Repository navigation
FRS benefit units keep some qualifying young persons apart from the adults responsible for them #562
Description
What is wrong
policyengine_uk_data/datasets/frs.py keeps the FRS's own benefit units (pe_person["person_benunit_id"] = person.benunit_id). The FRS puts a 16- to 19-year-old in an adult's benefit unit only if they are a dependent child on the HBAI definition: unmarried, living with parents, and in full-time non-advanced education. Benefit law is wider, so some young people the law treats as an adult's qualifying young person (QYP) sit in a benefit unit of their own. policyengine-uk works out each programme per benefit unit, so the responsible adult's unit never counts them.
Who the law covers that the FRS split does not:
- Every 16-year-old until the 1 September after their 16th birthday, whatever their education.
- Universal Credit: UC Regs 2013 reg. 5(1)(a).
- Pension Credit: SPC Regs 2002 reg. 4A(1)(a).
- Child Benefit: Child Benefit (General) Regulations 2006 reg. 4, and through SSCBA 1992 s.142 the Housing Benefit and legacy "young person".
- Child Tax Credit: CTC Regs 2002 reg. 4.
- School leavers until their Child Benefit terminal date (reg. 7) or in the 20-week extension (reg. 5).
Evidence
16-year-olds outside education (research for PolicyEngine/policyengine-uk#2279, Enhanced FRS 2024-25 at uk-data 1.57.4, 2026):
- 196.7k weighted 16-year-olds are outside education (83 records).
- 190.6k of them (63 records) head a single-person benefit unit of their own.
- Every one of those 63 records shares a household with another benefit unit that has an adult aged 36 to 68.
- Adding the reg. 5(1)(a) route in #2279 therefore moves
universal_creditby £0: the route cannot reach the parents' claim.
16- to 18-year-olds in education in their own unit (run for PolicyEngine/policyengine-uk#2281, uk-data 1.58.0, 2026):
- 20 records, 5,339 weighted people, are in education, meet the Child Benefit QYP test, and are the only member of their own benefit unit.
- All 20 share a household with at least one other benefit unit.
- In records carrying 3,901 of the 5,339, another unit in the household has a member flagged
is_parent. - None of their own units reports Child Benefit; another unit in the household does in records carrying 1,974.
- The largest case is a 17-year-old in upper secondary education living with people aged 45, 51, 13 and 5.
- By the FRS definition these young people should be dependent children if they live with their parents, so either they do not (the other adults are not their parents) or
current_education(from TYPEED2 and FTED) says "in education" where the FRS's dependent-child flag says no. Which it is has not been checked.
Effect in policyengine-uk
- Until Child Benefit, Child Tax Credit and Pension Credit count a benefit unit's own claimant or partner as its child or qualifying young person policyengine-uk#2281, policyengine-uk paid these young people Child Benefit for themselves. After it, nothing is paid for them, where the law pays their parents.
- The UC child element, the two-child limit order, the work allowance, the benefit cap family rate, the Pension Credit child addition and Child Benefit all miss a 16-year-old who is in their own unit.
- The planned Child Benefit and CTC school-leaver routes (follow-ups to #2279) will have almost no effect in the microdata for the same reason.
What is needed
A way for policyengine-uk to see which adult is responsible for a young person who is in their own FRS benefit unit. Two candidate designs, not yet chosen:
- re-form benefit units on benefit-law dependency (keep a 16-year-old before their 1 September, and other QYPs living with a parent, in the parent's unit); or
- keep the FRS units and export a person-level link to the responsible adult's benefit unit, for policyengine-uk to read.
Either needs the FRS household relationship grid to identify the parent, and a policyengine-uk change to read it. Related: #486 (supply is_hbai_dependent_child from the FRS).
Activity
MaxGhenis commentedon Oct 10, 2026
Measured figures for this issue, from uk-data 1.58.0 and the raw FRS 2024-25 tables (aggregates only; the 1.57.4 numbers in the body are superseded).
16-year-olds outside education and training
frs_2024_25.h5@1.58.0enhanced_frs_2024_25.h5@1.58.0frs.py:902,951; head =UPERSON == 1at968). So the 19 FRS records that are outside both are adult-table records in a unit of their own.RELHRP. So the relationship grid gives a likely responsible adult for every one of them; kinship is a routing assumption, not established legal responsibility.traincolumn that no FRS release has #560, fixed in Read approved training from the FRS TRAINEE flag for dependants #561). After that fix every 16-year-old outside education and training is in a unit of their own, and policyengine-uk#2279's route (a) reaches no parent's claim at all.16- to 18-year-olds in education in their own unit (the second evidence block)
Traced to the raw tables: the 20 enhanced records are copies of 6 adult-table FRS records aged 16 to 18 (20.0k on GROSS4; two each at 16, 17 and 18) with
EDUCFT= 1.TYPEED2is 3 (3 records) or 8 (3 records).create_frsmaps those codes to lower or upper secondary by age (frs.py:1012-1047).TYPEED23 and 8 also appear on child-table records at 16 (19 records), so the code alone does not explain it, and the value labels are not in this repo.For the design choice
Measured effects of re-forming the units against keeping them (policyengine-uk#2279 on 1.58.0 with the 16-year-olds moved into the reference person's unit, and the same on the FRS grossing weights) are being run now; I will add them here.
Revised 11 October after an independent fact-check: kinship stated as an assumption, post-fix shares added.
MaxGhenis commentedon Oct 11, 2026
Measured effects for the design choice, as promised above. Every figure is a pair of policyengine-uk microsimulations on the same dataset, so each difference is #2279's effect on that dataset: base
3173baa8against policyengine-uk#2279 at6af70194.Datasets. uk-data 1.58.0 with #561's training fix applied, then three variants:
The last row of the table repeats this on the FRS grossing weights (
frs_2024_25.h5), because the enhanced weights for this cell rest on two households (#563).Each move lasts the whole year. It clears the person's head and claimant flags and removes their old unit; other inputs are left as they were. The figures are therefore #2279's effect on that edited dataset. They are not estimates of any particular implementation.
The child element and counts are for 2026-27. Pension Credit, Housing Benefit and the benefit cap reduction do not move in any row.
What this shows:
uc_assessable_capital.py:37-55). Its effect would differ and has not been measured.benunit_contains_household_head.py), but it cannot tell a child from another household member.add_universal_credit_reformdraws its new-claimant seed by benefit-unit position (policyengine_uk/scenarios/uc_reform.py:59-61). Comparing a re-formed dataset with the released one directly is therefore contaminated from 2026-27, where that draw is active: removing 59 units moves UC in 371 units, 361 of them outside the moved households. Base-against-branch pairs on one dataset are unaffected.Revised 11 October after an independent fact-check: the figures are labelled as effects on the edited dataset, the separate effect of moving the units is added, and linking is separated from merging.