Rethinking Remoteness: a Simple and Objective Approach

Rethinking Remoteness: A Simple and Objective Approach

ZHAO YUEJEN

Adjunct Senior Research Fellow

Institute of Advanced Studies

Charles Darwin University

GUTHRIDGE STEVEN

Adjunct Principal Research Fellow

Institute of Advanced Studies

Charles Darwin University

Postal Address:

Health Gains Planning, 4th Floor, AANT Centre

DHCS, PO Box 40596

Casuarina NT 0811

Email address:

Rethinking Remoteness: A Simple and Objective Approach

Abstract

This paper re-examines the characteristics and assumptions of current remoteness/accessibility classifications in Australia and proposes a simple and easily understandable alternative measure for remoteness. In this study, remoteness is redefined simply as “the average distance between two nearest people within an appropriate spatial unit where population distribution is assumed to be homogenous”. By definition, the most straightforward remoteness and incapacity index (RII) would be remoteness times a measure of the incapacity for social and commercial interaction, where remoteness is gauged by the square root of the area divided by the population, and incapacity is measured by the reciprocal of population.

Australian Bureau of Statistics Statistical Local Area (SLA) level population data and digital boundaries have been utilised for assessment of this index. The utility of the RII is demonstrated with two examples of activity measures for general practitioner services and businesses. At the State/Territory level, RIIs are negatively related to both general practitioner services per person (Pearson correlation coefficient r=-0.873), and the number of businesses per person (r=-0.546). The correlation can be further enhanced by normalising the distributions of the remoteness scores with a simple logarithmic function. The strong correlations confirm that remoteness has a substantial inverse impact on daily activities. Greater distance means longer time and higher costs for travelling, diseconomy of scale, and higher personnel costs. The RII provides an alternative measure of remoteness that is both intuitive and statistically straightforward, and at an SLA level, closely coincides with the commonly used but complex Accessibility/ Remoteness Index of Australia Plus (ARIA+). Significantly the RII is free of the service specific and policy sensitive adjustments justified by “accessibility” that have been introduced into existing measures.

Key words: Remoteness, Population Density, Geographical Classifications, ARIA, ASGC, Accessibility.

Acronyms

ABS Australian Bureau of Statistics

ACT Australian Capital Territory

ARIA Accessibility /Remoteness Index of Australia

ARIA+ Accessibility/ Remoteness Index of Australia Plus

ASGC Australian Standard Geographical Classification

CD Collection Districts

GISCA National Centre for Social Applications of Geographic Information Systems

GP General practitioner

MDD Mean distance deviation

NSW New South Wales

NT Northern Territory

QLD Queensland

RDR Relative dispersion ratio

RII Remoteness and incapacity index

RRMA Rural, Remote and Metropolitan Areas classification

SA South Australia

SARIA State based ARIA

SLA Statistical Local Area

TAS Tasmania

VIC Victoria

WA Western Australia

Introduction

This paper re-examines the assumptions and characteristics of current remoteness/accessibility classifications in Australia and proposes a simple and easily understandable alternative measure for remoteness. It is critical that agencies that operate in rural and remote areas have a transparent and objective remoteness measure to inform equitable resource management and service delivery.

In the early 1970s, Holmes (1973) applied a range of key geographical measures developed overseas (Bachi, 1962; Neft, 1966) to investigate population dispersion for Australian States. Subsequently, Faulkner and French (1983) developed a remoteness index by measuring the distances from 702 grid squares of Australia to the closest urban centres and transforming the distances to scores of a standardised normal distribution. In 1987, Holmes (1988) defined remoteness by using area divided by population (in km2 per person) and found it was highly correlated with the Faulkner and French index, but with an algorithm that was significantly simpler. In a later development (Department of Primary Industries and Department of Human Services and Health, 1994), the Rural, Remote and Metropolitan Areas classification (RRMA) was released as a remoteness classification based on 1991 Census and Statistical Local Area (SLA) boundaries. The RRMA classification introduced the additional concept of “accessibility” by the measurement of distance from an SLA to the nearest centre in each of three zones - metropolitan, rural and remote. In the same year, Griffith (1994) developed a detailed methodology for quantifying accessibility to a particular service or a defined group of services by combining population, economic resources and a composite element of distance, time and cost.

