By Erin Rolandsen
Next time, we should make it count.
The integration of data across multiple government departments means the Census risks becoming an expensive exercise in confirming what the government already knows.
The Person Level Integrated Data Asset (PLIDA), established in 2015, links data from agencies including the Australian Bureau of Statistics, Australian Taxation Office, Department of Education, Department of Health and Aged Care, Department of Social Services, Services Australia and the Department of Home Affairs.
Central to PLIDA is the Person Linkage Spine, which determines when records held in different government datasets relate to the same person.
The identifying information from the Census (name, date of birth and address) provide identity-matching coordinates that allow a Census respondent to be resolved to an individual on the Person Linkage Spine, thereby allowing Census responses to be linked to every relevant government dataset that pertains to that same person.
Under current ABS policy, the names obtained by the Census for this process are then destroyed within 18 months.
Person-level data linkage enables powerful longitudinal research, with the capability to answer questions such as whether childhood disadvantage predicts later employment, whether housing circumstances correlate with subsequent hospitalisation, whether welfare-policy changes influence outcomes, and so forth.
While it may feel uncomfortable to realise the panopticon has, in one sense, already arrived, there are strong legislative safeguards on how this information is used. PLIDA is protected by the secrecy provisions of the Census and Statistics Act 1905. PLIDA information is not to be used for enforcement or compliance, and only aggregate statistics are to be produced for release.
Nevertheless, PLIDA enables the ABS to construct an extraordinarily detailed picture of every administratively visible Australian. A person’s age, sex, address, occupation, taxable income, benefit receipts, visa status, international travel history, educational attendance, attainment and healthcare usage are all known to the government, even before the Census results are linked.
Some countries, including Denmark, have recognised this duplication of information and have largely abolished traditional censuses altogether in favour of population registers that rely on administrative records.
It would be a mistake, however, for Australia to follow that same path.
Not every person or issue in Australia is administratively visible. The members of society who have the most to lose are going to be those who work very hard to remain invisible to many governmental systems.
An overcrowded household is unlikely to disclose the true number of occupants to the public housing provider. A visa-overstayer is not likely to disclose their address to rental bond authorities. An international student working more hours than legally permitted is not likely to disclose any cash earnings to the tax office.
While politically unpalatable to admit these problems exist, the extent of these issues is precisely what government departments most desperately need to know.
Robinvale, in north-west Victoria, shows what this looks like in practice. The 2021 Census counted 3,740 residents, yet subsequent estimates using water consumption, banking and supermarket activity suggested the real population may fluctuate between roughly 7,000 and 8,800. A community planned as though fewer than 4,000 people live there cannot possibly be adequately provisioned for twice that number.
Crucially, many of Robinvale’s missing residents are not simply statistical accidents. Community workers have pointed to migrant workers, undocumented residents, severe overcrowding and fear of government as reasons people may avoid the Census or leave household members off it.
The ABS estimated that Australia’s Indigenous population was undercounted by approximately 17.4% (170,800 people) in the 2021 Census for many of these same reasons.
Oscar Wilde wrote: “Man is least himself when he talks in his own person. Give him a mask and he will tell you the truth.”
Anonymity is known to be a key psychological tool for facilitating honesty. Researchers studying undocumented communities repeatedly identify trust and fear of enforcement as major barriers to participation. An ANU submission in 2016 warned that privacy concerns could affect who does and does not complete the Census, creating data-quality problems.
Consider, then, the impact, when the first three compulsory fields of the Census are address, full name and date-of-birth or age for every occupant of each household.
What the Census has tried to gain in person-level precision is lost by corrupting the measurement at the source. The ABS asks vulnerable respondents to place enormous trust in legislated privacy safeguards, while requiring them to disclose the very identifiers that make their answers personally traceable.
Government does not need to know which particular visa overstayer lives in which overcrowded house in order to build the right number of clinics, schools or sewer pipes. It does need to know that those people exist.
Existing privacy techniques ask “How can we protect people after collecting highly detailed information”. Maybe instead, we should ask “Could stronger privacy at the point of collection cause people to give us better information in the first place?”
We should intentionally reduce person-level identifiability so vulnerable people have a credible reason to disclose their situation. Rather than the ABS saying “trust us not to misuse identifiable information”, perhaps it should be saying “we deliberately don’t collect information to identify you”.
Maybe we should let PLIDA become extraordinarily good at describing administrative Australia, and deliberately redesign the Census to describe everything PLIDA cannot see.
That would not make the Census redundant. It would give it a clearer purpose than it has had for decades.
The Census needs to be an aggregate fact-gathering exercise, not a person-level cross-validation. The more complete PLIDA becomes, the less sense it makes for the Census to imitate it. Instead, the Census should specialise in measuring its blind spots.
Every five years, we should give Australians a protected opportunity to tell us where official Australia diverges from real Australia.
Before the next Census, the ABS should experimentally test whether removing names and exact DOB, prohibiting person-level administrative linkage and explicitly promising an independent statistical Census improves truthful reporting of hidden circumstances.
Homelessness, overcrowding, modern slavery, visa overstaying and the black market economy are all critical issues which are notoriously difficult to statistically infer.
A Census that merely confirms administrative reality is quickly becoming redundant, but a Census designed to uncover the aggregate reality outside administrative records is more important than ever.
We have five years to redesign it. Next time, we should make it count.
Erin Rolandsen is the CEO of Angelassist and author of beyondtheragemachine.substack.com
