Unpriced

National Childcare Benchmarks

2022

This demo now shows two descriptive childcare benchmarks side by side. The narrower active-care bridge benchmark scales respondent-level active childcare time to the under-5 population. The broader annual-hours childcare account keeps active household care, active nonhousehold care, and supervisory care explicit in annual hours. Both preferred benchmarks use the direct-care-equivalent price layer, while the gross-market figure is retained as an upper bound.

Measure	Active-care bridge	Annual-hours account	Unit
Preferred benchmark	$28.36B	$1.47T	annual national value
Gross-market upper benchmark	$34.16B	$1.77T	annual national value
Excluded non-direct residual	$5.80B	$300.17B	annual national value
Direct-care-equivalent price	$9,217	$9.72	bridge: per slot-year; annual-hours: per hour
Gross market price	$11,102	$11.70	bridge: per slot-year; annual-hours: per hour
Quantity	3.08M	151.06B	bridge: annual slots; annual-hours: annual hours
Active childcare hours	152.7	40.97B	bridge: per respondent-year; annual-hours: annual national hours
Supervisory childcare hours	—	110.09B	annual national hours
Price-support population share	41.1%	41.1%	of national under-5 pop.

How the direct-care-equivalent price is derived

▶

The direct-care-equivalent price isolates the labor component of the gross market price:

direct_care_price = (wage × fringe_multiplier × annual_hours) / children_per_worker

where wage is the observed or imputed childcare-worker hourly wage from QCEW, weighted by the state-year provider-type and age mix from ACS. The implied wage is back-solved: wage = direct_care_price × children_per_worker / (fringe_multiplier × annual_hours). The raw labor-equivalent price is clipped at the gross market price—direct-care cannot exceed the total.

Assumption	Value	Source
Fringe multiplier	1.3763	BLS ECEC Table 4, June 2025 (compensation-to-wages for private-industry health care & social assistance)
Annual hours	2,080	BLS OEWS Technical Notes (52 × 40 FTE convention)
Center staffing (children/worker)	4 / 7 / 10	ACF/NCECQA Brief #1, Table 4 (infant/toddler/preschool most-common state ratios)
Home staffing (children/worker)	2 / 4 / 6	Head Start 45 C.F.R. §1302.23 (benchmark mapping)
Market hours/child/week	18.24	NCES ECPP 2019 + Digest 202.30

This is not cost accounting—it is a transparent assumption-based decomposition. The non-direct-care residual (gross minus direct-care) is a pooled bucket that can cover facilities, administration, meals, advertising, transport, profits, and market power rather than a separately identified markup. The annual-hours account is broader because it keeps active household care, active nonhousehold care, and supervisory care explicit; supervisory care is still additive and not yet overlap-adjusted.

Interpretation: Read both as accounting benchmarks, not as predictions of what would happen if all unpaid care were marketized. The active-care bridge is a population-scaled bridge quantity, not an under-5-only ATUS time slice. The annual-hours account is much broader and should not be read as interchangeable with the bridge benchmark.

Counterfactual Marketization Price

Headline

If some share (α) of currently unpaid childcare entered the paid market, what price would the sector clear at? α is the outsourcing share: α = 0 is the status quo, α = 0.50 means half of unpaid hours enter the market, and α = 1.00 means full outsourcing. The headline shown here uses the observed-price core sample and the project’s parsimonious household-controls demand setup: 556 scenario rows across observed price-support years 2014–2022. The separate demand-fit audit reported below is a quality check, not a second headline.

Measure	Value	Unit
Gross market price	$8,224	per child per year
Direct-care-equivalent price	$6,551	per child per year
Displayed non-direct-care remainder	$1,673	per child per year
Implied direct-care wage	$9.73	per hour
Marketization price (α = 0.50)	$8,428	per child per year
Demand elasticity	−0.143	at mean
Supply elasticity	4.035	weighted median
LOO state R²	−0.085	held-out fit
LOO year R²	−1.621	held-out fit
Bootstrap acceptance rate	92.0%	92 / 100 draws

Displayed non-direct-care remainders are the arithmetic remainder from displayed gross and direct-care medians so decomposition displays read cleanly. The separate report artifacts still retain component medians.

Status: proof-of-concept headline. The selected observed-price core run is economically admissible, and the latest checked-in real-mode audit accepted 92 / 100 bootstrap draws.

How the supply curve is estimated in this demo

▶

The headline supply curve is built from county differences in childcare prices and provider availability, not from a national causal instrument. The first step keeps only county observations with a positive annual price and a positive provider presence.

