Rural Population Density

8/6/2019 Rural Population Density

1/20

Journal of Rural Studies 18 (2002) 385404

Rural population density: its impact on social and demographic

aspects of rural communities

Peter J. Smailesa,*, Neil Argentb, Trevor L.C. Griffina

aUniversity of Adelaide, Adelaide, SA 5005, AustraliabUniversity of New England, Armidale, NSW, Australia

Abstract

Using the settled areas of South Australia as a case study, this paper seeks to demonstrate the specific importance of rural

population and settlement density as an important variable in understanding the social, population and settlement geography ofsparsely settled rural regions, where sparse and falling density presents both practical and conceptual problems for rural planners.

After a review of the literature on population density, the case is argued for the use of net rural rather than gross density in the

analysis of settlement patterns. The paper then tests a series of hypotheses on the empirical relationship between rural density as

independent variable and selected demographic and socio-economic indicators as dependent variables, at two specific points in time.

For the same region, points in time and set of indicators, it goes on to compare the predictive power of rural density as an

independent variable with that of three other important qualities of rural settlement patterns (remoteness, settlement size and urban

concentration). Rural density is found to be a significant explanatory variable, both in its own right and in comparison with the

three other variables tested. In conclusion, the findings are related to policy development measures for rural Australia.

r 2002 Elsevier Science Ltd. All rights reserved.

Keywords: Rural population density; South Australia; Rural communities; Socio-economic indicators

1. Introduction

By international standards Australia, like Canada, lies

at the extreme low end of the spectrum of gross national

population densities. Internally, large areas are practi-

cally uninhabited, yet peri-urban densities around the

few metropolitan cities are similar to those of other

Western countries. The resulting very large range of

rural densities presents many problems, both conceptual

and practical, for rural policy makers and planners.

Over the last two decadesparticularly since the ruralcrisis of the early 1990sagricultural restructuring in

rural Australia has with few exceptions involved a

substantial net reduction in the number of rural

holdings and rural employment. At the same time, the

continued but spatially restricted counterurbanisation

movement has brought a new exurban element into

some rural areas. But, while there has been much

research on the impact of falling rural population

numbers and/or disposable incomes on the country town

network and on major regional centres, there has been

very little written on the independent impact of falling

(or, indeed, rising) rural densities. This paper makes a

start on addressing the question, and seeks to place the

density variable at centre stage in rural geography,

comparing it with other critical variables used to

evaluate rural settlement patterns.

2. Specific aims

In this paper, we hope to:

(1) Provide a brief conceptual review of the density

concept as it applies to rural populations in Western

countries.

(2) Using a case study where as many complicating

variables as possible are held constant:

(a) test a series of hypotheses on the empirical

relationship between density as independent

variable and selected demographic and socio-

economic indicators as dependent variables, at

two specific points in time;*Corresponding author. Fax: +61-8-8303-3772.

E-mail address: [email protected] (P.J. Smailes).

0743-0167/02/$ - see front matter r 2002 Elsevier Science Ltd. All rights reserved.

PII: S 0 7 4 3 - 0 1 6 7 ( 0 2 ) 0 0 0 3 3 - 5


2/20

(b) for the same region, points in time and set of

indicators, compare the predictive power of

rural density as an independent variable with

that of three other important qualities of rural

settlement patterns (remoteness, settlement size

and urban concentration);

(c) based on this comparison, demonstrate thatrural density has an important impact on

demographic and socio-economic characteris-

tics of rural populations, that is not reducible

to the effects of other qualities of rural

settlement patterns.

(3) Relate the findings to policy development measures

for rural Australia.

Within the confines of a single paper, there are

limits to what can be attempted. Three caveats,

therefore, are needed. First, we purposely use a

relatively simple measure of rural density at a local

scale, but excluding urban populations and major

unpopulated rural areas from the equation. Eventually

our intention is to develop a more sophisticated measure

of effective density, at varied scales. Secondly, we

recognise that density itself is the product of a set of

other causal factors; these complex relationships are

acknowledged, but not investigated here. Third, while

we fully recognise the complexity of the causal relation-

ships between density and the demographic and socio-

economic variables considered, in this paper our

primary aim is to demonstrate the empirical strength

of the relationships, leaving the matter of the sequence

and possible reciprocality of causality to a subsequentpaper.

3. Existing work on rural population density in

developed countries

Population density is a quintessentially spatial phe-

nomenon, expressing the way that human beings

spread out over, and occupy, the earth. As such it is a

highly significant element in population geography,

social geography and settlement geography. Yet a

search of the literature reveals few analyses in depth;

and of the work that has been done, the majority has

been concerned with urban areas. An excellent review

from a town planning viewpoint is provided by Saglie

(1998); a general overview, including major German

language contributions, is provided by B.ahr et al.

(1992).

For specifically rural areas, an early contribution by

Robinson et al. (1961) examined the relationship

between rural farm densities and rainfall, percentage

of land under crop, and distance from urban centres in

the Great Plains, while Aangebrug and Caspall (1970)

classified Kansas counties by their pattern of density

change over time. Arguably the first work systematically

to investigate the impact of density variations on an

entire settlement system, however, was that of Berry

(1967). Although working within the constraints of the

rather rigid theoretical framework of central place

theory, Berry was able to show that the size of rural

service centres and their surrounding trade areas issystematically related to the broad regional population

density in which they are embedded. Whatever the

density, centres tend to form a discrete spatial hierarchy,

but as density drops, the absolute size of places at each

level falls, while trade areas increase in size to

compensate partlybut only partlyfor the falling

density. As a result, particular types of service found at

the lowest hierarchical level under high-density condi-

tions will move up to the next higher order when density

falls. Berry also introduced the concept of a phase shift

in the spatial patterning of trade centres with abrupt

discontinuities in density, as between suburban areas

and the peri-urban countryside, or between irrigation

areas and broad-acre farming. Beavon (1977) later

introduced the concept of density changes over time to

central place theory, but only in an intra-metropolitan

context. An extremely interesting but little known paper

(Irving and Davidson, 1973) on density in an urban

context (but with strong rural relevance) introduced the

idea of social density, expressing the amount of person-

to-person interaction taking place in a given unit of area

per unit of time. This was found not to be a simple

function of physical density of population.

In Australia, important contributions were made by

Holmes (1977, 1981) who introduced the idea of criticaldensity thresholds for particular kinds of service centre

network, relating density levels to broad types of

primary production land use, e.g. the marginal density

zone where normal daily schooling of children using

buses becomes impracticable, and gives way to distance

education and school of the air, and normal ambulance

coverage of patients gives way to the Flying Doctor

service. Holmes introduced the suggestion that a certain

critically low density (8 km2/resident person) marked the

approximate limit of the Australian ecumene, separating

the settled areas from the sparselands, and produced a

loose classification relating population density bands to

various forms of land tenure, farm size, land use, town

spacing and patterns of access to services. Later the

Australian Bureau of Statistics used a figure of 0.057

dwellings/km2 to define the limits of the sparsely settled

areas for sampling purposes in their surveys (Hugo et al.,

1997, p. 105).

The terminology used by Holmes to express this

important phase shift from ecumene to non-ecumene, or

settled areas to sparselands, did not receive wide

adoption, and was gradually replaced by a distinction

(along a very similar geographic boundary) between

rural and remote Australia, culminating in the

P.J. Smailes et al. / Journal of Rural Studies 18 (2002) 385404386


3/20

adoption by the Commonwealth Government of the

Rural, Remote and Metropolitan Areas (RRAMA)

spatial classification system (Departments of Primary

Industries and Energy and Department of Human

Services and Health, 1994). The RRAMA system is

interesting here in that it incorporates a density

measure termed personal distance expressing theaverage distance between residents within a given

spatial data unit. Recently, Hugo and Wilkinson

(2002) have argued that while the concepts of density,

rurality and accessibility are closely related, they

should be kept conceptually distinct, and a spatial

classification that recognises all three of the rural/

urban, density and accessibility/remoteness dimensions

is required. The latter dimension is very well captured by

the Accessibility/Remoteness Index of Australia

(ARIA), described by Hugo and Wilkinson (2002,

pp. 1619). This index does not seek to measure density

directly.

