POTENTIAL MODELS IN HUMAN GEOGRAPHY

D .C. Rich

ISBN 0 86094 044 6

ISSN 0306 6142

1980 0.0. Rich

CATMOG
(Concepts and Techniques in Modern Geography)

CATMOG has been created to fill a teaching need in the field of quantitative CONCEPTS AND TECHNIQUES IN MODERN GEOGRAPHY No. 26
methods in undergraduate geography courses. These texts are admirable guides
for the teachers, yet cheap enough for student purchase as the basis of
classwork. Each book is written by an author currently working with the POTENTIAL MODELS IN HUMAN GEOGRAPHY
technique or concept he describes.
by
1. An introduction to Markov chain analysis - L. Collins D. C. Rich
2. Distance decay in spatial interactions - P.J. Taylor (Macquarie University, Australia)
3. Understanding canonical correlation analysis - D. Clark
4. Some theoretical and applied aspects of spatial interaction
shopping models - S. Openshaw CONTENTS
5. An introduction to trend surface analysis - D. Unwin Page
6. Classification in geography - R.J. Johnston I INTRODUCTION 3
7. An introduction to factor analytical techniques - J.B. Goddard & A.Kirby
II SOCIAL PHYSICS
8. Principal components analysis - S. Daultrey
(i) Background 4
9. Causal inferences from dichotomous variables - N. Davidson
(ii) Gravity 6
10. Introduction to the use of logit models in geography - N. Wrigley
(iii) Potential 7
11. Linear programming: elementary geographical applications of the
transportation problem - A. Hay III LINES OF DEVELOPMENT
12. An introduction to quadrat analysis - R.W. Thomas (i) Population and income potentials 8
13. An introduction to time-geography - N.J. Thrift (ii) Market potential and its derivatives 11
14. An introduction to graph theoretical methods in geography - K.J. Tinkler
IV OPERATIONALISATION
15. Linear regression in geography - R. Ferguson
(i) A simple example 13
16. Probability surface mapping. An introduction with examples and
Fortran programs - N. Wrigley (ii) The universe 16
17. Sampling methods for geographical research - C. Dixon & B. Leach (iii) The zones 17
18. Questionnaires and interviews in geographical research - (iv) Mass 17
C. Dixon & B. Leach (v) Distance 18
19. Analysis of frequency distributions - V. Gardiner & G. Gardiner
V CASE STUDY 1 - NEWS CIRCULATION IN THE UNITED STATES 22
20. Analysis of covariance and comparison of regression lines - J. Silk
21. An introduction to the use of simultaneous-equation regression VI CASE STUDY 2 - MANUFACTURING INDUSTRY IN SCOTLAND 27
analysis in geography - D. Todd
VII PROBLEMS AND PROSPECTS
22. Transfer function modelling: relationship between time series
variables - Pong-wai Lai (i) Outstanding problems 32
23. Stochastic processes in one-dimensional series: an introduction - (ii) Some recent advances 33
K.S. Richards
BIBLIOGRAPHY 35
24. Linear programming: the Simplex method with geographical applications -
J.E. Killen
continued on inside back cover
Acknowledgement
I INTRODUCTION
I would like to thank Peter Matwijiw and two referees for their
comments on a previous draft of this monograph. The potential model has appeared frequently in the human geography
literature as an index of the intensity of possible interaction between
Figure 1 is reprinted in adapted form from Geographical Review, Volume social or economic groups at different locations. This potential interaction
37, 1947, with the permission of the American Geographical Society. may be of goods, telephone calls and other forms of telecommunications,
migration, and a whole range of other social and economic contacts. If we
Figures 5,6 and 7 are reprinted in adapted form from A.R.Pred's examine the situation within a city, we could add to this list many other
City systems in advanced economies, with the permission of the author types of interaction, such as journeys to work, shop, school or entertain-
and the publishers, Hutchinson of London and John Wiley of New York. ment. This concern with 'potential for interaction' is quite distinct from
and should not be confused with 'potential for the future' in the sense of
capacity for change.

The basic notion is, then, one of flows - of goods and services, of
ideas and information, and of people. With this emphasis, the potential
model has close conceptual, empirical and historical associations with the
gravity model. Whereas the gravity model is concerned with analysing or pre-
dicting an observed pattern of spatial flows, the potential model is more
concerned with the opportunity for interaction between groups, created by
their sizes and locations, than with the interaction itself. The links be-
tween the two models are explored in more detail in section II.

All applications of the potential model share this same background.

However, the model is used in a variety of guises and for a number of pur-
poses; consequently, it is probably more appropriate to refer to 'potential
models' than to a single model. Ignoring, for the moment, the details of the
construction and application, the versions of the basic model vary in three
main ways: in their interpretation; in the phenomena dealt with; and in the
reasons for using them.

a) Interpretation: potential models have been interpreted variously as

giving a measure of the 'influence' of one place or group on another, as a
generalised measure of concentration or density (for example, of population),
as an index of the nearness of groups to a point and, most commonly, as an in-
dicator of the geographical position, or accessibility, of groups in different
places relative to each other.
b) Phenomena: potential models have been used in connection with a variety
of social, economic and demographic phenomena. Consequently, the term
'potential' has been qualified by a wide range of terms: ethnic, income,
supply, demand, market, economic, agglomeration, polarisation and particip-
ation potentials have all appeared in the literature as variants of the basic
population potential model.
c) Function: potential models have been used in three main roles. First,
they have been used to fulfil a descriptive or illustrative function: maps
of potential values have been used to illustrate the broad spatial distrib-
utions of such phenomena as total population, scientists, and the availability
of information. Second, they have been used to provide explanatory variables
in statistical analyses of the spatial distribution of phenomena as diverse
as industrial location, per capita incomes and the price of onions. Finally,
potential models appeared in one part of the debate in the 1950s and 1960s
on the most appropriate methodology for geography, and were used to support

3
calls for a more 'scientific' approach to the subject. For example, Stewart
and Warntz (1958A) saw the status and progress of geographical research as (1)
being reduced by its concentration until that time on 'microscopic' studies
of individual areas; they proposed an alternative 'macroscopic' methodology
involving not the collection of more case studies nor the examination of where
larger areas, but rather a heightened level of abstraction and generalisation
as a means of securing greater understanding of the processes operating to
determine spatial relationships. They saw distance, time and numbers of
people as being basic elements or dimensions of socio-economic systems, just
as the physical sciences had long recognised a number of fundamental dimen-
sions. They argued that a thorough appreciation of these basic factors is
Both gravity and potential models are based on two broad notions related
essential for an understanding of the workings of human society, and they put
to the D and M terms in this equation. The first is that the probable fre-
a lot of effort into trying to derive theoretical or empirical generalisations
quency of interaction between people in different cities is inversely pro-
about the spatial structure of societies using these dimensions. They further
portional to the difficulty of movement between them. This difficulty is the
argued that the progress of the macroscopic approach had hitherto been im-
result of a 'frictional' effect and is directly proportional to the interven-
peded by the absence of a sufficiently abstract and subtle measure of geo-
ing geographical distance between the two cities. The frictional effect is
graphical position; the potential model was proposed as such a measure and
the product of many factors, such as transport costs (whether for people or
consequently as a cornerstone of a new 'macrogeography'. It is rarely
goods), travel time, the availability of information in one place about the
accorded such a central position in geographical research today, but it is
contact opportunities existing in the other, and the probability of making
still frequently used in its descriptive and explanatory roles. This mono-
chance contacts. The frictional effect serves to inhibit contacts and means
graph concentrates on the latter two functions, and an example of each type
that spatial interaction commonly exhibits a distance decay pattern, with a
of application is discussed in sections V and VI.
high volume of contacts over short distances but progressively fewer over
longer distances.
The text is as far as possible self-contained. However, the discussion
assumes a basic knowledge of simple statistical methods, particularly cor-
The second notion is that any person in a city may be considered to
relation and regression analysis. A background knowledge of the gravity model
generate the same interaction as any other. Consequently, the volume of
is also useful but not essential; an excellent introduction is provided by
Taylor (1975) in this CATMOG series.

