1367.2 - State and Regional Indicators, Victoria, Mar 2009  
ARCHIVED ISSUE Released at 11:30 AM (CANBERRA TIME) 14/05/2009   
   Page tools: Print Print Page Print all pages in this productPrint All


Estimated Resident Population
Estimated Resident Population for Statistical Local Areas
Developing the Models
Limitations and Implications
Roles of the ABS Victorian Regional Office in the Validation Process
Revising SLA Population Estimates
Data Availability


Population estimates are one of the major outputs of any statistical office. They are critical for a wide range of planning and policy decisions. While population can be simply defined, such as the 'total number of persons living in an area', the concepts associated with its measurement are complex.

In Australia, the key population measure is estimated resident population (ERP), which is defined in terms of the number of people who usually live within Australia and its states and territories. The "usual residence" population concept refers to all people, regardless of nationality or citizenship, who usually live in Australia, with the exception of foreign diplomatic personnel and their families. It includes usual residents who are overseas for less than 12 months and excludes overseas visitors who are in Australia for less than 12 months.

This article explains how the ABS derives ERP including the difference in the methodology used for national/state ERP and ERPs for geographic areas below the state level. The role of the Victorian regional office in producing these estimates is also discussed.


There are several legislative requirements for the ABS to produce population estimates. For example sub-section 9(2) of the Census and Statistics Act 1905 requires the quarterly estimation of the population for each state.

ERP is used in legislative decision making such as determining the number of seats each state/territory has in the House of Representatives and as the relative distributional basis for Commonwealth grants to states/territories and local government authorities. ERP is also used for per capita measures such as fertility and mortality rates, survey benchmarks, federal electoral boundary redistribution and budget planning.

The ABS conducts a Census of Population and Housing every 5 years, however this does not provide information about a population's size and composition in the years between Censuses, defined as the 'intercensal period'. The initial Census year estimate is determined after a process called 'rebasing' which involves a number of steps:
  • obtaining place of usual residence data from the Census: in the Census usual residence means the address where a person has lived or intends to live for 6 months or more in the year of the Census (excluding overseas visitors in Australia for less than 12 months);
  • adjusting for over or under enumeration;
  • demographic adjustment;
  • adjusting for residents temporarily overseas on Census night; and backdating to 30 June from Census night: this is necessary to get a consistent reference date for population estimates post-Census.

After establishing the 30 June Census year base, quarterly ERP over the intercensal period are calculated by ageing the base and then adjusting for subsequent components of population growth, i.e. adding natural increase (births minus deaths) and net overseas migration (estimated using passenger cards), and for the states and territories, net interstate migration (estimated using Medicare data). This method is known as the 'cohort component method' and uses the 'demographic balancing equation', which is recognised as the ideal approach for estimating population.

Quarterly updating of the population estimates for Australia, states and territories continues until the next Census results are available and the rebasing process is repeated to obtain a new starting point for the next intercensal period. The diagram below shows how population estimates are updated for one quarter.

Quarterly Preliminary Population Estimate, Australia - 30 September 2006
Diagram: Quarterly Preliminary Population Estimate, Australia—30 September 2006

Note: The first release of the September 2006 estimate was based on the 2001 Census. When the June 2006 ERP was available as 'rebased' from the 2006 Census, the September 2006 ERP was recalculated from this new start population. Following this, the estimates undergo several revisions as more up-to-date data on births, deaths and migration becomes available. Statistics in the above diagram are the latest available.

The Census year ERP based on the current Census also provides a measure of how accurate the ERP was for the previous intercensal period. For example, the preliminary population estimate at 30 June 2006 calculated from the 2001 Census base using births, deaths and migration as described above was compared to the rebased June 30 estimate calculated from the 2006 Census. Using the cohort component method over the 5 years between 2001 and 2006 resulted in an underestimation of Australia's population. This difference is known as the 'intercensal error'.

After the intercensal error is determined, all quarterly ERP data for the previous intercensal period are revised using information from the Census on interstate migration and then spreading the remaining discrepancy evenly across the quarters. The initial revisions are 'preliminary rebased' estimates which are then updated again due to revision of the components in the intercensal period. The intercensal estimates are then updated once more to produce 'final rebased' ERP and no subsequent revisions are made after this process.

For further information on the rebasing process, see the Feature Article Final Rebasing and Revision of Australia's Population Estimates, September Quarter 2001 - June Quarter 2006 in Australian Demographic Statistics, Dec 2007 (cat. no. 3101.0).


'Sub-state' or 'small area' population estimates relate to areas below the state/territory level of geographic disaggregation. They are widely used by all levels of government, as well as business and the community. Estimates are produced for statistical local areas (SLA) which build up to local government areas (LGA) and statistical divisions (SD). For further information, see Australian Standard Geographic Classification (ASGC), July 2008 (cat. no. 1216.0).

While the cohort component method is considered the ideal method for estimating population in non-Census years, there are no reliable migration data at the SLA level making it very difficult to estimate SLA population using natural increase and net migration.

The method used to compile Census date SLA population estimates is similar to that used for national and state/territory estimates. However, as the demographic balancing equation can not be applied for post-Census estimates, mathematical models are used instead.

