Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

GEOGRAPHICA 73

MARCHAND, B. (1977) Los Angeles 1940-1970. Transformation d’une structure urbaine, These pour le Doctorat
d’Etat, Universitk de Paris I.
SNYDER, C. and LAW, H. (1979) ‘Three-mode common factor analysis: procedure and computer programs,’
Multivariate Behavioral Research, 14: 435-41.
TAYLOR, P.J. and PARKES, D.N. (1975). ‘A Kantian view of the city: a factorial ecology experiment on space and
time,’ Environment and Planning A , 7: 671-88.
TUCKER, L. (1966) ‘Some mathematical notes on three-mode factor analysis,’ Psychometric, 31: 279-31 I .
WALSH,J.A. (1964) ‘An IBM 709 program for factor analyzing three-mode matrices,’ Educational and Psycho-
logical Measurement, 24: 669-73.
WALSH,J.A. and WALSH,R.A. (1976) ‘A revised Fortran IV program for three-mode factor analysis,’ Educa-
tional and Psychological Measurement, 36: 169-70.

[ A N D RLANGLOIS,
~
Universite‘ d Ottawa]

THE GRAVITY MODEL AND HIERARCHICAL SPATIAL


SYSTEMS: THEORETICAL AND
EMPIRICAL OBSERVATIONS

The gravity model in its basic forms postulates that the volume of interaction between two places will
be positively related to the populations (masses) of the two places and inversely related to their
distance apart. Although it is not strictly speaking a hierarchical model, it is often used to explain
flows among places that constitute a hierarchical spatial system. This paper is specifically concerned
with the effect of hierarchical spatial structure on the performance of the gravity model. Accordingly,
I will examine first the effect of hierarchical structure on the volume of interaction between places and
second the degree to which the gravity model incorporates these hierarchical effects. Third,
suggestions will be made for the improvement of the gravity model’s performance by an explicit
incorporation of a hierarchical element. The performance of the gravity model will be examined for
both one-way and two-way flows between places.
Although, in the last decade or so, a number of sophisticated spatial-interaction models have been
developed (Openshaw, 1976; Wilson, 1970, 1971, and 1974), and although there has been some
controversy about the gravity model, only the basic or classical forms of the gravity model will be
considered here. This suits the purpose of illustrating the degree to which gravity-type interaction
models describe the hierarchical structure of spatial flows, not only because the classical model is
adequate from an analytical point of view but also because it is widely understood by geographers and
other social scientists. Recent debate about the gravity model’s theoretical foundations (Curry, 1972;
Senior, 1979; Wilson, 1970), and the relationship between the map pattern (the prevailing spatial
structure of origins and destinations) and the distance-decay parameter (Cliff et al, 1974, 1975, and
1976; Curry et al, 1975; Johnston, 1973, 1975, and 1976), are not relevant to my research objective
and will not be considered. With respect to the map-pattern problem, the important effect of spatial
structure on interaction has been demonstrated empirically (Griffith and Jones, 1980; Haining, 1982)
but the focus in this study is not on the distance-decay parameter, so it is unnecessary to separate the
map-pattern and friction-of-distance effects. In any event, further work is required before geometric
patterns can be explicitly incorporated into interaction models (Griffith and Jones, 1980). Finally,
reference may be made to Goodchild and Kwan (1978), who used a three-dimensional model to
describe the hierarchical structure of the flow of news among seventeen major Canadian cities. In their
study the hierarchical importance of a given place was not determined independently by means of
some surrogate, such as population, but was itself dependent on the flow structure. In contrast,
hierarchical importance is here considered to be an independent variable and is used to predict flow
structure.
The following formulations of the gravity model will be considered in this paper:

