KidCost: Methodology & Sources

Generated 2026-07-13 from the build artifacts (data.json built 2026-07-13 · spec 2026-07-10-kidcost-v2-design). 704 provenance-carrying values, 600 source-registry entries, 120 disclosed gaps across 26 cities. Every number on the calculator traces back to a row on this page.

How values are produced

Every city × category value is produced by a written evidence dossier (pipeline/kidcost/dossiers/<city>/<category>.yaml) built to a fixed protocol, then machine-validated on every build (validate.py: a value missing sources, derivation, or a confidence grade fails the build). The protocol:

1. Enumerate candidate sources per city: national statistical office, ministry/regulator fee schedules and subsidy scales, municipal data portals, central-bank and household surveys, established research institutes, peer-reviewed papers. An explicit exclusion list is enforced structurally by the registry: marketing surveys, listicles, bank client surveys, and crowdsourced aggregators cannot be cited at all.
2. Independence requirement: a value graded B or better needs at least 2 genuinely independent sources — different collecting institutions, not two articles citing the same survey — and 3 where the market is heterogeneous (childcare, private K-12, housing).
3. Triangulation & disagreement handling: all candidates are converted to a common estimand and a common price year via national CPI (sub-index where available). If sources disagree by more than 25% after normalization, the dossier must explain why or widen the range to cover both — never average silently.
4. Confidence grading (rubric below) stored per value and shown in the UI; grade D is never displayed as a number — it becomes a disclosed gap with a reason.
5. Adjustment, vintage, and audit discipline: national-to-city adjustments use a documented, separately-cited factor or the value stays labeled national; no silent vintage splicing (a CPI bridge longer than 3 years downgrades confidence one grade); and every dossier records the candidate sources it REJECTED and why — the audit trail is part of the artifact.

Every value is an interval {low, central, high} (percentiles where the source publishes them, an envelope of independent sources otherwise, and low = central = high for statutory prices). Projections propagate the interval bounds through the category sum assuming perfect dependence across categories — a deliberate, conservative simplification: the shaded band you see is the widest honest reading of the sources, not a confidence interval.

Confidence grades

• A — statutory/administrative price or official statistical-agency microdata/tables: the law sets the number, or the agency measured it.
• B — official survey at coarser granularity (national, city-adjusted with a cited factor) or an established institute with published methodology (EPI, CPAG, College Board, KFF, OECD, NCCHD, KOSIS).
• C — single reputable source, unreplicated; or a reputable source requiring a long CPI bridge or a geographic transfer.
• D → gap — nothing reputable exists. A grade-D finding is structurally unrepresentable as a value: the UI shows “no data: reason” (with the nearest reputable proxy clearly labeled when one exists) instead of an invented number.

City tiers are an output of this process, not an input: a city is “verified” only if at least 7 of its 9 core cost categories reach grade B or better; everything else is “indicative” and banner-labeled as such.

Tax & benefit engines

Taxes and child-linked cash benefits are never hand-copied constants: they come from law-encoding engines run offline in CI to produce per-city gross→net + benefits lookup grids (income sweeps × filing statuses × 0–4 children) that the frontend interpolates. Law engines: PolicyEngine-US (federal + state + CTC/EITC/childcare subsidies), PolicyEngine-UK, OpenFisca-France (quotient familial, CAF/PAJE), GETTSIM (Kindergeld/Kinderzuschlag). Remaining cities use hand-coded manual engines, each with statute citations and golden tests cross-checked against the jurisdiction's official calculator. The 15 grid cities:

Source registry

The registry (pipeline/kidcost/sources.yaml) is the only place source ids are defined; a value can only cite ids listed here, and entries may only be of an allowed institutional type (official, statutory, institute, peer-reviewed). All 600 entries, with how many of the 704 built values cite each:

Coverage by city and category

Value count per cell with its confidence mix ((2A·1B) = two grade-A values and one grade-B), plus disclosed gaps (n gaps). An em-dash means the category has no values and no recorded gaps for that city yet.

What we cut, and why

v1 had 30+ cost sections; v2 keeps the 9 core categories plus one-time birth costs. The plan's scope-cut lists, verbatim:

CUT: FAFSA estimator, wedding fund, elder care, estate planning, life insurance, gap year, allowances, pets, milestone events.

FOLDED: tutoring/test-prep -> k12 supplementary line (sourced for East Asia via KOSIS sagyoyuk etc.); summer programs -> childcare school-age.

One line on each cut:

• FAFSA estimator — US financial-aid awards depend on family-specific inputs no calculator of ours can source honestly; a precise-looking wrong number is worse than none.
• wedding fund — a discretionary post-18 transfer, not a cost of raising a child to adulthood.
• elder care — belongs to the parents' own late-life planning, not to the child's cost trajectory.
• estate planning — legal-fee anecdotes only; no authoritative price source exists at city granularity.
• life insurance — premiums price the parent's underwriting profile, not the child.
• gap year — a pure scenario choice with no citable market price statistic.
• allowances — an intra-household transfer, not a market cost -- it buys things already counted in other categories.
• pets — not a cost of the child.
• milestone events — no reputable price statistics exist -- the circulating figures are vendor-marketing folklore, exactly the source class the registry excludes.

And each fold:

• tutoring/test-prep -> k12 supplementary line — where supplementary education is a structural share of child spending (East Asia) it is kept and sourced (e.g. KOSIS private-education survey); elsewhere it is a scenario choice inside K-12, not its own category.
• summer programs -> childcare school-age — same estimand -- paid care for a school-age child outside school hours -- so it is priced inside childcare instead of double-counted.

Currency conversion (FX/PPP)

Local currency is the source of truth; every USD figure is a labeled conversion. Rates come from Frankfurter (ECB reference rates), fetched at build time and cached fail-soft (current build rate date: 2026-07-13). Two currencies are absent from ECB reference rates but hard-pegged to the US dollar by statute/central-bank policy, so their conversions use the peg constants, not market rates: AED 3.6725/USD (Central Bank of the UAE, fixed since November 1997) and SAR 3.75/USD (Saudi Central Bank/SAMA, fixed since June 1986). The optional PPP comparison view uses the World Bank private-consumption PPP factor (PA.NUS.PRVT.PP); PPP is shown only as an alternative lens and never mixed into the headline totals.