Meanwhile, Swedish geographers, examining the

urbanisation of their large and in great part sparsely

settled country, had devised a novel method of expres-

sing the relationship between density and accessibility

(H.agerstrand and .Oberg, 1970). A family of isopleth

maps was produced showingfor any point in Swe-

denhow many people were contained within a series

of radii of constant length, each corresponding to a

typical travel distance. In Finland, early work by Hult

(1962) and Saviranta (1973) showed a distance decay of

population density in 5 km zones surrounding urban

centres after the manner of Th .unens rings. Ruotsala-

Aario and Aario (1977), however, in testing this aroundKuopio, a town of 73,000 people in central Finland,

found that while a strong relationship existed between

distance and population density (r 0:62), this was

due much more to the towns central location in the

most environmentally favourable part of its region, than

to the influence of distance per se.

These contributions apart the analysis of specifically

rural population density has had limited attention.

Fitzpatrick (1983) examined the concept of density in

relation to isolation, with particular reference to

education. Smailes and Mason (1995) extended the

H.agerstrand and .Oberg mapping approach to examine

densities of either total population, or population

subgroups (e.g. school age children, total workforce,

pensioners) in relation to service provision in Eyre

Peninsula, South Australia. Smailes (1996) showed that

total population growth/decline over a period is better

predicted by rural population density at the outset of the

period, than by the absolute population size at the

outset.

In the United States, Rank and Hirschl (1993)

examined the link between population density and

welfare participation, but only in the context of

comparing urban, mixed and rural US counties. Lester

(1995) finds a negative relationship at the level of whole

States between density and suicide rates (r 0:5)

while Fonseca and Wong (2000) find that 19801990

density increase by State is positively related to the

1980 density. In the British context, an important

recent paper by Coombes and Raybould (2001)

provides a review and critique of the formulae usedfor the allocation of funding to local government.

They are particularly critical of the use of gross

population density as an indicator in such formulae,

arguing that it subsumes a number of important

characteristics that are better measured separately. In

our discussion (below) of population density as a

concept, we return in more detail to the significant

conceptual issues raised by these writers. Two other

important descriptors of certain elements of national

settlement patterns (isolation and clustering of medium-

sized towns) developed by Portnov et al. (2000) are

related to, but distinct from, population density and are

not discussed further here.

In summary, then, while the above review is certainly

incomplete, the literature on the density of population

and settlement in rural areas thus far appears to have

been fragmented (spatially and by discipline) and

desultory (over time), whether density is treated as a

dependent or as an independent variable. As an

independent variable, its influence on social, economic

and demographic qualities of rural districts has often

been implied, but rarely subjected to detailed investiga-

tion. A number of authors have recognised its intrinsic

importance as a fundamental aspect of settlement

systems, with some exploring its practical significancefor planning, but to date there appears to have been no

systematic or concerted investigation of how net rural

densities influence the socio-economic and demographic

composition of communities.

4. Why is density important, and why is falling

density a problem?

In developing countries, excessively high rural den-

sities are a frequent concern in terms of overpopulation

and pressure of population on the environmental

carrying capacity. But when population density gets

too low, it also has adverse impacts in rural areas. Farm

amalgamations not only reduce rural numbers, but also

inevitably increase the spacing between homesteads, the

ratio of clustered (country town) to dispersed popula-

tion, and the distance inputs per capita required to

provide the remaining households with services, social

functions and human companionship. Ladd (1992)

found in a study of 247 large US counties that the per

capita cost of providing public services followed a

J-curve with its lowest point at about 250 persons/mile2

(ca. 98 persons/km2). As density fell below this level, per

P.J. Smailes et al. / Journal of Rural Studies 18 (2002) 385404 387


4/20

capita costs rose quite sharply. The gross density

spectrum represented by the South Australian rural

communities in this case study is almost an order of

magnitude below this, yet still shows a huge range (from

58.59 to 0.13 persons/km2). Costs per capita are likely to

rise quite sharply (and/or quality of service decline

sharply) as density falls, if indeed the service can bemaintained at all.

For many services, too, as density drops not only

does the local population fall below some demand

threshold, but it becomes impossible to compensate

for this by, say, amalgamating local government areas

to achieve some arbitrary population target. In effect,

it becomes unrealistic to collect enough demand within

a reasonable travel distance to run the service at

some minimum viable level. Aasbrenn (1998) deals well

with the conceptual problem in a country where

population densities are comparable to Australia,

with a 1998 gross density of 0.9 persons/km2 in his

East Norwegian mountain study area. As in Australia,

such communities are threatened not by complete

depopulation, but by thinning out of both people

and the settlement pattern. Aasbrenns Norwegian term

uttynningssamfunn for such places has no direct

and accepted English equivalent but translates as

something like sparsifying communities or thinning

communities. He points out that the problems of

sparsification breaks down into a distance problem

and a scale problem: the extra cost in time, money

and convenience of overcoming distance as the

settlement pattern thins out, and the reduced opportu-

nities for remaining services to obtain scale economies.These two factors lead to five negative features of

the process of sparsification, listed by Aasbrenn (1998,

p. 73) as

* intensified ageing (relative to the national trend);* deterioration of social networks;* changes in demand (for services);* marginalised viability of service suppliers;* decay of physical infrastructure.

Together these form a syndrome often seen in areas of

falling density, and form part of a self-reinforcing

cycle of decline, giving rise to low morale and a

dispirited residual population (Smailes, 1997). The

extent of several of these elements in a particular

community is difficult to measure from secondary data

and requires fieldwork, yet to be undertaken in the

present study. However, we suggest below that a

number of important features of settlement and

population in a given community will tend to correlate

strongly with density at a given point in time, and also

with changes of density over time. First, though, we

need to consider the nature and measurement of the

density concept.

5. Population density as a concept

5.1. Population density as a dependent variable

In the analysis below, we shall treat rural population

density as an independent variable, which has a

hypothesised direct or indirect causal effect on othersocial variables. But first, we need to recognise: (a) that

any such causal relationships are at least partly

reciprocal and (b) that population density itself is also

a dependent variable, responding to a complex of more

fundamental causative factors. That is, it can be seen as

a reflection of the varied habitability and perceived

opportunity that a given territory offers to humans. The

strength and nature of the relationships between density

and these more basic causal factors (themselves heavily

intercorrelated) is not investigated here. However, in

Australia we would list the most important of them as

follows, in approximate order of primacy:

* rainfall;* soil quality;* remoteness (particularly from major cities);* land values;* farm type;* population potential.

In many places in Australia these variables will also

bear a close relationship to the age and duration of white

settlement, and hence the maturity and attractiveness (to

Europeans) of the cultural landscape. The dominant

environmental variables may vary from country tocountry. In the Great Plains of the United States rainfall

dominated, with a correlation of r 0:78 between rural

farm population density and average annual precipitation

(Robinson et al., 1961). In Finland, land quality (as

expressed by the percentages of land under forest, lakes

and peat bog) was more important (Ruotsala-Aario and

Aario, 1977). However, the causal sequence seems to start

with natural factors and the resulting potential for

primary production, but to be reinforced by a remote-

ness/accessibility dimension caused by the initial place-

ment of the dominant urban centres in the most

environmentally favoured locations. In Australia, the

subsequent fanning out of settlement and routeways from

a few dominant capitals has produced a particularly clear

and powerful access/remoteness effect, so that one can

conceive of a series of more or less concentric, over-

lapping and mutually reinforcing invisible surfaces of

intensity. These surfaces express variations in land values,

mean farm size, farm type, remoteness/accessibility,1 age

1The concepts of remoteness and accessibility are often seen as

mirror images or opposite poles on a continuum. In the case of journeys

to consume services, though, remoteness is often used as a quality of

the point of origin (e.g. a rural homestead) whereas accessibility is a

quality of the target destination (e.g. a shopping centre).



