SPATIAL INTERACTION PATTERNS
Waldo Tobler
July 1975
Research Reports are publications reporting on the work of the author. Any views or conclusions are those of the author, and do not necessarily reflect those of IIASA.
Contents
I. Introduction 11. Background
111. Algebraic Development
IV. From Vectors to Potentials V. Examples
VI. Caveats
L i s t o f F i g u r e s
E x p o r t a t i o n o f S t u d e n t s by S t a t e , 1968. 11 F i e l d o f A s y m m e t r i c S t u d e n t F l o w s , 1 9 6 8 . 12
W e i g h t e d S t u d e n t Flow F i e l d , 1968. 14
Flow F i e l d Computed f r o m a Random I n t e r a c t i o n T a b l e . 15 I n t e r p o l a t e d S t u d e n t Flow F i e l d , 1968. 17 S c a l a r P o t e n t i a l o f S t u d e n t F l o w s , 1968. 19 V e c t o r P o t e n t i a l o f S t u d e n t F l o w s , 1968. 20 G r a d i e n t F i e l d o f E d u c a t i o n a l E x p e n d i t u r e s , 1968. 2 1
B.A. t o Ph.D. F l o w F i e l d , 1968. 2 2
Ph.D. t o Employment Flow F i e l d , 1968. 23
Commuting F i e l d f o r Munich, 1939. 25
Commuting F i e l d f o r B e l g i u m , 1970. 26
E u r o p e a n T o u r i s t e x c h a n g e s , 1973. 2 7 T e l e p h o n e Flow F i e l d f o r Z u r i c h , 1970. 28
B u s i n e s s C o n t a c t F i e l d f o r Sweden. 29
F l o w s Between P s y c h o l o g i c a l J o u r n a l s . 3 0
Spatial Interaction Patterns Waldo Tobler*
Abstract
An algebraic examination of spatial models leads to the conclusion that a convenient description of the pattern of flows implicit in a geographical interaction
table is obtained by displaying a field of vectors computed from the relative net exchanges. The vector field approxi- mates the gradient of a scalar potential, and this may be invoked to explain the flows. The method can be applied to asymmetrical tables of a non-geographical nature.
I. Introduction
Empirical measurements of the interaction between geo- graphical areas are often and conveniently represented by
"from-to" tables, usually with asymmetric entries. Many mathematical models offered as descriptors of these geo- graphical interaction patterns do not treat this situation adequately. Specifically, they quite frequently can predict only symmetrical interaction tables, a glaring contrast to the empirical observations. In the present essay an attempt is made to overcome this difficulty through the introduction of a flow field, which one may wish to think of as a "wind."
This wind is interpreted as facilitating interaction in par- ticular directions. The algebra allows one to estimate the components of this hypothetical flow field from the empiri- cal interaction tables. Plotting the flow field provides a simple, convenient, and dramatic cartographic representation of the asymmetry of the exchanges, even for extremely large tables of interaction data. A table of county-to-county interactions in the United States, for example,. would yield nearly 10' numbers, an incomprehensible amount. A flow field, on the other hand, showing these data as a set of vectors might be more tractable. Going one step further, it should be possible to infer an estimate of the forcing function, the
"pressure," which might be said to have given rise to the
interaction asymmetry. Data collected for several time periods may allow one to infer the dynamics of the relation of the forcing function and the flows.
*This work was initiated in 1972 under National Science Foundation Grant GS 34070X, "Geographical Patterns of Inter- action", at the Department of Geography, The University of Michigan, Ann Arbor. My appreciation is also extended to the many individuals who provided data matrices and comments.
Each of them will be able to recognize their contributions in the pages which follow.
11
.
BackgroundThe foregoing objectives were motivated by previous
papers [27, 281 in which geographical locations were predicted from empirical interaction data by inverting models which
contained spatial separations as one of the explanatory variables.
The resulting spatial separations were then converted to latitude and longitude positions by a procedure analogous to trilateration, as practiced in geodesy. The empirical
data in each instance were indicators of the amount of inter- action between the locations in question. These interactions were given in the form of matrices, Mij, in which the rows are the "from" places and the columns the destination places.
For example,
if is the amount of migration from place i to place j, then the social gravity model predicts that
Mi j = kP.P.f (d. . )
,
1 3 1 3
where the P's denote the populations, d stands for distance, and k is a scale factor. Then the inversion is
From the adjustment procedures used in surveying one learns how to calculate the latitude and longitude coordinates of positions when
their separations have been measured [31]. A similar pro- cedure has recently been used in psychology [221. The social gravity model of course is symmetrical in the sense that if
= d j i
i j then Mij must equal Mji, and the converse. In prac- tice, however, interaction matrices are asymmetrical, and Mij # Mji. This would imply, if the model is inverted as was done above, that dij # dji, with the consequence that the
trilateration solution can result in more than one geometrical configuration [12], or that the standard errors of the posi- tion determination are increased. In order to overcome this difficulty it is natural to introduce a wind, or current of some type, which facilitates interaction in particular direc- tions. This vector field is to be estimated from the empirical data, and of course reflects their influence. At the moment the wind need not be given any interpretation other than that of a mathematical artifact which allows the problem to be solved. Later we can look for independent evidence which might confirm (or deny) its existence.
