Anchor bus fare distribution to DfT BUS05ai and NTS0705a

[vahid-ahmadi]

What changed

calibrate_bus_fare_spending now applies three anchors instead of one:

1. National total (existing): DfT BUS05aii passenger fare receipts, England → UK.
2. Within-England regional split (new): DfT BUS05ai receipts — London £1,347m / outside London £2,070m of England's £3,417m.
3. Income distribution (new): England quintile fare shares anchored to population-weighted NTS0705a local-bus trip rates (2024: Q1 47.6 … Q5 13.0 trips/person/year), assuming a common average fare per trip.

Anchors 2 and 3 alternate in a small IPF loop that preserves the national total. Devolved nations are untouched.

Why

• The LCFS imputation under-captures London bus fares (Oyster/contactless spend is poorly separated in household diaries). With national-only calibration the shortfall is smeared across the other English regions: the built dataset put London at ~£0.44bn vs BUS05ai's £1.35bn, overstating outside-London fare exposure by ~47%. This directly distorts any England-outside-London policy analysis (e.g. the 2027 £2 bus fare cap).
• The raw LCFS income gradient runs opposite to NTS evidence: richer households record more bus & coach spend, while NTS0705a shows the lowest income quintile makes ~3.7× the local-bus trips of the highest. Anchoring to trip rates flips the distributional impact of fare policies to the NTS-consistent direction.

This is the published-tables version of anchoring the imputation to the NTS; a follow-up could use NTS trip-level microdata (UKDS) to impute person-level trips directly.

Tests

tests/test_bus_fare_distribution.py validates the built Enhanced FRS via the existing baseline fixture (skips when the dataset has not been rebuilt), so the assertions run after a rebuild rather than against unbuilt code:

• national total within 5% of the BUS05aii target
• London share of England fares within 3 points of BUS05ai
• quintile fare totals decline with income per the NTS gradient

🤖 Generated with Claude Code