5/20

and duration of white settlement, and population

potential across the nation, with high points over the

capital cities, but with minor outlying peaks representing

naturally favoured locations of various types. The ability

of rural population density to reflect essential qualities of

the social catchment areas of country towns derives from

the fact that it appears to provide a single, convenientand perplexingly simple (though far from perfect) index

of all or most of these factors rolled into one.

5.2. Population density as an independent variable

5.2.1. Continuous or discrete variable?

As normally expressed, population density is a

discrete variable expressed as an average ratio for some

defined area. The above discussion implies that we can

also understand density as a continuous variable

forming an invisible surface over the settled areas. By

knowing the location of a community on this surface, we

argue that one should be able to predict quite a number

of things about the population and activities of that

community. To construct such a surface, however, we

need to have a lattice of control points, between which

isopleths can be interpolated, involving the calculation

of a specific local density for some defined region

surrounding each control point. The more numerous the

control points, and the smaller the defined areas, the

more difficult it will be to interpolate isopleths giving a

smoothly defined surface, for in many cases density

changes abruptly across a narrow borderas at the edge

of an irrigation area in an arid landscape, or at the foot

of a sharp fault scarp.

5.2.2. Local versus regional density: the scale

of resolution

As well as being capable of expression as either a

discrete or a continuous variable, density may also be

expressed at differing scales of geographic resolution.

For some strategic purposes, it is necessary to disregard

local peculiarities, and focus on where a community or

enterprise is located on the broad national density

surface. For detailed analysis of the importance of

density at the individual community level, however,

more specific local density measures are required. Thus,

H.agerstrand and .Oberg (1970, 23, 29) vary the radius of

the region around each control point to produce density

maps for different purposes, while Coombes and

Raybould (2001) suggest a similar moving window

approach, in effect using regionaldensity as an indicator

of urbanisation, with weighting to give the zones further

from the centroid diminishing impact. The latter

authors analysis (Coombes and Raybould, 2001, p.

236) shows, for instance, that their social deprivation

index correlates closely with local conditions, while

home-insurance levels respond more closely with the

broader setting around a locality. The present paper

concentrates purposely on localdensity, within carefully

defined spatial units corresponding to the functional

social catchments of country towns, and using depen-

dent variables expected to respond specifically to

conditions within these catchments.

5.2.3. Measured versus perceived densitySaglie (1998, Chapter 2), in an excellent discussion of

the density concept as applied to the built environment,

distinguishes between measured densitythe quantifi-

able relation between physical aspects of the built

environment and the number of inhabitantsand

perceived density, which also incorporates the subjec-

tively experienced impact of density within social space,

including perceptions of crowding and the like.

Although Saglies distinction between quantitative and

perceived aspects of density refers principally to urban

built environments, in principle the perception by rural

people of the increased isolation and remotenessengendered by the loss of neighbours, and the feelings

such perceptions engender are just as important in areas

of sparsification as in areas of overcrowding. While

measured density is a concept that can be applied at a

wide range of scales, perceived density is likely to apply

mostly to the local scale of daily, or frequent, interac-

tions. Assessment of perceived density must await

analysis in the field, however, and only measured

densities are dealt with below.

5.2.4. Gross density versus net rural density

Coombes and Raybould (2001) point out that gross

population density (including urban populations) has

long been used as a key variable in allocation models for

funding local government in the United Kingdom. As

they argue, however, gross density is a very blunt

instrument with which to analyse settlement systems,

since it tends to conflate and/or confuse more funda-

mental and specific measures, all of which impact on the

cost and difficulties of service provision and human

contact. These are identified as

(a) the size dimensionpopulation size of the spatial

unit and of the urban settlement(s) within it;

(b) the concentration/sparsity dimensionthe propor-

tion of population and dwellings in urban clusters;

and

(c) the remoteness/accessibility dimensionrelative

peripherality or centrality of the area in the relevant

space economy.

We would concur with Coombes and Rayboulds

judgment (2001, p. 226) that (gross) ypopulation

density is seen to be a proxy of first resort which is

standing in for a more specific measurement of one

or other of (these) three dimensions of population

distribution. Gross density is a particularly crude

measure when applied to spatial units of variable



6/20

size, temporally shifting boundaries, and sporadi-

cally distributed unusable land, and also where the

level of spatial resolution is ill-suited to the problem

at hand. Where multivariate statistical analyses are

applied to a relatively simple indicator like gross

density at increasing (or decreasing) scales of

resolution, spatial autocorrelation and the modifi-able areal unit problem (Openshaw and Taylor,

1981; Amrhein, 1995; Wrigley et al., 1996) can render

findings unreliable unless complex statistical controls

are employed (see Green and Flowerdew, 1996). This

is particularly the case where the spatial units used

form building blocks for larger units of analysis.

We would therefore agree that gross density is

indeed a crude concept that needs unpacking, and

that qualities (a)(c) above are essential and funda-

mental descriptors of settlement patterns. To these,

however, we would add (d) below as a quality of co-

equal importance.

(d) The net density of the rural (or more generally, non-

urban) population of the spatial unit.

We argue strongly that (d) is no mere proxy for

qualities (a)(c), but has equal or greater importance in

evaluating settlement patterns from a planning view-

pointparticularly in a strongly rural region such as

South Australia and when calculated for socially mean-

ingful spatial units at a scale of resolution appropriate to

the task on hand. Unlike the other three factors, it takes

direct account of the size of the communitys living

space, the habitability of the landscape in which the

town is embedded, and the likely cost of service deliveryand other forms of interaction requiring personal

human contact for non-urban dwellers. Thus, the nature

of the surrounding matrix in which the town is set is a

vital aspect of the settlement pattern, and may be

expected to have a substantial impact on the community

as a whole. Moreover, the social environment in most of

rural Australia (and similar countries) is radically

different from that of compact, highly urbanised

countries such as the United Kingdom where towns

are large, closely spaced, much of the national territory

is within commuting range of at least one major city,

and the residual farming population of minimal

importance. The median size of the main central town

in the spatial units used in the present paper is just 1120

persons or 470 households, and the central towns

median percentages of the total population and total

persons employed are just 52% and 46%, respectively.

In the 84 studied communities, the 1996 median

percentage of the workforce employed in primary

production was still 26.8% for the total community,

and 47.6% for the non-urban element, despite sub-

stantial labour shedding.

It should perhaps be pointed out that just because

gross density conflates aspects of both urban and rural

populations of a community and acts as a partial proxy

for variables (a)(c) above, it may give a superficially

higher correlation with various socio-demographic

qualities of a communitys population than do any of

net rural density, size, concentration or remoteness

acting on their own.2 This, however, does not overcome

its conceptual weakness and the consequent difficulty ofinterpreting the correlations, and it is not used in the

present paper.

While gross density may have been overstressed in

Britain as a measure of settlement patterns, in Australia

on the other hand there has been a widespreadindeed

almost universaltrend to focus on settlement size,

using the population size of country towns as a single

convenient indicator of the prosperity and/or likely

future trends for their communities. This reductionism

ignores the fact that while rural communities invariably

focus on a country town, that town in many cases

houses only about half of the total community popula-

tion and a tiny proportion of the living space. As we

shall demonstrate, while town size is indeed important

to rural communities, they also have many major social,

demographic and economic attributes for which town

size does not provide a suitable indicator.

We return later to the question of the relationship

between all four of the key factors, and their relative

influence on social and demographic characteristics of

the studied communities. First, though, we need briefly

to describe the study area.

