Professional Documents
Culture Documents
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Association of American Geographers, Taylor & Francis, Ltd. are collaborating with
JSTOR to digitize, preserve and extend access to Annals of the Association of American
Geographers
This content downloaded from 202.43.95.117 on Tue, 27 Aug 2019 21:39:06 UTC
All use subject to https://about.jstor.org/terms
THE RANDOM SPATIAL ECONOMY: AN EXPLORATION
IN SETTLEMENT THEORY
LESLIE CURRY
Arizona State University
HE spatial arrangement and spatial opera- methods. It has been phrased as a static
tion of human activities are very complex equilibrium problem with a maximum effi-
so that the development of the branch of our ciency postulate based on optimizing behavior
science concerned with their explanation is and competition. Rational behavior is often
inevitably slow. The difficulties may be a useful and indeed indispensable notion but
grouped under three heads: (1) spatial speci- is not necessarily apposite in this context. It
fication; (2) decision rules used to organize is all very well when one can supply the
behavior; (3) expression of the logical conse- parameters within which choices are made,
quences of the interrelations of the two factorsbut in any general location problem, particu-
above. In physical geography, progress is larly in a dynamic framework, one cannot
easier because invariant modes of conduct begin to comprehend the infinite number of
are easier to come by, and either very limited decisions, rarely coincident in time and sepa-
situations are handled or the phenomena dealt rately motivated under differing circumstances
with may be massively and yet meaningfully and degrees of information. The complex of
aggregated so that in any event the spatial physical capital appearing as settlement and
specification problem is simpler. This paper communications networks are "going con-
attempts to cut the Gordian knot posed by the cerns" with a locational stability greater than
complicated interrelations involved by assert- the individual units composing them; their
ing randomness of behavior and/or random- economic viability seems to be due to the lag
ness of location and using probability theory effect of individual decisions. Many industries
as a language. As it turns out, the results of have been located for reasons which are now
various models formulated resemble the real irrelevant and how many more have ante-
world. At first sight such findings may be cedents which are unrecorded and unknown?
bewildering but their reasonableness improves Even in a static situation and with a relatively
with acquaintance. simple location problem Koopmans and Beck-
This paper attempts to explain some fea- mann found locations were indeterminate if
tures of settlement in the U.S.A. Clearly the intermediate producers existed and if optimi-
large numbers of relevant variables have to be zation was the goal.' Such a result denies the
pruned and aggregated to the point where possibility of a general theory using these
they are analytically tractable. Randomness terms and intuition suggests that no change
is postulated as affording a convenient tech- in the mathematical formulation would yield
nique for investigating the overall properties a unique answer. Even if a solution is found,
of this field. Such a formulation is neutral as the notion of optimum spatial equilibrium
to rationality whether socially or economically may not be useful in a dynamic setting, even
oriented: every decision may be optimal from if the future is known with certainty. "Any
a particular point of view and yet the resulting given pattern of plants will only be optimal
actions as a whole may appear as random. at a given point in time [and] any expansion
Lack of information, social ties, and so on will plan will result in less than optimum opera-
change an economic optimizing solution but tions at any given time."2
not the randomness formulation. Clearly such
SPATIAL SPECIFICATION
reasoning is useful only in particular contexts,
The major location theorists, von Thuinen,
especially in general location problems, and
Weber, Christaller, and Lbsch, have all re-
leaves more restricted, and more urgent, prob-
lems unanswered. 1 T. C. Koopmans and M. Beckmann, "Assignment
LOCATIONAL DECISIONS Problems and the Location of Economic Activities,"
Econometrica, Vol. 23 (1957), pp. 53-76.
Settlement theory has been developed within
2F. A. Lindsay, New Techniques for Management
location theory and generally adopts its Decision Making (New York, 1958).