ically in their propensity to interact or in other ways; rather, the gravity

and potential models are concerned with an average pattern from which the
II SOCIAL PHYSICS deviations of individuals are assumed to cancel each other out. Thus, both
gravity and potential models consider aggregate patterns of interaction
(1) Background between groups of people, with the level of interaction being directly re-
lated to the sizes of the groups.
It is important to understand the nature of the simple gravity model in
order to come to grips with the potential model, because their early develop-
Both models rely on the argument that aggregate human behaviour is sub-
ment in the social sciences is closely intertwined. Both came to be used
ject to certain laws and that these laws are the same as those which physic-
because of a belief, held by some, in the value of quantitative represent-
ists found to govern the aggregate behaviour of molecules. Human beings are
ations of aggregate relationships in the social sciences in general and in
not like molecules, but the behaviour of human and physical aggregates are
geography in particular. This belief led to a quest for uniformities in
subject to similar laws of gravitation. There is, then, an analogy between
social behaviour which can be expressed in mathematical forms more or less
physical and social phenomena, so that the study of human interaction has
corresponding to the known patterns of physical science' (Stewart, 1952,
often been called 'social physics' - the application of physical laws to
p 110); both gravity and potential models came to the social sciences from
aggregate human behaviour.
physics.
The term 'social physics' was used as long ago as 1836 by Adolphe
The gravity model suggests that two separate groups of people, say in
Quetelet, a social statistician, and the search for analogies between physical
two cities, generate a mutual interaction in proportion to the product of the
and social phenomena goes back even further. At first, the term meant little
sizes of the cities, and that this interaction is impeded by the frictional
more than adopting in the study of social phenomena the quantitative approach
effect of the intervening distance over which it must take place. In other
that was already well established in the physical sciences. Probably the
words, the volume of interaction between the two cities is a positive function
first attempt to view human interaction in gravity terms was made by Carey
of their population sizes and an inverse function of the distance between
(1858-9), who argued as follows:
them. This relationship can be expressed algebraically as:

4 5
 Man, the molecule of society, is the subject area. However, the equations for demographic force and energy differ only
of Social Science .... The great law of in the exponent to which distance is raised; later social scientists have
Molecular Gravitation is the indispensable not felt constrained by the exponent of 1 suggested by the physical analogy.
condition of the existence of the being Equations (3) and (5) have been combined and generalised into a single equa-
known as man .... the greater the number tion representing the interaction, I, between i and j:
collected in a given space, the greater is
the attractive force that is there exerted... (6)
Gravitation is here, as everywhere, in the
direct ratio of the mass and the inverse
one of distance. where b is an exponent to be derived empirically. This equation is the basic
gravity model used in the social sciences, which is the focus of Taylor's
(Carrothers, 1956). This is fundamentally the same formulation as that monograph in this series (Taylor, 1975) and is therefore not further dealt
given in equation (1). with here.

(ii) Gravity (iii) Potential

The gravity concept made only sporadic reappearances in the social Lagrange added the notion of gravitational potential to Newton's concepts
science literature during the 80 years after Carey wrote. Our modern interest of force and energy. Whereas equation (4) represents the mutual energy
in the gravity and potential models stems from the early 1940s when J.Q. created between masses i and j, gravitational potential represents the energy
Stewart, an astronomer at Princeton University, generalised previous work on created at one mass by another. For example, the energy created by j at i
human interaction and formalised it in terms of Newton's planetary gravita-
tional theory. His interest was raised by an observation that students at
Princeton came predominantly from the local region, with progressively fewer
coming from successively further away (Stewart, 1941, 1942).

This observation led him (Stewart, 1948) to explore the analogy with
Newton's theory. Newton's original statement suggested that there is a
gravitational force, F, between two particles of mass, mi and m 1 , at a dis-

Population potential may be viewed in a number of ways. It is an index

of the nearness of people to one another or a measure of the influence of
people at a distance. Slightly more concretely, it indicates the intensity
of possible contact between people at i and those at all other locations.

6 7
But most commonly, it is regarded as an indicator of relative position, or
accessibility. Thus, equation (10) gives a measure of the accessibility of
people in i to the people in all parts of the area being examined.

A single potential value is difficult to interpret, and it is much more

useful to calculate the potentials of all the zones (subdivisions of the
study area) under investigation and to compare the scores of each zone.
(Potentials are often scaled by expressing them as a percentage of the highest
score of any zone in the group, to aid this comparison.) By computing
potentials for a number of zones, it is possible to construct a map showing
lines of equipotential, such as Figure 1. The resulting potential surface
gives quite a good visual impression of geographical variations in relative
accessibility. Places with relatively high potentials are those with many
opportunities for interaction with other places because they are highly
accessible to the major concentrations of population; on the other hand,
places with low scores have relatively poor opportunities for contact.

Potential is only one of a number of indices of accessibility (Neft,

1966). The various measures differ mainly in their weighting of the mass
term. In the case of potential, the mass at j is weighted by the inverse of
the distance between i and j (equation 10), so that distant masses have re-
latively little impact on the aggregate potential of i. It is also possible
to weight mass by distance directly (in which case accessibility scores are
dominated by distant masses) or to adopt one of a number of other procedures.
The various measures generally indicate somewhat different patterns of
accessibility, and even the point identified as being the most accessible is
likely to differ. Nevertheless, of the relatively simple measures that are
available, potential is the most useful in a number of respects (Neft, 1966;
Rich, 1978).

III LINES OF DEVELOPMENT

The basic themes and concepts of social physics have been refined and
applied in a great variety of ways. Much of the work in this sphere, and
certainly most recent geographical work on potential, falls into one of two
broad 'schools'. The first is really a linear extension and amplification
of Stewart's work while the second builds on the work of Harris (1954) and
uses potential to examine geographical patterns of industrial location and
regional development. Rather than trying to describe all work done on poten-
tial models, this section concentrates on these two major lines of research.
Further examples of recent research are described in sections V and VI.

(i) Population and income potentials

Population potential maps have been constructed for virtually every area
in the world'. Probably the earliest was Stewart's map of population poten-
tials in the United States in 1940 (Stewart, 1947), which is reproduced in
Figure 1. The major peak on the surface is at New York City, indicating that
this was the place most accessible to the American population in aggregate.
In rural areas east of the Sierras, the potential surface slopes downwards
from New York, reflecting generally declining accessibility. Superimposed
on this trend are local peaks at each major city, but these are too small to
show up on this relatively coarse-grained map. Rural areas near New York City

8 9
have higher potentials than urban areas in the Mid West because of the low One interesting extension of this work involved attempts by Warntz
base potentials on which such cities stand. Consequently, as Stewart (1957, 1959) and by Tegsjo and Oberg (1966) to use potential to explain the
(1948, p 39) points out: you are closer to people of the whole United States formation of prices of agricultural commodities, such as wheat, potatoes,
on a farm in Hunterdon County, New Jersey, than you are at the center of onions, strawberries and eggs, in the U.S.A. and Sweden. The aim was to
Omaha, Nebraska'. extend the simple notion that price is a function of supply and demand, by
suggesting that geographical variations in prices are direct functions of
Another early concern was with the relationship between population spatial variations in demand and inverse functions of spatial and temporal
potential and population density. If for the moment we regard people as being variations in supply. Accessibility of any point to national demand was re-
continuously distributed over an area instead of being gathered into discrete presented by income potential, while geographical variations in supply were
zones. potential is given by: measured by another variant of potential where the volume of local production
of the commodity under investigation formed the mass term. Temporal vari-
(11) ations in supply were represented by 'supply time potentials', a similar
notion to geographical potentials except that distance is measured in days,
where S is the population density in an infinitesimal element of the area, weeks or months since the last harvest of the product. The three potentials
dR. This formulation is of little practical value, because to quantify it were used as independent variables and in regression analyses were quite
requires knowledge of the precise location of each individual, but it does successful in statistically explaining prices. For example, in the case of
illustrate that potential is a sort of smoothed out density' (Stewart, wheat, Warntz (1959) formed the following regression equation:
1942, p 71), where the smoothing takes place in accordance with an inverse
distance formula. Population potential is therefore a weighted sum of the
population densities of the surrounding areas (Craig, 1972). The two concepts
are, however, complementary and not identical. Whereas density relates to an
area, potential has point significance and this is reflected in the fact that
(when a distance exponent of 1 is used) potential is measured in units of
people per kilometre (or, more generally, mass per unit of distance), whereas
density is measured in the more familiar notion of people per square kilo-
metre. Potential is a scalar quantity; that is, it has no direction in space.