6. The study area

The paper reports on findings from the settled areas

of South Australia, excluding the Adelaide statistical

division. Covering almost 150,000 km2, or roughly the

size of England and Wales, South Australias settled

areas were occupied by just 1.41 million people at the

1996 Census, of whom all but 26% (365,000) were in the

metropolitan statistical division. The non-metropolitan

area should thus be well suited to a study of the social

significance of rural density in sparsely peopled regions

where distance still presents a substantial barrier to

interaction, and the settlement pattern is dominantly

based on primary industry, tourism, retirement and

local services. Only 20% of the States employment in

secondary industry is outside the Adelaide statistical

division, and even this is concentrated into a very small

number of outlying towns. Very few other specialised

employment clusters (e.g. defence, railways, mining,

major tertiary institutions) exist to complicate the

2In the present paper, gross density used as an independent variable

gives an average correlation coefficient (disregarding sign) of r 0:582

for the 11 dependent variables, as against values of 0.480, 0.454, 0.449,

and 0.369 (net rural density, concentration, size, and remoteness

respectively) for 1996 data.



7/20

relationships between rural density and social character-

istics of the population. Fig. 1 locates the study area and

shows place names mentioned in the text.

7. The measurement of population density at the

local level

As Fonseca and Wong (2000, p. 505) observe,

population densitypopulation per unit of areais a

very straightforward idea until one actually starts to applyit. What population, and what area, go into the equation?

7.1. The population element

As people are recorded at their place of residence on

census night,3 and do not live as hermits scattered

individually over the paddocks, occupied dwellings

(which equate to households) are the appropriate

smallest units for measurement, applying to both rural

populations and rural settlement patterns. From a

statistical point of view using households rather than

persons as a density measure makes little or no

difference, since the two are so closely correlated

(r 0:999). In principle, however, households are more

locationally stable over time than are the individuals

within them. It is also the dwellings, and not individuals,

that form the nodes and end-points on the road, postal

delivery, telephone, electricity supply, school bus service

networksand indeed on almost all the physical

networks of transport and communication. (Mobile

phones and personal lap-top computers are significant

exceptions.) For all these reasons, households rather

than persons are chosen as the appropriate unit with

which to construct a basic density measure (independent

variable) against which other community characteristics

can be compared as dependent variables. Densities are

calculated as the number of inhabited ruraldwellings per

100 km2; except for the dwindling number of special

function towns such as Whyalla (steelworks) or Port

Fig. 1. Location of places mentioned in the text.

3Although the Australian Census provides some de jure data by

normal place of residence, the bulk of the data (including those used

here) is de facto: people are counted wherever they were on Census

night. Censuses are held on a week-night in winter to minimise the

count of people away from home.



8/20

Augusta (power station), this element together with

tourism and services to passing traffic functions as the

resource base upon which the remainder of the

settlement system is largely dependent in this study

region. Under rural households are included not just

the farm element, but scattered population of whatever

occupation. Constrained by Census data, rural alsoincludes dwellings in clustered settlements with less than

200 people, as well as dispersed houses.

Densities are calculated exclusive of the area occupied

by the town(s) within each community. These towns are

made up of one or more Census Collectors Districts

(CCDs), which in turn are defined, before each Census,

to correspond to the outer edge of the built-up area of

all clustered settlements expected to exceed 200 popula-

tion. While some CCDs may extend a little beyond the

current built-up area in order to include enough

households for a collectors workload, there is no

significant problem of urban definition. The median

proportion of the total community area occupied by

urban settlements is only 0.4% for our 84 communities,

and its removal has minimal impact on the calculation

of net rural density.

7.2. The area element

Since the object of this study is to identify the impact

of density upon rural society, socially meaningful spatial

units of analysis are required. Australia is a country of

dispersed settlement, and just about all social life is

concentrated on social organisations and facilities

centred in towns of various size. Country towns act asthe foci of social drainage basins. Even very small

centres are frequently nominated as the town of primary

social importance, while the nomination of places of

second and third most important social centres allows

the definition of social linkage patterns at various scales

(Smailes, 2000). The spatial units used in this study

therefore each consist of a socially significant town, plus

the surrounding households linked to the town, forming

a local interaction system. The aim was to identify

spontaneously evolved socially cohesive areas which

should be as small as possible, but still large enough to

be approximated by combinations of CCDs, splitting as

few as possible. Such areas were defined using a method

based on intensive fieldwork supplemented by two

State-wide postal questionnaire surveys, as described

in detail elsewhere (Smailes, 1999; Hugo et al., 2001).

The primary spatial units used here are thus empirically

validated approximations of the discrete social catch-

ment areas of 84 country towns, with areas of overlap

divided along break even lines to give a set of mutually

exclusive units covering the whole of the defined settled

areas of the State but excluding major empty areas

(water bodies, national parks and reserves, etc.).

Subjectively from field experience, empty areas of

at least 150km2 were judged large enough to

cause significant gaps in the settlement pattern and

communication network, often forming watersheds

in the social catchments. Typically, the central town

in each catchment would offer weekly shopping

facilities, police, banking, school, medical and ambu-

lance services, and in many (not all) cases, localgovernmentthough the number, variety and quality

of services naturally vary widely with population size

and affluence.

For each of these social units, which for con-

venience are referred to here as communities, demo-

graphic data have been calculated from the 1981

and 1996 Censuses for the entire community, its

major central town, any other small urban-like clusters

of over 200 persons included in the area, and the

rural balance. Census collection districts straddling

boundaries are allocated pro rata using the South

Australian Rural Area Property Identification Directory

(RAPID) to perform the allocation where possible, and

detailed topographic maps in areas not covered by

RAPID.

Before proceeding, we should demonstrate that the

density variable is not reducible to just population

numbers. It might be argued that any variance in the

dependent variables could be explained by the popula-

tion size alone, without the need for physical area to be

considered. If, for example, the areas of the 84

communities were all quite similar, but the populations

showed much greater variation, then obviously any

differences in density could be sheeted home to just

variations in population size. In fact, both the Pearsonand Spearman correlation coefficients between area and

the dependent variables to be analysed were stronger in

every case than those for population, showing that the

area over which the population is spread has a decisive

impact on any observed relationship, whether or not the

data are transformed.4

8. The pattern of local rural densities in South Australia

The 1996 pattern of local rural densities revealed for

the 84 communities has an extreme range from 787.5

occupied dwellings/100 km2 in Hahndorf, a suburbanis-

ing community some 40 min drive from Adelaide, to

only 2.5 occupied dwellings in Elliston, on the west coast

of Eyre Peninsula (Fig. 2). The median value is 29.1. If

all the occupied dwellings were farms, this median

density would equate to an average farm size of some

4Additionally, after transformation, there was no significant

difference between the statistical distributions of the two variables,

in either skewness or kurtosis.



9/20

344 ha. The class boundaries used in the choropleth

mapping reflect the approximately log-linear distribu-

tion of the community density values.

The overall basic pattern of rural density in the State

is probably better revealed by the construction of a

continuous surface, using the main towns of the 84

communities as control points (Fig. 3). The generally

very low densities stand out clearly, especially in the

peripheral broad-acre dry-farming areas, while even in

the most closely settled core zone, only a very small area

of South Australia has a rural density of more than one

occupied dwelling in every square kilometre. The

narrow strip of higher density along the eastern and

north-western coasts of Yorke Peninsula is due to the

presence of a string of small shack colonies and seaside

holiday and retirement homes, with less than 200

permanent residents. The outlier of high density in the

northeast (the Riverland) is due to the presence of a

string of major irrigation areas along the course of the

Murray; and in the southeast, high density is associated

with the urban influence of Mount Gambier, South

Australias second largest country town, and the fertile

basaltic soils and good, reliable rainfall of the Lower

Southeast. The tongue of slightly higher density

extending north-westward into the Upper Southeast

includes an area of better soils, some supporting

vineyards, reaching into formerly trace-element deficient

country cleared under the States last big land settlement

scheme during the 1960s (Marshall, 1972). The contour

of 25 occupied dwellings/100 km2 approximately out-

lines the older settled areas of the State, occupied for

farming by about 1880. The lowest density zone

corresponds with the most recently settled areas of Eyre

Peninsula and the Murray Mallee, and also includes the

northern marginal lands on the flanks of the Flinders

Ranges, from which overambitious settlers had to

withdraw in the late 19th and early 20th. centuries.