The models establish relationships between changes in population and changes in population indicator data between the two most recent Censuses for groups of SLAs. Post-censal changes in these indicators are then used to estimate changes in the population of SLAs for the years following the most recent Census.

Population indicators are data that can be used to estimate total population change over time. They need to be:
  • available for the entire estimation period;
  • consistently defined; available at the relevant geographic level (or able to be converted to the relevant level) and;
  • timely (available soon after the reference period).

Models are revised after each Census to ensure that the indicators used and the relationships established are providing the best outcome for SLA population estimation in each state/territory. Results from the 2006 Census enabled the ABS to develop new models for the 2007-2011 period. The most statistically robust model for each area is selected (i.e. the one that produces the ‘best’ estimates). This is the model which provided estimates for 2006 that are closest to the final (rebased) 2006 estimates.


A number of factors are taken into account when determining the best models to use. For example, characteristics such as population growth rates may vary quite considerably between SLAs. In acknowledgment of these differences, SLAs within a state are separated into subsets (known as strata) based on factors such as location (urban or rural) and population growth (high or otherwise).

More accurate estimates may then be calculated based on similarities existing within these subsets of SLAs and their relationship to particular combinations of indicators. Some indicators are more closely related to population change for some SLAs than others and usually a combination of indicators work better than a single indicator.

The selection of indicators varies across states and territories. For Victoria, the current set of models use indicator data from ABS dwelling approvals, Medicare enrolments and Australian Electoral Commission (AEC) enrolments.


Dwellings (approvals)

Dwelling counts from the last Census are used as the base number of dwellings by SLA. Updated estimates of dwellings are prepared by adding approvals to the Census base. Dwellings approvals are divided into houses and flats/apartments, which generally have different building lag times. A 6 month lag is applied to housing approvals and 12 months for flats/apartments.


Medicare enrolments for men, women and children are provided by postcode to the ABS by Medicare Australia. Generally, changes to the number of Medicare enrolments provide a reasonable indication of total population change.

Again, there is an assumed discrepancy between the time a person moves and changes their address, therefore Medicare enrolments are lagged by 3 months from the reference date.

Australian Electoral Commission

Counts of people by sex on the Commonwealth electoral roll are provided to the ABS by the Australian Electoral Commission (AEC). The data are provided at collection district (CD) level and aggregated to the SLA level. AEC data are also lagged by 3 months and not all strata models use these data.

Different weights are applied to each indicator in different models, as the relative importance of each indicator changes according to the stratum to which an SLA belongs. For example, births can have a greater impact on population growth in the fast growing suburban areas of Melbourne that are attracting families. Therefore the model used for these SLAs may give a greater weight to the number of children enrolled in Medicare. Similarly, some models may use AEC data while others don't, as this indicator may have a closer relationship to population change in some areas compared with others.


Using the indicators to estimate change in population under various assumptions does have some limitations due to the time lag, coverage and quality of the indicator data sets:
  • The length of lag time for dwelling approvals has been tested, however in some large building development projects a longer lag may occur;
  • Accounting for permanent residents of serviced apartment buildings that were completed during the intercensal period is not incorporated in the models. Anecdotal evidence provided by state and local governments suggests that serviced apartment buildings in some inner Melbourne areas are being partially occupied by permanent residents. However, only approvals for residential buildings are used as inputs into models. Serviced apartment buildings are classified as 'non-residential buildings', therefore people who are usual residents of serviced apartments may not be directly picked up by the approvals data and may need to be accounted for in other indicator data;
  • Even though people can only reside in occupied dwellings, the base dwellings count from the Census incorporates both occupied and unoccupied dwellings. This is consistent with the unknown eventual occupancy status of an approval. Using dwelling approvals as an indicator of population change assumes that the occupied/unoccupied ratio of approvals is the same as that of Census dwelling counts, which has limitations if the assumption does not hold over the estimation period;
  • Medicare data are currently only available by postcode and converted to SLA using a concordance. The quality of the data is highly dependent on the quality of the postcode to SLA concordance;
  • The models rely on the accuracy of datasets maintained for administrative purposes by Medicare and the AEC. While adjustments are made to the data such as applying a time lag to the reference period, there may still be implications for the statistical quality of the data;
  • Undercoverage of some sub-populations. There are issues for enumerating some sub-sections of the population in the Census, which affect the base population for ERP.These population groups may also be under-represented in the indicator data. Overseas students are one example, which may have an impact on Victoria due to the large volume of overseas migration for study purposes. For example, areas around Inner-Melbourne present a particular issue for enumerating growing populations of overseas students. The AEC and Medicare indicator data set may not capture overseas students who are considered to be usual residents of Australia for the purpose of ERP. They may also be left out of the Census base count depending on how the question for usual residence is interpreted;
  • In using models, we are making an assumption that the relationship between past population change and indicator data will continue into the future - if the relationship breaks down during the intercensal period then the models cannot account for it.

Aside from these known issues and limitations, the models are generally effective and accurate for the majority of SLAs. The use of Medicare and/or AEC along with dwelling approvals data goes some way to counteracting some of the problems described. However, awareness of indicator data characteristics and the potential impact on estimates within each state is a crucial part of producing local level ERP. Therefore, validation of ERP produced by the models is a significant part of the process for estimating resident population.


