Week 6 Session: Conflicting Evidence

Two Australian agencies count homelessness and get different answers — the work is diagnosing which question each of them answered

29 March 2027

Two agencies, both careful, both official, both counting homelessness in Australia. Bring nothing but attention.

The two records#

Two publications, two figures, two periods. The periods are half the session, so they travel with the numbers everywhere below.

Australian Bureau of Statistics — Estimating Homelessness: Census, 2021. The ABS estimates 122,494 people experiencing homelessness on Census night in 2021, a single night in August of that year.

Nobody was asked whether they were homeless. The ABS applies rules to what people reported about their housing that night and sorts them into six operational groups. One group is people in severely crowded dwellings, and the rule defining it is exact:

“Usual residents in dwellings needing 4 or more extra bedrooms under the Canadian National Occupancy Standard (CNOS).”

A count of people, produced from a count of bedrooms.

Australian Institute of Health and Welfare — Specialist Homelessness Services annual report. The AIHW records almost 289,000 clients assisted by specialist homelessness agencies across the financial year 2024–25, up from 236,000 in 2011–12. Among the subset of clients whose housing status was known at the beginning of their first support period — not the full 289,000 — the report splits them:

“More clients were at risk of homelessness (51% or 139,000), than experiencing homelessness (49% or 133,000).”

So the number to carry forward from the AIHW is 133,000 — the clients it counted as already experiencing homelessness, rather than the full client count.

The AIHW then prints its own caution, which is exactly the one a careless reader needs:

“It is important to note that the number of clients supported by SHS agencies … reflects agency engagement and does not necessarily represent the underlying prevalence of homelessness …”

Before anything else, say the two periods out loud. One night in August 2021. One financial year running July 2024 to June 2025. Three and a half years apart, and one is an instant while the other is twelve months. Nothing that follows makes sense if that slips.

The axes#

Work in pairs, publications open, and sort the difference between these two records onto the axes below. For each one, quote or point to the thing in the material that supports you.

• Definition. What has each agency declared it is counting? The ABS has six operational groups and a bedroom-shortfall rule. The AIHW has people who became clients of a funded agency, split between homeless and at risk.
• Coverage. Who could possibly appear in each figure? The Census reaches people who did not ask anyone for help. The AIHW collection reaches people who approached a service that reports to it, and no one else.
• Method. One number is derived from responses about housing circumstances. The other is assembled from records of support provided. Neither agency asked anybody the question “are you homeless”.
• Timing. A night against a year. A count of who was in a state at an instant, against a count of who passed through a service across twelve months. These are different kinds of quantity before you even get to the subject matter.
• Unit. Persons in one case, clients in the other, and the client count is not a count of visits. A client may have many support periods across the year and across different agencies, and the collection is designed to join them into one client rather than count each separately — “Matching of data from individual clients who presented at different agencies and/or at different times requires a valid statistical linkage key (SLK)”. Designed to, and largely does: 99% of support periods carried a valid SLK in 2024–25, which is how you know this is linkage that works well rather than a guarantee about every record. The report adds its own caveat when it breaks the count down, “noting that clients may have accessed services in more than one state or territory throughout the year”, which is why the jurisdictions do not sum to the national figure. The unit is a client observed through service engagement across a year, not a person occupying a defined housing situation on one night.

Some of these you can settle from the documents. Others you will have to reason to. Keep the two apart on the page: what the publication states in one column, what you concluded in the other. Reasoning is welcome. Reasoning dressed as sourcing is not.

The trap#

Look at 122,494 and 133,000.

They are close. Around eight per cent apart, which for two independent official estimates of a hard-to-measure population would ordinarily be reassuring. Two agencies, two methods, roughly the same answer.

Do not accept it, and spend the time working out why not.

The figures are close in the way that two thermometers in different cities in different seasons might read the same. One is everybody the Census placed in six housing situations on a single night in 2021. The other is people who walked into a funded service at some point across twelve months of 2024–25 and were homeless when they first arrived. They overlap, certainly. Neither contains the other. And the years are different, so even the overlap is between two different populations at two different times.

A number that agrees with yours for reasons you have not examined is more dangerous than one that disagrees, because disagreement at least prompts a question. This pair will pass unchallenged in any room where somebody wants them to agree.

The move to refuse outright: averaging. There is no quantity of which 122,494 and 133,000 are two measurements, so their mean is a number about nothing.

Pick one, for something#

Three people need a figure by the end of the day. In your pair, decide which source each should use, and — the part you have to justify, and the part we will press hardest on — write the sentence each of them must not say.

1. A community organisation writing a funding submission for more crisis accommodation beds in one city.
2. A journalist writing that homelessness in Australia is rising.
3. A service manager rostering staff for the coming winter.

The answers differ, and for at least one of the three the honest response is that neither source supports the claim being made and a third is needed. Say so where it is true.

Leaving#

One line each, out loud: name the axis you would check first, next time two numbers you trust refuse to agree.

Nothing is due. Next week somebody wants your number to come out a particular way.