Then the model looks within each state-year. A state-year counts as a direct estimate only if it has at least five county observations and enough variation in both prices and provider availability to trace a usable upward-sloping relationship. Each usable state-year produces one slope.

Those positive state-year slopes are then summarized with a population-weighted median, using the under-5 population when that information is available. If the within-state-year evidence is not usable, the demo falls back in order to a year-adjusted national slope, then a pooled national slope, and only then to a conservative default of 0.5.

The current headline uses a supply elasticity of 4.035. Read that as a local benchmark around today’s market: in the constant-elasticity version of the demo, a 1% higher price is associated with roughly 4% more paid-care capacity.

How the Model Works

The core intuition is a standard supply-and-demand framework. The observed market price (P₀) is the current equilibrium. When unpaid childcare enters the paid market, demand shifts right from D to D′. The new equilibrium price P(α) depends on how elastic supply and demand are.

The Outsourcing Shift

A stylized supply-and-demand diagram. The current canonical fit still implies inelastic demand (ε = −0.143) and elastic supply (ε = 4.035), so outsourcing shifts move mostly through quantity rather than explosive price increases. When unpaid care enters the market, demand shifts right (D → D′), moving the equilibrium from E₀ to E′ with a modest price increase.

This is a conceptual diagram illustrating the model logic. For the actual estimated curves with real data, see Solver-Implied Curves below.

Price Decomposition by Alpha

The gross market price is shown as a direct-care labor component plus a displayed non-direct-care remainder (facilities, administration, meals). The direct-care component grows with alpha as the equilibrium wage rises.

Direct-care component

Displayed non-direct-care remainder

Gross price is labeled above each bar. The amber segment is the displayed remainder from the rounded gross and direct-care medians so the bars close visually. The wage stability ($9.73–$10.16 across the full alpha range) still makes it a useful anchor.

Scenario Intervals

Each point is the median marketization price across canonical scenario rows at a given alpha. Shaded bands are bootstrap 10th-to-90th percentile intervals from resampling the demand and supply elasticities.

The marginal point is the shadow price at an infinitesimal outsourcing increment. Intervals widen as alpha grows because larger demand shifts amplify elasticity uncertainty. At α = 1.00, the median price rises about 4.8% from the baseline in the current canonical run.

Dual-Shift Marketization Frontier

This is the main medium-run sensitivity layer. The canonical headline above is a short-run fixed-supply benchmark where marketization shifts demand only, so positive α raises price by construction. The dual-shift layer also allows paid-care capacity expansion and provider cost pressure, so the price effect can rise, stay flat, or fall.

Generated pooled headline-alpha frontier at α = 0.50. The horizontal axis shows additional provider cost pressure at headline alpha, and the vertical axis shows additional paid-care capacity expansion at headline alpha. The overlaid line marks the median zero-price frontier.

At α = 0.50, the grid spans about 0% to 112% extra supply expansion and 0% to 11% extra cost pressure. The short-run pooled headline remains canonical; this figure is an additive medium-run sensitivity layer.

Same data as the frontier above, reframed as lines. Each line is a provider-cost-pressure scenario; where it crosses the dashed zero line, price stays roughly flat.

Read each line left to right: as paid-care capacity expands, price falls. Higher provider cost pressure (warmer colors) requires more capacity expansion to offset the price increase. The zero crossing is the break-even point.

How much expansion keeps price roughly flat at α = 0.50?

Provider cost pressure	Needed paid-care capacity expansion
About 0%	About 28%
About 5%	About 58%
About 11%	About 93%

Optional Method Mechanics

The next two blocks are method sensitivities that help interpret assumptions. They are diagnostic layers and do not replace the short-run canonical pooled headline.

Solver-Implied Curves

The solver constructs constant-elasticity curves through the observed baseline: Qd(P) = Q₀ × (P/P₀)^(−|εd|) and Qs(P) = Q₀ × (P/P₀)^(εs). Both curves pass through the baseline point (Q₀, P₀) by construction. Unlike the stylized diagram above, these use real estimated elasticities and data-derived baselines.

At α = 0.50, price rises about +$204. At α = 1.00, about +$394. These are solver-implied curves from the canonical elasticities, not nonparametric empirical schedules.

Piecewise Supply Extension

Maybe supply is easier to expand when prices are below today’s level and harder once prices rise above it. This demo allows exactly one kink in the supply curve. Below today’s baseline price, supply is more elastic (η = 5.78); above it, supply is stiffer (η = 2.60). That pushes predicted marketization prices up by about +$92 at α = 0.50 and +$182 at α = 1.00 relative to the constant benchmark.

