Cheaper twins: compare nearby areas and their price gaps

• England postcode sectors with at least 150 property rows and 40 property rows with non-null recorded £/m²; positive, non-null median estimated £/m² and non-null centroid coordinates and station distance.
• Centroids at most 3 km apart, using the original local planar approximation: x = longitude × 111.320 × cos(mean index latitude), y = latitude × 110.574.
• Same modal property type; finite median construction years, excluding years ≤1800, differing by at most 30 years.
• Good+ primary and secondary school catchment counts each differ by at most 1.5. This does not require the same schools or guarantee admission.
• Mean nearest-station distance at most 1.5 km in each sector, with a difference of at most 0.9 km. Per postcode, this is the minimum straight-line distance to rail, Tube, DLR or tram/metro stops. It does not require the same station or commute.
• Serious-crime annual averages over the source’s seven-year window within a factor of 1.5 in either direction. These are counts, not population-adjusted rates.
• Owner-occupied share differs by at most 22 percentage points; degree-or-higher share by at most 22 percentage points. Deprivation was not a separate matching variable.
• Median age differs by at most 12 years when both values are finite; otherwise this check is skipped.
• Ratio of higher-price sector median floor area to lower-price sector median floor area between 0.72 and 1/0.72 (about 1.389), when both are finite and the latter is positive; otherwise this check is skipped.
• Estimated £/m² gap between 15% and 45%, with an illustrative gap of at least £20,000 at 90 m². For each higher-price sector, retain the qualifying candidate with the largest percentage gap, then deduplicate unordered pairs.

Property statistics are medians over property rows. Area metrics and centroids are means weighted by the number of property rows in each postcode, excluding null metric values from the denominator. School counts are modelled catchment counts, not school quality scores. Missing values in the original implementation were not uniformly rejected; a published pair is not a complete-data guarantee.

The archived run produced 415 candidate pairs. The published subset took the six largest illustrative 90 m² gaps overall, eight largest among terraced/semi-detached/detached sectors outside E, EC, WC, W, SW, SE, N and NW postal areas, and six largest among those property types within those postal areas, removing duplicates. This selection favours large gaps and is not a representative sample of England.