system being examined is a part and changes in whose elements will
bring about direct changes in the values of the elements contained
in the system under examination. Again, there seems no objective
way of defining an environment in such a way that everyone will
agree that a particular definition of the environment is correct.
Defining the environment in an operational sense poses as much
difficulty as defining the system contained within it. This is not to
say that the definition is entirely arbitrary. Perhaps the best way to
approach this problem is to ask what are the relevant elements in the
environment for the operation of some system? We may then close
off some ‘meta-system’ which is made up of relevant elements of the
environment in interaction with the system under consideration. The
irrelevant elements in the environment (and they are presumably
infinite in number) are then discarded. Again we must abstract and
close our model before any analysis can proceed. In most cases it is
surprising how easily agreed definitions of the environment can be
arrived at, e.g. that the economy constitutes the environment of a
firm, etc. In other cases the environment is made up of two rather
different systems intersecting with respect to the system under in
vestigation. A farm system, for example, may havefor its environment
the biosphere and the economy. This flexible approach to the con
cept of environment in systems analysis is particularly useful to geo
graphy, which has made considerable use of the notion of enviion-
ment but has tended to use it in rather a rigid fashion. It is useful
at this juncture to clarify the usual meaning of the terms open and
‘closed’ system. The former refers to models which incorporate inter
action with the environment, while the latter refers to models built
without any interaction. The terminology is a little unfortunate since
analysis requires closure of the meta-system in any case.
the system means that, if the system is not initially in a preferred state, the system
w ill so act as to alter its state until one of the preferred ones is achieved.
has much to learn from systems analysis and in particular from infor
mation theory, which has the power to distinguish the degree of
organisation (the degree of departure from randomness) and thus to
provide objective measures of pattern and a technique of pat
tern recognition. Those geographers who focus attention upon
spatial organisation invariably invoke systems analysis as Haggett’s
(1965A) account of locational analysis in human geography demon
strates.
Yet it seems that the employment of systems concepts and systems
analysis has not yet achieved powerful operational status in geo
graphy. In part this must be attributed to the complexity of systems
analysis itself, which, if it is to be fully employed, involves mathe
matical techniques beyond the reach of most geographers. The solu
tion to this difficulty is, of course, for the geographer to learn more
mathematics, but this is perhaps easier said than done. But there
are other difficulties. Operationalising systems analysis involves many
evaluative judgements regarding the closure of the system, the
definition of the elements, the identification of the relationships, and
so on. The greater our experience of some problem and the more
information we possess, the easier it is to make such evaluative judge
ments with some degree of confidence. Our general lack of experience
with systems analysis, together with the relatively weak development
of theory, does not allow us to make such evaluations with any degree
of confidence, except in those cases where we can easily make assump
tions regarding the structure and behaviour of some system. We are,
in short, very much in the stage of a priori model use in our attempts
to apply systems concepts to geography. But experience in cognate
disciplines such as economics (e.g. Orcutt, et al., 1961), psychology
(Miller, 1965), political science (Deutsch, 1966), urban economics and
planning (Meier, 1962), business economics (Forrester, 1961), and
so on, suggests that the attempt to use systems concepts will be worth
while, if only because it provides the necessary framework for asking
the kinds of question that seem particularly relevant to the study of
the ‘organised complexity’ with which geographers deal. But we
should be foolish to think that we require merely to wave the magic
wand of systems analysis over the subject-matter of geography for
all to be revealed. As Hagen (1961, 151) has pointed out, each disci
pline has to grope its way through its own substantive problems
before it can see how and why systems analysis is relevant and
applicable:
This is w hy each discipline slowly and stumblingly rediscovers concepts con
cerning m ethod already discovered long ago in other disciplines—why, for
example, economics clumsily and painfully groped its w ay to the concept of
MODELS FOR EXPLANATION IN GEOGRAPHY
marginal productivity and only subsequently realised that it was merely applying
elementary calculus to its problems. . . .
Systems analysis thus provides us with a convenient calculus for
examining geographical problems. But to use that calculus we require
geographical concepts that allow us to find an interpretation for that
calculus in geographical context.
A Systems modelling
Consider two systems, Sx and S2, whose structure and behaviour are
known. By this we mean that we can identify the elements in each
system, the links, the outputs, and so on. Given this information, it is
possible to establish in what respects 6^ and S2 are similar and under
what conditions S2 can be used as a model of Sx and vice versa. Now
there may be several respects in which Sx and S2 are similar and
several in which they are dissimilar. They may exhibit similarity of
structure, of behaviour, of composition of the elements, of mathema
tical form of the relationships, and so on. It is therefore important to
distinguish the respect in which we are modelling the system,
whether, for example, we are modelling its structure, its behaviour,
or the system as a whole. We will first examine the most general case
and then examine the more specific aspects of systems modelling.