Only 21 of 107 eligible state-years have directly estimated positive slopes on both sides of the baseline. Most rows (80.4%) borrow a pooled fallback on at least one side. This is a methodology demonstration, not a richly estimated piecewise supply system.

How the one-kink supply version is estimated

▶

The piecewise version keeps the current statewide price as the bend in the curve. Within each state-year, county observations are split into counties at or below that baseline price and counties above it.

A separate price-responsiveness slope is estimated on each side when there is enough usable positive variation. If one side cannot be estimated cleanly, the demo borrows a pooled positive slope from the same side of the market in the broader sample.

The two branches are forced to meet at today’s price, so the curve is continuous but changes steepness at the baseline.

Only 21 of 107 eligible state-years have directly estimated positive slopes on both sides, and 80.4% borrow a pooled fallback on at least one side. That is why this is labeled as a method demonstration rather than a state-by-state structural system.

Sample & Data Support

Observed support is reasonably strong through 2022; this section audits the headline by showing where identification stability and out-of-sample fit remain fragile.

Pipeline Provenance

All inputs are free public data. The pipeline joins six sources into county-year and state-year panels.

Sample Ladder

Three samples are tested. The observed core is the main demo headline: 209 retained demand rows, 21 states, restricted to years with observed NDCP prices and strong labor support. A stricter internal review can still flag it as fragile without removing it from the proof-of-concept site.

Broad Complete

228

22 states, 2008–2022 (unbalanced panel)

LOO state R²: −0.077 LOO year R²: −1.920

Observed Core

209

21 states, 2008–2022 (headline, unbalanced)

LOO state R²: −0.085 LOO year R²: −1.621 Real bootstrap accept: 92%

Low-Impute

55

13 states, 2014–2022 (unbalanced panel)

LOO state R²: −0.396 LOO year R²: −0.164

Observed Price Support Window

20082022

Headline sample 2014–2022

NDCP observed price support: 2008–2022. All 556 canonical scenario rows stay within this window. No post-2022 nowcasts.

Specification Comparison

The demand elasticity is the most consequential parameter in the model. It determines how much price rises when outsourced care enters the market. The checked-in variants all use the same current instrument: wages available outside childcare. They differ mainly in which control variables are included. A separate births-based sensitivity was tested and set aside because it produced the wrong sign and weak first-stage fit in the headline sample.

Specification	Elasticity	LOO State R²	LOO Year R²	Status
Household controls	−0.143	−0.085	−1.621	canonical
Instrument only	−0.139	−0.034	+0.003	sensitivity
Labor controls	−0.137	−0.054	−0.505	sensitivity
Full controls	−0.489	−0.114	−2.705	quarantined

LOO R² (leave-one-out R-squared) measures out-of-sample prediction: each state or year is held out, the model is re-estimated on the rest, and R² is computed on the held-out data. Negative values mean the model predicts worse than a simple mean—confirming this is demonstration-grade, not a high-powered empirical design.

Quarantined specifications produce economically inadmissible demand responses. The instrument-only version leaves the wage-based instrument to absorb confounders it cannot plausibly handle on its own. The full-controls version adds median income, which likely soaks up the price signal rather than clarifying it. Both are kept for diagnostic comparison but excluded from the headline. The canonical version adjusts only for parent employment and single-parent share. A births-based sensitivity was also set aside because it implied a positive elasticity (+0.150) with weak first-stage fit (R² = 0.148).

How demand is estimated

▶

The model does not take the raw correlation between high childcare prices and low unpaid care at face value, because expensive states may differ in many other ways. Instead it uses an instrumental-variables design at the state-year level.

In the first step, the model asks whether wages available outside childcare predict childcare prices after adjusting for parent employment and single-parent prevalence. In the second step, it uses only the price variation explained by that outside wage measure to estimate how unpaid childcare hours respond to price.

Survey weights are used when the time-use data provide them; otherwise rows count equally. The reported demand elasticity converts the estimated price coefficient into a percent response at the weighted average price and the weighted average unpaid-care level.

For a result to be eligible as a headline candidate, the public pipeline also requires basic support and sign checks: at least 100 observations, at least 15 states, at least 6 years, a first-stage F-statistic of at least 10, and a demand effect that does not turn positive.

What the current IV evidence can and cannot support

▶

The project has a usable proof of concept, but not a strong causal design yet. On the demand side, the headline still relies on the outside-wage instrument, the leave-one-out diagnostics are negative, the price data are county-based while the causal core is state-based, and the headline fit keeps only 209 rows in the headline sample.

Passing the internal admissibility rules therefore means the estimate is coherent enough to display, not that it is ready to treat as a durable national causal parameter.

