Outlined below are issues to consider in choosing an appropriate series of Aboriginal and Torres Strait Islander population estimates, for the purpose of analysing or reporting on this population over time.

# Guide to using historical estimates for comparative analysis and reporting

Guide to choosing an appropriate series of Aboriginal and Torres Strait Islander population estimates for comparative analysis and reporting.

## Introduction

## Why the ABS recasts Aboriginal and Torres Strait Islander population estimates after each Census

The Census of Population and Housing (Census) provides the basis of Australia's official population estimates. For the total Australian population, quarterly estimates can be produced between Censuses by applying components of population growth (births, deaths and migration) to the latest Census-based estimate. Generally speaking, the accuracy of the Census and component data means that only minor adjustments to the population figures occur after each Census.

However, Aboriginal and Torres Strait Islander Census counts and the quality of data on births, deaths and migration of Aboriginal and Torres Strait Islander people do not support the use of the standard approach to population estimation. Instead, the ABS uses assumptions about future fertility, paternity, life expectancy and migration to project figures out from each Census. At the same time, it is also necessary to backcast the Aboriginal and Torres Strait Islander population, because each Census count is sufficiently different from the last, due to a range of non-demographic factors, such as changing levels of identification in the Census.

The uncertainty in the Census and component data occurs because:

- Aboriginal and Torres Strait Islander population is very small compared with the total Australian population (around 3.8%), and
- Aboriginal and Torres Strait Islander population requires people not only to be counted in the Census and component datasets, but also to be identified (and identified consistently) as being of Aboriginal or Torres Strait Islander origin.

The differences between the Aboriginal and Torres Strait Islander population estimates derived from each Census thus tend to be larger than demographic changes can account for. Each Census-based estimate reflects both the desire of people at that time to be identified as being of Aboriginal or Torres Strait Islander origin, and the ability at that time for the information to be effectively collected in the Census.

The Australian Census Longitudinal Dataset (ACLD) indicates that a certain proportion of the population will be counted as Aboriginal or Torres Strait Islander in 1 Census, but as non-Indigenous (or not stated) in the next, and vice versa. For further information see Understanding change in counts of Aboriginal and Torres Strait Islander Australians: Census for further information.

The graph, below, shows the percentage increase in estimated resident population (ERP) between Censuses for both the Aboriginal and Torres Strait Islander population and the total Australian population, for the last 5 intercensal periods. The change in the total population between Censuses remains lower than 10%, and is largely consistent with birth, death and migration rates observed between Censuses. This is not the case for the Aboriginal and Torres Strait Islander population, for which implausible demographic trends would need to have occurred to account for the intercensal difference.

Between 2016 and 2021, the Aboriginal and Torres Strait Islander Census counts increased by 25.2% (163,600 persons). Of this increase:

- 43.5% can be accounted for by explainable demographic factors of population changes - that is, births, deaths and migration
- changing identification was a major contributor to the remaining 56.5% increase, not due to demographic factors.

For more information see Understanding change in counts of Aboriginal and Torres Strait Islander Australians: Census and Occasional Paper: Population Issues, Indigenous Australians (cat. no. 4708.0)

Other collections, such as surveys and administrative datasets rely on a person's Indigenous status to be accurately and consistently reported and recorded, and therefore is susceptible to differing rates of identification, which may also vary over time. The changing rates of identification of Aboriginal and Torres Strait Islander people in the Census and other datasets create challenges when comparing data about this population over time. For further information, see Information Paper: Perspectives on Aboriginal and Torres Strait Islander Identification in Selected Data Collection Contexts (cat. no. 4726.0) for further information.

## Comparison of historical series

After each Census, the ABS creates a time series for the population based on that Census count, by projecting and backcasting around this estimate. The backcast series shows what the Aboriginal and Torres Strait Islander population (based on the latest Census) would have been in previous periods, based purely on demographic change. Similarly, the projected series shows how the currently identified Aboriginal and Torres Strait Islander population may change in future years, if certain demographic assumptions were to hold. Each Census-based series is independent from the others.

The graph, below, shows the 4 different population series that cover the period 2006 to 2021, as well as how each year's estimated population would change if a simple, straight-line interpolation was applied between successive Census-based estimates.

Often, population estimates for Aboriginal and Torres Strait Islander people are used as the denominator for rates, to compare data for the Aboriginal and Torres Strait Islanders population with the non-Indigenous population, or to create time-series indicators that shows how various aspects of Aboriginal and Torres Strait Islander people's wellbeing are changing over time. The use of different series of population estimates shown in the above graph will clearly impact on such indicators.

Following are some examples of the impact that the choice of Aboriginal and Torres Strait Islander population estimates can have on a rate. The examples use 6 different series to demonstrate this:

- the 2021 Census-based series
- the 2016 Census-based series
- the 2011 Census-based series
- a series which uses each Census-based series for the 2 years before and after that Census (option 1)
- a series which transitions between Census-based series in a way that minimises the 'jump' in the transition year (option 2)
- linear interpolation between Census estimates.

### Example 1: Child mortality rate

The use of different Census-based Aboriginal and Torres Strait Islander population estimates, while showing a similar pattern of change over time, has a considerable impact on the level of historical child mortality rates. When using a combination of series the downward trend in the rate is more obvious.