(1) The model of a system
Klir and Valach (1967, 108) point out that the principle of modelling
Sx and S2 is based on the concepts of isomorphy and homomorphy. Two
systems are isomorphic if the elements in Sx can be uniquely assigned
to the elements in S2 and vice versa, and if for every relationship
{ r f in Sx there exists an exactly similar relationship in S2 and vice
versa. The isomorphic relation between two systems is symmetrical,
reflexive, and transitive. A system through which water flows may
thus be made isomorphic with a system through which electric
current flows. Two systems are homomorphic when the elements in Sx
can be assigned uniquely to elements in S2 but not vice versa, and the
relationships in Sx can also be assigned uniquely to relationships in
S2, but not vice versa. A good example of homomorphic relationships is
that between a map and the countryside, in that every element on
the map can be assigned to an element in the countryside (but the
countryside contains many elements not recorded on the map) and
every geometric relationship portrayed on the map also holds in the
countryside (with respect to physical distance that is) but there are
many geometrical relationships which actually exist in the country
side which are not portrayed on the map. In such systems the
relationships between them are not symmetrical, reflexive, or transi
tive. We may treat the map as a model of the countryside, but we
cannot treat the countryside as a model of the map.
Isomorphic models of total systems are particularly useful where
we can construct them, since it frequently happens that one kind of
system (an electric circuit, say) is much easier to construct and use
than another (a water supply system, say). But most relationships
472 MODELS FOR EXPLANATION IN GEOGRAPHY
Von Bertalanffy thus attempts to unify the body of facts, laws, and
theories, produced by science, under one integrating theory. One
problem with such a theory is that
Its validity w ould depend on the deducibility o f disciplinary theories from it and,
hence, in turn, on the validity of the disciplinary theories. It is thus twice removed
from the experimental and applied aspects of science.
law-like statements, and so on, but for the sake of exposition we will
Call these statements laws from now on). The question then arises as
to how we can discover and control the use of law statements. Laws,
it is generally agreed, should be reasonable with respect to experience
and consistent with respect to each other. Some also prefer them to be
powerful (very general) on the grounds that explanation will be a
much more efficient process if we possess statements of great general
ity. It is generally agreed that we approach this problem by con
structing and validating a theory. But theories are speculative con
structs and speculation is a metaphysical and philosophical enterprise
whether we like it or not. The geographer cannot proceed, therefore,
without some clear notion as to what it is he is supposed to speculate
about. Even if we choose to view a theory as an abstract syntactical^
structure (a calculus or a specially constructed language), we still
have to face the problem of interpreting that theory by way of a text
which, among other things, relates that theory to a particular domain
of events. The philosophical implication is that the geographer needs
to identify the particular domain (or set of domains) with which he is specifically
concerned. Methodological considerations lead me to conclude, how
ever, that it is easiest to identify such a domain (or set of domains)
when we possess a well-articulated and well-validated theory. Hence
the significance of the comment (p. 129) that the nature of a disci
pline can be discerned through an examination of the theories which
it develops. The point of view of geography is thus embodied in
geographic theory, and the subject-matter is identified through the
texts of these theories which relate them to given domains. I specu
lated a little regarding the nature of geographic theory and sug
gested that we possess indigenous theory regarding spatial form and
derivative theory concerned with temporal process and that general
theory in geography amounts to a theory examining the inter
actions between temporal process and spatial form. This suggestion
is undoubtedly controversial and will be disputed by many. But I am
prepared to put it forward as one basic tenet of geographical thought.
Since academic activity frequently involves division of labour, in
dividual geographers may specialise in various aspects of this general
theory and some may not be aware of the general structure to which
they are contributing. This view of general theory in geography helps
us to pin down the point of view of the geographer but it does not
help a great deal in establishing the domain of geographical thought.
I find it difficult to state with any certainty the domain of geography,
but a number of issues emerged in subsequent chapters which at least
provide certain clues as to how we might set about solving this prob
lem. At various points, for example, three interrelated problems in
484 MODELS FOR EXPLANATION IN GEOGRAPHY
H a r t S o m e 4 R . r i 9 5 8 f ‘T h e c o n c e p t o f g e o g r a p h y as a sc ie n c e o f sp a c e , fr o m K a n t
a n d H u m b o ld t to H e t t n e r ’, Ann. Ass. Am. Geogr. 48, 97 108.
H a r ts h o r n e , R ., 1959, Perspective on the nature of geography (C h ic a g o ).
