One month, two reputable publishers, and a 5.6-point disagreement about how many Australians are out of work. This brief decomposes the gap — what each “unemployment rate” measures, and why two valid numbers diverge.
For April 2026 the ABS reported unemployment at 4.5% (seasonally adjusted / SA; survey reference period 29 Mar–11 Apr 2026; 15+ civilian population) — 692,500 people. For the same calendar month, Roy Morgan’s “real unemployment” estimate was 10.1% (actual / not seasonally adjusted / NSA; April 2026 monthly estimate; aged 14+) — 1,635,000 people. Two different defined quantities measuring one labour-market idea. A gap of about 5.6 percentage points, roughly 2.2× — RAOSCAFF arithmetic on the two published rates. Both are correct under their own published definitions; the gap is a property of how each defines the question.
For April 2026 the Australian Bureau of Statistics reported the unemployment rate at 4.5% (seasonally adjusted / SA) — 692,500 people, on a survey covering 29 March to 11 April (15+ civilian population). For the same calendar month, Roy Morgan’s “real unemployment” estimate was 10.1% (actual / not seasonally adjusted / NSA) — 1,635,000 people (April 2026 monthly estimate; aged 14+). These are two different sample frames and two different reference windows — not the same population or workforce — and a gap of roughly 5.6 percentage points (RAOSCAFF arithmetic: 10.1 − 4.5 = 5.6ppt) — about 2.2× higher under Roy Morgan’s number (RAOSCAFF arithmetic: 10.1 ÷ 4.5 ≈ 2.2).
The word that travels is a single “unemployment rate”. But the two publishers are not measuring the same quantity twice — they are measuring one labour-market idea with two different defined quantities. The ABS counts a person as unemployed only on a three-part test; Roy Morgan asks a single, broader question. This brief decomposes the gap and stops there: magnitude, not direction. It is not a criticism of either publisher — both figures are correct under their own published definitions — and it takes no view on which number is the “true” one, nor on whether either rate is high or low.
Place both readings on a single axis that runs from 0 to 12%. The ABS caliper clamps at 4.5% (seasonally adjusted, 15+ population, survey period 29 Mar–11 Apr 2026); the Roy Morgan caliper clamps at 10.1% (actual/NSA, 14+ workforce, April 2026 monthly estimate). The span between them — about 5.6 percentage points (RAOSCAFF arithmetic) — is the gap: same calendar month, two different sample frames, two different defined quantities.
The ~5.6ppt gap (RAOSCAFF arithmetic) decomposes along three observable axes. The definitional screen is the largest; seasonal adjustment and the sample frame are also observable but smaller.
The largest observable axis is how each publisher defines “unemployed”. The ABS three-part screen — no work (not even one hour), active search in the prior four weeks, availability, with employment assessed first so one hour of paid work moves a person to “employed” — is narrower than Roy Morgan’s single “looking for work, no matter when” criterion. A reader is comparing a narrow actively-searching-and-available count against a broader looking-for-work count; these are two different defined quantities measuring one labour-market idea, and the full ~5.6ppt gap is observed between them. Two further axes are also observable but smaller: the ABS headline is seasonally adjusted (SA) while Roy Morgan’s is actual / not seasonally adjusted (NSA), and Roy Morgan surveys Australians aged 14+ (a workforce base of 16,091,000) against the ABS 15+ civilian population — two different sample frames.
| Axis | ABS | Roy Morgan | Contribution |
|---|---|---|---|
| Definitional screen | no work (not 1 hr) + active search (prior 4 wks) + available | “looking for work, no matter when” | largest observable axis — the full ~5.6ppt gap (RAOSCAFF arithmetic) is observed across these two differently-defined quantities |
| Seasonal adjustment | seasonally adjusted (SA) | actual / not seasonally adjusted (NSA) | observable, smaller |
| Sample frame | 15+ civilian population (ABS reference period 29 Mar–11 Apr 2026) | aged 14+ (workforce base 16,091,000; Roy Morgan April 2026 monthly estimate) | observable, smaller — two different sample frames |
| Broad-slack family | underutilisation 10.2% SA (media release) / 10.3% SA (detailed page) | unemployment + under-employment 20.3% NSA (3,274,000) | broad-slack gap ~10.1ppt (RAOSCAFF arithmetic: 20.3 − 10.2) — ratio ~2×; ppt gap is wider than headline |
The divergence is not confined to the headline. On broad-slack measures the two figures remain far apart: ABS underutilisation is 10.2% (SA, seasonally adjusted) in the media release — and 10.3% (SA) on the ABS detailed latest-release page (a 0.1ppt rounding/revision-timing difference) — while Roy Morgan’s combined unemployment plus under-employment is 20.3% (actual/NSA) (3,274,000 people, of which under-employment is 10.2% / 1,639,000). That is a broad-slack gap of roughly 10.1 percentage points (RAOSCAFF arithmetic: 20.3 − 10.2 = 10.1ppt) — about 2× in ratio terms. Note: widening the lens does not narrow the ppt gap — the headline gap is ~5.6ppt and the broad-slack gap is ~10.1ppt; the ratio narrows only slightly, from ~2.2× at the headline to ~2× at broad slack. The divergence runs through the whole measure family, not just the headline. Gap figures are RAOSCAFF arithmetic on the two publishers’ own released rates.