Importantly, Figs. 2 and 3 illustrate that the pattern of

rural densities is not just a function of relative

remoteness from the metropolitan core.

Fig. 2. South Australian communities: rural population density, 1996.



10/20

9. Rural population density as an indicator of

demographic and socio-economic structure

Given the correspondence of the density zones

discussed above with all the fundamental aspects of

the States space economy and both relative and

relational space within it, we are now in a position to

test a series of hypotheses using rural density as

the primary independent variable. If we are right, in a

fairly simple space economy dominated by primary

industry, retirement, tourism, recreation and service

industry such as that of South Australia, rural popula-

tion density ought to correlate with a whole series

of socio-economic variables. Where a community is

located on the general density surface ought to have a

major influence on its fundamental makeup, as should

its local, specific density. The hypotheses outlined below

relate to synoptic relationships measured at two

different points of time. In testing these hypotheses,

we do not suggest a simple causal relationship, for

causality is likely to be complex. In the first instance, we

wish to establish the degree of correlation between

density and the selected variables. Later, we investigate

the extent to which that correlation is independent or

otherwise of settlement size, urban concentration and

remoteness.

9.1. Hypothesised relationships

(1) The lower the rural population density, the greater

will be the spatial extent of communities and the distance

between neighbouring towns. This is a fundamental tenet

of central place theory, and its general validity can be

confirmed by a casual glance at any topographic or road

map. We cannot expect the relationship to be simple or

perfect, however, for various reasonsincluding the

sporadic distribution of mineral resources, historical

accidents of town sitingand the fact that coastal

centres lose half their potential catchment areas to the

sea.

(2) The greater spatial extent of local social systems in

sparsely peopled areas will compensate only partially for

the low density. Thus, the lower the density, the smaller

will be the total population size of communities.

Hypothesis 2 has been firmly established empirically

by Berry (1967) in the United States, and there is no

reason to expect that Australia will differ. The reasoning

only applies where towns are dominantly service centres;

Fig. 3. South Australia: rural population density surface, 1996.



11/20

however, large mining, manufacturing or other special

function enterprises are likely to stand out as outliers in

the relationship.

(3) Low density will be associated with a low labour

force participation rate. This is expected mainly because

in sparsely populated rural areas, numbers of females in

the formal job market are likely to be low for a varietyof reasons, including the relative shortage of both full

time and part time non-farm jobs, particularly for

women, and long commuting distances for the few jobs

available. Among the farm population, the extent to

which female partners report themselves as farmers

rather than homemakers in the Census is uncertain, as is

the extent of underemployment.

(4) Low density will tend to be associated with a low

number, but a high proportion of the labour force

engaged in agriculture. A major impact on rural density

is the potential productivity of the land. In densely

peopled rural areas, farms are likely to be smaller and

the agricultural workforce therefore larger in absolute

numbers. But at the same time, a dense rural population

is likely to provide opportunities for other types of

enterprise to operate profitably, so that the community

as a whole becomes less reliant on farming alone.

(5) High rural densities will be associated with high

levels of both occupational and industrial diversity of the

population. Following the same line of reasoning as for

Hypothesis 4, the non-farm element of the population

will not only be larger, but will also have a greater

diversity of livelihoods, whether measured by occupa-

tion or by industry. Densely peopled areas will bring

more people and businesses into proximity, and willtend to produce a greater range of niches that can be

exploited by both full-time and part-time enterprises.

(6) Low rural density will be linked with a low

proportion of the workforce unemployed. This is expected

because in sparsely settled areas, jobs are scarce and

amenities and public services few. People who become

unemployed there, and want to find a new job, are often

more or less forced to move out. Also, such areas (with

some exceptions, such as remote places with a good surf

beach) have few attractions for the intentionally

unemployed, or near-unemployable, elements of the

workforce. Higher-density areas with more amenities

and services, on the other hand, are likely both to attract

this element as in-migrants, and to have a better chance

of retaining the local-origin unemployed. (Hugo and

Bell (1998) provide a full discussion of welfare-led

migration in Australia.)

(7) In areas of low rural density the masculinity ratio of

the population (males per 100 females) will tend to be

high. This hypothesis is consistent with Hypothesis 4

above for much the same reasons.

(8) Low population density will be associated with a low

proportion of the population born overseas. The concen-

tration of first and second-generation migrants particu-

larly those from non-English speaking countries in the

major cities and labour markets of Australia is a well-

known demographic phenomenon. Most rural areas of

low population density are subject to net out-migration

and cultural homogeneity with few opportunities for

people of minority cultures to become established. There

are some exceptions such as remote outback miningsettlements. The few concentrations of non-English

speaking overseas born in the present study area tend

to be in the higher-density areas as with Southern

European-born people in the irrigation-based fruit

blocks of the South Australian Riverland.

(9) In areas of low rural density the proportion of the

population who have changed address in the last 5 years

will also be low. This hypothesis results from the

expectation that areas of sparsification will also be areas

of net out-migration attracting relatively few in-

migrants. The resident population of such areas is likely

to have a smaller proportion of people who have

recently changed address than is that of regions of

higher density which are expected to attract a higher

proportion of in-migrants.

(10) Low rural density will tend to be associated with a

high fertility ratio (children under 5 per 100 women aged

1544). This is expected because of the declining but still

noticeable tendency for Australian rural birth rates to

exceed those in metropolitan and other large urban

areas. Low density is likely to correlate with high

rurality, isolation, a low proportion of exurban in-

migrants and the longer retention of established

behaviours.

(11) Low rural density will be associated with a low proportion of the population aged under 15. This

hypothesis appears counter to the reasoning behind

Hypothesis 10. However, the proportion of school-age

children in the population reflects not only the commu-

nitys fertility rate but also the proportion of young to

middle-aged adults who are their parents. The latter is

expected to be more important than residual fertility

differentials.

9.2. Approach to testing

The above propositions were tested using rural

population data derived from the 1981 and 1996

censuses and the areas of the spatial units in km2

calculated from the GIS database. The two years

chosen, 1981 and 1996, represent conditions before

and after the severe rural crisis of the mid to late 1980s

and early 1990s, as one of the main aims of the research

is to pick up any changes which may have occurred in

the influence of population density over time. Use of the

parametric statistic Pearsons r with n 84 requires a

reasonably normal distribution of the correlated vari-

ables. As many of our variables have a highly skewed

distribution appropriate transformations (mostly log 10)



12/20

were applied to normalise the distributions before

correlation analysis using Pearson. For comparison the

non-parametric Spearmans rho statistic was also

calculated based on non-transformeddata sets. In almost

all cases the values for rho were slightly higher than the

Pearson r values. However, since we shall later need to

use multiple regression methods to establish one link inour argument only the simple Pearson correlations are

presented below.

9.3. Results

In this type of analysis we do not expect very high

correlations of the order one would look for when

investigating physical processes. We know from the start

that population density results from a number of other

key causative variables and that it is still fairly crudely

measured. Moreover, at this stage of the analysis all

country towns and their immediate communities includ-

ing industrial centres and regional capitals have been

included. In many cases these form outliers that reduce

the strength of the correlations but it was considered

important initially to discover which of the relationships

were robust enough to be valid for all communities

before identifying groups for separate analysis.