H a rv e y , D . W ., 1 9 6 6 A , ‘T h e o re tic a l c o n c e p ts a n d th e analysis o f la n d u se
p a tte r n s ’, Ann. Ass. Am. Geogr. 56, 3 6 1 - 7 4 . _
H a r v e y D . W ., 1966B , ‘G e o g r a p h i c a l p r o c e s s e s a n d p o i n t p a t t e r n s : t e s t i n g m o d e l s
o f d iffu s io n b y q u a d r a t s a m p lin g ’, Trans. Inst. Br. Geogr. 40, 8 1 - 9 5 .
H arvey D. W ., 1967 A, ‘Models of the evolution of spatial patterns in hum an
geography’’, in Chorley, R . J . and H aggett, P. (eds.), (1 9 6 7 ).
H a rv e y , D . W ., 1967B , ‘B e h a v io u ra l p o stu la te s a n d th e c o n stru c tio n o f th e o ry
REFERENCES 495
in h u m a n g e o g r a p h y ’, Semin. Pap., Ser. A. N o . 6, D e p t , o f G e o g ., U n iv .
o f B risto l (B r isto l). (F o r th c o m in g in Geographica Polonica.)
H a r v e y , D . W ., 1 9 6 8 A , ‘S o m e m e th o d o lo g ic a l p r o b le m s in th e u se o f th e N e y m a n
T y p e A a n d n e g a tiv e b in o m ia l p r o b a b ility d istr ib u tio n s in th e a n a ly sis o f
s p a tia l se r ie s’, Trans. Inst. Br. Geogr. 4 3 , 8 5 - 9 5 .
H a r v e y , D . W ., 19 6 8 B , ‘P a tte r n , p ro cess, a n d th e sc a le p r o b le m in g e o g r a p h ic a l
r e se a r c h ’, Trans. Inst. Br. Geogr. 45, 7 1 - 7 8 .
H e m p e l, G . G ., 1 9 4 9 , ‘G e o m e tr y a n d e m p ir ic a l s c ie n c e ’, in F e ig l, H . a n d S ella rs,
W . (e d s .), Readings in philosophical analysis (N e w Y o r k ).
H e m p e l, C . G ., 1 9 5 1 , ‘G e n e r a l sy stem s th e o r y a n d th e u n ity o f s c ie n c e ’, Hum.
Biol. 23, 3 1 3 - 2 2 .
H e m p e l, G . G ., 1 9 5 9 , ‘T h e lo g ic o f fu n c tio n a l a n a ly sis’, in G ross, L . ( e d .),
Symposium on sociological theory (E v a n s to n ).
H e m p e l, C . G ., 1 9 6 5 , Aspects o f scientific explanation ( N e w Y o r k ).
H e n s h a ll, J . D . a n d K in g , L . J . , 1966, ‘S o m e str u c tu r a l c h a r a cte ristic s o f p e a sa n t
a g r ic u ltu r e in B a r b a d o s’, Econ. Geogr. 42, 7 4 - 8 4 .
H e s se , Μ . B ., 1 9 6 3 , Models and analogies in science (L o n d o n ).
H ilb e r t, D . , 1962, Foundations o f geometry (E n g lish e d itio n , L a S a lle ).
H ilb e r t, D . , a n d C o h n -V o s se n , S ., 1952, Geometry and the imagination (E n g lish
e d itio n , N e w Y o r k ).
H o lm e s , J . , 1 9 6 7 , ‘P r o b le m s in lo c a tio n s a m p lin g ’, Ann. Ass. Am. Geogr. 57,
7 5 7 -8 0 .
H o m a n s , C ., 1 9 5 0 , The human group (L o n d o n ).
H o w a r d , R . N . , 1 9 6 6 , ‘C la ssify in g a p o p u la tio n in to h o m o g e n e o u s g r o u p s’, in
L a w r e n c e , J . R . ( e d .) . Operational research in the social sciences (L o n d o n ).
H o w a r d , I . P . a n d T e m p le t o n , W . B ., 1966, Human spatial orientation (N e w Y o r k ).
H u g h e s , H . S ., 1 9 5 9 , Consciousness and society (L o n d o n ).
H u x le y , J . , 1 9 6 3 , ‘T h e fu tu re o f m a n : e v o lu tio n a r y a sp e c ts’, in W o lste r h o lm e , G .
( e d .) , Man and his future (L o n d o n ).
I m b r ie , J . , 1 9 6 3 , ‘F a c to r a n d v e c to r a n a ly sis in p r o g r a m s for a n a ly z in g g e o lo g ic
d a t a ’, Tech. Rep. 6, ONR Project 3 8 9 - 1 3 5 , D e p t . G e o l., C o lu m b ia U n iv ., N e w
Y ork.