111. A l a e b r a i c Development
A s a s i m p l e example, p o s t u l a t e t h a t a t r a v e l e f f o r t ( t i m e , c o s t , e t c . ) ti i s a i d e d by a f l o w
ai
i n t h ed i r e c t i o n of movement f r o m p l a c e i t o p l a c e j . Then w e c a n w r i t e t h a t t i j = d . . / ( r
+
-b c i j ) , where r i s a r a t e o f t r a v e l ,1 3
i n d e p e n d e n t o f p o s i t i o n a n d of d i r e c t i o n , and i s i n t h e same u n i t s a s -b c. An i n t e r p r e t a t i o n m i g h t b e t h a t t i j i s t r a v e l t i m e f o r someone rowing on a l a k e , r i s t h e rowing s p e e d i n m e t e r s p e r s e c o n d , and c -b i s a c u r r e n t i n t h e w a t e r ; o r t h a t j i s u p h i l l from i , and t h a t t h i s r e s u l t s i n a
d i f f e r e n c e i n t r a v e l s p e e d ; o r t h a t t h e r e e x i s t s a g r a i n , a s when s k i n s a r e p l a c e d u n d e r n e a t h s k i s , and movement i s r e n d e r e d e a s i e r i n o n e d i r e c t i o n . Whatever t h e i n t e r p r e t a t i o n , s o l v i n g f o r
a
one o b t a i n s-b -b
Here u s e h a s b e e n made o f t h e r e l a t i o n c i j =
-
c j i which m u s t h o l d f o r c u r r e n t s , a n d of t j i = d j i / ( r+
-b c j i ) .The same a r g u m e n t c a n b e a p p l i e d t o t h e g r a v i t y model.
- 1
S u b s t i t u t e t f o r d i n t h a t model, w i t h f ( t ) = t i j f o r s i m - p l i c i t y ,
P . P P . P .
= k A = k k ( r + :
,
Mi j
ti j di j i j
and
S o l v i n g f o r -b c i j , r e c a l l i n g t h a t d i j
-
- d j i , o n e f i n d s- c o n v e n i e n t c h o i c e o f u n i t s w i l l make r E 1 and t h e n
It is encouraging that this quantity has already been found useful in studies of migration [21], albeit without the present derivation. The original objective, inversion of the model, follows immediately:
Reversing i and j does not change this quantity. Algebra- ically
which is the same result as would have been obtained if
had been assumed initially. A solution to the original problem has thus been achieved, in the sense that any
asymmetric interaction table can be made to yield a unique distance estimate to be used in further computations involv- ing locations.
Two difficulties remain. First, only one interaction model has been examined. Secondly, can a reasonable inter- pretation be provided for the
zij
when the interaction con- sists of, say, telephone calls between exchanges?In the first instance, a more general gravity model might be written using
Mi.
+
M.i P.P2 1 = k A , a di j
the development of which is straightforward. In a similar vein, an exponential model
- b d . . / ( r + g
I
1 I i j Mij = k(Pi
+
P.) e3
yields
-P
In M S i
-
In Mi.'ij = In Mij
+
In Mji-
2 i n k (Pi+
P.)3
and this is a much more complicated result. One could con- tinue further by, for example, considering the entropy
model elaborated by Wilson, or the migration model published by Lowry [30]. These models are in fact already more general in that they do yield asymmetrical interaction tables, but they also require supplementary information before one can solve for the distances. The Lowry model is
u.
W . P.P Mi j = k 2 l l j ,u. wi
7
where U is related to unemployment and W to wages. This can be rewritten as
and implies, if dij = dji, that
and also (solve for a and substitute) that j
Thus the second half of the interaction table carries no information. Furthermore, if the distances are known, one can infer the wage and unemployment ratios from the empiri- cal migration data. Such a result has recently also been achieved for another model by Cordey-Hayes [ 4 1 . In the pre- sent instance there are n(n
-
1 ) / 2 equations of the formand n unknowns, the a i l s The system is overdetermined unless some of the equations can be shown to be dependent.
In a comparable manner, given only an empirical inter- n
action table, then the row sums 0 =
1
Mij and column sumsn i=l a
D =
1
Mij can all be computed. The simplest sort of model j i=1is then that Mij = kO .D. f (d. . )
,
and Mji = kO .D. f (d.. ) where1 3 1 3 3 1 1 1
the origin and destination s&s now take the place of the populations. In order to obtain a consistent value for
- - -
dij
-
dji it is necessary that M../Mji = O.D./ojDi, and this1 3 1 3
is a hypothesis which can be tested.