On an empirical level, potential was found to be 'very useful in des-

cribing the variations in density which occur between areas such as the
counties of the United States' (Anderson, 1956, p 176). As earl: as 1941, The values used were averages for the period 1940-49. All three explanatory
Stewart pointed to the correlation between population potential and rural variables were significant and the regression coefficients had the anticipated
population density in 37 U.S. states east of Colorado. As well, a whole signs, thus supporting Warntz's hypothesis about the effects of supply and
series of other variables are closely but non-linearly correlated with popu- demand on prices.
lation potential, including rural non-farm population density, rural non-farm
rents, farmland values, density of railway track, and death rates. The Given this apparent success with historical data, Warntz suggested that
general form of these relationships was found to be: his approach might be useful for projection purposes. For example, given a
planned boost to production of a particular commodity, perhaps because of a
(12) new irrigation programme, what is the impact on the geographical pattern of
prices for wheat? Any production change will alter the product supply space
where Y is the variable concerned, c is a constant, P is population potential potential surface. Insertion of these new scores in the regression equation
and z is an exponent. This is a 'power' relationship and is linear when both will allow a rough estimate of the new pattern of prices to be produced.
Y and P are logged. However, the results so obtained would have to be used with considerable
caution, because of difficulties with the statistical methodology and possible
However, not all areas create the same intensity of interaction per limitations of the potential concept in this context; alternative models of
head of population. People differ in their ability to interact, so that in price formation are available (Chow, 1961) and may give superior results.
calculating potentials, population is often weighted accordingly. Warntz
(1957; Stewart and Warntz, 1958A) argued that weighting population by per (ii) Market potential and its derivatives
capita income is the best means of representing each area's ability to inter-
act. He at first called the resulting potentials 'gross economic population Many people have suggested that potential is a useful tool in studies
potentials' but the more usual term is income potential. Further analysis of industrial location and regional development. Harris (1954) was the first
has shown that income potential has close power relationships with many var- to use it in these fields. He argued that the distribution of the U.S.
iables such as the density of roads, rail tracks and telephone wires, and national market exerts a strong control over the location of American manu-
(inversely) with the average geographical sizes of farms, counties and states facturing industry. He adapted Stewart's basic concept to produce a measure
in the 48 conterminous states of the U.S.A. However, most interest has been of market potential, which aimed to represent the accessibility of each point
focused on the relationships between income potential on the one hand and to the national market, and thus to illustrate the relative attractiveness of
income density and income per capita on the other (e.g. Warntz, 1965, 1973; alternative locations. Harris was content to produce a map of the market
Coffey, 1978; Coffey and Matwijiw, 1979).
11
10
potential surface and discuss its implications; similar descriptive exercises
have been carried out for other areas such as Florida (Dunn, 1956) and
Southern Ontario (Kerr and Spelt, 1960). More recently, market potential has
been used as an explanatory variable in regression analyses of the geograph- IV OPERATIONALISATION
ical pattern of industrial growth and decline in areas such as Southern
Ontario (Ray, 1965), Nigeria (Olagbaiye, 1968; Abumere, 1978) and Great A number of decisions must be made when constructing and using a poten-
Britain (Keeble, 1976). tial model, such as how best to quantify the distance and mass terms. Three,
often conflicting, criteria might help to guide such decisions: the theoret-
Harris's assumption that market access is the major determinant of in- ical appropriateness of the particular choice; the usefulness of the resulting
dustrial location has been increasingly questioned. In reality, businessmen potential model in the empirical situation for which it is being constructed;
may seek to attain a number of different goals besides sales maximisation or and the ease and practical convenience of construction. Because of the wide
distribution cost minimisation, and a wide range of factors may influence range of circumstances in which potential models can be used, there is often
location decisions (Smith, 1971). Many of these factors, such as labour no universal 'best' course of action. Therefore, the emphasis here is on
supply, agglomeration economies and sources of components, are likely to be outlining the range of choices available and the consequences of taking them,
closely related to population size, as of course is the distribution of the but first a simple worked example is given to illustrate the procedures used
final market. It therefore makes little sense to interpret potential simply in computing potentials.
as an index of market access. Rather, it represents the relative ease of
interaction between a manufacturing plant and a whole range of activities
(Clark et al., 1969; Rich, 1978) which collectively form not only the market
for outputs but also the sources for many of the firm's inputs. Consequently,
Clark preferred the term economic potential to Harris's earlier name.
Section VI discusses an example of the use of economic potential as an ex-
planatory variable in a regression analysis of industrial growth and decline
in Scotland.

Economic potential has itself spawned a number of offspring. Blair

(1976) used a very similar model, which he termed polarisation potential, to
examine the extent to which local agglomeration economies in part of Philadel-
phia provide a favourable environment for nurturing new manufacturing firms.
Richardson (1974) suggested another variant, agglomeration potential. He
argued that the most important links between a firm and its economic environ-
ment vary between industries and even between firms, and that each of these
links may have a different distance-decay pattern due to varying spatial
frictions. The model therefore involves identifying each individual component
of interaction and evaluating its particular influence on the firm's loc-
ational preferences. The data requirements of such an approach are clearly
vast and the full model has not been used in practice.

Westaway (1974) developed the same basic theme in a slightly different

direction. He argued that the prime factor determining the location of the
headquarters of large firms, and thus a major influence on patterns of
regional growth, is the availability of high quality specialist information.
This information might relate to technical and commercial innovations, Fig. 2 Hypothetical universe and zones
finance, the behaviour of competitors, government policy and the like. Much
of this information cannot be adequately handled by telecommunications, so (i) A simple example
the opportunities for easy face to face contact between business executives
and information sources is a major factor in locating company headquarters. Six stages are involved in the construction of a potential surface.
Westaway devied a measure of contact potential, taking into account the A. Define the universe over which potentials are to be calculated and
geographical clustering of likely information sources, to represent spatial divide it into a number of zones.
variations within Britain in the relative ease of information acquisition
through personal contact. The six zones shown in Figure 2 are used in this example.
B. Obtain suitable data for the mass term, M.
C. Define a centroid for each zone and obtain a suitable measure of the
distance, D, between each pair of centroids. In this example, the distance
from a centroid to itself is taken to be an arbitrary, non-zero value;
12 13
the appropriate enumeration of this 'self-potential distance' term raises a
number of practical problems which are outlined in section IV(v).
The centroids are marked with crosses in Figure 2.
The hypothetical mass and distance data are:

Zone Mass Distance (in kilometres) to zone j

(population
i in '000s) j=1 j=2 j=3 j=4 j=5 j=6
1 5 1.5 4.2 6.3 3.8 6.7 5.0
2 20 4.2 1.5 2.3 2.4 4.2 5.1
3 12 6.3 2.3 1.5 3.2 3.1 5.1
4 15 3.8 2.4 3.2 1.5 3.0 2.7
5 8 6.7 4.2 3.1 3.0 1.5 3.4
6 7 5.0 5.1 5.1 2.7 3.4 1.5

Similar calculations give the total potentials of each zone (see

table opposite)
E. Express the potential scores shown in the final column opposite as
percentages of the highest score, (here 29.26).

