The Gradient Inside the Gate

Every APN Sentinel™ analysis published to date has ranked places against the nation. This one asks a different question: how much safety variation exists inside a single council area — between suburbs that share one mayor, one rates notice, one local newspaper? The pre-registered hypothesis was two-part: that within-council dispersion varies enormously, and that the widest internal gradients belong to the large metropolitan councils. The first part is confirmed emphatically — nearly half the measurable councils in the country contain a thirty-point safety range inside their own boundary. The second is rejected, and the rejection is the story. Internal safety inequality is not a big-city phenomenon. It is a regional-city and disadvantaged-fringe phenomenon; the wealthiest inner-metro councils turn out to be the most internally uniform places in the dataset; and what within-council dispersion actually measures, on this evidence, is not size or crime volume but socio-spatial segregation — specifically, how far a council’s floor drops.

A four-fold range, and where most councils sit

Across the 163 New South Wales and South Australian councils with at least ten directly-scored member suburbs, internal dispersion — the standard deviation of suburb-grain Sentinel readings within each LGA — runs from 2.8 to 11.8. Most councils cluster in the middle of that range: seventy-six of the 163 sit between 6 and 8, with thin tails at both ends (Figure 1). The tails are where the information is. Twenty-one councils are tightly uniform (below 6); fourteen are highly dispersed (10 or above); and the two groups look nothing alike demographically — which is the substance of everything that follows.

Two kinds of council

At the uniform end sit North Sydney (spread 2.8 — its thirteen suburbs span just nine points, 50.3 to 59.7), Woollahra (3.7) and Randwick (4.4) in Sydney, and Holdfast Bay (4.4) and the Adelaide Hills (4.6 — forty-nine suburbs, every one between 61 and 81) in South Australia. These are councils where knowing the LGA tells you nearly everything: safety is a shared local condition, end to end.

At the other end, a single council can contain most of its state’s safety spectrum. Playford, on Adelaide’s northern fringe, spans 15.3 to 50.0 across thirty-two suburbs — readings near the least-safe in South Australia and readings at the state median, inside one boundary. Port Pirie runs 15.4 to 59.6. Port Stephens spans forty points (32.3–72.3), Murray River forty-five. The regional cities cluster here too: Wollongong’s sixty-three suburbs span 35.8 to 71.1, Wagga Wagga’s 25.2 to 64.9, Albury’s 19.8 to 55.6.

Anatomy of a gradient: Playford, suburb by suburb

The most dispersed council in the sample rewards a full look, because its internal structure is legible street by street (Figure 3). The bottom of Playford’s distribution is the Elizabeth cluster — Elizabeth South, Elizabeth Grove, Elizabeth Vale, Elizabeth North, Elizabeth East, Davoren Park — ten contiguous suburbs wall-to-wall between 15 and 20, among the lowest Sentinel readings in South Australia. The middle band is the growth corridor: Andrews Farm, Munno Para West, Blakeview, Craigmore — the new-estate belt reading 25 to 36. And the top is the council’s semi-rural hills fringe — Angle Vale, One Tree Hill, Humbug Scrub, Yattalunga, Sampson Flat — reading 40 to 50, at or near the state median. One council; three recognisably different Australias; thirty-five points between its own suburbs. Playford’s LGA-average safety statistic describes none of the three.

Size doesn’t drive it. The floor does.

The metropolitan half of the pre-registered hypothesis fails cleanly, and Figure 4 shows the failure in full: across all 163 councils, dispersion is essentially uncorrelated with the number of member suburbs (r = +0.079). Councils with a dozen suburbs and councils with more than a hundred produce the same range of spreads. The eight largest metro councils in the sample average a middling 6.9. New South Wales and South Australia average nearly identically (7.8 and 7.4) — this is not a state effect either.

What does correlate with dispersion is the council’s floor. The spread–minimum correlation is −0.381: the wider a council’s internal range, the lower its least-safe suburb reads — a materially stronger relationship than spread against the council’s mean (−0.159). In plain terms, high-dispersion councils are not places where some suburbs are unusually safe; they are places where some suburbs are unusually unsafe while the rest are ordinary. The gradient is dug downward. That is the statistical signature of concentrated locational disadvantage — an Elizabeth cluster, a historic town core — sitting inside an otherwise unremarkable council, and it is why within-council dispersion functions on this evidence as a segregation gauge: it measures how much of the state’s social distance one boundary contains, and where that distance lives, it lives at the bottom.

What the gradient means in practice

The practical implication cuts at a habit shared by data journalism, product design and policy targeting alike: describing places by their LGA-average safety statistics. The dispersion evidence says that habit fails selectively, and fails worst exactly where accuracy matters most. For North Sydney or the Adelaide Hills, the council average is the place — any suburb-level product would add little. For Playford, Port Pirie, Wollongong or Wagga Wagga, the average is a statistical fiction sitting between suburbs that share little beyond a boundary: it overstates safety in the Elizabeth cluster and understates it in the hills fringe, simultaneously, by twenty points or more in each direction. Any crime map, insurance proxy, school-catchment narrative or funding formula built on LGA-grain figures inherits that double error — and inherits it disproportionately in regional cities and disadvantaged growth fringes, which is to say, in the places least likely to have the analytical resources to notice. It is also, candidly, the empirical justification for why the Sentinel pillar scores at suburb grain in the first place: forty-four per cent of measurable councils contain an internal range wide enough to make the council-level number analytically misleading.