- Rates are for total population of NSW, QLD, WA, SA and NT.
- Uses 5-year average deaths, and population for the middle year of the 5-year period.
- Projected data is based on the medium series.

- Rates are for total population of NSW, QLD, WA, SA and NT.
- Uses 5-year average deaths, and population for the middle year of the 5-year period.
- Projected data is based on the medium series.
- Option 2 uses the 2016 Census-based estimate for 2013 to 2017, and the 2021 Census-based estimate for 2018 onwards.

### Example 2: School enrolments

In the case of school enrolments data, the use of previous Census-based series generates impossible results, with rates exceeding 100%. In contrast, when using a combination of Census-based population series, the enrolment rates remain plausible throughout, although there is no longer a clear trend in the time-series. Further analysis would be needed to determine the most appropriate choice of Aboriginal and Torres Strait Islander population estimates. In particular, the choice may depend on an understanding of how identification rates in the enrolment data (numerator) had changed over time.

- Option 2 uses the 2011 Census-based estimate for 2011 and 2012, the 2016 Census-based estimate for 2013 to2017, and the 2021 Census-based estimate for 2018 onwards.

### Example 3: Standardised death rates

In the case of age-standardised death rates, the use of different Aboriginal and Torres Strait Islander population estimates changes the direction of the trend.

- Rates are for total population of NSW, QLD, WA, SA and NT.
- Uses 5-year average deaths, and population for the middle year of the 5-year period.
- Projected data is based on the medium series.
- Not stated age at death have been prorated and included in analysis.

- Rates are from total population of NSW, QLD, WA, SA and NT.
- Uses 5-year average deaths, and population for the middle year of the 5-year period.
- Projected data is based on the medium series.
- Not stated age at death have been prorated and included in analysis.
- Option 2 uses the 2016 Census-based estimate for 2013 to 2017, and the 2021 Census-based estimate for 2018 onwards.

In each of the examples above, of the 3 methods of transition between population estimate series shown, the time-series rate is smoothest when using a linear interpolation between Census year estimates. Changing from 1 Census-based series to the next results in a sharp drop in the rate at the changeover point. This can be seen at points 2014, 2018 and 2019. 2023 would be the equivalent 'decision point' for the current intercensal period. The extent of the 2023 transition will not be known until after the 2026 Census-based ERP series becomes available.

## Which series to use

The question of which historical population series to use depends on the purposes of the analysis. The following provides guidelines to help in making these decisions.

The 2021-based series is the most accurate and up-to-date series currently produced by the ABS in terms of the methodology used in the Census collection and in the estimation/projection processes.

The 2021-based series can generally be used for any of the following scenarios:

- For demographic analysis of the Aboriginal and Torres Strait Islander population.
- For forward-looking analysis (that is, from 2021 onwards), for any purpose.
- As the denominator of rates being compared over time, if:
- the time series is short, generally no further back than the penultimate (2016) Census
- the numerator data are known to have been fairly consistently high over time
- analysis is restricted to remote geographies, where the impact of increased identification is much lower than in non-remote areas, or
- when there is a known quality issue in the data from earlier Censuses.

Using a range of different Census-based population estimate series could be considered when:

- transitioning from 1 series to the next will maximise consistency of identification levels and coverage between the numerator and the denominator
- historical rates are thought to be artificially deflated when using the 2021-based population estimates (usually due to lower identification rates in the numerator series in earlier years)
- there is a known relationship between identification in the numerator data and the Census data. Data linkage may be used to inform on this relationship – for example see Linking Death registrations to the 2016 Census. This relationship may be applied to the 2021-based series as a factor, or be used to inform the choice of a transitional series
- data confrontation using supplementary data sources indicates that the 2021-based population estimates do not produce a plausible result.

In some cases, an alternative data source may exist which is more comparable/consistent with the numerator data, and could be used rather than a population estimate series. For example:

- Survey data should be compared to the survey population. That is, re-weighting past surveys to the latest population estimate series is not generally recommended.
- Analysis of Census data should be based on Census counts rather than population estimates see Estimates of Aboriginal and Torres Strait Islander Australians methodology, Technical note: Methodology used in Aboriginal and Torres Strait Islander population estimates.
- Numerator data could be compared to a denominator that is internally consistent, rather than to population estimates - for example retention rates (15 year-old students in 2023 as a percentage of 12 year-old students in 2020) produces a more coherent time-series than enrolment rates (15 year-old students as a percentage of 15 year-old population).

If in any doubt, the ABS recommends that data users conduct their own analysis, for example similar to that illustrated in the examples above. This will highlight the impact of non-demographic changes on their particular area of interest.

### Impact of future changes

A new series for Aboriginal and Torres Strait Islander population estimates will be produced following the 2026 Census. Depending on how the 2026 Census-based estimate compares with the 2021 Census-based estimate, current indicator trends may change as a result of rebasing any population estimates to the new Census.

Aboriginal and Torres Strait Islander population projections are produced to illustrate potential future demographic changes to the currently identified population. The 2021-based projections therefore do not include any assumption variable on changing identification. Anyone considering the future population should consider their requirements for a projection based on a changing identification, and note that the ABS projections cannot be used for this purpose.