Isa rd , W ., 1 9 5 6 , Location and the space economy (N e w Y o r k ).
Isa rd , W ., i 9 6 0 , Methods o f regional analysis (N e w Y o r k ).
Is a r d , W . a n d D a c e y , M . F ., 19 6 2 , O n th e p r o je c tio n o f in d iv id u a l b e h a v io r in
r e g io n a l a n a ly sis, P a rts I a n d I I ’, J . reg. Sci. 4, N o . 1, 1 - 3 4 , N o . 2, 5 1 - 8 3 .
Iw a n ic k a -L y r a , E ., ‘T h e d e lim ita tio n o f th e W a r sa w a g g lo m e r a tio n ’, Pol. Acad.
Sci. Stud. 17, 1 6 3 - 7 2 .
J a m e s , P . E . a n d J o n e s , C . F ., 1954, American geography: inventory and prospect
(S y r a c u se ).
J a m m e r , M ., 1 9 5 4 , Concepts o f space (C a m b r id g e , M a s s .).
J a r v ie , J . C ., ‘L im its to fu n c tio n a lism a n d a lte r n a tiv e s to it in a n th r o p o lo g y ’, in
M a r tin d a le , D . ( e d .) , Functionalism in the social sciences (P h ila d e lp h ia ).
J e ffr e y , R . C ., 1 9 6 5 , The logic o f decision ( N e w Y o r k ).
J e n k s , G . F ., 1 9 6 3 , ‘G e n e r a liz a tio n in s ta tistic a l m a p p in g ’, Ann. Ass. Am. Geogr.
53» 15-26.
J o h n s to n , J . , 1 9 6 3 , Econometric methods (N e w Y o r k ).
J o n e s , E ., 1 9 5 6 , ‘C a u se a n d e ffe c t in h u m a n g e o g r a p h y ’, Ann. Ass. Am. Geogr. 46,
369 - 77 ·
J o y n t , G . B . a n d R e s c h e r , N . , 1961, ‘T h e p r o b le m o f u n iq u e n e ss in h isto r y ’,
History and Theory 1, 1 5 0 -6 2 .
496
REFERENCES
Carnap, R ., 37-9, 45, 66, 95, 108, 179- Devons, E., 123
181, 184-5, 215-17» 227, 239, 242, Dickinson, R . E ., 353
247-8, 373 D ilthey, W ., 47, 50, 64
Casetti, E ., 128, 346, 381 D odd, S. C., 108, 112, 121
Cassirer, E ., 19-21, 192-3 Donagan, A ., 49
Cattell, R . B., 315, 341, 343, 347 Downs, R . M ., 315, 322
Cavalli-Sforza, L. L ., 345 Dray, W . H ., 48-50, 53-4, 56, 61
Caws, P., 19-21, 23, 30, 43, 59, 299, Duncan, O. D ., 278, 352-3
301 Dury, G. H ., 93
Chamberlin, T. C., 295 D utton, J. A ., 381
Chapman, J. D ., 65
Childe, V . G., 195
Chisholm, M . D . I., 353, 476, 478-9 East, W . G., 68, 108-9, 369, 4° 9, 4l6
Chorley, R . J ., 18, 66, 81 -3 , 87, 109, Edwards, A. W. F., 345
121-2, 128, 141, 155-7» *59» i6 2 - Einstein, A ., 87, 183, 188, 195-7» 202,
164, 168, 175, 211, 214, 292, 350, 2 2 4 ,2 2 6
379-80, 427, 460, 468, 479-80 Eisenstadt, S. N ., 194
Choynowski, M ., 273-5, 376 Ekman, G., 374-5
Christaller, W ., 105, 118-19, 222 Eliot, T. S., 293, 411-12
Christian, R . R ., 327 Ellis, B., 316, 318
Churchman, C. W ., 9, 34~ 5» 39_ 4°j Empson, W ., 375
57-8, 62, i8 g , 232, 237-8, 251, Euler, L., 2x9
2 5 8 ,3 2 4 Evans, F. C., 382
Clark, A. H ., 408 Eyre, S. R ., 115, 467
Clark, K . G. T ., 134, 402
Clark, P. J ., 382
Clarkson, G. P. E ., 94-5, 138 Feller, W., 233-4, 237, 243» 246, 266
Cliff, A ., 382 Ferm at, P., 233
Cline, M . G ., 348 Fishburn, P. C., 57, 95, 231-2, 238, 240,
Cochran, W. G ., 237, 357 286, 310, 317, 320
Cohen, R ., 90 Fisher, R. A ., 235, 241, 248, 250, 252-4,
Cohn-Vossen, S., 217-18 256-8, 280, 365-6
Coleman, J. S., 186-7, 267-8 Forde, C. D ., 442
Coleridge, S. T ., 295 Forrester, J. W ., 457» 4^°» 4^9
Collingwood, R . G ., 51, 56 Fourier, J ., 381, 385
