Review of 2021 Census Topics

The Health Performance Council is South Australia’s only external review body with a statutory function to provide expert advice to the state’s Minister for Health and Wellbeing on the operation of the state’s health system and of health outcomes for South Australians. To discharge our functions effectively, we require access to good, accurate, useful data.
We are concerned that the question asked on the 2016 census ‘Is the person of Aboriginal or Torres Strait Islander origin?’ conflates multiple concepts into a single, ambiguous, question and hence is inadequate to allow collection of accurate, useful data for analysis of the numbers and social-economic attributes of Aboriginal and Torres Strait Islander peoples.
We suggest that much valuable nuance in the data collected can be had simply by splitting the question into two, along the lines of the following:
1) Is the person of Aboriginal or Torres Strait Islander origin? [response options as on 2016 Census]
[Only if one or both ‘yes’ boxes marked in response to question 1]
2) Has the person always known that they are of Aboriginal or Torres Strait Islander origin?
a. Yes – they have always known.
b. No – they have not always known.
c. Unsure.
The addition of this second question would allow for a critical distinction to be made in analyses between all people of Indigenous origin and those people carrying a particular burden of enduring trans-generational trauma.
We have provided further information in support of our submission against the assessment criteria in the questions that follow.

Knowledge of the size of the Aboriginal and Torres Strait Islander population is used in the generation of public policy. The cross-jurisdictional Closing the Gap initiative continues to seek to make progress against targets for reducing inequalities of outcomes for Aboriginal and Torres Strait Islander people. State and territory governments have also set and monitor progress against targets for opportunities and outcomes for Aboriginal people [example: limited monitoring data for Aboriginal life expectancy in South Australia is available at https://data.sa.gov.au/data/dataset/sasp-target-53-aboriginal-employees]. Setting and monitoring progress against these targets requires good, meaningful, high quality data.
It has been estimated that there are more than 1,000 different Indigenous programs funded by government and NGOs [Hudson S, 2016. Mapping the Indigenous program and funding maze. The Centre for Independent Studies]. The planning, delivery, funding and evaluation of these programmes is dependent on there being good estimates of the number of Aboriginal people who may be eligible to receive their services.

The South Australian Health Performance Council has a duty to report on health outcomes not only for South Australians in aggregate but for particular population groups – especially those often considered marginalised and vulnerable. The Aboriginal population is, in any case, small: the 2016 Census estimates 2.8% of the Australian population, and just 2.0% of the South Australian population [Based on ABS data, 2017. 2016 Census Counts - Aboriginal and Torres Strait Islander Peoples, cat. no. 2075.0].
We know that in remote and lightly populated areas there are pockets of poor service provision and poor outcomes [example, not specific to Aboriginal people: Health Performance Council [South Australia], 2017. Hotspots of potentially preventable hospitalisations in South Australia’s pubic hospitals]. Without universal coverage to allow for small area analyses, we would not be able to produce similarly insightful reports at geographically local levels.
We cannot perform our duties effectively to monitor and evaluate performance and outcomes for these and other small subgroups of our population without the universal coverage that only the Census provides. A peril of such small numbers is an inability to drill down to perform subgroup analyses, allowing much disparity to be hidden in the high level averages; the collection of data from a whole of population Census at least mitigates these issues as far as is possible.

Official estimates of Aboriginal populations have been made from the Census since 1971 [Source: ABS, 2018. Census of Population and Housing: Consultation on Topics, 2021, cat. no. 2007.0] and the single question on identified origin has been on the Census in its current form since 1981 [Source: ABS, 2016. Census of Population and Housing: Census Dictionary, 2016, cat. no. 2901.0].
The subsidiary question that we propose echoes much of the wording of the existing, well tried and tested, question. It asks for personal information that can reasonably be assumed to be able to be answered without pause or the need to refer to other sources of knowledge.