CANADIAN XXVII,
GEOGRAPHER, 1, 1983
14 THE CANADIAN GEOGRAPHER

and

where T,j is the predicted flow from area i to area j ; P i and Pj are the populations of areas i and j
respectively; Dij is the distance between areas i and j ; and k , a,y, K , a , b, and care parameters to be
estimated. The values of the parameters will be determined by multiple linear regression of the
logarithmically transformed variables. Taking logarithms of both sides of equations 1 and 2
respectively will yield the following:

and
log TfJ= K' + d o g P I + blog P, - clog D,. (4)

THE DATA BASE

The empirical data consist of the number of telephone calls exchanged among the fifteen largest cities
in the province of Saskatchewan, for twenty-four-hour periods on ten consecutive weekdays (5-9 and
12-16 July) in 1971; the data were made available by Sask Tel, the Saskatchewan telephone
company. Population data were obtained from the Census of Canada, 1971, while the distances
between city pairs were measured by road. (Road distance was preferred to air distance because it was
considered to be a more reliable measure of the true friction of distance between the communities.)
The generally accepted opinion is that telephone traffic, as a composite index of intercity contacts,
is one of the best indicators of the multifunctional associations among cities (Nystuen and Dacey,
1961). Since the data for this study were collected only on weekdays, they would perhaps be more
representative of economic than of social interaction. But that possible bias is also advantageous,
since economic flows tend to be hierarchically structured. Saskatchewan is predominantly an
agricultural province, and its cities are primarily engaged in the provision of central place services. It
is therefore to be expected that intercity telephone traffic will be hierarchically structured.
The fifteen largest cities resulted in 210 observations of one-way telephone flows but, in order to
avoid the randomness normally found at low levels of interaction, it was decided to eliminate all city
pairs with fewer than fifty calls in either direction. Accordingly, the number of observations was
reduced to 102 one-way flows and 51 two-way flows.

HIERARCHICAL STRUCTURE OF SPATIAL FLOWS

If population size is taken as a surrogate of hierarchical importance, one-way flows between places
constituting a hierarchical spatial system may be dichotomized into up-hierarchy (when Pj > Pi)and
down-hierarchy (when Pj < Pi) categories according to the destination-origin size ratio. Depending
on the type of flow, it may be expected, on the basis of existing theories of spatial interaction, that the
net directional flow between two places of unequal size will be either up the hierarchy or down the
hierarchy. In the case of the weekday telephone flows used in this study, it may be hypothesized that
there will be a net up-hierarchy flow since, in a central place system, there should be a net flow of
demands up the hierarchy.
If for two interacting places, i and j , the outgoing and incoming flows for place i are defined as TU
and Tjirespectively, the net hierarchical effect may be measured as the outgoing-incoming calls ratio
(i.e. Ti,/Tjj)and may be hypothesized as a function of PjIPi,the ratio between the populations of places
j a n d i:
TjjITji= P[Pj/Pil8 (5)
GEOGRAPHICA 75

TABLE 1
REGRESSION RESULTS FOR ONE-WAY FLOWS

Coefficients*
Sample Equation Constant P, PJ P,.PJ D,J P,/P2 R2 F:

-0.1721 0.6619 0.8067 - 1.6630 0.8031 133.253


All flows ( n = 102)
0.0945 0.7002 -1.6773 0.1548 0.8054 135.219
P, > P , ( n = 51) 4 -0.0821 0.4944 0.9360 - 1.6533 0.8150 69.026
Pi < P , ( n = 51) 4 0.2712 0.7740 0.5963 - 1.7014 0.8224 72.554

*Based on the t test all the regression coefficients are significant at the 0.01 level.
tAll the F values are significant at the 0.0001 level.

where p and 6 are parameters to be estimated. The constant of proportionality, p, has a theoretical
value of unity since flows in both directions should be identical when there are no differences in the
populations of the two interacting places. For calibration purposes the above equation may be
expressed in logarithmic form:
log Tfl - log TI, = p + 6(bg PI - log P,). (6)
Because the mean logarithm values of the two variables will be zero, the value of p in equation 6 will
always be zero; this conforms to the theoretical requirement that the constant of proportionality, p, of
equation 5 be equal to unity. The regression results were as follows:
log Tl, - log T,, = 0 + 0.1448 (log f 1 - log P,), (7)
with n = 102; r2 = 0.7491; F = 298.65; and significance = 0.0001. The results confirm the
expectations that the intercity telephone flows will be hierarchically structured and that there will be a
net up-hierarchy flow.