Zone Relative
Potential (%)
1 56.46
2 100.00
3 89.30
4 97.95
5 74.40
6 67.84

F. Select a suitable contour interval and draw lines of equal potential

on the map of the zones (Figure 3).
15
14
 consider each case on its merits and define the universe that most suits the
problem in hand. For example, in studies of manufacturing location in Britain
it might be increasingly useful to include the whole of the E.E.C. within the
universe because of the growing importance of economic links between Britain
and the mainland of Europe.

(iii) The zones

Conceptually, it might be possible to calculate potentials by treating

each person separately. In this way, each individual j would contribute to
the potential at i according to the reciprocal of the distance D. Besides
the obvious practical problems involved, this may not even be a theoretically
desirable approach, because below a certain level of spatial disaggregation,
the distribution of population is irrelevant when using a macroscopic in-
dicator such as potential (Craig, 1972).

Consequently, we need to divide the universe into a set of zones.

Since we usually rely on using aggregate data collected for some other pur-
pose to quantify the mass term, these zones often consist of statistical or
administrative units of one sort or another. The choice is thus not so much
one of selecting appropriate regional boundaries as determining the most
suitable level of spatial disaggregation and the optimum number of, zones.
Even here the choice is constrained by the level of disaggregation for which
mass data are available.

However limited the choice of scales, it is useful to know the effect

of changing the sizes and numbers of zones used. Various people have examined
Fig. 3 Potential surface over the hypothetical universe this problem empirically. For example, Stewart's first potential map of the
United States (Figure 1) used only 24 zones; later versions used 115 (Stewart
(ii) The universe and Warntz, 1958A) and more than 3000 (Warntz, 1964). In Britain, Craig
(1974) investigated the effect of scale on the East Riding of Yorkshire por-
Every point on earth can be regarded as contributing to the potential tion of the national population potential surface. Such experiments suggest
of all others, but practical considerations usually mean that one must re- that altering the number of zones within a given universe does not fundament-
strict the size of universe over which potentials are calculated. Any ally alter the geographical pattern of the potential surface. Smaller units
decision over what boundaries to choose is ultimately semi-arbitrary and simply provide more detail on the surface; larger zones produce a more
this is seen by some as the central problem in all macroscopic models' smoothed pattern which conceals urban peaks and rural troughs of potential.
(Lukermann and Porter, 1960, p 503). The extent of the universe selected The problem when selecting the most appropriate set of zones is therefore
will probably have some impact on the shape of the potential surfaces: one of balancing the amount of detail required on the potential surface
Houston (1969) showed that the city identified as most accessible in Soviet against the rapidly escalating time and cost involved in using successively
Asia depends on the geographical limits put on the region. However, places more zones.
of a given mass that are successively more remote create progressively less
potential in the area of interest and so will have declining impact on the (iv) Mass
potential surfaces.
The ability of areas to interact is a function of their volumes of
Selection of the appropriate universe requires some care. The important socio-economic activity. The mass term should be a reasonable measure of the
thing is that the universe should be a relatively closed social or economic level of activity if potential is to give an adequate indication of the
system: the phenomenon under investigation should have relatively few, weak contact-making ability of different zones. The most appropriate definition
or unimportant links outside the system boundaries. For example, in a study will depend on the problem in hand, and especially on the types of inter-
of the social structure of a city, the appropriate boundaries might be the action potential being investigated. In some studies, simple population
fringes of the metropolitan region concerned. Similarly, an industrial geo- totals have been used to quantify the mass term, but there have been many
grapher might choose to calculate potentials over a complete nation state, attempts to improve on this. One approach is to assign weights to the popu-
even when he is interested in only one region of the nation; thus, in a study lation (just as molecular weights are used in physics) so that potential is
of manufacturing change in Outer South East England, Keeble and Hauser (1971, given by:
1972) considered the whole of Great Britain to form a single universe, the
argument being that the regional economy is significantly more open to flows
of goods, capital and labour than the national economy. It is important to
17
16
 (14) distance is a slightly more complicated alternative. Both of these methods
are easy to use and for many purposes they give adequate approximations to
the effective distance separating centroids: Nordbeck (1964) found that when
multiplied by a constant, or extension factor, straight line distance gives
a very good statistical fit to shortest road distances within a city. The
difficulty is that in some circumstances there are geographical variations
in both the goodness of fit and the extension factor. For example, if we
consider distances from places in the central core of England to other places
in Britain, the extension factor is around 1.2, while if we look at distances
from Anglesey and various peninsular areas the figure is over 1.4. Straight
line distances therefore substantially underestimate the effective remoteness
of many peripheral areas from the rest of the country.

For this reason, some authors have experimented with a variety of alter-
natives such as the distance along major transport routes. Carrothers (1958),
for example, used railway distance, and Anderson (1956) suggested that it
might be useful to take account of differences in the predominant mode of
travel for particular types of interaction along particular routes. Many
others have tried to get away from using such purely physical notions of dis-
A second possibility is to replace population with another variable that tance and have turned to measures of economic or social separation, such as
appears to give a good measure of local contact opportunities. For example, travel time or cost. A study of access to job opportunities in Atlanta
in his study of the market orientation of U.S. manufacturing, Harris (1954) (Bederman and Adams, 1974) used the average travel time between home and work
used the value of retail sales to represent the size of the market. Similarly, zones. Harris (1954) used a measure of the costs of transporting goods,
Pred (1973, 1977) used the number of newspaper editions published locally each while a similar study in Western Europe (Clark et al, 1969) used a more com-
week in his study of the availability and circulation of information in pre- plex cost index incorporating terminal charges, movement costs and (where
telegraphic America; this study is examined in more detail in section V. appropriate) international tariffs.
A third approach, suggested by Anderson (1956) and others, is to raise
the weighted or unweighted mass term to some exponent, a. Potential is then
given by:

(15)

This formulation aims to take account of factors, such as agglomeration

economies, which might mean that an area's interaction capacity rises more
than proportionally with its social or economic size.

Recently, the emphasis has been on simplicity in defining the mass term Dii can be defined in a number of ways, though all are somewhat arbitrary.
and many studies have used a single variable, such as zonal population, em- Applications of potential using transport cost rather than physical distance
ployment or income, thought to give a reasonable indication of interaction- generally avoid the problem by using an estimate of intrazonal transport cost
generating capacity. There is generally less debate than in the past over or by adding a significant terminal charge to all costs. Anderson (1956)
which mass term to use since most of the readily available possibilities tend tackled the problem by adding a constant to all distances, to avoid Dii values
to be quite closely correlated and the choice among them has relatively of zero; both Ray (1965) and Houston (1969) used an arbitrary fixed value to
little impact on the resulting potential surfaces (Houston, 1969). For many represent the average distance over which intrazonal contacts occur. The
people, the distance term is more important and more interesting. latter approach, though convenient, is difficult to justify because it takes
no account of variations in the geographical sizes of the zones (which may be
(v) Distance considerable if administrative units are being used). A preferable method,
developed by Stewart (1947), uses the result that the self-potential of a
The distance term represents the effects of the geographical separation uniform circular disc at its centre is equal to the mass divided by half the
of different masses in impeding the interaction between them. The simplest radius: the self-potential distance is taken to be half the radius of a circle
measure to use is the straight line distance between the centroids of each of the same area as the zone concerned. Although this approach allows for
zone, which can be easily found using Pythagoras's theorem. When potentials variations in the size of zones, it does not allow for differences in their
are to be computed over large areas such as continents, the 'great circle' shape or in the distribution of the mass within them. Departures from cir-
cularity probably have a relatively minor impact onself-potential in most
18

19
cases (Stewart and Warntz, 1958B), but the distribution of the mass may be
more important. For example, the self-potential created by a uniformly dis-
tributed mass over a circle is only two-thirds that imposed by either a
conical or a Gaussian distribution and less than half that imposed by a neg-
ative exponential distribution around the centroid (King, 1969, pp 96-97):
the more concentrated the mass around the zone's centroid, the greater is
the self-potential created. A partial solution to this might be to modify
Stewart's approach by taking D ii to be some fraction, less than half, of the
zone's radius, to allow for the likely peaking of the mass around the centroid.
However, at present there is no clear consensus on which definition of Dii
is most appropriate.

