Approved training is never set: the FRS build reads a train column that no FRS release has #560

Summary

is_in_approved_training is false for every person in the FRS and enhanced FRS datasets. derive_is_in_approved_training_from_frs_person reads a train column, and returns False for everyone when it is missing (frs.py:367-376). No FRS release has that column: the adult tables for 2020-21, 2022-23, 2023-24 and 2024-25 carry TRAIN2 and TRAINEE, and the child tables carry TRAINEE.

Who it affects

DWP's FRS counts a 16- to 19-year-old as a dependent child only when they are "in full-time non-advanced education or in unwaged Government training", unmarried and living with their parents. So a child-table record aged 16 to 19 outside full-time education is a government trainee.

FRS 2024-25 confirms it. Of the 860 child-table records aged 16 to 19, 20 have EDUCFT = 2 (not in full-time education), and all 20 have TRAINEE = 1:

uk-data codes all 20 as NOT_IN_EDUCATION with is_in_approved_training false. In policyengine-uk they then fail the education-or-training condition of the qualifying-young-person tests for Universal Credit (reg 5(1)(b)), Pension Credit (reg 4A(1)(b)), Child Benefit and Child Tax Credit, each of which reads is_in_non_advanced_education | is_in_approved_training. In the released enhanced_frs_2024_25.h5 (uk-data 1.58.0) 74 persons are copies of these 20 records, with 12.2k of weight, and none is flagged.

No adult-table record has TRAINEE = 1 in 2024-25 (250 answer 2; the rest are blank).

Why it went unnoticed

• The function defaults to False when the column is missing, so the build never failed.
• Its tests pass a synthetic train column.
• The code list it applies (1 to 9 as schemes, "10 or 13 for None of these") does not fit TRAIN2 either: in 2024-25, 4,879 adult records have TRAIN2 = 9, 28 have 1 to 8 and 2 have 10. The value labels for TRAIN2 are not in this repo, so the fix does not use it.

Fix

Read TRAINEE = 1 ("Whether on a government scheme for employment training"; the FRS dictionary gives it the same yes/no value format as EDUCFT), and fail the build when the column is missing.

Found while checking how PolicyEngine/policyengine-uk#2279 (the UC reg 5(1)(a) route for 16-year-olds) reaches the data: the six 16-year-old trainees are the only 16-year-olds outside education who sit in a parent's benefit unit.