Comte, A ., 46-7 Freeman, D ., 95, 112
Cooley, W. W ., 343, 346 Freeman, T. W ., 234
Coombs, C. H ., 300, 320, 324 Freud, S., 47
Coxeter, H . S. M ., 198
Cramer, H ., 233, 236-7
Curry, L ., 128-9, 139, 160, 261-2, 266, Galton, F., 234
268-71, 286, 318, 382-3, 427-8, Galtung, J., 281-4, 312, 358-9
468 Garner, W. R ., 375
Garrison, W. L ., 125, 132-3» 219,
262, 266, 275, 277, 286, 337
D acey, M . F., 80, n g , 124, 128, 134, Gauss, K ., 201, 208-9, 219, 234
138-9, 166-7, 216-18, 225-6, 229, George, F. H ., 162
2 6 6 ,2 6 8 -9 ,2 7 7 , 283-5, 287, 370-4, Getis, A ., 2 11, 382-3
377-8, 381-3 Ghiselli, E. E., 320
Darby, H . C., 80, 295-7, 4°8» 4 J3, 421» Gilbert, G. K ., 291
4.25 Glacken, C. J., 293
Darwin, C., 44, 416-17, 423, 427 Gluckman, M ., 123
D avid, F. N ., 233 Golledge, R ., 120
D avis, W . M ., 85, 93, 109, 115, 122, Goodall, D . W ., 343, 349
163-4, 408, 416, 423, 427, 429 Goodfield, J ., 414-15» 43*
Descartes, R ., 206, 414 Goodm an, L., 279
Deutsch, K. W ., 469 Gould, P., 120, 194, 229, 322
AUTHOR INDEX 507
Gouldner, A. W ., 447 Hom ans, C., 99, 135
Gower, J. C., 315 Horton, R. E ., 109
Graybill, F. A ., 277-8, 357, 361, 367, Howard, I. P., 193
380 Howard, R. N ., 345
Gregg, J. R ., 90 H oyt, H ., 360
Greig-Smith, R , 334, 342, 357, 363, Hughes, H . S., 47
382-3 Hum boldt, A. von, 80, 207, 210, 402
Grigg, D. B., 72-3, 125, 162, 216-17, H um e, D ., 37
3275 333- 6, 349 Huntington, E., 71, 402, 416, 425, 427
Griinbaum, A ., 196, 209, 417 Hutton, J., 87, 108-9, 415, 426
Gurvitch, G ., 412 H uxley, J ., 123
Accuracy, 234, 306, 322, 357, 361-2 Autocorrelation problem, 188, 363-4,
Aggregation problem (see also Scale 368, 384-6
problem ), 73, 278-9, 353, 452-5 Axiom atic formulations (see also Cal
Agriculture, 130-4, 146, 225, 277, 352, culus; T heory), 28, 88-91, 96-9,
397~ 9, 4*3, 458, 46 2-3 151, 174, 184, 197-206, 217, 231,
Algebra, 187, 198 232, 291, 294
Allom etric law, 465 -6 in geography, 90, 116-17, 132-4, 174
Am biguity (see also U ncertainty), 19- Axioms, 88-92, 9 6 -9 , 146, 180, 182,
22, 28 1 8 4 ,1 9 3 ,1 9 7 -2 0 5 ,2 0 6 ,2 1 7 , 230-1,
in geographical concepts, 19-22, 28, 243 - 6 , 390
90-2, 98, 117, 134, i3 5 -4 0j l8 6 “ of geometry, 88, 184, 193, 197-206,
190, 226, 242, 258, 259, 277, 217
295- 6» 3° 2, 3° 4- 5» 3° 8» 373» 378» of probability theory, 231, 243-6
401, 437, 444, 447, 449, 452, 467,
475» 476» 479
in maps, 375-6, 377 Bayes’ theorem, 241, 249, 250, 253, 257
necessary, 304-5, 375 Behaviour
Analogy (see also M odels-analogue; adaptive, 436, 442, 459-61, 465
Organismic analogy), 14-15, 48, in space, 193-4, 209-12, 225, 227,
147, 162, 164, 169-72, 424, 474, 228
477» 478 purposive, 57, 438 _
negative, 150, 164 and scientific decision making, 9-23,
neutral, 150 3 b 39 - 43 » 55 -9 , 62, 174, 240-2,
positive, 150, 164 2 4 8-9, 273
Anthropology, 27, 68, 81, 95, h i , 121, study of, 4 6 -8 , 55-61, 9 4 -6 , 99, 108,
184, 194, 412, 433, 434, 44°» 44 b n o , 112, 119-21, 128, 132-4,
442, 443 139, 146, 193, 194, 268, 281, 316,
Archaeology, 121, 423 320, 322, 393, 429
Areal differentiation, 3 -4 , 70-5, 81, Binomial distribution, 246
1 1 4 ,^ 15, 125, 127 Biogeography, 65