THE GRAVITY MODEL: ONE-WAY FLOWS

The basic gravity model does not incorporate an explicit hierarchical variable, but when different
exponents are applied to the two population variables, as in equation (2), the model is capable of
explaining some proportion of the hierarchical interaction between places of unequal size. Thus, if
there is a net up-hierarchy flow structure (i.e. Til > Tji when f j > PJ, the coefficient for Pj will be
larger than that for P i . This has been observed in migration studies, for example (Clayton, 1977;
Schwind, 1971).
Table 1 illustrates the results of applying the gravity model - (equation 4) - to the 102 one-way
flows. As expected, since there is a net up-hierarchical flow structure (equation 7), the coefficient of Pj
is larger than that of Pi, indicating a net flow from smaller to larger cities. Despite the high level of
explanation ( R 2 = 0.8031), the model explained only the net (not the full) hierarchical effect. (It can
easily be shown that the exponent 6 of equation 5 is equal to the difference between the two population
exponents of equation 2; that is, 6 = b - a , )
The net hierarchical effect explained by the gravity model will be identical to the full hierarchical
effect only in the special case where up-hierarchy and down-hierarchy flows are affected in the same
manner by variations in the destination-origin population ratio; that is, the gravity model assumes that,
when Pi.Pj and D, are held constant, the same linear relationship between Tq and f j P ,will apply to
both up-hierarchy and down-hierarchy flows. It is, however, conceivable that, for many types of
interaction that are hierarchically structured, different linear functions should ideally be specified for
up-hierarchy and down-hierarchy flows.
Figure 1 illustrates a number of hypothetical relationships between Tq and P j P i when combined
population size (Pi.Pj) and distance (Dq) are held constant. For some types of interaction, such as
social exchanges, there will not be a hierarchical element; that is, flow volumes between places of
unequal size will be identical in both directions and the partial correlation between Tqand Pj/Piwill be
76 LE G~OGRAPHE.CANADIEN

/
Tij

FIGURE1. Some hypothetical relationships between T,J and PJIP,when P,.P, and D,Jare held constant

zero (Figure la). For such flows, which are not hierarchically structured, equations 1 and 2 will
produce identical and valid results. Figures 1b and Ic are examples of hierarchically structured flows
for which the same linear function may be applied to both up-hierarchy and down-hierarchy flows and
for which the gravity model is appropriate. For these two cases the use of two population variables for
the gravity model (equation 2 ) will fully capture the hierarchical effect. Figure Id is an example of a
symmetrical flow pattern (1.e. T , = Tji) that is hierarchically structured; since there are no net flows up
or down the hierarchy, the partial correlation between To and fj/fi will be zero and the gravity model
will fail to explain any portion of the hierarchical variation. A different linear function must be applied
to each of the two sets of hierarchical flows.
Finally, Figures l e and If are examples of near-symmetrical flow matrices characterized respect-
ively by net up-hierarchy and net down-hierarchy flows. The latter two patterns appear to be
descriptive of the hierarchical structure of many types of spatial flow. In migration studies, for
example, two observations have commonly been noted: first, the close correspondence between
counter-flow volumes (Schwind, 1971); and, second, a net up-hierarchical flow structure (Clayton,
1977; Schwind, 1971). In combination these two characteristics are likely to give rise to a pattern such
as Figure le. More recently, a counter-urbanization trend has been observed in a number of developed
countries, suggesting that the net up-hierarchical migration structure characteristic of previous years
may have changed to a net down-hierarchical structure (Berry, 1976; Roseman, 1977; Vining and
Kontuly, 1978). Thus, migration flows in developed countries may increasingly resemble the pattern
illustrated by Figure If rather than Figure le. Yet, in both cases, the gravity model will explain only
the net hierarchical effect, and different linear functions will have to be specified for up-hierarchy and
down-hierarchy flows in order to capture the full hierarchical effect.
In the case of telephone interaction among Saskatchewan cities, the relationship between T, and
fj/Pi, when Pi.Pj and D, are held constant, appears to be most similar to Figure le. The partial
regression and correlation results are shown in Table 2 for the combined data set ( n = 102) and for the
up-hierarchy (P, > f ,) and down-hierarchy (fj< Pi)subsets. Although the partial correlation for the
GEOGRAPHICA 77

