Can we trust Census data?

The 2016 Census didn’t run as smoothly as the statisticians would have liked. In response to privacy concerns, hardware failure and denial of service attacks on Census night, as well as ongoing criticism by media and the public, the Australian Bureau of Statistics (ABS) established an Independent Assurance Panel. The Panel’s task was to review and assure the quality of statistical outputs from the 2016 Census. Coinciding with the first major release of 2016 Census data, the Panel’s report  was released today.

The Panel included representation from academics, state government representatives and a number of eminent Australian and international statisticians:

  • Chair–Professor Sandra Harding, Vice Chancellor and President of James Cook University
  • Professor Lisa Jackson Pulver, Pro Vice-Chancellor Engagement and Aboriginal and Torres Strait Islander Leadership at the University of Western Sydney and member of the Australian Statistical Advisory Council
  • Professor Peter McDonald, Head of Demography within the Centre for Health Policy, University of Melbourne
  • Peter Morrison, former Assistant Chief Canadian Statistician who was responsible for running the Canadian Census
  • Dennis Trewin, ex-Australian Statistician (2000-2007)
  • Anton Voss, Deputy Secretary, Tasmanian Department of Treasury and Finance and member of the Australian Statistical Advisory Council

Its responsibilities were to ‘provide a written report no later than 16 June 2017 to the Australian Statistician giving its view on aspects of quality of 2016 Census outputs, having considered issues including, but not limited to Census design, enumeration, processing and quality assurance. The quality should be assessed using the quality of outputs from the 2011 and 2006 Censuses as benchmarks.‘

The Panel strongly supported the ABS’ move to online census collection, and noted more individual items were completed in the online version than the paper version. The online Census form was withdrawn by the ABS for some 42 hours after Census night. The Panel concluded this withdrawal may have resulted in more paper forms being completed, but had no significant impact on overall completion of the Census.

Regarding privacy concerns, the Panel noted an increase in the number of people reporting their age rather than date of birth, as well as people providing false names on the Census form. While these issues don’t directly affect the quality of Census data, they may impact on usefulness for future data linkage projects.

There was also a decrease in agreement to allow census results to be kept by the National Archives of Australia.

The Panel recommended an external privacy impact statement be prepared for future censuses.

The Panel also looked at the counting of Indigenous Australians in the Census and concluded that this group continues to be undercounted—net undercount was 137,750 people or 17.5 per cent. The Panel recommended the ABS should consider ways to improve the enumeration of Aboriginal and Torres Strait Islander people for future censuses, in consultation with Aboriginal and Torres Strait Islander communities and organisations.

For small area geographies, the Panel reported ‘…there is a broad expectation that accuracy of Census data at smaller area geographies is acceptable. However, this is a very tentative conclusion and requires further analysis by the ABS.’[1]

The person response rate for the 2016 Census was 94.8%, declining from 96.3% in 2011 and 95.8% in 2006.

Key findings for the 2016 Census were:

  1. Post-Enumeration Survey (PES) identified a net undercount of 1.0 per cent which was lower than the result for 2011 Census (1.7 per cent). The PES sample size was increased by 20 per cent for the 2016 Census to measure Census quality with greater accuracy.
  2. Census count aligns well with the Estimated Resident Population estimates.
  3. For key topics, the levels of characteristics matched well with 2011 Census results—the increase in ‘no religion’ response may reflect the change in that question.
  4. Overall response rate was lower than, but comparable with, results for censuses held in United Kingdom, Canada and New Zealand.

The Panel assessed data quality at the Australia/state/territory level, and concluded data quality was broadly in line with expectations, and results were of comparable quality to international benchmarks, as well as being fit for purpose, useable and useful.


Flagpost is a blog on current issues of interest to members of the Australian Parliament

Parliamentary Library Logo showing Information Analysis & Advice

Filter by



Tag cloud