Areal units, 351-4» 38 4, 385 Biology, 28, 4 7 -8 , 65, 68, 69, 90, 102,
artificial, 351 108, 121, 164, 185, 204-5, 331,
collective, 351-3 415-17, 423, 427, 429, 433, 434»
natural, 351 437, 439, 44° , 444» 449, 4 5 °, 452,
singular, 351-3 462, 464, 466, 467, 473, 477, 479
A s-if thinking, 9 0 ,1 0 2 -3 , I08» 109, II0 » Black box, 157, 472, 473, 479
112, 174, 264, 268, 393, 456 Botany, 108, 343
Assumptions, 34, 39, 71-4, 79~ 8b i i 2 -
113, I2 3» 127-8, 132-3, 139» J47,
165-7, 175- 6» i8 4- 9°» 208, 236, Calculus (see also Language; M athe
241-2, 250, 2 54-6, 276, 280-2, matics; T heory), 5, 7, 57, 88-9,
318, 338, 347» 378, 381, 40 3- 6, 103, 130, 139, 152-4, 165, 170,
423, 425- 9, 482 1 7 5 -6 , 179-90, 199, 200, 224, 228,
Astronomy, 124, 418 231, 243-7, 263, 266, 276, 470,
Atmospheric circulation, 137 485
Attributes (see also Properties), 306-15, designation rules in, 180, 373
316, 328, 332-48, 350- 1, 354-6, formation rules in, 8 8 -9 , 180, 184,
44b 455 373
SUBJECT INDEX
Calculus— cont. Concepts {see also Ideal types, Idealisa
interpretation of (see also Correspon tions, Geographical concepts), 18-
dence rules; T ext), 152, 179-90, 22, 46, 54, 68, 91-6, 117-27, 135,
199-200, 231, 263, 265, 485 147 -5 4 , 186-90, 191, 192-7, 199,
truth condition in, 180, 373 203, 206-29, 238, 245, 301-3, 3 i6 ~
Causal chain, 80, 81, 391, 400, 401, 17, 320, 370, 378, 393, 408, 4 1 0 -
402, 406, 407, 450 418, 428, 440, 447-63, 481, 48 5 -6
Causal ordering, 394-400, 406 m athematical, 179-90
Causal principle, 389, 390, 392, 401, ofspace, 192-7, 199, 203, 206-29
402 o f time, 410-18
Causal systems, 396-400 theoretical, 15, 91-6, 117, 151, 245,
reciprocal, 399-400, 406 302- 3, 316-17, 320
recursive, 397-9, 406 Confirmation {see Verification)
Causality, 134, 389, 402, 409, 439, 441, Contiguity, 216, 218, 382, 384
449 Continental drift, 137
Causation, 19, 56, 389 Co-ordinate System {see also Geometry,
circular and cumulative, 400, 461 Space-time languages), 73-5, 83,
Cause and effect, 80, 81, 159, 164-5, 9 6 ,1 9 1 , 198, 201-6, 212, 213, 215,
188, 227, 261-2, 331, 348, 389- 216, 219, 223, 224, 226, 353, 355,
406, 419, 422, 429, 433, 437-9> 368, 378, 410
441, 448, 459, 46 7, 482 Correlation and regression, 16, 159,
and determinism, 80, 389, 400-6, 439 167, 188, 234, 235, 277-9, 319,
Censuses, 215, 355, 377 340-2, 343, 347, 353, 366, 397-400
Central lim it theorem, 246-7, 269-70 ecological, 278-9
Central place theory {see also Settle Correspondence rules {see also T ext),
m ents), 80, 104-5, n o , 118-19, 89-96, 180, 302-3, 317, 390, 391
123, 124, 126, 128-9, 138-9, 163, Covering laws, 36, 48-54, 6 9 , 75-9, 109,
211, 222, 225-6, 268, 277, 283, 130, 132-4, 447, 482
28475, 382, 433, 434- 6, 446, 468 and state determined systems, 447
empirical status of, 119, 138-9 Cultural geography, 66 ,1 1 1 -1 3 ,1 1 5 ,4 0 8
geometry of, 80, 225-6 Cultural relativism, 54 -9 , m - 1 3
Chance (see also Probability; U ncer Cultural stages, 157, 423
tainty) Culture, 18-23, 54~ 9, n o - 1 3 , 135, 194,
equal, 237, 238, 245, 384 227, 228, 302, 375, 4 12-14, 433
independent, 237, 238, 245, 384 Cybernetics, 450, 455, 462, 473, 479
mutations, 417, 427 Cycles