138
This content downloaded from 202.43.95.117 on Tue, 27 Aug 2019 21:39:06 UTC
All use subject to https://about.jstor.org/terms
1964 RANDOM SPATIAL ECONOMY: EXPLORATION IN SETTLEMENT THEORY 139
This content downloaded from 202.43.95.117 on Tue, 27 Aug 2019 21:39:06 UTC
All use subject to https://about.jstor.org/terms
140 LESLIE CURRY March
Since p (x) is a probability function randomness, the very difficulty of this work
and the early results mentioned above can
Jk O(x)dx= 1 reasonably allow neighboring settlements to
it may be shown that C = h/70 5 and thus be regarded as randomly distributed to a first
approximation. It may be noted that a normal
+(x) = (h/7ro.5 ) e 2X2 distribution could be obtained by taking
the normal density function. second nearest neighbors or any other con-
The probability of a particular value of y sistent set of measurements. Richardson, for
occurring, given that there is a probability of example, used all possible distances between
x, such that x2 + y2 = r2 has been obtained. In all points of a distribution in his work on
other words, given the amount of northing turbulent diffusion and obtained the same
between adjacent settlements, the probability distribution.7
of, say, easting between these settlements is
THE LOG NORMALITY OF NEAREST NEIGHBORS OF
also known. This result can be used to obtain
THE SAME SIZE
the probability of a settlement being a certain
distance from its nearest neighbor.5 The Assume for the moment that settlements are
settlement at P will be north of the OX axis distributed uniformly along a line with regular
and east of the OY axis by coordinates chosen spacing between them and that they are dis-
from normal distributions. Since it is known tinguishable into two groups with the two
that P is to be measured from 0 in positive groups randomly mixed. If each settlement is
xy coordinates, the y and the x will be nor- examined in turn, p is the chance of obtaining
mally distributed. Because this is homoge- a settlement of a given size range and q =
neous space, the coordinates chosen at random 1 - p is the chance of finding one outside that
will be from the same normal series. Note range. The number of trials to obtain the
now, however, that it is not required that first success, here a city in the given range, is
X2 + y2 = r2, i.e., when x is large y must be pqlO.S Thus the probability distribution of the
small: they could both be large or small. It nearest neighbor distance k between towns of
may be shown that the probability of the the same population size (in Thomas' termi-
distance of the settlement being between Ju nology) is in this case the geometric distribu-
and JL + 8fk from 0 is tion. Here discrete variables are being dealt
with, but it is easy to show that in the con-
P = 2ph2e-A27,2h8
tinuous case the distribution of nearest neigh-
which is a normal series. This result could bor distances is described by the negative
equally well be obtained by realizing that the exponential series, e'k.
convolution of two normal series is itself It is clear that both of these distributions
normal. are linear in their logarithms. Now although
This result only gains significance by reason a uniform arrangement of settlements on a
of Dacey's work on nearest neighbors.6 He line has been used for simplicity of exposition,
has shown that urban places having popula- the results will apply equally well to the
tions of 2,500 or more in the states of Arizona,
settlement arrangements derived previously.
of Connecticut, and of Iowa, areas chosen
However, instead of the regular spacing be-
deliberately because of their apparent con-
tween settlements used above it is necessary
trast, all show the distribution of nearest
to take account of the normal distribution of
neighbors to be closely fitted by a normal
nearest neighbor settlements. Thus the dis-
series. While it is equally true that Dacey's
tribution of nearest neighbor distance between
later more refined methods do allow one to
towns of the same population size will graph
distinguish the degree of randomness, i.e., the
tendency to regularity or to clustering from 7L. F. Richardson, "Atmospheric Diffusion on a
Distance-Neighbour Graph," Proceedings of the
b Levy and Roth, ibid., p. 131. Royal Society of London, Ser. A, Vol. 110 (1926),
6 M. F. Dacey, "Analysis of Map Distributions by pp. 709-37.
Nearest Neighbor Methods," Discussion Paper No. 1. 8W. Feller, An Introduction to Probability Theory
Department of Geography, University of Washington, and its Applications, 2nd ed. (New York, 1950), p.
1958. 156.