A further issue is the selection of an appropriate exponent for the

distance term. Newtonian theory suggests an exponent of 1.0, and Stewart
(1947) drew faithfully on the analogy, arguing that higher exponents would
render the resulting potential scores meaningless (Stewart and Warntz, 1958B).
Olsson (1965, p 59) pointed out, somewhat cynically perhaps, that a variable
exponent would complicate the search for social laws and constants into
which Stewart and Warntz put so much effort. In contrast, most other authors
have recognized the value of a more flexible approach in social and economic
applications of potential. There has been no general agreement on any single
exponent of general applicability, although values between 1.0 and 2.5 are
commonly used.

Variations in the exponent represent differences in the effect of dis-

tance on the volume of interaction. High values indicate that distance is a
major impediment to contact and that the bulk of contacts are highly local-
ised. Conversely, low values indicate that interaction is little constrained
by distance. Thus, different exponents are appropriate in different circum-
stances since some types of interaction are more significantly impeded by
distance than others. Journey to work and to school may be much more local-
ised than social or recreational travel and some groups of people may be more
mobile than others, and the movement of goods is generally more localised
than flows of information. Further, exponents seem to have shown some ten-
dency to fall through time as transport and communication facilities improve
and relative costs decline.

The value of the exponent has a significant impact on the nature of the
potential surface. This is illustrated in Figure 4, which shows potential
surfaces for Great Britain with exponents of 1.0 and 2.0. The lower value
is associated with interaction that is little constrained by distance so that
effective geographical variations in accessibility are relatively small;
correspondingly, the potential surface is relatively flat, without major
local variations in scores. Conversely, the larger exponent, indicating that
distance is a major impediment to interaction, means that geographical vari-
ations in potential are substantial; the potential surface is strongly peaked
around major population concentrations and more distant areas, by contrast,
are seen to have a very low capacity to interact.

There remains the issue of how to select the distance exponent. Possible
approaches are to choose a value arbitrarily or to duplicate a value used in
another study. A more satisfactory method is to examine suitable interaction
data, to fit a gravity model to them using a regression method described by
Taylor (1975) and so derive an exponent empirically. Ray (1965) used this
approach in his study of manufacturing in Ontario by regressing the tonnage

20 21
 Pred constructed such information accessibility maps for 1790, 1820
and 1840 (Figures 5, 6 and 7). Potential values were computed for all coun-
of commodity flows on distance; he obtained a value of 1.42, although various ties having newspapers, with the mass term being represented by the number
technical and conceptual problems led Houston (1969) to question this result. of editions appearing weekly. Intercounty distances were measured between
A more pragmatic approach is to use a trial and error procedure to find the the population centroid of each county in 1960 (on the argument that any
'best' exponent in particular circumstances. When the potential surface is small shift in the population centre of gravity in the intervening period
being constructed to act as an explanatory variable in regression analysis, would have only a very slight effect on potential scores). Self-potential
'best' may be defined as the surface providing the largest statistical fit distance was derived by taking the mean of the particular county's two
to the dependent variable; this is the method used in the study described longest perpendicular axes; this had the effect of undervaluing the self-
in section VI. This method indicates that there is no universally optimum potential, and hence the total potential, of those counties, such as New
exponent, although values much over 3.0 are rarely appropriate, and that York and Philadelphia, where there was a significant level of purely local
quite small changes in the exponent may have quite a substantial impact on exchange of news. The distance exponent was set to unity throughout. For
relative potentials and consequently on the statistical explanation achieved. ease of comparison, all potentials for 1790, 1820 and 1840 were converted to
a single base, with New York's value in 1790 being set at 100.

The highest levels of accessibility to newspapers in 1790 occurred in

V CASE STUDY 1 - NEWS CIRCULATION IN THE UNITED STATES the belt from Baltimore to Boston (Figure 5). Within this zone, accessibility
peaked sharply at New York and Philadelphia. In contrast, the public-
One interesting application of potential by Pred (1973, 1977) used the informational remoteness of the south-east and the western interior was
model to illustrate geographical variations in the availability of business marked, with large areas having potentials less than five per cent of that
information in the United States between 1790 and 1840. Pred was concerned in New York, reflecting poor access to the nation's newspapers, the time lag
to understand the processes involved in the growth of cities, and particularly involved in their circulation, and a low probability of contact with business
the mechanisms producing the long term stability in population size rankings information. Pred further underlined the extreme nature of geographical
of major cities that is a characteristic of many urban systems. He argued differences in information accessibility by comparing maps of information
that the growth of the system of cities reflects an accumulation of decisions potential and population potential. Public information access declined very
determining the location and size of public and private investment. Such much more rapidly from New York than would be expected from the pattern of
decisions depend on the stock of relevant specialised information, since accessibility to population.
decision makers can only choose from among the alternatives of which they
are aware. Commercial information, for example on costs, raw material supp- With a marked increase in the numbers of newspapers published in the
lies, demand and business opportunities, is almost never universally avail- thirty years to 1820, general levels of access to business information ad-
able. Rather, its distribution is spatially biassed because of the nature vanced considerably. Nevertheless, important spatial variations in inform-
of the contact system through which news circulates, and because of the time ation access remained. Whereas New York's potential had increased to an
lag involved in its geographical dissemination (particularly in the era be- unmapped value of 302 (Figure 6), large areas of the country had failed to
fore telecommunications). Investment decisions are thus made with incomplete reach that city's 1790 score. New York was thus well on the way to establish-
knowledge of all the opportunities existing, and their locations reflect the ing its pre-eminent position in the American news circulation network.
information that is available. Information-rich areas tend to attract more
investment, while the commercial links so created provide additional avenues By 1840, most of the well-settled area west of the Appalachians had at
for news dissemination, thereby further intensifying spatial biasses in the last surpassed the public-information access that had been reached in New
availability of information. Consequently, initial advantages in terms of York 50 years before. Zones of moderately high potential had grown up
information availability are translated, through a positive feedback process, around Cincinnati, Louisville, St. Louis and Pittsburg (Figure 7), which
into long term economic superiority. News of high quality, in the sense of were already dominating the regional system of cities in the Ohio and Upper
being commercially relevant, accurate and up to date, is most readily avail- Mississippi valleys. However, by this time the dominance of New York (with
able in large volumes in the major cities. The pattern of investment that an unmapped value of 825) over the circulation of domestic news in the U.S.A.
results from this is a critical factor moulding the development of a system was complete. The city's importance as a news source was further accent-
of cities. uated by its role as the primary point of entry of foreign news into the
country. New York had thus achieved major competitive advantages over its
In the United States of the post-independence, pre-telegraphic era, commercial rivals in terms of access to business information. This advantage
newspapers were the main means of disseminating commercial information, in- was a major factor allowing the city to emerge as the undisputed business
cluding advertising, shipping intelligence and wholesale prices. Further, capital of the U.S.A., which in turn further increased the volume of commer-
in a period before the development of a news wire service and the appointment cially useful information readily available there.
of non-locally based correspondents , the main source of domestic and foreign
information was the reports carried in other journals. This practice of
'mutual journalistic plagiarism' (Pred, 1977, p 40) was well established as
a major feature of the network by which news circulated. Consequently, a
potential map of variations in accessibility to all U.S. newspapers illus-
trates spatial biasses in the availability of public commercial information.
23