set-ups, 238-9, 247, 265, 368 o f erosion, 109, 163-4, 4 2 3 -4
Chemistry, 28, 48, 65, 102, 103, 108, in history, 51, 416, 423
1 0 9 ,1 2 1 ,4 7 7
Classification {see also Grouping pro
cedures; Logical division), 33-5, D 2 statistic, 342
5 1, 54, 56, 79, 9 1, !25, 126, 175, D ata {see also Areal units; Information),
234, 273, 297-300, 309, 314, 326- 15, 18—21, 3 1 -3 , 35, 79, IOO> n o ,
349, 362, 373, 393, 4° 9,483 136-41, 142, 154, 184, 188, 2 1 4 -
as form of explanation, 33-5, 79, 300, 215, 217, 218, 2 5 7 ,2 6 0 ,2 6 5 -7 ,2 6 9 -
326, 348 270, 273, 278, 282, 286, 326, 3 3 9 -
general (‘natural’), 126, 331-2 343, 347, 350-86, 422
genetic, 348, 409 banks, 377
monothetic, 334, 335 collection {see also Sam pling), 2 6 9 -
polythetic, 334, 338 270, 273, 286, 350-76
purpose of, 326-7, 330-8, 348 m atrix, 214-15, 339“43, 347, 3 5 °-6 ,
quantitative techniques of, 338-49 379
specific (‘artificial’), 332 sense perception, 18-21, 31, 32, 35,
Climatology, 65, 109, 115 100, 136, 184, 249, 326
Communications theory, 450, 473 Decision making
Competition, 89, 92, 226, 460, 465 in science {see also Statistical infer
Computers, 76, 214, 305, 377 ence), 15, 3 9 -4 °, 62, 189-90
S U B JE C T I N D E X 513
study of, 119, 132,270-4 ,3 8 3 ,4 1 2 -1 3 , Economics, 44, 65, 68, 69, 89, 9 4 -5 , 99,
429 102, n o , 118-19, 128, 132, 137-9,
theory of, 39, 59-60, 232, 2 4 0-3, 258, 142, 148, 149, 153, 157, 158, 187,
270-4, 473 3° 7, 394, 4Ι 5» 441, 449, 456, 463,
Deduction, 32, 36-43, 76, 88-91, 9 6 -9 , 469, 485
180, 186, 202-3, 232, 245, 259, Ecosystem, 112, 115, 164, 467-8
265, 390, 400, 475 Empathy, 2 8 -9 , 56-9, 71, 87
in explanation (see also Covering Entropy, 427, 463-4, 468
laws), 13-14, 3 6 -7 , 75, 482 Environment, 12, i n , 115, 120, 271,
in theory construction {see also 293, 305, 322, 393, 4°2, 4° 9, 429,
Geometry; Probability theory; 430, 464, 468
Theory), 36-7, 88-91, 9 6 -9 , of a system, 4 4 7 ,4 4 8 ,4 5 1 , 4 5 6 , 457-8,
145, 148 460, 461, 466
Definition {see also M eaning), 3, 33, 35, Environmentalism, 77, 120, 130-2, 261,
9 2 -6 , 100, 131-3, 135, 180, 184, 402, 425, 431
297, 298-300, 301-6, 326, 333, Equilibrium, 89, 138, 148, 149, 428,
354? 393, 395, 452 442, 458, 460
intuitive, 92-6, 303 Ergodic hypothesis, 128-9, 269-70
lexical, 301 Errors {see also Accuracy; Precision)
operational, n o , 116, 135, 236-8, in measurement, 279, 282, 283, 284,
245, 3°3“4, 458 319-22, 324, 375- 4, 397
purpose of, 301, 305-6 in observation, 234, 321
theoretical, 92-6, 303 sampling, 280, 356
Degrees o f belief {see also Probability— term, 139, 188, 262, 284-5, 321, 399,
subjective), 240-1, 250, 259, 271-2 400, 448
Dem and theory, 9 4 -5 , 118, 127, 138-9, theory of, 139, 267
226 T ype I and T ype II, 252-6
Determinism, 7, 46, 232, 262, 277, 415, Event, 243-5, 264-5
425, 43 1, 439, 441, 442 elementary, 244
in geography, 7, 47, 69, 71, 76, 77, compound, 244
8o-x, 115, 130-2, 164-5, 262, m utually exclusive, 244, 329
400-4 Evolution, 44, 48, 80, 164, 415-21, 424,
Diffusion, 96, 121, 127, 135, 227, 266, 426
416 Exceptionalism, 70-8
Dimensional analysis, 318-19 Expectations, 10-13, 138, 249, 257,
Dirichlet region, 225-6 382 , 394 .