Another interesting model has been proposed by Somermeijer [23]. This is
Here Qij is the difference in attractiveness between areas i and j. Qij = Aj
-
Ai. Clearly Qij =-
Qji; solving for this quantity usingand adding, then subtracting, the equations for M ij and Mji, one finds
M m i
-
Mi.Qij
=.(
b~i~
+ MjiThis is a very interesting relation because, although there is much speculation in the literature, no one really knows by how much areas differ in attractivity. The model allows an estimate to be made of this quantity. One notices that Qii = 0, although usually Mii # 0, and a desirable property would be that Qi
= Qik
+ a
for all i, j,
k. In this case-
the attractivity of area j, call it A;, would simply be
>
A = Ai + Qij for some base level Ai. If this relation does not hold for all i and j then an approximate estimator must j be devised, which does not appear difficult. One may then wish to draw contour maps of the scalar field A(x,y), on the assumption that attractivity is a continuous variable.
Solving for distances in this model leads to
where r = a/b. This is remarkably similar to the equations obtained earlier.
A few interaction models have now been examined, and manipulated algebraically under varying assumptions. The
reader will hopefully find it fruitful to extend, and improve, these results. The class of hierarchical models [ 5 1 which
also might be used to approach interaction tables has been completely neglected here.
From a mathematical point of view it is an elementary theorem of matrix algebra that every asymmetric matrix can
be written as the sum of a skew-symmetric matrix and a symmetric matrix. This unique decomposition is given by
and
- - -
Mij
-
(Mij-
M.. )/2 = - Mji,
3 1
where M denotes the symmetrical, and M- the skew-symmetrical,
+
portion of the table. If the usual social gravity model is written as Mij = 'ji = kP
.
P ./d: j, then this implies thata 1 3
dij = 2 k P . p . / ~ i ~ , taking the symmetry literally. If this is
1 3
set equal to- the- comparable value
obtained from the Somermeijer model, then one finds that r = 2 is required for the two results to be consistent. The similarity of the answers obtained under differing assump- tions encourages one to believe in the robustness of the results. The recurrent appearance of the difference of the interactions in the two directions divided by their sum, M-/M+ the relative net interaction, is particularly striking.
This quantity first appeared as being proportional to the + c's introduced as currents.
'
i j was assumed to be the component of a current (or wind) flowing from i to j and which made interaction easier
in that direction. If the locations, in geographical two- space, of the positions of i and j are known (or estimated using the trilateration procedure from dij, as defined
by one of the above equations), then we can draw a small vec- tor at i towards j of length $ c
i-j' The direction at i away
-f -f -f
from j is used when c.. is negative. Since cij = -c
ji' one
1:
half of the vector magnitude is assigned to each of the points i and j
.
This happens also to make r = 2. It may be appropriate to weight the vectors by a quantity proportional to e-dij,
but this is a side issue. Analytically the calcu- lation of the vector components is a simple trigonometric computation if the latitude and longitude coordinates of i and j are known. Doing this for all directions which inter- act with each point, i.e., for each i performing the compu- tation for all j, one obtains a cluster of vectors at each point. The resultant vector sum gives an estimate of the wind field at that point. After the summation has been per-formed for all points, a vector field c(x,y) can be assumed + to have been defined for all x,y in the region of obser- vations.
IV. From Vectors to Potentials
Every scalar field a k f y ) has associated with it a vector field, grad a. The converse, however, is not true.
Nevertheless every vector field can be written as the sum of the gradient of a scalar field plus an additional vector field. These two parts are referred to as the scalar poten- tial and the vector potential, respectively. If the second field is everywhere zero, then, and only then, the original vector field is the gradient of some scalar field. In the latter case one should be able to recover this scalar poten- tial by reversing the gradient operation, i-e., by inte- gration. In the present instance it is necessary to decide whether the finite set of numbers, making up the vector field and obtained from the empirical interaction table, can be considered an exact differential [19]. The idea here is that a wind implies a potential function (the attractiveness), and we would like to infer this potential from the wind.
The observed vector field must be decomposed into diver- gence- and curl-free parts, and the scalar and vector poten- tials can then be calculated as follows [ 1 7 , 201. Recall that, for any vector field ;fc(x,y), the vector identities curl
+ -f
grad = 0 and div curl = 0 both hold. Write c = grad a
+
curl vo r c + = a c u r l f r e e p a r t
+
a d i v e r g e n c e f r e e p a r t , and t h e na p p l y t h e d i v e r g e n c e o p e r a t o r , t o o b t a i n d i v c + = d i v g r a d a
+
0.~ u t d i v g r a d =
v 2
=+ % .
Thus, by c a l c u l a t i n g t h e d i v e r - aygence of o u r v e c t o r f i e l d we o b s e r v e t h a t t h e s c a l a r f i e l d a can be o b t a i n e d from P o i s s o n ' s e q u a t i o n
which i s s o l v a b l e by known methods [ 2 ] t o o b t a i n an e s t i m a t e of t h e s c a l a r p o t e n t i a l .