TABLE 2
PARTIAL REGRESSION OF T,JWITH P,/P, (P,.PJ A N D D, HELD CONSTANT)
~~ ~

Cases included in regression*


ALL(^ = 102) PJ> P,(n= 51) PJ<P,(n = 51)

Constant 0.0000 -0.11 I6 -0.1086


Regression coefficient 0.0724 0.2118 -0.0641
r 0.26231 0.3933t -0.1369
F 7.3906 8.9680 1.0687

*For each of the three regressions the partial variables were those derived for
the 102 observations.
tSignificant at the 0.01 level.

combined data set is positive (because of the net up-hierarchical flow structure), the correlations for
the two subsets are of opposite sign: a positive relationship for up-hierarchy flows and a negative one
for down-hierarchy flows. (Although the correlation for the down-hierarchy set is not statistically
significant, the difference between the up-hierarchy and down-hierarchy correlation coefficients is
significant at the 0.01 level.) In other words, when other variables are held constant, population
inequality (or hierarchical distance) between origin and destination has a positive effect on both
up-hierarchy and down-hierarchy flow volumes. Because the gravity model (equation 2) assumes a
single hierarchical function for both hierarchical flows, its prediction errors will be systematically
biased; the model will generally underpredict the volume of interaction between places that are
substantially unequal in population and overpredict interaction between places with similar popula-
tions.
For the telephone data of this study it may in fact be more valid to assume a symmetrical flow
matrix, such as Figure Id, than to assume the linear function (such as Figure lb) of the traditional
gravity model. Consider, for example, a modified gravity model incorporating a hierarchical element
for a symmetrical flow matrix:

where PI and P2 are respectively the larger and smaller of the two interacting places and G,
b l , b2, and b3 are parameters to be estimated. In logarithmic form the equation becomes:
log Ti, = G' + bl (log P; + log P,) + b2 (log PI - log P2) - b3 log D;,. (9)
The results for equation 9 indicate a slightly better explanation than that provided by the traditional
gravity model (Table l)!
Except for symmetrical flow patterns (equation 8), there do not appear to be any means of
modifying the gravity model in order to incorporate different linear functions for up-hierarchy and
down-hierarchy flows, and the most useful procedure may be the separate modelling of the two
hierarchical flows. Table 1 illustrates the results of fitting equation 4 to the two data subsets; for each
of the subsets the performance of the gravity model, as measured by R 2 , is better than that for the
combined set. Thus, the separate modelling of up-hierarchy and down-hierarchy flows achieved a
better level of explanation than did the combined-set modelling. I
In summary, the hierarchical structure of a flow matrix will normally be fully captured only when
the gravity model is separately fitted to down-hierarchy and up-hierarchy flows. Such a procedure is
justifiable on both theoretical and empirical grounds. Apart from analytical convenience, there is no a
priori justification for assuming the same linear relationship between Tjj and P j / P j for both
up-hierarchy and down-hierarchy flows, when P,.Pj and DGare held constant. Even when a net flow
up or down the hierarchy is theoretically to be expected, there is often a strong tendency towards a
symmetrical flow matrix; this appears to be the case for both migration and telephone flows.
TABLE 3
REGRESSION RESULTS FOR TWO-WAY FLOWS

Coefficients+
Sample Equation* Constant P, PJ PI '2 't"J D,J P,IP2 R2 F*
~ ~ ~ ~ ~ ~ ~ ~ ~

