Problem
would_claim_uc is drawn once per benefit unit in create_frs (policyengine_uk_data/datasets/frs.py, the take-up block around lines 1487 and 1540-1544). The draw uses assign_takeup_with_reported_anchors: benefit units with a UC reporter are forced True, and everyone else is filled so that 55% of all benefit units are True (parameters/take_up/universal_credit.yaml, which cites no source). The UC anchor holds when the draw is made, but the stage-2 QRF later rewrites universal_credit_reported on SPI-synthetic rows and never re-anchors it (#491). Legacy-benefit reporters get no special treatment: HB, tax credits, income-related ESA, IS and income-based JSA.
Every working-age legacy benefit has now closed in PolicyEngine UK:
Income-related ESA ended on the same dates, by SI 2025/1148 art 3A (inserted by SI 2026/409) and SR 2025/176 art 3A. PE-UK does not model that yet.
A legacy reporter whose draw came out False therefore loses the legacy benefit and gets no UC. On the Enhanced FRS 2024-25 (uk-data 1.56.16, sha256 e433e532…), in FY 2026-27, 175.2k working-age benefit units report a legacy means-tested benefit and have would_claim_uc False. 47.9k of them also report UC, on SPI-synthetic rows where the anchor broke. Of the 175.2k:
- 81.3k report HB, and 28.0k of them were still paid HB through the continuing-award route (£0.184bn);
- 62.7k report tax credits.
Real runs (clean main at 1c5b4d04, current weights) give the effect of switching would_claim_uc to True for these groups:
- all 175.2k: UC +£0.683bn, HB −£0.184bn;
- the HB reporters alone: UC +£0.470bn, HB −£0.184bn.
The real world looks very different from the draw. DWP's Move to UC statistics to the end of March 2026 (published 12 May 2026) report 1,822,374 households sent a migration notice, of whom 87% claimed UC. 814,703 households were awarded transitional protection, 53% of those eligible. The claim rate varies widely by legacy cohort (Stat-Xplore, dataset "Households invited to Move to Universal Credit", tables MtUC Households 3 and 6, retrieved 2026-09-30):
| Legacy cohort (GB) |
Notices |
Claimed UC |
Claim rate |
Transitional protection, share of claimants |
| HB only |
41,300 |
30,900 |
74.8% |
22.3% |
| HB and income-related ESA (incl. with CTC) |
532,271 |
522,112 |
98.1% |
63.4% |
| HB and any other legacy benefit |
750,093 |
730,696 |
97.4% |
58.5% |
| Income-related ESA only |
335,190 |
321,075 |
95.8% |
31.6% |
| IS only |
19,067 |
17,719 |
92.9% |
29.1% |
| JSA only |
6,199 |
5,867 |
94.6% |
24.1% |
| Tax credits only (CTC, WTC, both) |
650,635 |
454,700 |
69.9% |
58.1% |
| All |
1,822,966 |
1,580,761 |
86.7% |
51.6% |
Rows overlap: the third row includes the second. "ESA" is Stat-Xplore's label. Migration notices go only to claimants of an "existing benefit", which covers income-related ESA but not contributory ESA. The protection share is of all claimants.
Northern Ireland's Department for Communities table 28 (UC statistics, May 2026) shows the same pattern: ESA & HB 98%, HB only 76%, all cohorts 88%.
In the dataset, 82% of working-age HB reporters who do not also report UC have would_claim_uc True (weighted). The share is 79% for the largest group, HB plus income-related ESA (182k units, 633 records), whose observed claim rate is 98%. Among all working-age HB reporters the share is 72%, pulled down by SPI-synthetic rows with the broken anchor. On FRS-original rows the draw treats HB reporters like any other non-reporter (unweighted, about 48%); the weighted shares come from calibration.
Proposal
Order. Land PE-UK's income-related ESA abolition, and make ESA(IR) exclusive with UC, first. Today PE-UK pays 137k units both ESA(IR) (£1.6bn) and UC. Moving more ESA(IR) reporters onto UC before that fix would widen the double payment.
Option (a), data only.
- Add
parameters/take_up/uc_managed_migration.yaml, with claim rates by legacy cohort from the table above and its source.
- In the take-up block, anchor working-age legacy reporters who do not report UC to
would_claim_uc True with their cohort's probability.
- Draw from a separate seeded generator, so that the
seed=100 sequence for every other flag is unchanged.
- Property tests: every UC reporter is True, the cohort shares land within sampling error of the rates, and flags outside the legacy cohorts are identical to before.
Option (b), data plus model. This is the alternative if 2024-25 and 2025-26 legacy receipt must stay intact until each benefit closes.