This content downloaded from 202.43.95.117 on Tue, 27 Aug 2019 21:39:06 UTC
All use subject to https://about.jstor.org/terms
1964 RANDOM SPATIAL ECONOMY: EXPLORATION IN SETTLEMENT THEORY 141
log normally. Thomas' work on Iowa cities9 negative outcome "for fixed a (0 < a < 1) and
and King's on sample cities of the United n -> oc, the probability that the fraction k/n
States10 give significance to these results. of time spent on the positive side be < a tends
The two previous sections have been intro- to 27r-1sin-a0O-5." What is being done is to
duced solely to demonstrate that considera- make a running summation of the positive and
tions of randomness are relevant to settlement negative results at each trial and then to count
distribution. That immediately adjacent towns the number of subtotals which are positive =
are placed without regard to each other's k. The total number of partial sums = total
position save for maintaining a mean distance number of trials = n. For illustration, take a
apart is not unexpected. The location of one single settlement representing a large number
settlement can be only a weak determinant of of such towns. If the spin of a coin shows
the site of its neighbor. However, numerical heads, the town will add a manufacturing
values of the parameters of the distributions employee. The next addition will be de-
discussed above will be most useful in relating termined by another spin plus the previous
to other aspects of areas. Obviously only the spin. If tails, the subtotal is clearly zero, but
fringes of the problem of settlement distribu- we adopt the convention that a return to zero
tion have been touched, since there may still from the positive side will denote the addition
be considerable pattern or order in the spac- of a manufacturing employee. If the next spin
ing of the total system of towns. is tails, a service (nonmanufacturing) em-
ployee is added and so on. A running mean is
THE ARC-SINE NORMALITY OF URBAN
being formed and it is whether this is positive
MANUFACTURING EMPLOYMENT
or negative which decides what type of em-
King used an arc-sine transformation on the ployee will be added. The proportion of
fraction (persons engaged in manufacturing/ manufacturing (or of service) employees is
total populations) for his sample of towns in very likely to be very high or very low. If
the U.S.A. to obtain the normal distribution a number of these sets of trials are made for
necessary for regression and correlation stud- each town, the proportion of employees in
ies." It is reasonable to believe that the frac- manufacturing will be distributed according
tion mentioned above is a variable representing to the arc-sine law.
the convolution of two independent series, This provides a type of specialization
the normal and that approximated by an arc- process in which what has occurred in the
sine function. past affects the future only to the extent that
That there is apparently an arc-sine com- the past is summed up in the present. Eco-
ponent is surprising and most interesting. nomic structure is likely to change only slowly
Graphed, this is a U-shaped curve implying but it can change and occasionally very
that very few towns have a medium amount quickly. Above all, specialization, at least in
of this component of manufacturing, most the highly aggregated categories considered
towns having a lot or only a little, i.e., there here, may be purely a result of random im-
are specialized manufacturing towns or non- pulses over time. It may be seen that the
manufacturing towns with few mixed types inarc-sine law is not affected by the number of
this sector. The simplest model leading to the trials, provided that at least 20 be made. Thus
arc-sine law is given by Feller.'2 In a series of towns can be of any size. Indeed it is not
n consecutive trials with additive results, each necessary to have equally sized increments
having a 50: 50 chance of a positive or a occurring over time. Erdbs and Kac have
given a more general derivation of this law
9E. N. Thomas, "The Stability of Distance-Popu-
lation-Size Relationships for Iowa Towns from 1900- in which the dichotomous win-lose (Bernoulli)
1950," Proceedings of the I. G. U. Symposium in trials are replaced by drawings from any
Urban Geography (Lund, 1960), ed. K. Norborg independent random series having zero mean
(Lund, 1962), pp. 13-29.
10 L. J. King, "A Multivariate Analysis of the Spac-
and a variance of unity.'3 Nor need the vari-
ing of Urban Settlement in the United States," Annals,
Association of American Geographers, Vol. 51, No. 2 13 P. Erdbs and M. Kac, "On the Number of Posi-
