Abstract

The Census in Australia is run every 5 years and uses a traditional approach, collecting information directly from respondents. Whilst the Australian Census continues to face risks and costs like other countries, it continues to have the strong support of the community, Government and stakeholders, and enjoys high a response rate when compared internationally. While this continues, there is no compelling case for stopping or reducing the frequency of the Census.

Meanwhile, in line with other National Statistical Organisations (NSOs), the Australian Bureau of Statistics (ABS) has been developing integrated administrative data assets over the last decade or so. These data assets provide great insights into both population and housing and show increasing potential to meet some needs traditionally provided by the Census – particularly accurate counts of population and basic demographics.

As these assets have developed, the ABS has steadily increased its use of administrative data to improve the efficiency, quality and utility of its traditional Census. For example, the ABS developed an Address Register for the 2016 Census using administrative data which greatly reduced data collection costs. In 2021, the ABS used data from its Person Level Integrated Data Asset (PLIDA) to improve the quality of Census counts to add a set of new income variables. In addition, although it was not deployed, administrative data was prepared as a contingency for unforeseen events such as bushfires and floods.

Building on this work, the ABS recently released an experimental Census-like product built solely from administrative data. This product - the Administrative Data Snapshot of Population and Housing - was created to reflect the same time point as the 2021 Census so the two could be compared. It aligned closely to official estimates of the population and housing stock at a national level.

Throughout these developments the ABS has been considering the future form of the Census: whether information should continue to be directly collected from respondents; extracted from administrative data; or some combination of the two. Given the strong case for retaining direct collection coupled with the increasing value of administrative data, the ABS recently adopted a hybrid model as a vision for its 2031 Census. This vision explicitly recognises the potential of administrative data to provide more frequent output between 5-yearly direct collection exercises.

Whilst this vision is broadly compelling, it presents some significant challenges. For example, administrative data is recorded on a usual address basis across time, while Census data primarily captures where people were located on Census night at a single point in time. This presentation will sketch further details on what a 2031 hybrid Census could look like, giving special attention to how intercensal output from administrative data would shape the model and work with this challenge.

Two broad directions for intercensal output will be briefly explored: 1) starting with a combined Census dataset and updating this with administrative data over time; and 2) updating an administrative population that includes directly collected Census characteristics.