Greater Manchester 2026 · Net Change
152 ward flips · 213 declared wards · 10 boroughs

Labour 146 → 34.
Reform 0 → 104.
The map redrew itself.

The companion page to Winners asked which Census variables predict where each party won. This page asks where each party gained, lost, or held. Of the 213 wards with both a declared 2026 winner and a known prior party, 152 changed hands — a 71% turnover. Labour shed 112 of its 146 prior seats; Reform took 104 from a standing start; the Greens went from 5 wards to 30. The structural signal is sharp where the sample allows it: flipped-to-Reform tracks +0.64 with apprenticeships and −0.50 with degree-holding; flipped-to-Green inverts the polarity with +0.67 with under-29s, +0.65 with private renters, +0.59 with density — gains in the dense, young, renter-heavy inner cores. The before/after cards in §3 show how each party's *surviving* coalition shifted demographically (Labour's 34 remaining wards skew slightly more graduate / SOC1-3; Greens' 30 are vastly denser and more renter-heavy than their 2022 footprint). Smaller parties (LibDem 5 flips, Other 6, Labour 3 gains, Conservative 3, Independent 4) sit too close to the noise floor to read off correlation cards cleanly.

Method: "Prior party" is the top-of-poll party the last time each seat was contested before 7 May 2026 — for the nine thirds boroughs that's the May 2022 election (the same seat slot, four years apart); for Salford (all-out cycle) it's May 2021. Prior winners scraped from the per-borough Wikipedia articles for those years; see scripts/01b_extract_prior.py. Where a 2026 ward has no clean 1:1 prior because of the 2023/2024 boundary reviews in Bolton, Stockport, Trafford, and Wigan, we use the closest-predecessor ward (~31 wards, flagged ≈ in the appendix). One ward (Manchester · Miles Platting & Newton Heath) is still pending and is excluded from the analysis; Bury · Moorside (cancelled, rerun expected June) appears with no 2026 winner. The §3 before/after table also carries each party's mean ward income (ONS small-area income estimates for FY ending March 2023, "Net income before housing costs", aggregated from MSOA to ward via Census-2021 OA-count weighting — see scripts/03b_aggregate_income.py). Per-party correlation cards below show only parties with at least 5 flipped wards (MIN_FLIPS = 5); Pearson r on smaller samples is essentially noise. The headline shift band below groups Labour, Green, and Lib Dem as "left" and Conservative and Reform as "right"; Independent and Other are excluded as heterogeneous.
Left → Right flips
Wards that flipped from Labour, Green, or Lib Dem to Conservative or Reform.
Right → Left flips
Wards that flipped from Conservative or Reform to Labour, Green, or Lib Dem.
Shift ratio
Left→right flips per right→left flip. Independents and small "Other" parties are excluded as heterogeneous.
§ 1 Census variables × flipped-to-party Pearson r · 213-ward analysis sample

Read each column as: "if a ward flipped to this party, how does each Census variable correlate with that flip?" Positive r in green, negative in red, opacity scaled to |r|. Cells with — are below the n=3 minimum for Pearson.

§ 2 Top correlations per "flipped-to" party Top 5 positive · top 5 negative · MIN_FLIPS = 5
§ 3 Each party's coalition · 2022 → 2026 Mean Census profile of held wards · Δ = 2026 − 2022

Each card shows how a party's structural footprint shifted between the prior and 2026 elections. Reform and Other have no 2022 baseline (Reform UK didn't win any GM ward in 2022; "Other" is a 2026 catch-all for Workers Party / Oldham Group / etc.) so their cards show 2026 means only. Δ for percentages is in percentage points; for density it's residents/km²; for age it's years. Bold positive Δ in green, bold negative in red — colour intensity scales with magnitude relative to typical inter-party spread.

§ 4 All wards · prior → 2026 Sorted: flips first · borough · ward

"Prior" is the top-of-poll party the last time the seat was contested (2022 for thirds boroughs, 2021 for Salford all-out). flip = ward changed party hands; hold = same party as before; no prior = boundary change, no clean predecessor. ≈ next to a ward name marks an approximate prior mapping.