The f i r s t p a r t of t h e problem i s t h u s r e s o l v e d . Now a p p l y t h e c u r l o p e r a t o r t o t h e o r i g i n a l e q u a t i o n , o b t a i n i n g
+ +
c u r l c = 0
+
c u r l curl v. I f c u r lG
i s a v e c t o r i n t h e x , y p l a n e t h e nG
and c u r l c u r lG
are p e r p e n d i c u l a r t o i t , and t h u s c h a s no components, n o r any v a r i a t i o n , normal t o t h e -b-+ ; ac ac
x,y p l a n e . T h e r e f o r e , c u r l c =- i (=
-
-1, and a n a l o g o u s l y-b ay
f o r c u r l c u r l v. Thus
i . e . , P o i s s o n ' s e q u a t i o n i s a g a i n t o be s o l v e d , t h i s t i m e f o r v -+ = Zv.
I t i s n e c e s s a r y t o s o l v e t h e s e e q u a t i o n s by f i n i t e d i f f e r e n c e methods a t a n i r r e g u l a r s c a t t e r of p o i n t s i n two dimensions. Assuming f o r t h e moment t h a t t h i s c a n b e done, t h e r e remains t h e problem of i n t e r p r e t a t i o n . The s c a l a r f i e l d a i s r e a d i l y viewed a s a " p r e s s u r e " which i n d u c e s t h e flow.
I would e x p e c t t h i s t o b e t h e l a r g e r of t h e two components of t h e o b s e r v e d f i e l d . The most r e a s o n a b l e i n t e r p r e t a t i o n which I c a n see f o r t h e v e c t o r p o t e n t i a l i s t h a t of a s p a t i a l impedance. I n o t h e r words, t h e r e i s a mismatch o r incon- s i s t e n c y between t h e f l o w f i e l d and t h e f o r c i n g f u n c t i o n . T h i s can p e r h a p s b e t h o u g h t of a s a v i s c o s i t y , o r as a n un- e x p l a i n e d component i n t h e s t a t i s t i c a l s e n s e o f a r e s i d u a l . I t i s e a s y , t o o e a s y , t o i n v e n t e x p l a n a t i o n s f o r t h e m i s - match i n d i c a t e d by t h e v e c t o r p o t e n t i a l . But suppose t h a t one h a s o b s e r v a t i o n s ( i . e . , i n t e r a c t i o n t a b l e s ) f o r two, o r more, t i m e p e r i o d s . I n t h i s dynamic s i t u a t i o n one would l i k e t o c o n s i d e r t h e e x i s t e n c e of l a g g e d p o t e n t i a l s . One would i n f a c t l i k e t o c a l c u l a t e t h e s e from t h e d a t a [ 1 3 ] . U l t i n s t e l y one would l i k e t o c o n s i d e r p o l i c y changes which
could be used to modify the flow field. One foresees a'
-
spatially continuous, temporally dynamic input-output scheme [I11
Other questions come to mind readily. If there is a flow of interactions in one direction, for example, must these be balanced by a counterflow of some other quantity in order to close the system: money when products flow, or
decongestion (negative population density) when people move?
The interpretation of interaction asymmetries as being in- duced by a "wind" may also be reversed. Recall that in
~agerstrand-type Monte Carlo diffusion studies [9] the mean information field is usually symmetrical. But diffusion may also take place in a field of winds [16], and this is a natural way of obtaining asymmetrical information fields.
Thus explicitly introducing these "winds" may improve the accuracy of predictions obtained from spatial interaction models. Included here are models of the gravity type as modified in this presentation and widely used in practice,
as well as others not treated here.
V. Examples
Several sets of data have been examined in order to
clarify the foregoing concepts. The primary illustration uses a 48 by 48 table of state to state college attendance [14].
The accompanying map, Figure 1, indicates by a plus sign those states which are sources (net exporters) of students;
a minus sign indicates student importers, with the area of the symbol proportional to the volume of import or export.
Exportation here means that college-age residents of a state go to another state for their college education. One could have reversed the interpretation to say that some states export education to non-residents; but the students actu- ally move, and this has been labelled an exportation. New Jersey was, in 1968, by far the largest exporting state by this criterion, with 90,000 students studying in other states.
This state, compared to its educational facilities, had a surplus of educationally motivated residents.
'
The computed vector field is shown in the next illustration, Figure 2.The vectors have been positioned at the approximate center of gravity of their resolution elements (states), which introduces an error of less than one half of the resolution element. he components of the vector are computed as
'I£ one separates the 48 by 48 college attendance table into symmetrical and skew-symmetrical parts, then the latter portion contains 24% of the total variance.
where
The vector ci is plotted at the map location (Xi Yi), -b
after an arbitrary scaling appro9riate to the particular map, using plane coordinates on a local map projection [ 2 5 1 . The foregoing formula is derived by simple trigonometry from the interpretation of ci as the components of a vector bound at -f
i, directed from i to j.