If the above hypotheses are to be supported firstly the

sign of correlations must be as expected. Secondly, for

n 84 a relatively low correlation coefficient can be

statistically significant so we require a coefficient with

the predicted sign plus a fairly stringent probability

levelhere set at pp0:001before considering any

hypothesis supported. As well as calculating thecoefficients, scatter plots were produced for each pair

of variables. Because of space limitations just one of

these is included as an example (Fig. 4). The relative

location of towns on the plots gives many clues to the

hypothesised relationships. Given the above caveats

many of the coefficients in Table 1 are surprisingly and

consistently high particularly for 1981.

The first two hypotheses not surprisingly are strongly

upheld by the Pearsons r values. The positive relation-

ship between density and total population of the

community has even strengthened between 1981 and1996, no doubt partly because of the shrinkage in the

population of large industrial towns like Port Augusta

and Whyalla which are highly anomalous outliers in

otherwise low-density regions (Fig. 4).

Hypothesis 3 suggested that the workforce participa-

tion rate would be highest in the high-density

regions because it was expected that the latter would

have more opportunities for women to enter the

workforce. In fact in 1981 there was a significant

moderately strong negative correlation between density

and participation rate. In that year, before the onset

of the rural crisis, most people living in the outlying

areas of sparse population were evidently there because

they had jobs and the percentage of older people not in

the workforce had not yet been exacerbated by selective

out-migration of the young. Female workforce partici-

pation may indeed have been low but this was evidently

less significant than the low proportion of aged

dependents relative to the numbers of people in the

workforce.

By 1996 a radical change had occurred. The former

coupling of low density with high workforce participa-

tion had almost totally gone and in fact the 1996 r value

of only 0.107 is the lowest in Table 1. The sparsely

peopled areas had by then suffered accelerated ageingand growing concentrations of retirees and other non-

work seekers in mostly coastal communities at a wide

variety of densities had produced a very diffuse relation-

ship.

Fig. 4. Relationship between rural population density and total population for 84 South Australian rural communities, 1996.



13/20

To test whether the above conclusions are different

when the retirement age groups are excluded, for 1996

the workforce participation rate was calculated for the

population aged between 15 and 64 only. The rates with

and without the retirement age groups were in fact very

closely correlated (r 0:883). The 1996 r value of

0.107 was raised to 0.211 when the retirement age-

groups were excludedstill not a statistically significant

relationship.

Hypothesis 4, however, is strongly supported and in

this case the relationship has also gained strength over

time. The denser the rural population, the smaller is the

proportion of the workforce engaged in primary

production as the denser population provides opportu-nities for a greater variety of enterprises and business

types. The study area does contain a number of special

function communities with exceptionally low propor-

tions engaged in farming; if it were not for these the

relationship would be even stronger.

Hypothesis 5 extends the above line of reasoning

suggesting that in densely peopled areas not only would

the proportion of the workforce in agriculture tend to be

smaller but the non-agricultural population would also

tend to be more diverse both by occupation and by

industry producing a more complex and varied social mix.

To test this hypothesis an index of diversity was

devised using a variant of the Gini coefficient based on

the Lorenz curve, which measures the degree to which

the proportions of a population in a set of subgroups

differ from a theoretical situation where all the

subgroups are equal in size. The index of diversity

produced ranges from a value of zero (if the whole

population were in a single category) to 100 (if the

population were evenly divided between all the cate-

gories). For the diversity index used in Table 1 index

values range from a maximum of 55.6 to a minimum of

22.9 (for communities with rural densities of 279 and 10

occupied dwellings/100 km2, respectively).

The subgroups of interest here are the occupational

and/or industry categories into which the workforce is

divided. The social structure of a population is in

principle better expressed by occupation than by

industryi.e. by analysing what jobs people actually do

rather than what type of firm they work for. However, in

this case the breakdown of occupational data available at

Collection District level is restricted to just 10 categories

and includes farmers in the general category of managers

and administrators and farm hands among general

labourers and related workers, etc. The industry data

on the other hand provide 13 categories which do

distinguish primary industry workers adequately and

which are comparable between 1981 and 1996. Thesewere therefore used in the results in Table 1.5 Clearly the

hypothesis is strongly supported. There is a very strong

tendency for the more densely peopled communities to

have a more diversified workforce and again this

tendency has strengthened over time.

Hypothesis 6 relates to the distribution of unemploy-

ment expecting that there would be a positive relation

with rural density because people losing jobs would be

obliged to move out of the sparsely peopled areas

whereas those who lived in more densely peopled areas

would be more likely either to find another job locally or

to remain there for amenity reasons. This hypothesis

held good in 1981 but by 1996 the general economic

stress and associated general increase in unemployment

had all but destroyed the relationship, which was not

significant.

Hypothesis 7 is strongly supported in both 1981 and

1996: the sparser the population, the greater the

Table 1

Tests of the 11 hypotheses: simple Pearson correlation coefficients between rural density and the dependent variables, 1981 and 1996

Variable r (1981) r (1996)

Spatial extent of community (km2) 0.760** 0.766**

Total population of community +0.407** +0.530**

% Workforce participation (persons aged 15+) 0.396** 0.107

% Employed in agriculture 0.547** 0.622**Industrial diversity index +0.489** +0.572**

% Unemployed +0.360** +0.183

Masculinity (males/100 females) 0.616** 0.581**

% Born overseas +0.632** +0.675**

% Changed address, last 5 years +0.322* +0.493**

Fertility index (persons 04/100 females aged 1545) 0.489** 0.496**

% Population aged o15 0.453** 0.253

Mean value of r (disregarding sign), all 11 variables |0.497| |0.480|

**Coefficient significant at the 0.001 level.

*Coefficient significant at the 0.01 level.

5Experimentation with the diversity index showed that, for 1996,

both occupational and industry subgroupings gave very similar results.

Actual values of the diversity coefficient are sensitive to the number of

subgroups used. Compacting the data to 10 subgroups for each, the

indices based on industry and occupation were very strongly correlated

(r 0:885).



14/20

masculinity ratio. Females tend to either remain in, or

migrate into, areas of high density and to move out of

sparsely peopled areas; relatively few rural communities

actually have a female majority in their population, but

there are some sparsely peopled areas with a ratio of

over 120 males to every 100 females. This relationship

had weakened by 1996 since during the rural crisis manyyoung men were forced to leave farms which were no

longer able to provide enough income for more than one

generation.

Hypothesis 8, expecting that rural density would be

positively related to the percentage of overseas born in

the population, was also strongly supported and has

gained strength between 1981 and 1996. Again a few

outliers, notably the steel city of Whyallas internation-

ally recruited workforce and one or two resort towns,

reduce what would otherwise be an even stronger

relationship.

The ninth hypothesisthat in-migration and local

residential mobility would be positively related to rural

population densitywas not significant at the 0.001

level in 1981. However, by 1996 this situation had

changed and the hypothesis is quite strongly supported.

The reason for this radical change is almost certainly the

heavy out-migration from the sparsely peopled areas in

the early 1990s, with very few new in-migrants to replace

those leaving, coupled with a reduction in people

changing address within their own local government

area during economically difficult times.

Hypothesis 10, suggesting a negative relationship

between density and fertility ratio, is also firmly

supported. From the States slowly diminishing ruralurban differences in fertility one might have expected the

relationship to weaken slightly between 1981 and 1996.

However, by 1996 the r value had if anything

strengthened slightly.

Finally, Hypothesis 11 is definitely not supported and

in fact resembles Hypothesis 3 in that the negative sign

of the relationship is opposite to the expected positive

and the strength of that negative relationship has

declined sharply since 1981. Contrary to the proposition

of this hypothesis, the percentage of the population aged

under 15 in a community appears to have been affected

more by the local fertility rate than by the in- or out-

migration of young families. In the pre-crisis conditions

of 1981 the gap between urban and rural birth rates was

still substantial; moreover, the selective out-migration of

young families from the farms had not yet been speeded

up. By 1996 the general ageing of the rural population

was more advanced. The relationship between rural

density and percentage of children in the population was

still negative but no longer significant.

