Impute is_apprentice on the UK spine so the apprentice employer NICs zero rate reaches 21- to 24-year-olds #1372

Problem

policyengine-uk #2505 applies the employer NICs zero rates for earners under 21 (SSCBA 1992 s.9A) and for relevant apprentices under 25 (s.9B). The apprentice rate reads the person input is_apprentice, which defaults to False. No UK dataset supplies it:

• Microcosm: no stage in uk/spec/sources.yaml or uk/source_stages.json writes is_apprentice.
• policyengine-uk-data: datasets/frs.py only takes it as an optional argument to the non-advanced-education proxy and passes zeros.

So in microsimulation no one aged 21 to 24 gets the apprentice relief. policyengine-uk#2505's model cost for the two reliefs is £1.42bn against HMRC's £2.47bn for 2025-26.

The FRS has no apprenticeship variable

I checked the public UKDS variable listings for FRS 2022-23 (SN 9252), 2023-24 (SN 9367) and 2024-25 (SN 9563). None of them mention "apprentice". The nearest fields record government employment schemes, which are not apprenticeships:

• TRAIN2 (Work Club, New Enterprise Allowance, Work Experience, Work Programme, Kickstart, and so on)
• NI3TRAIN (Northern Ireland schemes)
• TRAINEE ("on a government scheme for employment training")
• EMPSTATB code 8 ("Work related Govt training")

So is_apprentice cannot be derived from the survey. It has to be imputed from administrative statistics.

How much of the gap this explains

HMRC's tax relief statistics (January 2026), Table 2, splits the £2.47bn into two lines:

HMRC notes that the apprentice relief takes precedence for apprentices under 21. So the apprentice line covers every apprentice under 25, including those under 21 whom the model already relieves by age. Imputing apprentices aged 21 to 24 can close at most part of £570m, so it cannot close the £1.05bn gap on its own. Most of the gap sits in the under-21 line: the model's under-21 relief is well below HMRC's £1.9bn even before apprentices are counted. That gap needs a separate look at earnings and headcount for under-21 employees on the spine.

Approach (draft PR to follow)

Draw is_apprentice in frs_person_draws for employees aged 21 to 24 only. Employees are people whose employment_status (from frs_employment) is FT_EMPLOYED or PT_EMPLOYED.

• Rate. DfE's England participation per head of population aged 19 to 24, from the 2024/25 Apprenticeships data set "Geography Region - starts, achievements, participation, population by age, level" (national rows). In 2024/25, 252,960 participants over a population estimate of 4,221,504 gives 5,992 per 100,000. The series runs from 2019/20 to 2024/25 and goes into take_up_contract.json as apprenticeship_participation_19_to_24.

• Conversion to the eligible group. At the frame's household weights, the stage sets p = min(1, r * W(age 21 to 24) / W(employees aged 21 to 24)). It flags an eligible person when their identity-keyed uniform is below p, so the weighted apprentices equal r times the weighted population aged 21 to 24.

• Why only 21 to 24. Under-21s already get the s.9A zero rate, so flagging them adds no fiscal effect. It would also change their other policyengine-uk outcomes, because is_apprentice feeds:

  • is_in_non_advanced_education, which sets qualifying young person status for child benefit and Universal Credit
  • minimum_wage and minimum_wage_category
  • Care to Learn
  • the 16 to 19 Bursary Fund

  Restricting the flag to employees also keeps minimum_wage unchanged for the self-employed, whose Universal Credit minimum income floor reads it.

Caveats

1. DfE gives no single-year split of ages 19 to 24, so the rate is assumed uniform across those ages.
2. DfE ages are age at the start of the apprenticeship, while the relief uses current age.
3. DfE participation counts everyone in learning at any point in the academic year, which is more than the number in learning at a point in time. Together with the uniform split across ages 19 to 24, this likely overstates the point-in-time share, although age at start works the other way (18-year-old starters now aged 21 to 24 are not in the 19 to 24 count).
4. England's rate is applied UK-wide. The devolved series run lower per head: Skills Development Scotland reports 7,172 Modern Apprentices aged 20 to 24 in training at Q2 2025/26, of 38,902. The Welsh (Medr) and Northern Ireland (ApprenticeshipsNI) series also need adding. A follow-up could draw by nation.
5. The national solve calibrates to the OBR Class 1 employer NICs total, so it will absorb part of the change. The same caveat is in policyengine-uk#2505.

Validation once a spine is built

Compare the 2025-26 apprentice relief for ages 21 to 24 with HMRC's NIC - NS1 line (£570m, which includes under-21 apprentices). Also compare the weighted apprentices aged 21 to 24 with about 6% of the population aged 21 to 24, and check that ni_class_1_employer for flagged earners falls to the expected rate x max(0, earnings - AUST) above the threshold.

🤖 Generated with Claude Code