22
Fig. 5 Public-information accessibility in the United States, 1790
(after Pred, 1977) Fig. 6 Public-information accessibility in the United States, 1820
(after Pred, 1977)

24
25
 VI CASE STUDY 2 - MANUFACTURING INDUSTRY IN SCOTLAND
The second example, illustrating the use of potential as an explanatory
variable in statistical analysis, is a study of the geographical patterns of
growth and decline in manufacturing employment in Scotland between 1961 and
1971. The work is part of an examination of the processes leading to the
existence of the 'regional problem' in Britain. Two main research hypotheses
were tested:

1) areas of Scotland that provide maximum accessibility to economic activ-

ity located within the region and in other parts of Britain are the most
satisfactory locations for manufacturing; consequently, there is a positive
linear relationship between accessibility and industrial growth; and
2) the fringes of the core areas of the Scottish economy experience the
most rapid industrial growth and central areas fail to secure a share of new
development proportional to their existing share of activity; consequently
there is a quadratic relationship between accessibility and net industrial
growth.

The idea behind the first hypothesis is that accessible areas are
attractive to new industries and provide favourable environments for the
expansion of existing activities, and that consequently they tend to have
good records of employment growth compared with other areas. The possible
advantages of centrality are many (Keeble, 1976, pp 46-71), but they may be
conceived as resulting from the fact that the concentration of economic
activities there provides manufacturing with a wide range of interaction
opportunities. In general, firms seek locations offering a large volume of
high quality potential contacts at a low cost, and because central areas
offer this they have competitive advantages over remote regions. These ad-
vantages involve input and output links of all kinds; not only forward links
with customers, but also backward links with sources of raw materials, com-
ponents, labour, capital, information and more general urbanisation economies
may be important.

The links influencing locational decisions will vary, but both inter-
and intraregional contacts are likely to be relevant, so accessibility-seeking
behaviour might occur at either scale in conditions approximating a free-
market economy. However, British regional policy has long operated to offset
the advantages of locating in the country's economic core, running from
south-east to north-west England; consequently, accessibility-seeking be-
haviour might be more evident within regions, where public intervention might
possibly have had less effect on broad patterns of location. This was one
reason for testing the two hypotheses at the intraregional scale.

The contact advantages of accessible locations stimulate the economic

growth of these areas. The advantages are cumulative, since by locating
there firms add to the stock of interaction opportunities available to new
locators. There is thus a situation of cumulative causation of the sort
Fig. 7 Public-information accessibility in the United States, 1840 identified by Myrdal (1957), and by Pred (1973, 1977) in the study discussed
(after Pred, 1977) in section V, whereby initial locational advantages are continually rein-
forced. There is a long term positive feedback relationship between central-
ity and regional growth. The first hypothesis concentrates on one part of
26
27
this relationship by suggesting that in the short term, say five to ten
years, growth is determined by patterns of accessibility; the feedback from
regional growth to modify these patterns is ignored here.

The first hypothesis also ignores the probable emergence of spread

effects which gradually develop (Hirschman, 1958; Friedmann, 1966) to modify
the cumulative growth process. More concretely, the hypothesis overestimates
the net benefits of locating in the most accessible areas of Scotland because
of the peaking there - as in the cores of other regions - of land, labour and
congestion costs. Further, the hypothesis ignores the gradual weakening of
geographical constraints on the location of industry produced by economic
and technical changes in industrial production and in transport and commun-
ications of all kinds. Hypothesis two thus suggests that conceptually there
is a trade-off between the need to maximise potential contacts and the desire
to avoid the diseconomies of excessive development.

Regression analysis was used to test the two hypotheses. The dependent
variables examined were the employment changes (in jobs per 100 km 2 ) between
1961 and 1971 in manufacturing as a whole and in the 14 manufacturing orders
(using the 1958 version of the Standard Industrial Classification). The
independent variable was economic potential, scaled so that the highest value
in Great Britain was 100.0. Great Britain was used as the universe over
which potentials were calculated, although only the Scottish scores were used Fig. 8 Possible relationships between employment change and potential
in the regression analysis, because it was felt that links both within Scot-
land and with other regions might influence industrial location patterns in declining industries. Six of the seven with expanding employment follow
Scotland. A variable exponent version of potential was used to represent as the quadratic pattern suggested by hypothesis 2, while the contracting in-
accurately as possible the contact-impeding effects of distance in different dustries exhibited much more varied patterns of change. Not one of the 14
circumstances; the value of the exponent, b, was selected by trial and error industries supports the first hypothesis, although two have an inverse
to maximise the statistical fit of the regression model to each dependent linear pattern of change, with their worst employment records in central
variable. Two of the potential surfaces are shown in Figure 4. areas.

The regression functions suggested by the two hypotheses are sketched The details of the results, and their implications for regional develop-
in Figure 8. Hypothesis 1 implies that the greatest employment growth is ment, are not important here, but the analysis is consistent with the notion
in the most accessible regions of Scotland, namely around the Clydeside that, in growing industries, there is some sort of trade-off between centri-
conurbation (see Figure 4). Hypothesis 2 suggests that the peak employment fugal and centripetal locational pressures. However, the same geographical
growth rates are found in near-central areas, such as the old counties of patterns could conceivably be produced by other processes. Geographical
Ayrshire, the Lothians, Fife and Stirlingshire. variations in the balance between nationally expanding and declining
activities, and the spatially varying effects of regional policy are the
The actual pattern of employment change in manufacturing as a whole two most obvious candidates; a closer examination indicates that although
(Figure 9) does bear some resemblance to that suggested by hypothesis 2: they do reinforce the growth of near-central areas, even when this is allowed
the major employment losses were around Clydeside and the largest increases for there is still a clear quadratic relationship between potential and
were in East and West Lothian, Ayrshire and Fife. On the other hand, a num- employment change. This strongly suggests that the maximisation of contact
ber of areas performed better (e.g. Aberdeenshire) or worse (e.g. Midlothian) opportunities and the minimisation of congestion costs are, in combination,
than suggested. Regression analysis of these data provides little support important factors in moulding industrial change in growing activities and
for the linear hypothesis but shows that the quadratic model is of the form that economic potential can be used as a reasonable surrogate for the relative
hypothesised and provides a reasonable fit to the data. The quadratic re- attractiveness of different areas on these criteria. On the other hand,
gression equation is: the very different geographical patterns of change in industries with de-
clining employment suggests that although these industries may be influenced
by contact and congestion factors, other mechanisms are also important.