Discovery in science {see also Theory Experience {see also Perception), 9-13,
form ation), 31, 38, 42, 292-3 18-23, 142, 182, 184, 192-3, 199,
Discriminant analysis, 346-7 207, 227, 257, 2 9 5 -6 , 401, 481, 483
Distance {see also Space) perceptual, 4 -5 , 18-23, 3 2 -6 , 292,
measurement of, 195, 196, 197, 2 0 1 - 295- 6, 3° ! , 410
203, 206-7, 210-12, 213, 221, Experiment, 46, 76, 139, 264, 365, 420,
223-4, 234, 3I 4-I5 472 n
and grouping procedures, 340-8 Experimental designs, 35, 78, 79, 136,
taxonom ic, 339 139, 146, 153, 257-8, 320, 365,
D om ain (of laws and theories), 3, 22, 396 , 4*-*4
89-96, 100, 104, n o , i n , 116, Explanation {see also Cause and effect;
123, 136, 146, 149, 152-4, 160, Covering laws; Functional Ex
161, 170, 249, 257, 277, 333, 372, planation; Genetic explanation;
393, 429, 473, 475, 477, 483, 485 Systems explanation; Temporal
modes o f explanation), 9-23, 64-7,
169-76, 4 8 1 -6
Ecology, 81, 408, 409, 441, 467, 468 in history, 44-61, 75, 430
Economic concepts in geography, 118- in physical science, 30-43
' 119, r32, 137- 9, χ48, 149 in social science, 44- 6i , 75
Economic geography, 65, 115, 118-19 mechanistic, 44, 4 6 -8 , 80, 441
Economic man, 92, 94, 120, 304 routes to 32-6, 151-3, 175-6
514 SUBJECT INDEX
Information, 20, 22, 30, 32, 92, 121, Laws {see also Generalisations; Geo
171, 183, 219, 226, 265, 271, 297, graphical laws; Hypotheses), 14,
320, 326, 327, 332, 334, 339, 344) 28, 31-61, 80-1, 91, 100-6, 107-
354> 370, 373) 375) 377) 3^5) 396, 113, 130-4, 136-7, 142, 145, 169-
429, 465, 468, 469, 471, 473 176, 247, 260, 389-90, 420, 422,
filters, 22, 298-300, 305, 326-7, 348 425-6, 447, 477, 482, 483
systems (see also Language; M aps), and facts, 3 1-6
214-17, 219, 377 and generalisations, 31-43
theory, 385, 450, 462-4, 469 and hypotheses, 31-43
Initial conditions (in explanation), 36, and theories, 31-43, 91, 101, 104-6,
41-2, 49, 57, 482 142
Input-output, 148-9, 456-7, 459, 461, as universal statements, 33, 50, 53,
472 101-4, 109, 112, 247, 389
Intensions (explanations by), 56—7, 67 empirical, 31, 33, 53, 91, 101-4
Introspection, 28, 9 4 -6 , 137 theoretical, 31, 91, 101-5
Intuition {see also Im agination), 31, 35, Likelihood, 250-1
87-8, 173-4, J92, 200, 202, 207, Linear models, 188, 399, 456
39b 393) 446) 4Sl Linear Programming, 148, 465
Inverse distance law {see also Gravity Location {see also Coordinate systems;
m odel), 110-11, 146 Space)
Isomorphism (similarity), 19, 145-54, in space and time, 70, 72, 90-1, 96,
162, 184-7, 263-4, 371, 420, 428, 103, 191, 283, 293
470, 471-3, 474-8 theory {see also Central place theory),
66, H 5 -I7 ) 118,129, 132- 4) 19b
209-10, 211, 225-6, 276-7, 304,
Journey-to-work, n o , 264, 461 3 2 2 ,4 6 4
Logic of justification, 6, 45
Kantian thesis, 70-9 Logical division, 330, 334-7, 338, 352