Why anxiety does not track deprivation across English local authorities.

01 — The decoupling finding

Structural indicators barely predict anxiety.

For most public health outcomes deprivation is a dominant predictor. Obesity, cardiovascular disease, life expectancy and chronic illness typically correlate with the Index of Multiple Deprivation at r values of 0.5 to 0.8 across local authorities. Self-reported anxiety does not follow that pattern.

The five structural factors tested in this analysis each correlate only weakly with adult anxiety prevalence. The strongest, recorded depression at GP level, has a correlation of r = 0.15, where 1.0 would be a perfect match and 0 would be no relationship at all. The deprivation score itself reaches only r = 0.10.

A statistical model built from all five factors produces predicted anxiety values spanning only 21.5 to 26.8 percent. Observed anxiety spans 10.4 to 33.8 percent. The gap between what the model predicts and what survey data actually shows is what the rest of this report focuses on.

02 — What it looks like at area level

Anxiety runs ahead of structural factors in unexpected places.

If anxiety tracked deprivation, the areas where anxiety runs furthest ahead of what structural factors would predict would cluster in the most deprived places, and the lowest in the least deprived. Neither holds. The cards below show three areas with the largest positive residuals in the dataset alongside their deprivation and income context.

Conversely, the most deprived areas in the dataset do not sit at the top of the above-expected ranking. Stoke-on-Trent (IMD 33.6) is within 0.2 percentage points of its predicted anxiety value. Knowsley (IMD 36.2) is +2.1pp above prediction, well short of Southampton's +9.4. These places have high anxiety in absolute terms; their anxiety is closer to what their structural circumstances would lead a model to expect.

The four-segment classification, built on a two-way split of residual and median household income, distributes the 126 areas as follows.

03 — Where elevated anxiety concentrates, adjusted for uncertainty

A shorter, defensible priority list.

The unadjusted ranking inside the Primary segment is sensitive to single-year survey noise. Several named areas have so much uncertainty around their figures that their order reflects that noise as much as any real difference. Milton Keynes has an uncertainty range stretching 12.3 percentage points either side of its central estimate, the widest in the segment. Re-ranking by above-expected anxiety, adjusted for how reliable each estimate is, produces a shorter, more defensible priority list.

Each area on the priority list combines above-expected anxiety, a narrower-than-typical uncertainty range, and (for four of the seven) ONS statistical significance against the England average. Brighton & Hove is absent from the unadjusted top 10 but earns a place on the adjusted ranking through the tighter uncertainty range and the ONS flag. Milton Keynes, Southwark, Richmond upon Thames and Ealing all drop ten or more places when judged on the lower end of their plausible range, and are better treated as candidates for further investigation than headline priority entries.

The robust priority list

04 — Segment view

The areas inside each quadrant.

The four segments group local authorities by how much their anxiety runs above or below what the structural factors predict, and by household income (above or below £38,814, the dataset median). The quadrant scatter below positions every area on both axes; the tabs that follow list the top areas in each segment.

05 — What the data can and cannot show

Two limitations shape the read.

Sampling uncertainty

The ONS Annual Population Survey carries substantial sampling error at local-authority level. Across the 33 Primary-segment areas, the typical 95% confidence interval (the range within which the true value most likely sits) spans 15.6 percentage points. Several named areas (Milton Keynes, Westminster, Southwark, Ealing, Brent, Islington) have uncertainty ranges so wide that their position within the segment reflects single-year survey noise as much as any real difference. The uncertainty-adjusted priority list in section 03 is the more reliable read of where elevated anxiety sits.

The Enfield and Newham anomaly

Enfield (10.40%) and Newham (10.43%) report 2022/23 anxiety prevalences roughly half their own 12-year averages and well below every other London borough. The values match the official ONS release exactly. Cross-checking against the ONS Personal Wellbeing local authority time series (version 4, 28 November 2023) shows both as 12-year lows on confidence intervals wide enough to swallow the change.

The headline finding (that elevated anxiety is widely distributed and not concentrated by deprivation) is robust to both caveats. The case for any single area being more of a priority than its neighbour is weaker than an unadjusted ranking implies.

06 — A reading for public health planning

Three readings for service planning.

The 63 areas classified as Primary or Social span the full range of deprivation and household income in England. They include Westminster, Wiltshire, York, Bath, Southampton, Croydon and Brighton alongside Wakefield, Derby, Kingston upon Hull and Doncaster. The set is geographically and socioeconomically diverse: no consistent regional, urban-rural, or income pattern unites them.

For population health planning

Service planning anchored to deprivation quintiles will systematically underweight higher-income areas where anxiety prevalence is higher than the structural factors would predict. An approach that prioritises the most deprived deciles will route resources away from areas where elevated anxiety is least addressed by existing IMD-targeted provision. The Primary group as a cohort, not its internal league table, is the operationally useful unit: elevated anxiety holds across all 33 areas, but ranking within the group is sensitive to single-year survey noise.

For NHS commissioners and population mental health

The Social segment carries the strongest statistical evidence of elevated anxiety in the dataset: 17 of its 30 areas are ONS-flagged significantly worse than the England average, the highest such concentration in any segment. These areas combine high unmet need with limited household capacity to fund private support. Population-scale interventions (NHS Talking Therapies expansion, integrated care board commissioning, employer-funded mental health support, low-cost digital delivery, voluntary-sector capacity) are the natural fit; approaches that depend on individual willingness or ability to pay will under-serve this group.

What the data does not show

The absence of a strong structural predictor is consistent with several explanations: that the drivers of anxiety sit outside what IMD measures (isolation, comparison pressure, workplace insecurity, housing precarity), that the survey captures a daily emotional state that crosses class lines more readily than chronic physical conditions do, or that the protective effect of income operates differently for mental than for physical health. The data is consistent with each of these readings but does not, on its own, establish any of them. Causal explanation is the next research question; this dataset rules out the simplest one.

07 — At a glance

Dataset, model, segments.

08 — Where this work fits

Anxiety is one example. The pattern is wider.

A single dataset can rule out a hypothesis. This one rules out the simplest deprivation-led account of where adult anxiety sits in England. The wider observation Rare's analytical practice runs into is the same across condition after condition: standard deprivation models capture less of population health than they are credited with, and decisions taken on those models alone leave real need unaddressed.

Rare's work runs along both sides of the resulting question. The public side: condition-level prevalence and unmet-need analysis for commissioners, integrated care boards, public health teams and charity researchers. The private side: provider-level intelligence for clinics, manufacturers, brand teams and investors who need to understand where services already exist and where they do not. The two surfaces share a dataset and a method; they ask different questions of the same evidence. We publish abbreviated findings like this one to put the structural question in front of the people who can act on it, in either market.