As we note in our response to assessment criteria 3, a question on this topic has been on the Census since 1971 and in its current form since 1981. The addition of a secondary question, subsidiary to that already asked, does not on its face disclose any reason for its being found unacceptable to respondents any more than would the existing question. Furthermore, splitting out two conflated concepts from one existing question should make it easier for respondents to answer accurately without having to apply so much subjective interpretation as at present.

The proposed additional question seeks information which can be expected to be known and answerable from objective knowledge. Accordingly, it should not require extensive explanation or instruction.
There is a straightforward response coding disclosed directly by the proposed question, almost analogous to the coding categories for the existing Indigenous Status variable, being one of six options: 1 = always known; 2 = not always known; 3 = unsure; 9 = both always and not always marked; & = not stated; V = overseas visitor.
It is accepted that there would exist a challenge in determining how best to code either or both Aboriginal origin data items for Census returns with certain inconsistencies, for example where a Non-Indigenous response is given to the current question but a substantive recency of knowledge response is given to the new question.

The Health Performance Council believes that the need for data on this topic will be enduring for the reasons given in the responses against the other assessment criteria, and especially because of its being of ongoing national importance.

We know of no other existing data collection that is allows for the particular subgroup analyses that we believe are necessary. As we note in our response to assessment criteria 2, it is necessary that this data be collected through the Census because of the need to enable small area analysis and cross-tabulation analysis with other Census data items that go to a range of social and economic determinants of health.

It is well established that there has been a long-standing and continuing problem with the population Census undercounting the Aboriginal population of Australia. Although the Census population counts generally have been summarily considered to be broadly fit for purpose and able to be used with confidence, the net undercount of the Indigenous population in the 2016 Census, at 17.5%, was thrice that of the non-Indigenous population and the need to improve the enumeration of the Aboriginal population has been clearly recommended [Harding, Jackson Pulver, et. al., 2017. Report on the Quality of 2016 Census Data. Census Independent Assurance Panel to the Australian Statistician]. The sizable undercount, in both absolute terms and relative to that of the population count more generally, has been a sustained and unsolved issue [ibid.].
The systematic historic undercount of Aboriginal people in past Census collections has devalued the utility of the compiled statistics for point-in-time preparation and evaluation of public policy. The growth over time in the number of people identifying in successive Censuses as of Aboriginal and/or Torres Strait Islander origin is welcome inasmuch as it suggests better willingness for respondents to identify as Aboriginal and their better knowledge of their ancestry, but it is also problematic for trend analyses. To the extent that growth is driven by greater awareness by people of their backgrounds, there is therefore a growing data quality dissonance: on the one hand, the compiled statistics can be ever-better approximations of the true population count of people of Aboriginal and/or Torres Strait Islander origin; on the other hand, there is an ever worse ability to use this data item as a good proxy for people living with the sustained life-long inter-generational trauma inherent in having lived with knowledge of such origins since birth. The coherence of time-series analysis is being undermined as is the ability to develop and monitor the implementation of policies that rely on this data.
The impact of the current data quality problem is particularly noticeable when monitoring outcomes measures at the whole-of-Aboriginal population level. All else being equal, a rise across the board in people identifying an Aboriginal origin would manifest as an apparent improvement in outcomes measures: for instance, it will be seen that employment participation rates will rise and apparent life expectancy gaps will narrow. Yet these apparent improvements will actually be biased by the defects in the Census question, the desired time-series polluted by dint of the changes in the underlying population being measured, changes which are hidden by the conflation of the multiple concepts in the single Census question.
We have noted, in our response against Assessment criteria 1, that good temporally-coherent data is necessary to be able to set and monitor progress against important opportunity and outcomes targets and to monitor and evaluate inequities which exist for subsets of our population. We have noted the sizable and growing data quality concerns which result from the current, single, question on Aboriginal origin and the impact on developing and monitoring public policy and achieving greater societal equity. We submit that the widening schism in the data quality issues arising from conflating multiple issues in a single question may be neatly addressed by the addition of the second question that we propose. Furthermore, we submit that this split of the existing single question is likely to improve the quality of the data collected in the census by removing ambiguity and embodying in each question just a single concept. We urge its inclusion in the next Census.