Double-count 3 0.1038 0.7353 -1.6218 0.7991 196.970


( n = 102) 4 0.1038 0.7353 0.7353 -1.6218 0.7991 129.987
11 0.3886 0.8642 0.5334 - 1.6691 0.8187 147.539
13 0.3886 0.6988 -1.6691 0.1654 0.8187 147.539
Single-count 3 0.1038 0.7353 - 1.6218 0.7991 95.500
( n = 51) 4 0.0535 0.7907 0.6840 - 1.6038 0.8070 65.511
11 0.3886 0.8642 0.5334 - 1.6691 0.8187 70.758
13 0.3886 0.6988 -1.6691 0.1654 0.8187 70.758

*Except for equation 11 places are randomly labelled P , and P J .


tBased on the t test all the regression coefficients are significant at the 0.05 level.
$All the F values are significant at the 0.0001 level.
GEOGRAPHICA 79
THE GRAVITY MODEL: TWO-WAY FLOWS
In the modelling of two-way flows, Tje*, (i.e. Ti, t T j J , the basic gravity model assumes, at least
implicitly, that there will be no hierarchical effect; that is, when P j . f j and D , are held constant, there
will be no relationship between propensity to interact and inequality in population. However, for some
types of interaction the volume of two-way flows may at times be positively (e.g. Figure Id) or
negatively related to the degree of population inequality between interacting places.
Since in the modelling of two-way flows the two population variables, P I and f j , are meaningless in
terms of origin and destination, equations I and 2 should yield identical results; indeed, there is no
theoretical justification for using equation 2. Yet in modelling two-way flows, Pi and P, labels are
often allocated in some random way to interacting places. Alcaly (1967), for example, applied these
labels according to the alphabetical order of the two place names. To ensure a truly random labelling,
however, the two-way flows should be counted twice, with the P and f j labels applied alternately to
the paired places. The results of a double-count model ( n = 102) are presented in Table 3 for equations
3 and 4. Also shown are the results for a single-count model ( n = 51) in which paired places are
randomly labelled Pi and f j . As is to be expected, the results for the single-count and double-count
models are identical for equation 3. However, for equation 4 the single-count model provides a
slightly better explanation than does the double-count model (Table 3). Because the Pi and Pj labels
were randomly applied to interacting places in the single-count model it is impossible to interpret
meaningfully the parameter estimates; there is, for example, no theoretical significance to the fact that
the P i coefficient is greater than the f j coefficient (Table 3). It would appear that, depending on how
places are labelled, a single-count model's performance will generally be equal to or better than that of
the double-count model when equation 4 is calibrated. Still, this is theoretically unsatisfactory, since
it suggests that both the parameter estimates and the level of explanation are functions of the labelling.
Moreover, the fact that a random labelling may be expected to produce a better explanation suggests
that an appropriate systematic labelling might provide even better results.
In order to capture fully the hierarchical element it is necessary to label the interacting places
systematically according to their relative sizes. Equation 2 may therefore be rewritten as follows:

where PI and P are respectively the populations of the larger and smaller of the two interacting
places; and M , b l , b2,and b3are parameters to be estimated. For calibrationpurposesequation 10 may
be transformed as follows:

In order to make the hierarchical element more explicit, equations 10 and 11 may also be rewritten
respectively as:

and

As Table 3 illustrates, the full hierarchical effect, if present, will be captured by equation 1 1 or 13,
irrespective of whether two-way flows are counted once or twice and of whether P I and Pj labels are
randomly or systematically applied; equations 1 1 and 13 perform better than equations 3 and 4 for both
single-count and double-count models. The empirical results confirm that two-way flow volumes are
in part a function of the hierarchical associations between interacting places; that is, when PI.PJand D,J
are held constant, there is a significant positive relationship between two-way flow volumes (T,Uj)and
the population inequality of the two interacting places ( Pl / P 2 ) .
80 LE G ~ O G R A P H ECANADIEN