On the supply side, the headline supply elasticity is not instrument-based at all. It is a reduced-form price-versus-provider gradient. The broader licensing-based supply instrument is kept as diagnostics only: its full panel has a first-stage F-statistic of 1.24 across 41 states and 5,013 county-year observations.

A sharper four-state pilot is much stronger on paper, with a first-stage F-statistic of 83.3, but it still identifies only a small treated-state setting. It is informative as a local design, not a national replacement for the current headline.

Evidence Quality & Decomposition Layers

This section is an audit layer: it tracks support quality and decomposition credibility so readers can interpret the canonical pooled headline, not replace it.

Beyond the pooled headline estimates, the project maintains an evidence-quality layer that tracks how much underlying data rests on direct administrative evidence versus inferred, proxy, or fallback sources.

IV Designs Tried

The project has tried instrumental-variables designs on both the demand and supply sides. Some are useful as diagnostics or local demonstrations; none currently replace the public headline supply curve, and only one remains in the headline demand workflow.

Use case	Instrument	Main source stack	Status	Why it is or is not used
Demand headline	Wages available outside childcare	BLS OEWS, NDCP, ATUS, ACS	used with caveats	It passes the project’s admissibility checks, but leave-one-out fit is still weak and the geography is mismatched.
Demand sensitivity	State births	CDC WONDER, NDCP, ATUS, ACS	not used	It implied the wrong sign for demand and fit weakly, so it was quarantined.
Supply full panel	Licensing shock exposure	State licensing rules, ICPSR archives, NDCP, CBP, NES, ACS	diagnostics only	The first stage is weak and the resulting supply estimates are not headline-usable.
Supply pilot	Licensing shock exposure in a four-state reform setting	State licensing reforms, NDCP, CBP, NES, ACS	local demo only	The design is sharper locally, but it identifies only a small treated-state setting rather than a national parameter.

Demand IV details: what was tried and why only one version remains

▶

Current headline demand design. The instrument is a state-year measure of wages available outside childcare. In practice that comes from BLS occupational wage data for non-childcare jobs, with county and state childcare wage information used to align and fill the series where needed. The endogenous price is the state childcare price index built from county childcare prices. The outcome is unpaid childcare time from the American Time Use Survey.

The current headline control set is intentionally lean: parent employment rates and single-parent prevalence. Those come from state-year demographic controls. Alternative versions also tried adding unemployment and median income.

The main source stack is: county childcare prices from the National Database of Childcare Prices; unpaid childcare time and weights from ATUS; state demographic controls from the American Community Survey; and wage instruments from BLS occupational wage data. Even in this preferred version, the design remains fragile because prices start at the county level while the causal core is estimated at the state-year level, and held-out fit remains negative.

Births-based demand sensitivity. The project also tested state-year birth counts as an alternative instrument, using CDC WONDER natality data. It used the same childcare price measure, the same unpaid-care outcome, and the same control framework. That version was not used because it produced a positive demand elasticity of +0.150, had weak fit in the first stage (R² = 0.148), and performed worse in held-out checks. In plain terms: it did not behave like a credible demand instrument in this sample.

Supply IV details: what the licensing designs use and why they are not headline estimates

▶

Full multi-state licensing design. The supply-side instrument is not a simple single variable. It combines two things: a county’s pre-existing exposure to center-based providers, and the size of state-year licensing changes that make labor requirements tighter or looser. Intuitively, places that were already more exposed to center care should react more when center staffing rules change.

The price variable is county childcare price, and the main supply outcomes are provider density and employer-establishment density. The design uses county and state-year fixed effects so it relies on within-county changes around state-year regulatory shocks. The source stack is: county prices from NDCP; county employer establishments from County Business Patterns; county nonemployer firms from Nonemployer Statistics; county under-5 population from ACS for density scaling; and harmonized state licensing rules from state regulatory documents and ICPSR licensing-study archives.

This full-panel version is diagnostics only because the first stage is weak: F = 1.24 across 41 states and 5,013 county-year observations. Neither supply outcome yields a headline-usable estimate, so the public site does not use this IV for the canonical supply elasticity.

Four-state pilot. The project also keeps a sharper local pilot built around a smaller set of real licensing reforms. The treated reforms are Virginia’s 2018 group-size caps, Montana’s 2018 group-size guidance, and Louisiana’s post-2017 group-size requirements; South Carolina stays in the panel as overlap support. This pilot is stronger on paper, with a first-stage F-statistic of 83.3, but it still rests on only 3 treated states and 4 state clusters. That makes it useful as a local proof of concept, not as a national replacement for the main supply curve.