To summarise on the set of 11 hypotheses so far we

have found that rural population density at a given

point in time bears a significant negative relationship to

the spatial extent of the community (km2), the percen-

tage of the workforce employed in primary production,

the masculinity ratio and the fertility ratio. Prior to the

recent rural crisis there was also a significant negative

relationship with the workforce participation rate and

the percentage aged under 15, but by 1996 these were no

longer significant. Rural density bears a significantly

positive relationship to total population size, theindustrial diversity of the workforce, the proportion of

the population born overseas and (in 1996) the propor-

tion who have changed address over the preceding 5

years. The relationship between density and unemploy-

ment was weakly positive in 1981 and insignificant by

1996. Overall, the impact of the rural crisis of the mid-

1980s to the mid-1990s had weakened or destroyed the

former links between density and unemployment work-

force, participation and percentage of children in the

population, while the earlier relationships between

density and five other variables had strengthened.

10. Densitys explanatory power in comparison with that

of the other independent variables

So far we have demonstrated the importance of the

density variable in analysing settlement patterns. How-

ever, we also need to demonstrate, rather than simply

claim, that rural density affects the quality of rural

settlement, landscape and population in a way that is

not simply reducible to some combination of population

size of the main town, urban concentration or remote-

ness. We also need to test whether one or more of these

three might have even stronger relationships with thedependent variables than has rural density. Both of these

factors are influenced by the degree to which the four

independent variables themselves are intercorrelated.

10.1. Approach

The intercorrelation matrix between size, concentra-

tion, remoteness and rural density is given in Table 2.

For this purpose,

(a) size of the main central place is measured simply by

its 1996 population;

(b) concentration of the population into clustered

settlements is measured by the 1996 proportion of

the whole community population resident in places

of 200 or more population;

(c) remoteness is measured by the ARIA score. This

index, to be adopted by the Australian Bureau of

Statistics for the 2001 Census, expresses the distance

of a given place from the nearest urban centres of a

varied range of population sizes (Hugo and Wilk-

inson, 2002). It takes the form of a continuous

variable to two decimal places ranging from 0.00

(most accessible) to 12.00 (most remote);



15/20

(d) density is the 1996 net local density of the defined

community areas excluding major unpopulated

areas and expressed as the average number of rural

households (occupied dwellings) per 100 km2.

Three of the six possible pairs of relationships are

strongly correlated. Not surprisingly the degree of

concentration into urban settlements is strongly linked

to the size of the main town in the community. There is a

somewhat weaker but still highly significant positive link

between rural density and size of the main town. It might

be expected that this correlation would be even stronger

if all country towns were based mainly on local service

functions; however, the sporadic distribution of indus-

trial and mining activity, etc. limits this association.

Finally, there is another very strong link between density

and remoteness. It should be noted that the latter twomeasures are totally independent in their calculation: the

remoteness index for any given point is based on

distance-to-services measures only and pays absolutely

no regard to the nature of the countryside at that point.

The other pairs of correlations are not statistically

significant at the 0.01 level; but do these three strong

correlations render any of the four variables redundant?

To test this a multiple correlation analysis was

performed to analyse the combined effects of all the

four independent variables upon each of the 11

dependent variables (from Table 1) in turn. The back-

ward elimination variant of the model was employed in

preference to the more common stepwise procedure.6

Collinearity tolerance calculated for the four indepen-

dent variables showed that the degree of intercorrelation

did not approach the low point (ca. 0.100) where the

influence of any one variable would be reduced to just a

linear combination of the others thus rendering it

completely redundant.7 All the four independent vari-

ables are therefore initially entered into the multiple

regression equation. In each subsequent iteration the

partial coefficient is calculated for each independent

variable along with its significance level. Partial coeffi-

cients express the correlation between density and each

dependent variable after controlling for or partialling

out the separate linear impact of the other three

independent variables (size concentration and remote-

ness). As the iterations proceed, variables whose partial

(measured by Snedecors F) fails to reach a probability

level of po0:10 are eliminated one at a time, weakest

first. The resulting regression models retain only thoseindependent variables which make a significant con-

tribution to multiple R2 (Tables 3 and 4).

10.2. Results: density as an independent variable in

multiple versus simple correlation

A comparison of the partial coefficients with the

simple Pearson coefficients for rural density is presented

in the first two columns of Tables 3 (1981) and 4 (1996).

This clarifies the extent to which density has an

autonomous influence, once the separate influence of

Table 2

Intercorrelation matrix between rural density and three other independent variables, for 84 South Australian rural communities, 1996

Size (populatn. of main town) Concentration Remoteness Rural density

Size 1.000

Concentration

Sig. (2-tailed)

+0.758** 1.000

0.000Remoteness

Sig. (2-tailed)

0.269 0.154 1.000

0.013 0.161

Rural density

Sig. (2-tailed)

+0.448** +0.191 0.689** 1.000

0.000 0.082 0.000

Relationships significant at 0.001 level italicised and marked with **.

6In the stepwise model, even if there are two independent variables

with almost the same value of simple r; the slightly stronger one first

enters the multiple regression model. Thereafter the model calculates

which of the remaining variables gives the greatest increase in F (and

multiple R); that variable then joins the multiple model. If the first two

variables are intercorrelated, the variance associated with the second-

ranking one may be eliminated prematurely from the further stages in

the model. The backward elimination procedure gives a better measure

of the relative importance of the independent variables, by first forcing

them all into the equation, and then eliminating the ones whose

partials fail to make a significant contribution to multiple R2:

7In multiple correlation analysis, the collinearity tolerance of an

independent variable ranges between 0 and 1, expressing the

proportion of the variance in the dependent variable that is not

explained by the collective influence of the other independent variables

in the equation. The lowest tolerance was 0.324 for 1996 and 0.379 in

1981, in both cases for the size of the main town, when all four

variables were initially input into the models; tolerances for variables

remaining in the final models after the backward elimination process

were naturally higher, ranging up to over 0.900. Collinearity between

town size and urban concentration as independent variables may

cause a problem of interpretation in the case of one dependent variable

(total community population): the sign of its partial with urban

concentration is negative in Tables 3 and 4, as against the intuitively

expected positive.



16/20

one or more of the other three variables is removed.

Most frequently, the partial correlation between an

independent and a dependent variable will tend to

be lower than their simple Pearson correlation. The

difference between the two is a function of the extent

to which the simple relationship of that variable with

density is affected by the joint influence of population

size, concentration and remoteness. In 1981, for all 11

dependent variables simple r was significant at the

0.01 level or better, and in only two cases was the

influence of the other variables sufficient to reduce

the partial coefficient below this significance level

(Table 3). In 1996, Pearsons r for eight of the 11

variables was significant at the 0.001 level (nine at

0.01), and none of the equivalent partials dropped below

0.01.

Table 3

Comparison of the explanatory power of density, concentration, size and remoteness on 11 dependent variables: partial coefficients from multiple

regression model (backward elimination), 1981, with summary comparison of Pearsons r

Variable Rural density Urban concentration Size (main town) Remoteness R (var. 14) R2 (var. 14)

(simple r) (partial) (partial) (partial) (partial)

Population +0.407** +0.254 0.748** +0.958** 0.974 0.949Area 0.760** 0.834** 0.760** +0.798** +0.348** 0.934 0.872

% Prim. Indust. 0.547** 0.699** 0.857** 0.902 0.814

Ind. diversity +0.489** +0.501** +0.628** 0.734 0.539

Masculinity 0.616** 0.621** 0.490** 0.727 0.528

% born o/seas +0.632** +0.355** +0.409** 0.221 0.716 0.513

Wkfce participn. 0.396** 0.447** 0.582** +0.292* 0.694 0.482

Fertility index 0.489** 0.237 0.314* +0.241 0.596 0.355

% o15 years 0.453** 0.518** 0.425** +0.355** 0.592 0.350

% Unemployed +0.360** +0.324* +0.429** 0.538 0.290

% Chge. Address +0.322* +0.285* +0.347** 0.460 0.212

|Mean| simple ra 0.497 0.463 0.425 0.409

Cases of pro:001 10/11 9/11 8/11 8/11

|Mean| partialb 0.355 0.441 0.280 0.129

pp0:01:

pp0:001:aMean of simple Pearsons r for all 11 dependent variables including those not statistically significant, disregarding sign.bMean of all 11 partials including those not statistically significant, disregarding sign.