CONCLUSIONS

This study has addressed the problem of incorporating hierarchical effects into spatial interaction
models; specifically, I have been concerned with the performance of the gravity model in the
prediction of flows characterized by a hierarchical structure. Within hierarchical spatial systems it is to
be expected that, if population size is taken as a surrogate of hierarchical importance, for many types
of interaction the flows between any two places (T,) will be a function of the destination-origin
population ratio (P,/P,) when the product of the two populations (P,P j )and the distance (Dij)are held
constant. However, it has been demonstrated that the gravity model will fully capture this hierarchical
effect only in those cases where the same linear function relating Tijto Pj/Pimay be specified for both
up-hierarchy and down-hierarchy flows (e.g. Figures Ib and Ic). In those cases where it is
inappropriate to specify the same linear function for the two hierarchical flows the gravity model will
be capable of explaining only the net hierarchical effect (measured as the ratio between the reciprocal
flows, T,and Tji). For symmetrical flow matrices, such as Figure Id, the model will fail to explain any
of the hierarchical effect, while for near-symmetrical flow matriczs, such as Figures le and If, the
model will explain only the net hierarchical effect.
Ways of improving the performance of the gravity model have been suggested for both one-way and
two-way flows. For one-way flows the most appropriate procedure appears to be the separate
modelling of up-hierarchy and down-hierarchy flows. For two-way flows the hierarchical effect may
be fully captured either by the systematic labelling of the two interacting places according to their
relative populations (equation 10) or by the addition of a larger-to-smaller population ratio variable to
the basic gravity model (equation 12). For both one-way and two-way flows, the above procedures
resulted in a significant improvement in explanation over the traditional gravity model.
In conclusion, this study highlights the need to incorporate hierarchical effects into spatial
interaction models and suggests how this may be done with respect to the gravity model. By explicitly
incorporating hierarchical effects not only will the explanatory power of spatial interaction models be
increased but also the parameter estimates of the models will more accurately reflect the influence of
each variable on interaction and will enrich our understanding of spatial interaction. For example, this
study raises several theoretical questions that will require further research. 1 How does the
hierarchical effect vary among the different types of spatial interaction? 2 How does the effect of the
destination-origin size ratio vary between up-hierarchy and down-hierarchy flows for a particular type
of interaction? 3 How does the effect of distance vary between up-hierarchy and down-hierarchy
flows for a particular type of interaction?

NOTE

1 The combined goodness-of-fit measure for the two subset models may be estimated by regressing the 102
observed values (A,J against the predicted values ( T J of the two models. In this case the rz value was 0.8204,
which was, as expected, higher than the R2 value for the combined-set model. As noted by Wilson (1974), the
correlation between actual and predicted values is an accurate measure of variance explained only if the values
of a. and a , in the following equation are equal to zero and unity respectively:
A , = ao + aiT,
Since both sets of predictions are derived by least-squares linear regression, the values of a, and a , will always
be zero and unity respectively.