Similar regression analyses showed that the patterns of change in 13 of

the 14 individual industries were related to potential (Table 1) but, more
interestingly, there is a clear distinction between the growing and the

28 29
Fig. 9 Manufacturing employment density changes in Scotland, 1961-1971

30 31
 statistical goodness of fit between potential and another spatially dis-
tributed phenomenon, and again means that conventional tests are inappropriate.
One alternative approach, used in the Scottish study, is to test whether
synthetic, spatially random data would be likely to have as strong a relation-
VII PROBLEMS AND PROSPECTS ship with potential as that observed between potential and the dependent
variable; if not, then one might be reasonably safe in suggesting some sort
(i) Outstanding problems of structuring in the dependent variable of the type hypothesised. However,
at present there is no consensus on how to deal with this difficulty and it
Potential models have been widely and often successfully used, but is probably best to regard it as an unresolved problem.
there are problems to be faced in their application. A number of technical
decisions, outlined in section IV, have to be made during their construction. (ii) Some recent advances
Some authors (e.g. Lukermann and Porter, 1960; Houston, 1969) have cited
the difficulty of making these, and the semi-arbitrary nature of some of the There is continuing interest in potential models and new applications
decisions, as grounds for avoiding potential models if at all possible. appear fairly frequently. For example, potentials have recently been con-
However, these construction problems need not prevent potential models from structed to indicate the geographical distributions of scientists around the
making a useful contribution to geographical studies if they are used sen- world (Inhaber, 1975), physicians in Seattle (Schultz, 1975), dentists in
sibly and carefully. But there are two other, perhaps more important, areas Newcastle-upon-Tyne (Bradley et al., 1978) and ethnic groups in Winnipeg
of difficulty. (Matwijiw, 1979). Perhaps more interesting than these new applications of
old models are attempts to refine the concepts and methods involved.
The first is that by their very nature, potential models are used in
the course of macro-level geographical work and suffer from the problems of In the income potential field, the power relationships between potential
all macro-analytical techniques. Simply, potential models deal with aggre- and density and between potential and per capita income have been reinter-
gates - of people, firms and the like - and consider average patterns: they preted in entropy terms by Fein (1970), Warntz (1973) and others. This has
have no direct behavioural input and tell us nothing about any one individual. stimulated renewed research. In particular, the social and economic processes
Further, potential does not represent any single process, but is a fairly underlying the spatial structure of the city have been explored by focussing
crude surrogate for a whole range of spatially varying phenomena. Con- on the varying 'integration' of different groups, that is their potential to
sequently, it is often difficult to infer, and ultimately impossible to interact with people and activities located elsewhere within the city.
'prove', the processes influencing individuals and producing observed aggre- Boston, Toronto and Sydney have all been examined in this fashion (Coffey,
gate patterns. For example, in the Scottish study just outlined, it is 1977, 1978; Coffey and Matwijiw, 1979). The interpretation of the power re-
possible to say that patterns of industrial change are consistent with an lationship, and especially its exponent, are still subject to debate but it
hypothesis about the factors influencing individual firms and to discount does seem likely that further investigation might lead to new insights about
other factors that might conceivably have produced similar patterns of change, the processes moulding the structure of cities, and particularly the social
but using potential as an explanatory variable it is not possible to go much and geographical distribution of income within them (Coffey, 1978).
further. Thus, the use of the potential model is not a substitute for de-
tailed micro-level analysis of individual cases. Rather, the two approaches The application of potential in industrial geography is continually
are complementary: while any use of potential must be firmly based on the being refined. Richardson's interesting proposal of an agglomeration poten-
results of relevant previous research, it is often useful in generating or tial model has already been mentioned. Tybout and Mattila (1977) have used
i mproving hypotheses that can later be further tested by other methods. a somewhat similar but slightly more simple approach to investigate the
intrametropolitan location of 11 manufacturing industries in Detroit. They
The second problem area relates to the use of potential models in used multiple regression analysis to examine the distribution of employment
statistical hypothesis testing, especially their use as explanatory variables in the industries in terms of a number of explanatory variables, including
in regression analysis. Difficulties here mean that the conventional tests agglomeration economies, potential supply of inputs and potential demand for
in inferential statistics are probably not appropriate. One aspect of the outputs for each firm. Each of these three explanatory variables was quant-
problem is that the so-called independent and dependent variables may not, ified using potential; the mass terms of the latter two were weighted by the
in fact, be independent of one another. For example, because of the func- relevant transaction coefficients of the U.S. national input-output table
tional relationship between density and potential, some correlation is to be to take account of the relative importance to an industry of links with each
expected for any distribution of population potential and population density, particular type of activity. The use of input-output data to disaggregate
so that the suggestion from t or F tests that the correlation is 'significant' the interaction between a firm and its environment in this way is an interest-
may be entirely spurious. Testing procedures generally need to be much more ing line of research and promises to add considerably to the value of poten-
conservative in this situation. tial models in industrial location studies.

A different aspect of the same problem is that conventional inferential By way of conclusion, it is useful to point out that planners have also
tests require random, independent samples (Gould, 1970), a requirement that made a number of uses of potential models. Hansen (1959) developed a version
runs counter to many geographical hypotheses. Potential is a highly spatially of potential to examine opportunities for residential growth.
autocorrelated phenomenon: potential scores of neighbouring areas are, by
definition, not independent of one another. This will tend to inflate the
33
32
His model differed from simple population potential in that:
a) unlike population potential, it is used to predict likely population
growth;
BIBLIOGRAPHY
b) it includes a measure of the inherent attractiveness or physical Abumere, S.I. (1978), 'Multinationals, location theory and regional develop-
capacity of each zone; and ment : case study of Bendel State of Nigeria'. Regional Studies,
12, 651-664.
c) whereas population potential measures the mutual accessibility of
groups of people, Hansen's model represents the accessibility of resi- Anderson, T.R. (1956), 'Potential models and the spatial distribution of
dential areas to places of work, on the argument that work-place and population'. Papers and Proceedings, Regional Science
journey to work are major influences on the location of residential Association, 2, 175-182.
development (Smith, 1974).
Anscombe, F.J. (1960), 'Rejection of outliers'. Technometrics, 2, 123-145.
As an initial stage, Hansen argued that the likely residential development,Q, Bederman, S.H. and Adams, J.S. (1974), 'Job accessibility and under-
of zone i is proportional to the proximity of the zone to employment, M,
employment'. Annals, Association of American Geographers,
and to the inherent attractiveness, or 'holding capacity', H, of the zone.
64, 378-386.
In other words, potential scores are weighted by a measure of zonal attrac-
tiveness. Blair, J. (1976), 'Industrial polarization and the location of new manu-
facturing firms : an empirical application'. Regional Science
(18) Research Institute, Discussion Paper Series, No. 89.
or
Bradley, J.E., Kirby, A.M. and Taylor, P.J. (1978), 'Distance decay and
(19) dental decay : a study of dental health among primary school
children in Newcastle upon Tyne'. Regional Studies, 12, 529-540.
The development potential of i is then expressed as a proportion of the Carey, H.C. (1858-9), Principles of social science. (Philadelphia :
total development potential of all residential zones and multiplied by the Lippincott).
total residential population, N, in the particular universe, so that: Carrothers, G.A.P. (1956), An historical review of the gravity and poten-
tial concepts of human interaction'. Journal American Institute
of Planners, 22, 94-102.
Carrothers, G.A.P. (1958), 'Population projection by means of income poten-
tial models'. Papers and Proceedings, Regional Science
Association, 4, 121-152.
Hansen's model has been used frequently, particularly as the residential
part of Lowry's 'model of metropolis' (Lowry, 1964). It has been successively Chow, G.C. (1961) Review of Warntz, W. (1959), 'Toward a geography of price'.
revised and was eventually reformulated by Wilson (1970) in terms of observed Journal American Statistical Association, 56, 209-210.
rather than potential journey to work trips. The wheel had then gone full
Clark, C., Wilson, F., and Bradley, J. (1969), 'Industrial location and
circle in that a potential model had been integrated back into an enlarged
economic potential in Western Europe'. Regional Studies, 3,
gravity model: More significantly, perhaps, in doing so Wilson gave the
197-212.
newly extended gravity model a much more rigorous theoretical base. This,
like the analogy with Newton's theory, came from physics, but this time it Coffey, W. (1977), 'A macroscopic analysis of income regions in Metropolitan
was developed from the more modern concepts of statistical mechanics. Boston'. Professional Geographer, 29, 40-46.
Coffey, W. (1978), 'Income relationships in Boston and Toronto : a tale of
two countries?'. Canadian Geographer, 22, 112-129.
BIBLIOGRAPHY Coffey, W. and Matwijiw, P. (1979), 'Measures of demographic potential
applied to the definition of income regions in Metropolitan
Abumere, S.I. (1978), 'Multinationals, location theory and regional develop-
Sydney'. Tijdschrift voor Economische en Sociale Geo-
ment : case study of Bendel State of Nigeria'. Regional Studies,
grafie, 70, 98-109.
12., 651-664.
Craig, J. (1972), 'Population potential and population density'. Area,
Anderson, T.R. (1956),''Potential models and the spatial distribution of
4, 10-12.
population'. Papers and Proceedings, Regional Science Association,
2, 175-182. Craig, J. (1974), 'How arbitrary is population potential?'. Area, 6, 44-47.
Anscombe, F.J. (1960), 'Rejection of outliers'. Technometrics, 123-145. Dunn, E.S. (1956), The market potential concept and the analysis of loc-
ation'. Papers and Proceedings, Regional Science Assoc-
iation, 2, 183-194.