- Add a PE-UK input
would_claim_uc_at_legacy_closure, drawn here at the cohort rates.
- Add a PE-UK
claims_universal_credit variable for the five would_claim_uc consumers: would_claim_uc | (legacy award closed in the period & the new input).
- Guard it with
require_variable and raise the PE-UK floor.
Either option moves the calibrated weights (calibration runs at FY 2025-26 against OBR HB and UC targets), so the effect has to be measured from a CI build, not scaled from the counterfactual above. It also bears on #452 (UC caseload about 40% below administrative counts).
Transitional protection is not modelled in PE-UK, so option (a) or (b) would pay migrants plain UC. See the linked PE-UK issue.
Problem
would_claim_ucis drawn once per benefit unit increate_frs(policyengine_uk_data/datasets/frs.py, the take-up block around lines 1487 and 1540-1544). The draw usesassign_takeup_with_reported_anchors: benefit units with a UC reporter are forced True, and everyone else is filled so that 55% of all benefit units are True (parameters/take_up/universal_credit.yaml, which cites no source). The UC anchor holds when the draw is made, but the stage-2 QRF later rewritesuniversal_credit_reportedon SPI-synthetic rows and never re-anchors it (#491). Legacy-benefit reporters get no special treatment: HB, tax credits, income-related ESA, IS and income-based JSA.Every working-age legacy benefit has now closed in PolicyEngine UK:
gov.dwp.tax_credits.active);gov.dwp.income_support.active,gov.dwp.JSA.income.active);Income-related ESA ended on the same dates, by SI 2025/1148 art 3A (inserted by SI 2026/409) and SR 2025/176 art 3A. PE-UK does not model that yet.
A legacy reporter whose draw came out False therefore loses the legacy benefit and gets no UC. On the Enhanced FRS 2024-25 (uk-data 1.56.16, sha256
e433e532…), in FY 2026-27, 175.2k working-age benefit units report a legacy means-tested benefit and havewould_claim_ucFalse. 47.9k of them also report UC, on SPI-synthetic rows where the anchor broke. Of the 175.2k:Real runs (clean
mainat 1c5b4d04, current weights) give the effect of switchingwould_claim_ucto True for these groups:The real world looks very different from the draw. DWP's Move to UC statistics to the end of March 2026 (published 12 May 2026) report 1,822,374 households sent a migration notice, of whom 87% claimed UC. 814,703 households were awarded transitional protection, 53% of those eligible. The claim rate varies widely by legacy cohort (Stat-Xplore, dataset "Households invited to Move to Universal Credit", tables MtUC Households 3 and 6, retrieved 2026-09-30):
Rows overlap: the third row includes the second. "ESA" is Stat-Xplore's label. Migration notices go only to claimants of an "existing benefit", which covers income-related ESA but not contributory ESA. The protection share is of all claimants.
Northern Ireland's Department for Communities table 28 (UC statistics, May 2026) shows the same pattern: ESA & HB 98%, HB only 76%, all cohorts 88%.
In the dataset, 82% of working-age HB reporters who do not also report UC have
would_claim_ucTrue (weighted). The share is 79% for the largest group, HB plus income-related ESA (182k units, 633 records), whose observed claim rate is 98%. Among all working-age HB reporters the share is 72%, pulled down by SPI-synthetic rows with the broken anchor. On FRS-original rows the draw treats HB reporters like any other non-reporter (unweighted, about 48%); the weighted shares come from calibration.Proposal
Order. Land PE-UK's income-related ESA abolition, and make ESA(IR) exclusive with UC, first. Today PE-UK pays 137k units both ESA(IR) (£1.6bn) and UC. Moving more ESA(IR) reporters onto UC before that fix would widen the double payment.
Option (a), data only.
parameters/take_up/uc_managed_migration.yaml, with claim rates by legacy cohort from the table above and its source.would_claim_ucTrue with their cohort's probability.seed=100sequence for every other flag is unchanged.Option (b), data plus model. This is the alternative if 2024-25 and 2025-26 legacy receipt must stay intact until each benefit closes.
would_claim_uc_at_legacy_closure, drawn here at the cohort rates.claims_universal_creditvariable for the fivewould_claim_ucconsumers:would_claim_uc| (legacy award closed in the period & the new input).require_variableand raise the PE-UK floor.Either option moves the calibrated weights (calibration runs at FY 2025-26 against OBR HB and UC targets), so the effect has to be measured from a CI build, not scaled from the counterfactual above. It also bears on #452 (UC caseload about 40% below administrative counts).
Transitional protection is not modelled in PE-UK, so option (a) or (b) would pay migrants plain UC. See the linked PE-UK issue.