Using administrative data to fill potential data gaps in the Census

Preparing the 2021 Census for unexpected events like bushfires and floods


Being prepared for unexpected events

The COVID-19 pandemic and 2020 bushfires are reminders that the Census, one of Australia’s largest peacetime operations, must be ready to adapt to unexpected events.

To make sure we can still deliver the highest quality Census, the ABS is preparing to use administrative data to fill any significant gaps that might be caused by unexpected events.

A bushfire, for example, might make it hard for residents in an affected town or area to complete the Census, while a national emergency like a large scale uncontrolled outbreak of a pandemic could lower response rates across the country.

Administrative data can help improve the population counts and fill in the gaps in some of the other information collected on Census forms.

This approach is guided by international precedent. Some other countries have successfully used administrative data to fill in gaps in their Censuses.  For example:

  1. To achieve a high response in their 2020 Census, preliminary results show that the US Census Bureau used administrative data to count people in about 6% of houses. This was needed even after the time for collecting Census forms was increased from three to six months, due to COVID-19 delays.
  2. Stats NZ used administrative data to fill in gaps in their 2018 Census after a lower than expected response. Data for 89% of individuals in the Census dataset came from a 2018 Census form and 11% from administrative data. This produced higher quality counts compared to previous Censuses, although not all gaps in Census information could be filled.
  3. In Canada, a large bushfire at Fort McMurray interrupted the 2016 Census, with about 100,000 people evacuated from the area. For dwellings in the evacuated areas, if no response from field collection was received, Statistics Canada used administrative data to fill in basic information where possible.

Principles guiding the use of administrative data for the Census

We think it’s important to be prepared, and would only use this type of approach if it was really needed.

Maintaining public trust is of utmost importance, and in 2020 we conducted a privacy impact assessment on taking this approach.  The assessment found that people’s privacy is well preserved, but recommended that ABS publish information on how we would decide whether or not to use administrative data in this way.

In response to this recommendation and our commitment to transparency, the ABS has developed the principles below to guide when and how administrative data would be used to fill gaps in the Census.  Figure 1 provides further detail on how these principles would be used to guide the decision-making process.

  1. Census data quality is substantially affected
There must be a large enough impact to the quality of Census data to justify the use of administrative data.  For example, data for a town or local region can’t be obtained, or gaps for particular populations are large enough that Census can’t provide accurate enough information for planning and policy decisions.
  1. Administrative data is of sufficient quality
Administrative data must be shown to be of sufficient quality to address the impacts to Census data quality.
  1. Delays to Census outputs are considered acceptable
The benefits to data quality must clearly outweigh the costs, particularly any delays to Census outputs.
  1. Custodians of the administrative data are supportive
Before using administrative data to fill gaps in the Census, the custodians of the administrative data (e.g. the Australian Taxation Office) must agree to the use of their data for this purpose.
  1. Transparency and ensuring public trust

ABS is transparent about the potential use of administrative data in the Census and the benefits.

Privacy impacts must have been assessed and minimised before any final solution is implemented.


Using administrative data

If unexpected events affected the Census, the ABS could fill gaps using combined administrative data from the Tax, Medicare and Centrelink administration systems.

The first step to filling any gaps would be to find any records from the combined registrations of Tax, Medicare and Centrelink administration systems that appear to be missing from the Census and are located in the area or population affected by the unexpected event.  The administrative data would supply at least the age, sex and area of residence information, which are key outputs from the Census.

The next step would be to fill in further Census information for these records using both administrative data and 2016 Census data where possible. For example, we may be able to identify family relationships from Centrelink and Medicare records, and/or fill in information from the 2016 Census that wouldn’t have changed over time, such as Country of Birth.

We are also developing a potential approach to filling in Indigenous Status if counts of Aboriginal and Torres Strait Islander persons were significantly affected. Any adoption of such an approach would only occur after substantial consultation with Aboriginal and Torres Strait Islander groups.

Figure 1: Decision process guiding when to use administrative data to fill gaps in the Census

    Flow diagram which steps through the process for deciding whether to continue normal output processes for the Census, or to use administrative data to fill gaps in the Census. Each step in the process relates to the principles outlined in the table.
    Decision process:

    Step 1 - Has Census data quality been sufficiently affected?
    Is a whole area missing, or are there big gaps for important populations?
    Are important planning and policy decisions going to be affected?
    If no, continue normal output processes for the Census.
    If yes, continue to Step 2.

    Step 2 - Is there admin data of sufficient quality to address the impacts?
    Does admin data cover the gaps for the affected areas or populations?
    Does it have the right information?
    If no, continue normal output processes for the Census.
    If yes, continue to Step 3.

    Step 3 - Do benefits outweigh the costs, particularly delays to critical outputs?
    What is the delay to critical outputs like population counts?
    Are the additional costs to data processing feasible?
    If no, continue normal output processes for the Census.
    If yes, continue to Step 4.

    Step 4 - Do the data custodians support the way their administrative data is being used?
    Are signed agreements in place?
    Are data custodians being kept informed of plans?
    If no, continue normal output processes for the Census.
    If yes, continue to Step 5.

    Step 5 - Has the ABS been transparent, and is privacy preserved?
    Has there been a privacy impact assessment of this approach?
    Has the approach been made public?
    If no, continue normal output processes for the Census.
    If yes, use administrative data to fill gaps in the Census.