The next figure, Figure 3, shows a spatially weighted version of the same data, computed as
with
[(X
-
Xi), (Y-
Yi)]j j
The motivation here is that local influences should carry more weight than distant ones. But the difference.between the figures is comparatively minor, and this seems to be the case for other examples which have been computed. As a contrast it is additionally useful to display a map,
Figure 4, constructed from a 48 by 48 table of non-negative random numbers. The pattern computed from this random table lacks structure, the degree of spatial autocorrelation with- in the vector field being small. This does not appear to be the case with maps computed from real interaction data.
Using the forty-eight vectors located at state centroids one can estimate values at a regular spatial lattice to obtain a somewhat more legible picture of the vector field, Figure 5.
In the present instance this spatial interpolation captures better than 85% of the pattern of the irregular field. The positioning of the lattice is rather crude, with several vectors falling outside the United States; this is of course
not meaningful since there are no observations in that region.
But the main purpose of the assignment of values to the
lattice positions is to facilitate the solution of Poisson's equation by finite difference methods. While this could in principle be done for data arrayed at state centroids, the additional computer programming effort did not seem warranted at this stage. A program which iterates by finite differences for data at a regular array of points seemed adequate as an initial attempt. Both the scalar and vector potential functions have been calculated by such a program, which also plots
contour maps of these functions. The boundary conditions are of course that no movement can cross the border of the region, since the system of student flows is closed as far as these data are concerned.
One might argue that the potentials should have been computed from a field based on net flows, rather than on
relative net flows. aut if one examines the relation obtained from the Somermeijer model
M .
-
Mi.= A - A i =
c (
li*ij j b M~~
+ Mji I)
and postulates the existence of a scalar function ~ ( x , y ) , then a finite approximation to the directional derivative based on observations at i and j would be
But the directional derivative in any direction is the component of the gradient in that direction, and thus the gradient is a linear combination of directional derivatives.
The, f ~ r m u l a used to compute the components
Fi
of theCI
vector field is just such a linear combinationL. Thus the
*1n principle the gradient can be computed from the basis formed by two independent directional derivatives.
But which two? It is easily possible that one can derive a better approximation to the gradient than the averaged value used here. The problem is one of finding the tangent plane to a surface from finite measurements. The essential argument of this essay is not dependent on the particular approximating equation used, although the point requires further clarification.
Somermeijer model s u g g e s t s t h a t t h e r e l a t i v e n e t i n t e r a c t i o n i s t h e more a p p r o p r i a t e e x p r e s s i o n .
S u p e r i m p o s i t i o n of t h e map of t h e v e c t o r f i e l d on t h a t of t h e s c a l a r p o t e n t i a l , F i g u r e 6 , shows t h e d e g r e e t o which t h e f i e l d c o r r e s p o n d s t o t h e g r a d i e n t of t h e s c a l a r f u n c t i o n . P e r f e c t agreement would h o l d i f t h e v e c t o r s were everywhere o r t h o g o n a l t o t h e c o n t o u r s . T h i s o f c o u r s e would imply t h a t c u r l v + = 0 , a s p r e v i o u s l y n o t e d . Beckmann [ I ] , s e v e r a l y e a r s ago, a s s e r t e d t h a t f l o w s would be p r o p o r t i o n a l t o g r a d i e n t s of some s c a l a r f u n c t i o n , and a c r u d e t e s t of t h i s a s s e r t i o n i s now p r o v i d e d . The map of t h e v e c t o r p o t e n t i a l c a l c u l a t e d as a residuum i s a l s o shown, F i g u r e 7.
The most n a t u r a l h y p o t h e s i s f o r t h e movement of s t u d e n t s i s t o p o s t u l a t e t h a t t h e computed scalar p o t e n t i a l , F i g u r e 6 , i s r e l a t e d t o t h e d e p a r t u r e , by s t a t e , from t h e n a t i o n a l e d u c a t i o n a l e x p e n d i t u r e , p e r c a p i t a . T h i s h y p o t h e s i s c a n be approached i n s e v e r a l ways. The t a c k u s e d h e r e i s t o compute a n e x p e c t e d s t u d e n t - f l o w f i e l d . An e s t i m a t e o f t h e 1968 p e r c a p i t a e x p e n d i t u r e f o r h i g h e r e d u c a t i o n a l p l a n t , by s t a t e , i s r e a d i l y a v a i l a b l e [ 7 ] . On t h i s b a s i s , l e t t i n g Ai r e p r e s e n t t h e e d u c a t i o n a l e x p e n d i t u r e i n s t a t e i and d i j t h e d i s t a n c e (km) o f s t a t e i from s t a t e j , t h e q u a n t i t y
w a s formed a t e a c h i as a v e c t o r i n t h e d i r e c t i o n 8, t o w a r d s j. The r e s u l t a n t of t h e s e v e r a l v e c t o r s a t e a c h p o i n t w a s t h e n used t o o b t a i n a " c o n t i n u o u s " v e c t o r f i e l d , F i g u r e 8.