34 35
Fein, E. (1970), 'Demography and thermodynamics'. American Journal of Nordbeck, S. (1964), 'Computing distances in road nets'. Papers and
Physics, 38, 1373-1379. Proceedings, Regional Science Association, 12, 207-220.
Friedmann, J. (1966),Regional development policy : a case study of Olagbaiye, J.A. (1968), 'Towards manufacturing locational analysis in
Venezuela. (Cambridge, Mass. : M.I.T. Press). Southern Nigeria : a population potential model'. Nigerian Geo-
graphical Journal, 11, 11-20.
Georgescu-Roegen, N. (1971), The entropy law and the economic process.
(Cambridge, Mass. : Harvard University Press). Olsson, G. (1965), 'Distance and human interaction'. Regional Science
Research Institute, Bibliography Series, No. 2.
Gould, P. (1970), 'Is 'Statistix inferens' the geographical name for a wild
goose?' Economic Geography, 46, 439-448. Pred, A.R. (1973), Urban growth and the circulation of information:
the United States system of cities, 1790-1840.
Hansen, W.G. (1959), 'How accessibility shapes land use'. Journal American (Cambridge, Mass. : Harvard University Press).
Institute of Planners, 25, 73-91.
Pred, A.R. (1977), City systems in advanced economies : past growth,
Harris, C.D. (1954), 'The market as a factor in the location of industry in present processes and future development options.
the United States'. Annals, Association of American Geo- (New York : Wiley).
graphers, 44, 315-348.
Ray, D.M. (1965), 'Market potential and economic shadow'. Universi ty of
Hirschman, A.O. (1958), The strategy of economic development. (New Chicago, Department of Geography, Research Papers,
Haven, Conn. : Yale University Press). No. 101.
Houston, C. (1969), 'Market potential and potential transportation costs: Rich, D.C. (1978), 'Population potential, potential transportation cost and
an evaluation of the concepts and their surface patterns in the industrial location'. Area, 10, 222-226.
U.S.S.R.'. Canadian Geographer, 13, 216-236.
Richardson, H.W. (1974), 'Agglomeration potential : a generalization of the
Inhaber, H. (1975), 'Distribution of world science'. Geoforum, 6, 231-236. income potential concept'. Journal of Regional Science, 14,
Inhaber, H. and Przednowek, K. (1974), 'Distribution of Canadian science'. 325-336.
Geoforum, 19, 45-54. Schultz, R.R. (1975), 'A space potential analysis of physician location'.
Keeble, D.E. (1976), Industrial location and planning in the United Proceedings, Association of American Geographers, 7,
Kingdom, (London : Methuen). 203-208.
Keeble, D.E. and Hauser, D.P. (1971), 'Spatial analysis of manufacturing Smith, D.M. (1971), Industrial location : an economic geographical
growth in outer South-East England. 1: Hypotheses and variables'. analysis. (New York : Wiley).
Regional Studies, 5, 229-262.
Smith, P. (1974), 'The use of the potential surface technique in sub-regional
Keeble, D.E. and Hauser, D.P. (1972), 'Spatial analysis of manufacturing planning'. (University of Reading, Department of Geography,
growth in outer South-East England. 2: Methods and results'. Geographical Papers, No. 30.
Regional Studies, 6, 11-36.
Stewart, J.Q. (1941), 'An inverse distance variation for certain social
Kerr, D. and Spelt, J. (1960), 'Some aspects of industrial location in influences'. science, 93, 89-90.
Southern Ontario'. Canadian Geographer,15, 12-25. Stewart, J.Q. (1942), 'A measure of the influence of population at a
King, L.J. (1969), Statistical analysis in geography. (Englewood distance'. Sociometry, 5, 63-71.
Cliffs, N.J. : Prentice Hall). Stewart, J.Q. (1947), 'Empirical mathematical rules concerning the distrib-
Lowry, I.S. (1964), A model of metropolis. (Santa Monica : Rand Corpor- ution and equilibrium of population'. Geographical Review,
ation). 37, 461-485.
Lukermann, F. and Porter, P.W. (1960), 'Gravity and potential models in Stewart, J.Q. (1948), 'Demographic gravitation : evidence and applications'.
economic geography'. Annals, Association of American Geo- Sociometry, 11, 31-58.
graphers, 50, 493-504. Stewart, J.Q. (1952), 'A basis for social physics'. Impact of Science
Matwijiw, P. (1979), 'Ethnicity and urban residence: Winnipeg, 1941-1971'. on Society, 3, 110-133.
Canadian Geographer,23, 45-61. Stewart, J.Q. and Warntz, W. (1958A), 'Macrogeography and social science'.
Myrdal, G.M. (1957), Economic theory and under-developed regions. Geographical Review, 48, 167-184.
(London : Duckworth). Stewart, J.Q. and Warntz, W. (1958B), 'Physics of population distribution'.
Neft, D.S. (1966), 'Statistical analysis for areal distributions'. Regional Journal of Regional Science, 1, 99-123.
Science Research Institute, Monograph Series, No. 2. Taylor, P.J. (1975), 'Distance decay models in spatial interactions'.
Concepts and Techniques in Modern Geography, No. 2.

36 37
 Tegsjo, B. and Oberg, S. (1966), 'Concept of potential applied to price
formation' . Geografiska Annaler, 48B, 51-58.
Tybout, R.A. and Mattila, J.M. (1977), 'Agglomeration of manufacturing in
Detroit'. Journal of Regional Science, 17, 1-16.
Warntz, W. (1957), 'Geography of prices and spatial interaction'. Papers
and Proceedings, Regional Science Association, 3, 118-129.
- Warntz, W. (1959), Toward a geography of price : a study in geo-
econometrics . (Philadelphia : University of Pennsylvania Press).
Warntz, W. (1964), 'A new map of the surface of population potentials for
the United States, 1960' . Geographical Review, 54, 170-184.
Warntz, W. (1965), Macrogeography and income fronts. (Philadelphia :
Regional Science Research Institute).
Warntz, W. (1973), 'New geography as general spatial systems theory - old
social physics writ large?'. In: R.J. Chorley (ed.) Directions
in geography, (London : Methuen).
Westaway, J. (1974), 'Contact potential and the occupational structure of
the British urban system 1961-1966 : an empirical study'.
Regional Studies, 8, 57-73.
Wilson, A.G. (1970), Entropy in urban and regional modelling.
(London : Pion).

38
 contd.

25. Directional statistics - G.L. Gaile & J.E. Burt

26. Potential models in human geography - D.C. Rich

This series, Concepts and Techniques in Modern Geography

is produced by the Study Group in Quantitative Methods, of
the Institute of British Geographers.
For details of membership of the Study Group, write to
the Institute of British Geographers, 1 Kensington Gore,
London, S.W.7.
The series is published by Geo Abstracts, University of
East Anglia, Norwich, NR4 7TJ, to whom all other enquiries
should be addressed.