In 1999, the National Centre for Social Applications of Geographic Information Systems (GISCA) completed a commission by the Australian Department of Health and Aged Care to develop a new classification system, the Accessibility /Remoteness Index of Australia (ARIA)(Dunne et al, 1999). The ARIA approach had four steps: first, define four categories of service centres based on subjective thresholds of population size; second, measure the road distances from any particular point to the nearest four centres; third, divide the distance by the mean and truncate the maximum score at 3; and finally, aggregate the scores to a single index with a maximum value of 12. This approach results in five categories of regions: highly accessible, accessible, moderately accessible, remote and very remote (Department of Health and Aged Care, 2001). Since then the classification has been further modified for specific purposes, for example ARIA+ and State based ARIA (SARIA). ARIA+ adopted five categories of service centres and therefore used a spread of scores 1-15, whereas SARIA maintained the ARIA+ algorithm except that it measured road distance from each populated locality to the nearest service centre in the same State/Territory (GISCA, 2004). The Australian Bureau of Statistics (ABS) has adopted ARIA+ as a remoteness classification (ABS, 2003).

For ARIA, remoteness has been defined in terms of distance of a location from service or population centres (Department of Health and Aged Care, 2001). Importantly the definition does not consider the distances between individuals or between service centres. In this study, “remoteness” is redefined simply as “the average distance in space between two nearest people within an appropriate spatial unit where population distribution is assumed to be homogenous”. This measurement directly reflects the ease with which people can interact. The greater the average distance between people, the more remote is the area, the fewer the opportunities for social interaction and the less opportunities for both supply and demand. This definition for remoteness consists of only two elements, distance in space and number of persons. Remoteness is a characteristic of geographic localities in relation to their human inhabitants. Apart from distance, the ability to travel and interact is also clearly reliant on infrastructure, such as roads, public transport and communication, but these factors are themselves a direct consequence of population size.

Accessibility is a concept closely related to remoteness, but is deliberately excluded in the proposed measure for “ remoteness”. The reason is that once accessibility is incorporated into a remoteness measure, the measure is no longer simple and more importantly cannot be consistently generalised. Accessibility is highly influenced by “availability” and can only be defined in relation to a specific service. Without a clear definition of type of service, accessibility is unquantifiable (Griffith, 1994; 2002). In distinguishing service access models from geographic models, Griffith (2007) has recently defined the separate elements for both types of classification.

Method

Analogous to Holmes (1988), this study proposes a microgeographic approach to population density as a measure for remoteness. By definition, the most straightforward remoteness and incapacity index (RII) would be remoteness times incapacity, where remoteness is gauged by the square root of area divided by population, and incapacity is measured by the reciprocal of population in thousands. Thus,

RII=a1/2p-3/2 (1)

where a=area and p=population with a>0, p>0 and RII> 0, ranging from least remote (close to 0) to most remote (much greater than 0).The use of the square root transforms area to distance, with area assumed to be square. The RII represents the average distance that people have to travel over the capacity for interaction when, for example, a service provider delivers a service to a client. The RII is directly proportional to per capita average travelling distance for the area and inversely proportional to the size of the resident population. The population distribution is assumed to be homogeneous within the area of interest.

The RII is a different concept from Holmes’ measure of dispersion (Holmes, 1973), which measures the mean distance deviation (MDD) of the population from the median centre (intersection of the population middle values along the orthogonal axes) or mean centre. A related measure, the relative dispersion ratio (RDR), is the standard distance deviations for the population and the area, and is used to offset the affects of area size and population density. Holmes applied dispersion to describe population concentration. For example, the United Kingdom (RDR=0.86) was reported as more dispersed and less concentrated than Australia (0.72) (Holmes, 1973). This result highlights the difference between dispersion and remoteness and do not indicate that the United Kingdom is more remote than Australia.

Optimally, a remoteness measure would be: transparent and simple to calculate; easy to aggregate and disaggregate to cover all levels of spatial units; scientifically plausible; continuous, informative and comprehensive. The measure also needs to be based on a minimum of assumptions and be free of subjective manipulation.

Simplicity: This approach simply uses the average distance in kilometres per person as the basis to quantify remoteness adjusted by incapacity. It can be applied using a range of data sources, including publicly available ABS population and area data for SLA or census Collection Districts (CD). There are no difficulties in calculating RII and comparing it with other areas. The algorithm is transparent and easy to understand.

Aggregation: RII can be easily aggregated to a higher spatial or administrative unit without the need for weighting, and disaggregated to a lower spatial unit without losing comparability with neighbouring areas.