REFERENCES

ALCALY,R.E. (1967), ‘Aggregationand gravity models: some empirical evidence,’ Journal ofRegiona1 Science,
7: 61-73.
BERRY, B.J.L. (ed.) (1976), Urbanization and Counterurbanization (Beverly Hills: Sage).
CLAYTON, C. (1977). ‘Interstate population migration process and structure in the United States, 1935 to 1970,’
Professional Geographer, 29: 177-81,
CLIFF,A.D., MARTIN,R.L., and Om, J.K. (1974). ‘Evaluating the friction of distance parameter in gravity
models,’ Regional Studies, 8: 281-6.
GEOGRAPHICA 81
- (1975), ‘Map pattern and friction of distance parameters: reply to comments by R . J . Johnston and by
L. CUT, D.A. Griffith and E.S. Sheppard,’ Regional Studies, 9: 285-8.
- (1976). ‘A reply to the final comment,’ Regional Studies, 10: 341-2.
CURRY,L. (1972), ‘A spatial analysis of gravity Rows,’ Regional Studies, 6: 131-47.
CURRY,L., GRIFFITH, D.A., and SHEPPARD, E.S. (197% ‘Those gravity parameters again.’ Regional Studies,
9: 289-96.
GOODCHILD,M.F. and KWAN,M.Y.C. (1978), ‘Models of hierarchically dominated spatial interaction,’
Environment and Planning A 10: 1307-17.
GRIFFITH, D.A. and JONES, K.G. (1980), ‘Explorations into the relationship between spatial structure and spatial
interaction,’ Environment and Planning A 12: 187-201.
HAINING,R. (1982), ‘Interaction models and spatial diffusion processes,’ Geographical Analysis, 14: 95- 108.
JOHNSTON, R.J. (1973), ‘On frictions of distance and regression coefficients,’ Area, 5 : 187-91.
- (1975), ‘Map pattern and friction of distance parameters: a comment,’ Regional Studies, 9: 281-3.
- (1976), ‘On regression coefficients in comparative studies of the “frictions of distance”,’ Tijd.schrft voor
Economische en Sociale Geografie, 67: 15-27.
NYSTUEN,J.D. and DACEY,M.F. (1961), ‘A graph theory interpretation of nodal regions,’ PapersandProceed-
ings of the Regional Science Association, 7: 29-42.
OPENSHAW, S. (1976). ‘An empirical study of some spatial interaction models,’ Environment and Planning A 8 :
23-41.
ROSEMAN, C.C. (1977), Changing Migration Patterns within the United States (Washington, DC: Association of
American Geographers, Commission on College Geography, Resource Paper No. 7712).
SCHWIND,P.J. (1971), Migration and Regional Development in the United States, 1950-1960 (Chicago:
Department of Geography, University of Chicago, Research Paper No. 133).
SENIOR,M.L. (1979), ‘From gravity modelling to entropy maximizing: a pedagogic guide,’ Progress in Human
Geography, 3: 175-210.
VINING,D.R., JR and KONTULY,T. (1978). ‘Population dispersal from metropolitan regions: an international
comparison,’ International Regional Science Review, 3: 49-73.
WILSON,A.G. (1970), Entropy in Urban and Regional Modelling (London: Pion).
WILSON,A.G. (1971), ‘A family of spatial interaction models and associated developments,’ Environment and
Planning 3: 1-32.
- (1974), Urban and Regional Models in Geography and Planning (London: Wiley).

[GEORGEA. NADER,
Trent University]

USER REACTIONS TO A VISITOR DISTRIBUTION PROGRAMME


IN KILLARNEY PROVINCIAL PARK, ONTARIO: A COMMENT

A recreation manager is sensitive to user opinions in deciding upon approaches to the regulation of
wilderness use. Different methods of regulation will likely be viewed as resulting in more or less
satisfaction for the user. To determine recreational satisfaction with a particular approach, then, it is
essential that an appropriate portion of the user population should be sampled. Otherwise, manage-
ment policy might promote recreational behaviour inappropriate to the management objectives for the
setting, while displacing users whose behaviour is more in keeping with that setting.
Establishing whom to interview is critical when management policy is based on response to user
surveys (Heberlein, 1973;Schreyer, 1980). Unfortunately, Morrison and Priddle (1981), in reporting
a favourable response to a method of regulating wilderness use, appear to have sampled an
inappropriate portion of the recreationists using Killamey Provincial Park.
Various regulatory procedures are available to alleviate congestion in a recreational area. These
include use limits, regulations on the type of group, technology regulations, campsite assignments,
limits on length of stay, travel method regulations, group size limits, zoning of uses (spatially and
temporally), blocking off access points, and so forth. In the visitor distribution programme
implemented at Killamey Provincial Park, overnight destination zones were assigned with the
objective of balancing use throughout the park. When the user survey was conducted, however, the

CANADIAN XXVII,
GEOGRAPHER, 1, 1983

You might also like