Table 4

Comparison of the explanatory power of density, concentration, size and remoteness on 11 dependent variables: partial coefficients from multiple

regression model (backward elimination), 1996, with summary comparison of Pearsons r

Variable Rural density Urban concentration Size (main town) Remoteness R (var. 14) R2 (var. 14)

Simple r Partial Partial Partial Partial

Population +0.530** +0.277* 0.702** +0.953** 0.977 0.954

Area 0.766** 0.828** 0.735** +0.777** +0.386** 0.927 0.859

% Prim indust. 0.622** 0.715** 0.789** 0.877 0.769

% Born o/seas +0.675** +0.446** +0.340* 0.245 0.739 0.546

Masculinity 0.581** 0.310* 0.495** +0.290* 0.732 0.536

Ind. diversity +0.572** +0.563** +0.503** 0.706 0.498

% Chge. address +0.493** +0.359** +0.243 +0.205 0.692 0.479

Fertility index 0.496** 0.467** 0.305* 0.562 0.316

Wkfce participn. 0.107 0.185 0.485** +0.318* 0.505 0.255

% o15 years 0.253* 0.324* 0.379** +0.230 0.186 0.472 0.223

% Unemployed +0.183 +0.427** 0.427 0.182

|Mean| simple ra 0.480 0.454 0.449 0.369

Cases of pro0:001 8/11 10/11 7/11 6/11

|Mean| partialb 0.351 0.428 0.272 0.136

pp0:01:

pp0:001:aMean of simple Pearsons r for all 11 dependent variables including those not statistically significant, disregarding sign.bMean of all 11 partials including those not statistically significant, disregarding sign.



17/20

Less frequently, in a multiple correlation analysis the

partial coefficient may be higher than the original simple

Pearson coefficient once the confusing element of other

variables is controlled for, and this is the case for a

number of the hypotheses tested. In 1981, the partials

exceeded simple r in six of 11 cases, and in four cases in

1996. However, as expected, the two tables show that onthe average, controlling for the influence of the other

three independents did indeed reduce the value of simple

r (disregarding sign) for the relationships between

density and the 11 dependent variables. The influence

of the other variables was strongest in the cases of total

community population (both years), the percentage of

the population born overseas and the fertility index

(1981), and the masculinity ratio (1996). However, the

strength and degree of significance of the partials shows

that density is not merely acting as a proxy, but retains a

strong independent relationship in its own right in both

years.

10.3. Results: comparison of the relative importance of

density, urban size, urban concentration and remoteness

We turn now to the question: although we have

shown that rural density bears an important relationship

to the 11 social variables investigated, does one or more

of the other three critical independent variables have an

even greater influence? First, to compare the four

variables in the strength of their correlations using

simple r; summary data appear in the two second-last

rows of Tables 3 and 4. In both years, of the four

independents rural density had the strongest mean valueof r; disregarding sign. In 1981 it also had the greatest

number of the 11 relationships significant at the 0.001

level, and in 1996 the second highest.

A more important test, however, is provided by a

comparison of the partial coefficients. Columns 36 in

Tables 3 and 4 show which of the four independents were

retained in the final equation in respect of each

dependent variable, and the partial coefficient for each.

The significance levels of the relationships are indicated

by asterisks together with italicising of the appropriate

numbers. The rows in Tables 3 and 4 are sorted in

descending order of the strength of multiple R2; showing

the extent to which the variance of the dependent

variable is statistically explained, jointly, by the indepen-

dent variables remaining in the model. Multiple R2 gives

the percentage of the total variance statistically explained

by the variables retained in the model, reaching 0.5 or

above in six cases. The bottom row gives a comparison of

the general performance of the four independent vari-

ables in the multiple regression analysis, when all four

are initially included in the models, and before any are

removed by the backward elimination process.

Not surprisingly, over a 15-year gap spanning a major

rural crisis, the relative explanatory power of the four

independent variables changes somewhat, and the

average strength of the partials has declined slightly

for all except remoteness. However, in both years it is

clear that rural density and in particular urban concen-

trationthe percentage of the total community popula-

tion resident in clustered settlements of at least 200

personsare far more significant than the other twoindependent variables, in respect of the 11 hypotheses

under consideration. Density and concentration take

into account the rural matrix in which the town is set,

unlike the other two measures, which relate only to the

main town in the community.

The former two measures do, however, require data

for a meaningful surrounding rural area functionally

linked to the town. This at present is a major task, due

to deficiencies in the Australian Standard Geographical

Classification, as used in the Census. The ready

availability for many years of the town population size

variable explains its widespread use as a surrogate

measure of many aspects of rural communities. Popula-

tion size of the main town, though, is the strongest

explanatory variable only in the obvious cases of total

population and total area of the whole community (both

years) and in the proportion of overseas born persons

(1981). As to the remoteness score, despite its close

correlation with density (Table 2) after controlling for

the other independent variables it remains in the

multiple correlation model for only four of the 11

dependent variables (three in 1981), and never achieves

more than third or fourth importance. However, the

recent availability of the standard ARIA scores on the

Web will also make remoteness a very useful measurefor many purposes in the future, even though it does not

play a very important role in the present study.

The multiple correlation analysis can also answer one

further pertinent question: if density measures were

omitted from the multiple regression model altogether,

would the other three indicators alone (as proposed by

Coombes and Raybould, 2001) give equally good R2

values? A trial of the multiple correlation models with

and without net rural density shows that inclusion of the

density variables (a) raises the mean value of multiple R2

for the 11 dependent variables from 0.42 to 0.51 and (b)

gives a more realistic picture of the role of remoteness,

clarifying its close relationship with rural density

(Table 2). Including rural density in the backward

elimination process leaves remoteness as a significant

variable in just four, instead of eight, of the 11 models

for 1996.

11. Conclusions

The present paper has, we hope, sufficiently demon-

strated the importance of rural population density as a

significant phenomenon in social, settlement and



18/20

population geography, well worthy of investigation in

its own right. We argue that, at the local level, rural

density is a more sensitive and meaningful measure than

the more commonly used gross density that includes

both urban and rural components. Examining local

rural population density, we have demonstrated that it is

quite strongly associated with a large range of importantdemographic and socio-economic indicators in synoptic

studies at the 1981 and 1996 censuses, timed to show the

effects of the major rural downturn in Australia between

1985 and 1993. While three of the 11 significant 1981

relationships between density and the selected variables

had lost significance by 1996, the average value of r

dropped only marginally, from 0.497 to 0.480 (disre-

garding sign). Thus, rural density remained as a very

important indicator of the chosen community qualities.

A comparison of rural density with the other three

independent variables was carried out, controlling in

each case for the influence of the other three by the use

of partial correlation coefficients. This showed that in

both years the best overall predictor of the 11 dependent

variables in terms of the average partial coefficients was

urban concentration, closely followed by rural density;

for simple r, the inverse was the case. In both years these

two key variables were much less affected by controlling

for the influence of the other independents, than were

urban size and remoteness. The population size of the

largest town in the community had the highest partial

coefficient in only one of the 11 multiple regression

models, and the remoteness index p

Rural Population Density

Documents

Transcript of Rural Population Density