(June, 1961), pp. 222-33. tive Sums of Independent Random Variables," Ameri-
:: L. J. King, ibid. can Mathematical Society Bulletin, Vol. 53, No. 10
12 W. Feller, op. cit., p. 80. (1947), pp. 1011-20.
This content downloaded from 202.43.95.117 on Tue, 27 Aug 2019 21:39:06 UTC
All use subject to https://about.jstor.org/terms
142 LESLIE CURRY March
This content downloaded from 202.43.95.117 on Tue, 27 Aug 2019 21:39:06 UTC
All use subject to https://about.jstor.org/terms
1964 RANDOM SPATIAL ECONOMY: EXPLORATION IN SETTLEMENT THEORY 143
Two rather peculiar features of the arc-sine manufacturing employees they will add to
law are worthy of note. From equation (1), their employment. If the successive ratio in-
the mean number of boundaries crossed varies crements for each town are averaged, a normal
with the extent of the area or rather with the distribution of cities having various ratios of
number of settlements considered. Boundaries manufacturing will result. Both experiment
become less frequent farther out from the and the central limit theorem attest the
diffusion source, implying that manufacturing validity of this remark. The notion here is
or service zones are wider than they are close that no specialization process is involved, but
in. There is greater areal differentiation in that in each period an entirely independent
this component of manufacturing near the drawing uninfluenced by the past or present
source than farther out. Such a characteristic is made. A drawing once made is not affected
would certainly bear investigation if the arc- by future conditions, although there could
sine component can be identified. presumably be random failures. In this way,
The second feature, which Feller terms the the ratio involved need not be related to any
Second Arc Sine Law,'5 is scarcely credible specific characteristics of the city, so that the
but nevertheless follows logically from the type of industries involved appears to be
premises. In words, this states that the prob- those not requiring linkages to other industries,
ability distribution of the length of runs of the so-called "footloose" industries or those
positive sums (or negative sums) and of the catering to local markets for example.
position of maxima are almost the same. The To sum up then, the fraction a/p, (number
maxima referred to are the accumulated posi- of employees in "linked" manufacturing/total
tive or negative signs, i.e., the maximum or population) is distributed according to an
minimum diffusion of manufacturing (or of arc-sine series and the fraction b/p (number
service) towns. Thus, for example, assuming of employees in nonlinked manufacturing/
the diffusion source in the U.S.A. to be in the total population) is normally distributed.
New York-New England area, it is very likely Thus the fraction ab/p2 is distributed accord-
that the greatest concentrations of either ing to the convolution of these two series, i.e.,
manufacturing or of nonmanufacturing ratios arc sine normally.
would occur on the Pacific coast (assuming No great exception can be taken to the
diffusion to be occurring along an east-west specialization process described above and
line), and this without regard to seagoing or since it agrees with King's findings, it can be
to the length of period of development of the accepted until a less aggregated model is
different areas. Equally, the second law im- devised. However, the idea of continuous,
plies that towns will generally show an one-directional diffusion out from a point
increasing degree of manufacturing or non- source may be less digestible. In its defense
manufacturing specialization with time, al- it can be remarked that viewed from within
though any individual case may be decreasing the system the diffusion origin would not nec-
its ratio. essarily be apparent: due to the frictional
The normally distributed component of effect of distance, the influence of the source
manufacturing industry which is being re- is cut to a minute value fairly quickly, and the
garded as separately allocated by an inde- alternating zones of specialization mask it
pendent process will now be examined briefly. further. The one-directional nature of diffu-
Presumably, this distribution could arise in sion is, however, more repugnant; i.e., cities
various ways, so that a very general "explana-lying to the east of a city will affect that city
tion" is offered here in the sense that as few
but they will not be affected by it or by other
assumptions as possible are introduced, in
cities farther west. It is not too difficult to
order not to contradict other more plausible
describe more realistic processes but it has not
models which might be offered. Let there be
been possible to obtain an arc-sine normal
a probability density function, the form of
distribution from them. Perhaps the most
which is not required, from which cities draw
random independent samples a number of promising approach is to have diffusion
times to decide on each occasion the ratio of passing down the urban hierarchy, but not
enough knowledge exists of the branching in-
"5W. Feller, op. cit., p. 83. volved.