If you read one thing: 4.5% (SA) and 10.1% (NSA) are two different defined quantities measuring one labour-market idea — two definitions, two sample frames — not one number measured twice.
When a story cites “the unemployment rate”, the honest version names the definition behind it. For April 2026 the ABS and Roy Morgan are measuring one labour-market idea with two different defined quantities — a narrow three-part screen (4.5%, SA, 15+ population, survey 29 Mar–11 Apr 2026) and a broad looking-for-work criterion (10.1%, actual/NSA, 14+ workforce, April monthly estimate) — and placed on one axis, with their tests, sample frames, and seasonal basis attached, the two readings move the number by about 2.2× (RAOSCAFF arithmetic: 10.1 ÷ 4.5). Mirror Brief AU-14 reports that magnitude and the reasons for it: the definitional screen is the largest observable axis; seasonal adjustment and the sample frame are also observable but smaller. Every input figure here is taken directly from each publisher’s own April 2026 release; the gap and ratio are RAOSCAFF arithmetic, labelled as such. We make no claim about which figure is “true”, no judgement on whether either rate is high or low, name no party, government, minister, or central bank as a political actor, and draw no inference about policy or what anyone should do next.
Research approach. Mirror format — RAOSCAFF anchors on each publisher’s own filed release and decomposes the figures they print. AU-14 is a cross-publisher divergence brief: two reputable measures of one idea — the ABS Labour Force release (seasonally adjusted) and Roy Morgan’s Australian Unemployment Estimates (actual / not seasonally adjusted) — both for April 2026, with the ABS concepts-and-methods page supplying the three-part definition. The two rates are placed on a single axis and the gap is decomposed into its definitional, seasonal, and sample-frame components. No primary data collection, no analyst estimate, no extrapolation; the only derived figures are simple differences and ratios of the published rates, each labelled RAOSCAFF arithmetic.
Source standards & limitations. Every input figure traces to a named, dated publisher release in § H of the companion FACTS.md — the ABS media release and latest-release page (4.5%, 692,500, underutilisation ~10.2% / 10.3%), the ABS concepts-and-methods page (the no-work / active-search / availability test), and Roy Morgan’s April 2026 estimate (10.1%, 1,635,000, combined 20.3%, workforce 16,091,000, aged 14+, “looking for work, no matter when”). The two rates are different defined quantities and are never treated as one metric measured twice; each figure carries its publisher, its month, and its seasonal basis (SA for the ABS, NSA for Roy Morgan) wherever the two are placed side by side. The brief decomposes the gap only: it asserts no “true” rate, names no party, government, minister, or central bank as a political actor, adopts none of any publisher’s rate or policy commentary, makes no causal or normative claim about the labour market, and forecasts no future value. The 0.1ppt difference between the ABS media-release underutilisation (10.2%) and its detailed-page figure (10.3%) is a rounding / revision-timing difference; the brief cites both as distinct labelled figures (10.2% SA from the media release, 10.3% SA from the detailed page) rather than collapsing them.
H1–H4 are official publisher releases for April 2026 (ABS survey period 29 March–11 April 2026; Roy Morgan April 2026 monthly estimate), fetched from each publisher directly. H5 is Roy Morgan’s March 2026 release, included for series-continuity context only — no April 2026 figure is drawn from it. Full citations and the verified figures behind every number in this brief are listed in the companion FACTS.md, § H. Derived figures — the headline gap (~5.6ppt / ~2.2×) and the broad-slack gap (~10.1ppt / ~2×) — are RAOSCAFF arithmetic on the published rates and are labelled as such throughout.