CCDF Support Quality

Support Bucket	Rows	Share	Headline Eligible?
Explicit administrative	7	4.3%	Yes
Inferred from related fields	100	61.3%	Yes
Retained proxy	9	5.5%	No
Downgraded proxy	47	28.8%	No

107 rows (explicit + inferred) are eligible for headline-grade CCDF decomposition claims. 56 rows (retained + downgraded proxy) are excluded from headline use and retained only for diagnostics.

Licensing Harmonization & Causal Diagnostics

▶

State childcare licensing rules are harmonized from raw regulatory source documents into a structured panel: 4,692 harmonized rule rows from 16,338 raw audit rows, covering center-based, family home, and large group home provider types across all 50 states and DC.

Licensing IV status: diagnostics only. The full-panel multi-state licensing IV produces weak first-stage results (F-statistic = 1.24 across 41 states, 5,013 county-year observations). The 4-state pilot documented in the README uses a more concentrated identification strategy (3 treated states, F = 83.3).

Segmented Childcare Decomposition

▶

The segmented decomposition separates childcare quantities into three channels:

Public-admin care — directly provided or administered through public programs (Head Start, state pre-K)
Subsidized-private care — privately provided but publicly subsidized through CCDF vouchers or contracts
Unsubsidized-private care — privately provided and privately paid

These outputs are additive only. They do not replace the canonical pooled estimates. 139 segmented state-year rows are available for scenario decomposition alongside the pooled benchmark.

Release Contract

▶

The release package is a self-contained bundle with a machine-readable contract specifying which artifacts are headline-safe, which are diagnostics-only, and what publication rules apply.

Only explicit and inferred CCDF support rows are headline-eligible for decomposition claims.
Licensing IV outputs are diagnostics-only in their current form.
Segmented outputs are additive-only and should not be presented as the canonical marketization result.

The core pipeline is rebuildable from source data with no manual steps. Three optional inputs enrich the release bundle but require manual download: CCDF policy workbooks from ACF and ICPSR Child Care Licensing Study archives. The machine-readable manual_requirements output documents exactly which files are needed.

Home Maintenance

Preview

A secondary module estimates the cost of outsourcing unpaid home-maintenance labor using AHS housing survey data, NOAA weather-exposure controls, and the same ATUS time-use framework. The pipeline builds a CBSA-wave panel of outsourced home-maintenance prices and fits a DIY-vs-outsourced switching model.

This module is functional but not yet featured in the headline outputs. It demonstrates the framework’s generality beyond childcare.

Limitations

Interpretation boundaries for the demo.

The satellite account is still partial-equilibrium.

Even the preferred benchmark is an accounting construct, not a GE valuation. It says what unpaid childcare would be worth at a current marginal replacement price, not what would happen if all unpaid childcare actually moved into the paid market.

Causal identification is weak.

The current headline demand estimate still relies on wages available outside childcare as the instrument, and the demand side remains statistically weak. A births-based sensitivity was tested but is not used because it becomes economically inadmissible in the headline sample. Negative leave-one-out diagnostics confirm that the fit does not generalize well out of sample. This is the primary limitation.

Geography mismatch.

Prices are observed at the county level (NDCP), but the causal core is estimated at the state level (ATUS). County-to-state aggregation loses real variation.

Constant-elasticity assumption.

The main solver assumes one constant percentage response for supply and one for demand. The piecewise sensitivity relaxes this with a single kink, but real markets can still have additional capacity cliffs, regime changes, and local bottlenecks.

Price decomposition is benchmark-driven.

The direct-care / non-direct-care split depends on source-backed staffing and compensation benchmarks; it is not literal cost accounting estimated from repo data.

Temporal coverage.

NDCP prices end in 2022. The project does not extrapolate.

Sample size.

209 retained observed-core demand rows are sufficient for demonstration, not robust inference. The site intentionally keeps that demo visible and labels the weakness instead of hiding the output.

No behavioral response.

Scenarios assume families and providers respond only through price—no quality changes, informal-care substitution, parental labor-supply responses, or policy feedback.

Evidence-quality caveats.

107 of 163 CCDF rows are headline-eligible; 56 use proxy evidence. Licensing IV is diagnostics-only (F = 1.24). Segmented outputs complement but do not replace the canonical pooled model.

What the numbers do not mean

▶

The gross market price is not “what it costs to raise a child.” It is the annual price of a childcare slot.
The direct-care-equivalent price is not a verified cost-accounting split.
The marketization prices are not forecasts of what would happen if policy changed.
The demand elasticity is not a strong causal estimate in this project.

How to Read This Demo

Benchmark

Short-run canonical

Medium-run sensitivity

Diagnostics / audit