This content downloaded from 202.43.95.117 on Tue, 27 Aug 2019 21:39:06 UTC
All use subject to https://about.jstor.org/terms
144 LESLIE CURRY March
work which does not indicate its probabilistic P =Z!/11Zj!. (O <,i <-N)
i=o
basis.'9 Here random behavior is postulated
and concepts are used from information When the system is large, its entropy is
theory and from the study of self-organizing
systems. That appeal is made to the latter H = log P - Z log Z - :Zj log ZA.
needs no defense, since the existence of order Note that entropy is defined as log P.
is evident and no cosmic consciousness has H is maximized when
directed its development. As Vidal de la
Blache says: "Thus men establish new rela- Zi = (Z/n)e-4/n
tionships between scattered features. For the an exponential distribution, n being the mean
random effects of local circumstances they population per settlement N/Z. In terms of
substitute a new and coherent system of inter-the cumulative distribution function
locking forces."20
The results of an unrestrained random proc- Zi'<8= S(I- e /n) ,
ess can be defined as showing zero order. S being the size of the largest city. In this
case, Hmiiax= Z log (en.). The most probab
16 W. L. Garrison and B. J. L. Berry, "Alternate state of the system is that which gives maxi-
Explanations of Urban Rank-Size Relationships," mum entropy, i.e., when the sum of logarithms
Annals, Association of American Geographers, Vol.
48 (1958), pp. 83-911.
is a maximum. This corresponds to a situation
17 H. A. Simon, "On a Class of Skew Distributions in which, given the size of the largest city,
Functions," Biometrika, Vol. 42 (December, 1955;) the probability of the q + 1st city having a
reprinted in Models of Man (New York: John Wiley population which is a given ratio of the qth
and Sons, 1957), pp. 145-64.
city is constant.
18E. N. Thomas, "Additional Comments on Popu-
lation-Size Relationships for Sets of Cities," in
Quantitative Geography (Atherton Press, in press). 21 C. Shannon and W. Weaver, The Mathematical
19 B. J. L. Berry et al., Comparative Studies of Theory of Communication (Urbana, Ill.: Univ. of
Central Place Systems, mimeographed report to the Illinois Press, 1949).
U.S. Office of Naval Research, Geography Branch, 22 H. von Foerster, "On Self-Organizing Systems
1962. and their Environments," in Self-Organizing Systems,
20 P. Vidal de la Blache, Tableau de la Ge'ographie eds. M. C. Yovitts and S. Cameron (New York: Per-
de la France (Paris, 1903). gammon Press, 1960).
This content downloaded from 202.43.95.117 on Tue, 27 Aug 2019 21:39:06 UTC
All use subject to https://about.jstor.org/terms
1964 RANDOM SPATIAL ECONOMY: EXPLORATION IN SETTLEMENT THEORY 145
This content downloaded from 202.43.95.117 on Tue, 27 Aug 2019 21:39:06 UTC
All use subject to https://about.jstor.org/terms
146 LESLIE CURRY March
from the point of view of the locational struc- system and process as Blaut24 and Chorley25
ture as a whole their actions appear as random. have convincingly argued. If the individual
This situation may be described as though the decision maker acting rationally be taken as
places involved were making random deci- the unit to be considered, conclusions such as
sions or were acted on by random forces, "Corn Belt farmers grow corn because they
although this is only a convenient shorthand. live in the Corn Belt" are obtained but are not
That a given set of premises contains logical very helpful. Alternatively, a history of unique
consequences which are in agreement with events is written to describe development and
reality is no guarantee that a model is itself the lawfulness or generality of process is
realistic. Several quite different models may missed. Whether history is lawful is a matter
give the same result. This type of work can for those who write it to decide. That spatial
only be judged over a period of years when processes occur, to which the historian could
not contribute understanding but which are
separate models are being articulated and
the very stuff of geography, appears self-evi-
elaborated to the point where a whole theoret-
dent-or at least a worthy article of faith.
ical structure exists to explain many seemingly
isolated phenomena. In the meantime, cex-
24 J. M. Blaut, "Object and Relationship," Pro-
plorations" are all that are possible with the fessional Geographer, Vol. 14, No. 6 (November,
considerable chance that they lead only up 1962), pp. 1-7.
blind alleys. 25 R. J. Chorley, "Geomorphology and General
Systems Theory," Theoretical Papers in the Hydro-
But it is insisted that such explorations are
logic and Geomorphic Sciences, Geological Survey
necessary, if we are to approach geography as Professional Paper, 500-B (Washington, 1962).
This content downloaded from 202.43.95.117 on Tue, 27 Aug 2019 21:39:06 UTC
All use subject to https://about.jstor.org/terms