[vahid-ahmadi]

Closes #6. Supersedes #8, which GitHub closed when #7's branch was deleted; rebased onto main. Also includes the dashboard copy tightening, the corrected free-hours incentive bullets, the two-switch labour supply control, and the new title.

Max asked for an intensive-margin response using a −0.042 childcare price elasticity of hours worked, derived from Brewer et al. (2020). This adds it.

What it does. For each adult in the responsive population (youngest child 1-4, not excluded by the OBR framework) who is in work with a positive out-of-pocket childcare cost, the proportional change in that cost is multiplied by the elasticity to give a change in hours, and earnings move with hours at a constant wage. Out-of-pocket cost is childcare expenses less Tax-Free Childcare (or the subsidy), less the paid care that new free hours displace, less the realised Universal Credit childcare support — the UC award minus the award with the element's coverage rate set to zero, the same counterfactual the baseline benchmark uses, since the element's face value is mostly withdrawn by the taper for working families. The revenue is not an assumed rate: each leg is rerun with the changed earnings and the revenue is the extra earnings less the rise in household net income. The household gain is allocated back to the adults whose hours produced it and laid over the distribution alongside the participation response, so the Household effects view moves with it. Low and high scale both margins by the same multiplier.

Results, 2027-28 central, UK. Parents in scope (1.26m; 1.02m see a price change) face a 58% mean fall in out-of-pocket cost under both legs together, giving +2.4% hours: 30,565 FTEs, £1.67bn of earnings, £0.67bn back to the exchequer, against £0.02bn from the participation margin. Combined dynamic cost £6.64bn → £5.95bn (10.4% offset; 3.2% low, 21.1% high). Free hours alone: 2,073 FTEs, £0.07bn (only displaced paid care moves the price). Subsidy alone: 29,401 FTEs, £0.63bn. UC families sit in scope with a near-zero price change, because the subsidy leg pays them nothing.

Caveats, recorded in sources.py, the README and on the dashboard. I checked the derivation against the paper. The +0.600 is Table A.3 panel B (file page 41) and is measured over all mothers with zeros for non-workers; 14.319 is Table 1's all-mothers mean (file page 17). So −0.042 is a total-hours elasticity that already contains the participation channel — the paper's own +3.5pp employment effect at part-time hours accounts for most of the +0.6 — and adding it to the extensive margin counts that channel twice. The treatment was also 12.5 extra free hours a week (file page 8), not a 100% price fall. The dashboard says both.

Also in this PR. _build_with_earnings has to call reset_calculations() before set_input: employment_income is a derived variable upstream, and the other order silently left earnings unchanged (revenue came out equal to earnings). Three CostTab strings used --x, a JavaScript decrement that mutated the loaded results by £1 and flipped the displayed sign; now -x. The extended-flag comparison from #7 now compares static costs, since that scenario does not carry the hours margin. Methodology tab, assumptions.margin, the module docstring and the CLI description all now describe both margins.

Tests. Eight unit tests on a stub simulation (price change, rerun earnings, revenue, household allocation for one and two earners, country reconciliation, scale, no-change, the 100% bound, and a tapered-UC family whose price is measured on realised support rather than the element) and an artifact gate that the dynamic cost equals static less both margins, each margin's country split reconciles, and revenue sits strictly between zero and earnings. dynamic_cost.labour_supply_offset_bn keeps its meaning (participation only); hours_offset_bn and total_offset_bn are added beside it.

66 tests pass. Run time is up by about 16 simulations a year for the reruns and UC counterfactuals.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
free-childcare-reform	Ready	Preview	Sep 2, 2026 12:10pm UTC