After initial estimates of SLA ERP are produced by models it is considered vital to confront the results against local knowledge. The models have limitations as discussed above, so incorporating expertise in the regional offices of the ABS not only allows for local intelligence to be drawn into the process but also allows a wider use of resources for increased scrutiny of the initial estimates.

The validation process undertaken within the Victorian office is a major exercise. Significant effort is placed on gathering supplementary data and information leading up to the validation and analysis phases, prior to receiving modelled ERP.

The growth and contribution of indicator data to the population is considered more closely using trend analysis over time, local knowledge of the area and a number of validation resources. The key resources used are:
  • ERP Local Government Authority Survey; each year the Victorian Regional Office sends a survey to all local government councils requesting information relevant to population change within their Local Government Area for the previous year;
  • Regional intelligence; throughout the year information is gathered from media reports to develop an understanding of current proposals, completions and delays in housing developments to assist in assessing the underlying dwelling approvals data;
  • Consultation with state and local government agencies is undertaken as needed to further clarify the changes in population.

The models are used to estimate population for 200 SLAs in Victoria(footnote 1) , so assessing all of them in detail would involve a high overhead of resources and time; therefore efforts are concentrated on areas of significant change. The first stage of the validation process is to short-list SLAs which will be subject to more detailed scrutiny. During preparation, some of these are selected based on the results of regional intelligence, the ERP Local Government Authority Survey or other factors identified in previous validation rounds. To ensure that the majority of SLAs are selected based on the significance of change since the previous year's estimate, a validation tool is used, which performs a series of tests on the data. Based on these tests, particular SLAs are selected for detailed scrutiny.

The validation data sources and local information may be used to adjust the modelled estimate for a particular SLA. The aim of validation is to account for population changes which may not be (directly or indirectly) picked up by the models. For example, a dwelling approval for a large block of flats may have been taken into account in the initial estimate of the ERP for an SLA. However, if local knowledge indicates that the building has not been completed and occupied within 12 months of the approval, the population estimate may need to be adjusted downwards.

The Victorian Regional Office of the ABS regularly assesses validation sources and investigates others that may be useful in understanding population growth for small areas in Melbourne and Regional Victoria.


Part of the complex nature of estimating usual resident population is the revision cycle. To meet conflicting demands for accuracy and timeliness there are several versions of sub-state population estimates.

For sub-state estimates, preliminary data are normally available nine to ten months after the reference date, so for the year ended June 30 the data are released by April the following year.

Revised estimates are then provided 12 months later. This is because estimates at the SLA level are constrained to state/territory population estimates. When those estimates are revised following updates to components of natural increase, net overseas migration and net interstate migration at the state and territory level, the sub-state SLA populations are also adjusted to add to the revised state and territory totals.

Preliminary Rebased and Final Rebased estimates are calculated after the next Census. Once a Census is held, new population estimates for each SLA at 30 June in the Census year are calculated as described above under 'Estimated Resident Population'. The models have also calculated preliminary 30 June SLA ERP for that year based on the previous Census, which allows an assessment of the performance of the models over the intercensal period. As a result, all ERP for all sub-state areas for the intercensal period are rebased by apportioning the intercensal error evenly across the 5 years, but constraining it to state/territory totals.

A summary of the accuracy of the preliminary 2006 ERP is provided in paragraphs 12 to 15 of the Explanatory Notes of Regional Population Growth, Australia, 2007-08 (cat. no. 3218.0).


National and state/territory ERP is produced for each quarter and data are available five to six months after the reference date in Australian Demographic Statistics (cat.no.3101.0). The most recent release was for September quarter 2008, while ERP for the December quarter 2008 will be available in early June 2009.

The ABS publishes small area ERP annually in Regional Population Growth, Australia (cat. no. 3218.0). The most recent publication, released in April 2009, provides ERP by SLA, LGA and SD for the 2007-08 financial year. ERPs for other geographical areas are available on request.

ABS will also release an updated version of Population Estimates: Concepts, Sources and Methods (cat. no. 3328.0.55.001) in June 2009 which will provide further information on the ABS process for estimating resident population.

Australian Bureau of Statistics (ABS) 1999, Demographic Estimates and Projections: Concepts, Sources and Methods, 1999, cat. no. 3228.0.

ABS 2008, Feature Article: Final Rebasing and Revision of Australia's Population Estimates, September Quarter 2001 - June Quarter 2006, in Australian Demographic Statistics, Dec 2007, cat. no. 3101.0.

ABS 2009, Regional Population Growth, Australia, 2007-08, cat. no. 3218.0.

ABS 2009, Australian Demographic Statistics, Sep 2008, cat.no.3101.0.

1 There are currently 210 SLAs in Victoria, however in areas where indicator data are unreliable and migration can be assumed to be insignificant, population change since the previous Census may be estimated by adding estimates of natural increase (births minus deaths) since the previous Census. In some very small areas population change since the previous Census may be assumed to be zero in the absence of any reliable indicator data for these areas. <back