T h i s f i e l d , i f o u r h y p o t h e s i s h a s any m e r i t , s h o u l d a g r e e w i t h t h e one c a l c u l a t e d i n d e p e n d e n t l y from t h e s t u d e n t f l o w t a b l e . But t h e map which i s o f f e r e d i n e v i d e n c e o n l y
m o d e s t l y r e s e m b l e s t h e f i e l d of s t u d e n t m i g r a t i o n s . The
c o r r e l a t i o n (R') between a n u m e r i c a l estimate of t h e a t t r a c - t i v i t y ( s c a l a r p o t e n t i a l ) and t h i s v a r i a b l e ( e d u c a t i o n a l e x p e n d i t u r e ) i s o n l y 70 p e r c e n t . Thus, on two c o u n t s , one i s f o r c e d t o c o n j e c t u r e a d d i t i o n a l r e l a t i o n s h i p s . The p o i n t i s t h a t t h e map o f t h e s c a l a r p o t e n t i a l may of i t s e l f s u g g e s t l i k e l y r e l a t i o n s h i p s , p a r t i c u l a r l y i n s i t u a t i o n s more
o b s c u r e t h a n s t u d e n t f l o w s . A f l o w f i e l d computed from em- p i r i c a l d a t a might a l s o b e compared t o a f l o w f i e l d ob- t a i n e d from an i n t e r a c t i o n t a b l e g e n e r a t e d by some model;
t h e d i f f e r e n c e s i n t h e f i e l d s may t h e n a g a i n l e a d t o i n s i g h t s . The n e x t two maps, F i g u r e s 9 and 10, are of r e l a t e d co- h o r t s , and are b a s e d on t a b l e s [ 6 ] showing t h e number o f
p e r s o n s r e c e i v i n g a B a c h e l o r ' s d e g r e e i n s t a t e i who r e c e i v e d a Ph.D. d e g r e e i n s t a t e j , and t h e number o f p e r s o n s r e c e i v i n g
their Ph.D. degree in state i who took their first post- Ph.D.-degree employment in state j. The latter might be considered as a type of 'brain drain". All three sets of data are best considered together, even though they do not trace the paths of specific individuals. A number of inter- esting investigations are suggested by these tables, in-
cluding of course the structure of the transition probability matrices. A somewhat comparable effect is obtained by corn- paring the three vector fields.
It should be clear that the technique described can be applied to a variety of data. Interregional commodity flow tables collected for input/output analyses constitute one accessible source [18]. Migration tables are also of interest
[26]. Here it becomes obvious that, since people move for different reasons, a potential function calculated from an asymmetric migration table must be the sum of individual potential functions. Thus it would be advisable to dis- aggregate such tables by age, by occupational group, and perhaps by other categories. As related examples, Figure 1 1 shows the 1939 commuting pattern (journey-to-work) for the city of Munich, and another, Figure 12, shows the December 1970 commuting field in Belgium [8]. The pattern of commuting in Munich is very clear, and is as one expects. Belgium, due to the larger areal coverage, has several centers of employ- ment and presents a much more mixed pattern. Figure 13 re- presents tourist travel within that portion of Europe for which there are data [I51 and this again shows a rather coherent pattern.
The next two maps, Figures 14 and 15, show the flow of information. In the one case the field is computed from a 17 by 17 table [lo] of the maximum number of telephone calls from one exchange to another in the city of Zurich in 1970.
One is tempted to place a vortex just west of Parade Platz, where important banks are located. The second figure shows business contacts in Sweden, estimated as an origin-desti- nation table obtained from a sample of airline passengers
[29]. The major Swedish cities stand out clearly.
As a final, non-geographical application, the asymmetry between psychological journal citations has been analyzed.
The first step is to position the journals spatially. This could be done using one of the formulae derived earlier, or by treating the elements of
MI
as dissimilarities. In the-
present instance the data have already been located in a two- space by a multidimensional scaling algorithm [3] and it
only remains to use the asymmetries to assign vectors to
these data points. Interestingly the flow, Figure 16, is from the experimental journal to the clinical journal, but the vector field does not match the gradient of the scalar
M U N I C H
COMMUTING F I E L D , 1 9 3 9
F I G U R E 11
COMMUTING F I E L D F O R D E C E M B E R 1 9 7 0 , B E L G I U M
F I G U R E 1 2
T O U R I S T I C EXCHANGE BETWEEN SOME OECD C O U N T R I E S , 1 9 7 3 .
F I G U R E 1 3
F I G U R E 14
/ /
/ /
00 /
/
/ 0 /0
/
- -
/'
J E d P*
J C P P cl
P k a / /
0
//
4
" / ,
J A P 4 /0
/
/
I \ I
I I I I 1 I
I /F L O W S B E T W E E N P S Y C H O L O G I C A L J O U R N A L S
F I G U R E 1 6
p o t e n t i a l v e r y w e l l . The f u r t h e r t a s k o f i n t e r p r e t i n g t h i s d e s c r i p t i o n o f t h e p s y c h o l o g i c a l l i t e r a t u r e i s b e s t l e f t t o t h a t p r o f e s s i o n , and t h e sample s i z e i s s m a l l . But t h e p r o - c e d u r e c a n b e a p p l i e d t o any a s y m m e t r i c a l t a b l e .
