Our Methodology

Population Estimation

Data Sources

• 2017-2021 American Community Survey (ACS) 5-year detailed estimates table B17024
• 2017-2021 American Community Survey (ACS) 5-year detailed estimates table B23008
• 2017-2021 ACS Public Use Microdata Sample (PUMS) household and person-level files

Estimation Methodology 

Generate local child counts separately for household income levels and employment categories.

The 5-year ACS summary data tables B17024 and B23008 provide estimates of total children by age group by household income and employment, respectively. 3Si sourced these estimates at the census tract level. The two tables differ in terms of the age groups for which summary counts of children are reported. We process the data from these tables to create two separate tract-level tables of child population by age group and income bracket (based on ranges defined by percentages of FPL) and age group and percentage distribution by parent employment status, respectively.

Determine correlation between household income and employment at the state level.

Since subsidy eligibility is determined jointly by income and employment requirements, we use the household- and person-level PUMS tables to obtain unemployment counts by income level for households containing children ages 0-5 years. We use PUMS household weight variables to translate household-level survey data to aggregated state-level child counts by income bracket (as a percentage of the Federal Poverty Level or FPL) and parent employment status. With this, we determine the correlation between income bracket and parent employment status at the state level.

Adjust demographics of local child counts based on statewide income and employment correlation.

We apply the correlation between income (as a percentage of FPL) and parent employment determined from the 5-year PUMS data to merge separate aggregate counts by child age and income (derived from B17024) and age and parent employment (derived from B23008). For this, we employ an iterative goal-seek algorithm to preserve the overall counts of age and household income and the percent distribution of children by age and parent employment status from the source ACS summary tables while also matching the proportionate distribution of income levels and employment status observed in the PUMS data. The outcome from this estimation step are tract-level child counts in distinct, non-overlapping categories defined jointly by age group, household income and employment.

Create single-year age categories and estimate the counts of children in every age category in each census tract.

The estimation steps outlined above result in data on the counts of children by census tract, ACS age group, household income, and parent employment status. We then evenly pro-rate this population of children into single-year ages (for instance, by dividing the age group “Under 6” by 6). In all cases in which the total count of children in an ACS age group is not evenly divisible into single-year ages, we randomly assign the remainder (a number that ranges from 1 to 5) to single-year ages using a uniform probability distribution for the single-year age categories (i.e., the all single-year age categories have an equal probability of having a child assigned to them).

Create population table with a unique record for every child and assign addresses within the census tract to each.

We then use the aggregate estimates from the steps described above, to create a disaggregated, child-level table for every child in the population, including the age (by single-year age category), household income, parent employment status, and census tract for every child. We then assign an address to every child within the relevant census tract. For this, we leverage a consolidated address base of geocoded residential addresses sourced predominantly (but not exclusively) from the National Address Database and openaddresses.io.

Reverse geocode child addresses to estimate counts of children at different geographic levels (county-level for CUSP Public).

Finally, we reverse geocode the addresses in the child population base to determine the geographies they are associated with (for example, census tract, zip code tabulation area or ZCTA, county, township, legislative districts, school districts, etc.). The inherent flexibility of the CUSP Population Base model allows aggregation to any geographic unit for which reliable shape formats are available. Our approach of modeling to the child-level and aggregating up to different geographic units is designed to preserve counts by census tract and county. CUSP Public population data is available at the county level.

Convert ACS income brackets to continuous incomes to allow for flexible modeling.

In the next step, we model the ACS income brackets (in the form of ranges of incomes as percentages of FPL) as continuous incomes so that each child in the CUSP Population Base has a household income value (in addition to income bracket). For this, we leverage granular income data available in PUMS data to model a probability density function (pdf) of income values in every PUMA. We then use this probability function to develop multiples of FPL (as a continuous rather than discrete variable) that can be used to generate income values for every child while preserving the counts of children within every ACS income bracket. The modeling of income as a continuous variable rather than a categorical variable is crucial for flexible estimation of child counts by customizable income bins.

Translate FPL-based child-level income estimates to SMI to inform eligibility modeling.

Many subsidized early learning programs determine income-based eligibility using State Median Income (SMI) rather than FPL. To translate CUSP Public’s FPL-based income estimates into SMI-based estimates, we calculate the ratio between each state’s SMI and FPL, each for a family of four. We use SMI statistics from the same year as the rest of our model’s data. We multiply each child’s FPL-based income estimate by the SMI/FPL ratio, creating that child’s SMI-based income estimate.

Eligibility Requirements

Head Start

Children are eligible for Head Start starting at the age of 3 years to kindergarten if their family incomes are at or below the Federal Poverty Level.

Early Head Start

Children are eligible for Early Head Start starting at birth to 3 years old if their family incomes are at or below the Federal Poverty Level.

Child Care and Development Fund (CCDF)

The table below summarizes the Child Care and Development Fund (CCDF) eligibility requirements related to household income and employment status for children in all states in the US.