Continuity: RII is a continuous geographical measure for remoteness. It can be applied to any level of geography, including geocoded localities. By defining a centroid with appropriately sized area (for example, 10´10 km2) as an information capturing grid cell and scanning the whole of Australia, every locality has an RII value. For convenience, this study calculates the RII values using the existing Australian Standard Geographical Classification (ASGC) digital boundaries (CDs, SLAs and State/Territory), and ABS SLA level population data (ABS, 2002).

(Insert Table 1 here)

Results

The RII values for each Australian State and Territory are presented in Table 1. The Northern Territory (NT) is the most remote with an RII value of 13.057, followed by Tasmania (TAS), Western Australia (WA) and South Australia (SA). The Australian Capital Territory (ACT) outweighs Queensland (QLD), because of the small population and less capacity. New South Wales (NSW) and Victoria (VIC) are the least remote States with RII values less than 0.1. MDD and RDR are also presented using 2001 census data updates for comparison in Table 1. NT is ranked first by MDD and second by RDR in terms of dispersion, while the MDD and RDR are inconsistent in determining the least dispersed State/ Territory. Figure 1 shows the RII values mapped by SLA.

The application of RII can be simplified by categorising the scores into six broad conventional regions: urban (0<RII <0.002), suburban (0.002≤RII <0.008), regional (0.008≤RII <0.05), rural (0.05≤RII <0.5), remote (0.5≤RII <2), and very remote (RII ≥2). The population proportions for these categories, calculated at an SLA level are presented for each State and Territory in Table 2.

(Insert Figure 1 here)

Figure 1 Remoteness and incapacity index defined regions mapped by Statistical Local Areas (ASGC 2001)

(Insert Table 2 here)

The utility of the RII was tested in two contrasting examples by Pearson correlation between RII and activity. For the first example, general practitioner (GP) annual per capita services data by State/Territory between 2000-01 and 2005-06 was obtained from Medicare Australia’s online database (Medicare Australia, 2007). It was shown that the annual per capita GP services and RII were negatively correlated (Pearson correlation coefficient r=–0.873). This indicates the more remote and less capacity, the more difficult it is for residents to utilise Commonwealth funded primary care medical services. Model diagnostics showed the NT was both outlying and influential. To increase the robustness of the model, the logarithm of RII was used. Consequently, the correlation coefficient has increased substantially (r=-0.908).

The calculation for this example highlighted the wide variation of scores and that an adjusted index with a normalised distribution could be estimated in the form of k+log(a1/2p-3/2), where k can be estimated by solving the following equation with a known travelling cost relativity C:

. (2)

This equation can be rewritten as

. (3)

At the State/Territory level, using the distance ratio in a1/2p-1/2 between the most and least remote State/Territory as a surrogate measure for the cost relativity, it follows that C=12 in rounded terms, and then k=1.58. The resulting cost adjusted RII values are listed in the last column of Table 1.

For the second validating example, we utilised data on the number of businesses per person (ABS, 2001). Pearson correlation and regression were again used for analysing the relation between remoteness and business activity. It was found that the cost adjusted RII and the number of businesses per person was moderately related (r=-0.546). Again it is indicative that the more remote the area, the lower the business activity.

(Insert Figure 2 here)

Figure 2 Correlation between RII and ARIA+ scores by Statistical Local Areas, Australia, 2001

The RII is also consistent with existing but more complex measures. The ARIA+ scores by SLA were obtained from the GISCA website (GISCA, 2007) and merged with the RII data. At the SLA level, it is estimated that C=138 and k=4.00. A set of cost adjusted RII values were derived and a scatter diagram was plotted to study the relationship between ARIA+ and the adjusted RII values (Figure 2). It revealed that at the SLA level, adjusted RII and ARIA+ were closely correlated (r=0.700), after excluding those SLAs for which ARIA+ scores are ‘null’. If SLAs with a population less than 100 are excluded, r increases to 0.747. If SLAs with a population less than 500 are excluded, r increases only marginally to 0.749. The analysis also highlighted that ARIA+ scores were artificially concentrated around 0 and 3. Further scrutiny of Figure 2 suggested that RII outlying localities were likely new/outskirt rural suburbs or national parks with small population and relatively large area (eg. Willawong ARIA+ 0:RII 3.8, Pialligo and Yarra Ranges (B) 0.1:4.3, and Majura 0.2:4.4), but ARIA+ outlying localities were more likely remote mining or Indigenous townships with relatively large population and small geographical area (eg. Nhulunbuy 12:2.1, Torres 15:2.6, Weipa 12:2.5 and Tennant Creek 12.1:2.6).