V I . C a v e a t s
One of t h e r e a s o n s f o r i n c l u d i n g s o many examples i n t h i s p a p e r i s b e c a u s e t h e t e c h n i q u e works w i t h any s e t of d a t a . One c a n t a k e any asymmetrical t a b l e and from i t com- p u t e a v e c t o r map. The p r o c e d u r e c a n n o t f a i l , which immedi- a t e l y makes it somewhat s u s p e c t . F u r t h e r m o r e , t h e p o t e n t i a l s c a n n e v e r a c t u a l l y b e o b s e r v e d , b u t c a n o n l y b e deduced from t h e i r c o n s e q u e n c e s . Thus one c a n a s k w h e t h e r t h e p o t e n t i a l s s h o u l d b e computed from t h e r e l a t i v e n e t f l o w s ( a s h a s b e e n done h e r e ) , o r d i r e c t l y from t h e n e t f l o w s (which i n t u i t i v e l y seems more r e a s o n a b l e ) , o f from a t a b l e a d j u s t e d s o t h a t t h e rows sum t o u n i t y ( a s i n a p r o b a b i l i t y t r a n s i t i o n t a b l e ) , o r i n some o t h e r manner. The r e s u l t s w i l l i n e a c h case d i f f e r ,
a l b e i t o n l y s l i g h t l y . The p l a u s i b i l i t y o f t h e s e v e r a l example f i e l d s , t h e r e l a t i o n t o t h e Somermeijer model, and t h e l a c k of p a t t e r n i n a v e c t o r f i e l d computed from a random t a b l e , a l l combine t o s u g g e s t ( p e r h a p s more f o r t u i t o u s l y t h a n a l g e b r a i c a l l y ) t h a t t h e t e c h n i q u e h a s some m e r i t . Thus, s e v e r a l v e c t o r f i e l d p a t t e r n s , computed from a number o f
e m p i r i c a l i n t e r a c t i o n t a b l e s 3 , h a v e b e e n examined. I n
e a c h case t h e s e p a t t e r n s h a v e seemed p l a u s i b l e . T h i s means t h a t o n e c o u l d r e a s o n a b l y e x p e c t t h e o b s e r v e d p a t t e r n s on t h e b a s i s o f a p r i o r i knowledge of t h e phenomena a n a l y z e d .
The c o n c l u s i o n i s t h e r e f o r e t h a t t h e method i s v a l i d . But t h e e x i s t e n c e of a p o t e n t i a l f u n c t i o n i s a l r e a d y c o n t a i n e d i n t h e a s s u m p t i o n o f g r a v i t y model, and t h e domain o f
v a l i d i t y o f s u c h models i s r e s t r i c t e d t o m a c r o s c o p i c e f f e c t s . I t h a s f u r t h e r m o r e b e e n c o n v e n i e n t t o c o n s i d e r s p a c e , f i e l d s , a n d p o t e n t i a l s a s c o n t i n u o u s e n t i t i e s . The p r o x i m a t e n a t u r e o f t h e s e a s s u m p t i o n s s h o u l d b e k e p t i n mind i n any a p p l i c a t i o n s . On t h e o t h e r hand t h e method a p p e a r s t o b e i n s e n s i t i v e t o a l t e r n a t e r e g i o n a l i z a t i o n s o f t h e d a t a , and y i e l d s more d e t a i l e d r e s u l t s t h e f i n e r t h e g e o g r a p h i c a l r e s o l u t i o n o f t h e i n i t i a l i n t e r a c t i o n t a b l e .
5 (whose a c c u r a c y , i n c i d e n t a l l y , i s n o t known [ 2 4 ] .)
References
[I] Beckmann, M. "On the Equilibrium Distribution of Population in Space," Bulletin of Mathematical Biophysics,
19
(1957), 81-90; also seeW. Warntz, Macro Geography and Income Fronts,
Philadelphia, Regional Science Research Institute, 1965.
[2] Carnahan, B., Luther, H., and Wilkes, J. Applied Numeric'al Methods, New York, J. Wiley, 1969, pp. 429-530.
[3] Coombs, C., Dawes, J., and Tversky, A. Mathematical P ~ y ~ h ~ l ~ g y , Englewood Cliffs, Preiltice Hall, 1970, pp. 73-74.
[4] Cordey-Hayes, M. "On the feasibility of simulating the relationship between regional imbalance and city growth," pp. 170-195 of E. Cripps, ed., Space-Time Concepts in Urban and Regional Models, London, Pion,
1974, p. 193; see also: J.H. Ross, A Measure of Site Attraction, Ph.D. thesis, Department of
Geography, University of Western Ontario, London, Canada, 1972, p. 27 et seq; also see: F. Cesario,
" ~ ~ e n e r a l i z e d ~ r i p Distribution Model," Journal of Regional Science,
-
13, 2 (1973), 233-247.[5] Dacey, M., and Nystuen, J. "A Graph Theory Interpretation of Nodal Regions," Papers, Regional Science Assn., - 7 (1961),
pp. 29-42; M. Harvey, J. Auwerter, "Derivation and Decomposition of Hierarchies from Interaction Data,'' Discussion Paper 31, Department of Geo- graphy, Ohio State University, Columbus, Ohio, 1973; P.B. Slater, "College Student Mobility in the United States in Fall 1968: Graph-Theoretic Structuring of Transaction Flows, Proceedings, Social Statistics Section, Am. Stat. Assn., IX, 3 (1974)
,
pp. 440-446.[6] Provided by Professor W. Dekan of the University of Wisconsin, White Water, Wisconsin.
