SPI-synthetic rows carry imputed reported council tax reduction on every person #497

SPI-synthetic rows carry imputed "reported" council tax reduction (CTR). In policyengine-uk that drives CTR claiming, and for English working-age households it is the CTR output itself.

Evidence

• The code. datasets/imputations/frs_only.py lists council_tax_benefit_reported in FRS_ONLY_PERSON_VARIABLES. The stage-2 QRF predicts it for every person row of the SPI-donor subsample and clamps it at 0.

• Only the HRP should carry it. In frs.py, FRS-reported CTR sits only on the household reference person (hrpid == 1), because CTREBAMT is a household variable.

• Measured on Enhanced FRS 2024-25 (1.56.16). No FRS-original household has more than one person with reported CTR, but 140 synthetic households have 2 or 3.

• Weights (2026, weighted). Synthetic rows hold:

  • £0.44bn of the £1.35bn English working-age CTR;
  • £0.056bn of Scotland's and £0.027bn of Wales's working-age "reported" CTR.

  Mean employment income on these rows is about £78k to £90k.

• Mismatched inputs. The synthetic rows keep the donor's household council_tax, which is net of the donor's CTR (see the linked issue), and the donor's savings.

Why it matters

• In policyengine-uk, would_claim_council_tax_reduction is claims_all_entitled_benefits | council_tax_benefit_reported > 0. Synthetic high earners with imputed CTR are therefore treated as claimants.
• For unsupported English working-age schemes, council_tax_benefit is the reported amount, so these imputed amounts flow straight into household incomes and CTR totals.
• FRS-reported amounts are also a weak benchmark for spending. The median CTREBAMT is about 26-29% of the undiscounted band bill in every nation. Admin data say 85-87% of recipients pay no council tax at all (Welsh Government 2025-26: "218,414 (84.9%) paid no council tax at all"; Scottish Government 2025-26: "87 per cent ... in receipt of full CTR").

Proposed fix

Either:

• drop council_tax_benefit_reported from the stage-2 person list and impute CTR receipt and amount at household level onto the HRP only, conditioned on low income; or
• set it to 0 on synthetic rows, since SPI-income households are rarely CTR recipients.

Related: #491 (take-up flags not re-anchored after stage 2), #448, PolicyEngine/policyengine-uk#1962, and PolicyEngine/policyengine-uk#1831 (per-benefit-unit claims_all_entitled_benefits, held for a decision on CTR take-up). This is a dataset change, so the release is Max's call.