[71 Department of Commerce. Statistical Abstract of the United States, 1972, Washington D.C., G.P.O., pp. 130-132.
Fliedner, D. "Zyklonale Tendenzen bei ~evslkerungs- und Verkehrsbewegungen in stadtischen Bereichen untersucht am Beispiel der Stsate Gsttingen,
~ G n c h e n , und Osnabruck," Neues Archiv fur Niedersachsen, - 10 (15) 4 (April 1965), 277- 294, Table 2, p. 281; Belgium census commuting data provided by Mr. Frans Willekens of IIASA.
Hagerstrand, T. "A Monte Carlo Approach to Diffusion,"
in R . Berry & D. Marble, eds., Spati-a1 Analysis, New York, Prentice Hall, 1968.
Data provided by Prcfessor A. Kilchenmann, Geographisches Institut, Technische Universitat Karlsruhe, FRG.
Leontief, W., and Strout, A. "Multiregional Input Output Analysis," pp. 119-150, in T. Barna (ed. 1 ,
Structural Interdependence and Economic Development London, Macmillan, 1963.
Lingoes, J. The Guttman-Lingoes Non Metric Program
Series, Ann Arbor, Mathesis Press, 1973, pp. 81-111.
McQuistan, R. Scalar and Vector Fields, New York, J. Wiley, 1965, pp. 292-305.
National Center for Educational Statistics. Residence .---- and Migration of College Students: Basic State-to- State Matrix Tables, United States Department of Health, Education, and Welfare, Washington D.C., Government Printing Office, 1970.
OECD, "Nights Spent by Foreign Tourist-s," Tourism Policy and International Tourism in OECD Member Countries, 1973, OECD, Paris, 1974, pp. 95-141.
Pasquill, F. Atmospheric Diifusion: the Dispersion
-
of Windborne Material from Industrial and other Sources, -- New York, Van Nostrand, 1962.Pollock, H.N. "On the Separation of an Arbitrary Two Component Vector into Divergence Free and Curl Free Parts, and the Determination of the Scalar and Vector Potentials," mimeo, Ann Arbor, 1973,
~ P P -
Rodgers, J.M. State Estimates of Interregional
Commodity Trade, 1963, Cambridge, Lexington Heath Books, 1973, pp. 447. For examples see: W. Tobler,
"Commodity Fields," Internal paper, International Institute for Applied Systems Analysis, Laxenburg, March 1975.
Salas, S., and Hille, E. Calculus, Part 11, Toronto Xerox College publishing, 1969, pp. 659-666.
Schey, H.M. Div, Grad, Curl, and All That, New York, Norton, 1973.
Schwind. P. Miaration and Reaional Develo~ment in the united ~tites: 1950-1966, Chicago, ~ h i v e r s i t ~ of Chicaqo, Department of Geosraphv, Research P a ~ e r
1933, -1971 ;-W. Borejko, "~tud)
Gf
the ~ffective- ness of Migrations," Geographia Polonica, - 14(1968), 305-312.
Shepard, R. "The Analysis of Proximities," Psychometrika, 27 (1962), 219-46.
-
Somermeijer, W. "Multi-Polar Human Flow Models," Papers, Regional Science Assn., XXVI (1971), pp. 131-144.
Thompson, D. "Spatial Interaction Data," Annals,
Assn. Am. Geographics, - 64, 4 (Dec. 1974), 560-575.
Tobler, W. "Local Map Projections," The American Cartographer, - 1, No. 1 (1974), 51-62.
Tobler, W. "Migration Fields," Internal paper, International Institute for Applied Systems Analysis, Laxenburg, March 1975.
Tobler, W., Mielke, H., and Detwyler T. "Geobotanical Distance Between New Zealand and Neighboring Islands," Bioscience, 20, 9 (1970), 537-542.
Tobler, W., and Wineberg, S. "A Cappadocian Speculation,"
Nature, - 231, 5297 (7 May 1971), 39-42.
Tornqvist, G. Contact Systems and Regional Development, Gleerup, Lund, 1970, p. 77.
[30] Wilson, A. Urban and Regional Models in Geography and Plannin
_
g , New York, J. Wiley, 1974; I. Lowry, Mlgratlon and Metropolitan Growth, San Francisco, Chandler, 1966.[31] Wolf, P. "Horizontal Position Adjustment, 'I Surveying and Mapping, XXIX, 4 (1969), 635-44.
