Technical writing
Census SAIPE: The Small Area Income and Poverty Estimates Behind Federal Education Funding and County-Level Poverty Maps
The Small Area Income and Poverty Estimates program is the Census Bureau's annual release of poverty rates and median household income for every county, every state, and every school district in the United States. Unlike the Current Population Survey, which cannot produce statistically reliable county-level estimates, SAIPE uses model-based methods to generate single-year poverty figures for all 3,100-plus counties and 13,000-plus school districts — figures that directly govern how the federal government allocates roughly $17 billion per year in Title I-A education grants and $3.5 billion in Community Development Block Grants.
What SAIPE Is and Why It Exists
Congress enacted Title I of the Elementary and Secondary Education Act in 1965 to direct federal education dollars toward schools serving children from low-income families. The formula requires a poverty count for every school district in the country — a level of geographic detail that the flagship federal household surveys simply cannot deliver. The CPS requires three-year averages even for state-level estimates and cannot produce reliable county-level figures for small counties. The American Community Survey's five-year estimates cover counties and school districts but average across five calendar years, making them too stale for annual funding cycles. No direct survey could affordably interview enough households in every one of the country's 13,000-plus school districts each year.
SAIPE was developed in the 1990s precisely to fill this gap. The program's core insight is that administrative data — tax filings, SNAP enrollment records, population estimates — can be combined with survey data in a statistical model that produces annual poverty estimates with quantified uncertainty for geographies far too small to survey directly. The result is a dataset that is both timely and geographically granular enough to serve as the official basis for federal funding formulas.
SAIPE estimates are published each December for the prior calendar year. The 2024 SAIPE release, for example, appears in December 2025. This one-year lag reflects the time required to assemble administrative data from the IRS and SNAP agencies, run the statistical models, validate the estimates, and prepare the public release. The estimates cover the fifty states, the District of Columbia, all 3,100-plus counties and county equivalents, and all school districts for which geographic boundaries can be matched to administrative records.
Model-Based Methodology
SAIPE estimates are not survey estimates — they are model-based estimates that combine multiple data sources through a statistical framework called small area estimation. For county-level poverty, the Census Bureau fits a state-space model using a set of predictors drawn from administrative and survey sources:
- ACS microdata — one-year or five-year American Community Survey estimates provide the primary survey-based poverty signal for counties. For large counties with sufficient ACS sample, direct ACS estimates carry significant weight. For smaller counties the model relies more heavily on administrative predictors.
- IRS tax return data — counts of tax filers with income below a threshold who claimed the Earned Income Tax Credit, tabulated at the county level from the Statistics of Income program. EITC-eligible filers are a direct administrative proxy for low-income households, and the IRS tabulations cover nearly the entire population of tax filers, making them a high-coverage predictor.
- SNAP participation counts — monthly administrative records of households receiving Supplemental Nutrition Assistance Program benefits, aggregated to the county. SNAP is broadly targeted to households below 130 percent of the poverty line, and county-level SNAP participation is among the strongest administrative predictors of poverty.
- Intercensal population estimates — Census Bureau population estimates by age and county, used to convert poverty counts to rates and to adjust for population composition changes between decennial censuses.
- CPS state-level poverty estimates — the CPS Annual Social and Economic Supplement produces reliable poverty estimates at the state level, and these enter the county model as a constraint that ensures county estimates within each state aggregate to values consistent with the CPS state estimate.
The model produces point estimates and associated standard errors for each county. The standard errors are published alongside the estimates so that users can assess uncertainty, particularly for small counties where the model relies heavily on administrative predictors and the estimates carry larger uncertainty. For school district poverty, the estimation problem is harder: school district boundaries often do not align with county lines, so the Census Bureau allocates county-level model estimates to school districts using a proportional method based on the geographic overlap between district boundaries and the counties whose estimates drive the allocation.
Compared to the ACS five-year estimates, SAIPE has two key advantages for annual funding applications: it produces single-year estimates rather than rolling five-year averages, and the model-based approach produces estimates for all counties regardless of sample size. The tradeoff is that SAIPE estimates reflect modeling assumptions and administrative data coverage patterns rather than direct household interviews. In counties where SNAP enrollment is systematically low relative to poverty — due to administrative barriers or cultural factors — SAIPE may underestimate poverty.
Key Metrics the Program Produces
The SAIPE release provides several distinct measures at each geographic level. At the county and state level, SAIPE produces poverty rate and count for the total population and for children under 18 separately, along with median household income. At the school district level, SAIPE provides the number and percentage of school-age children aged 5 to 17 in poverty — the specific metric that drives the Title I-A funding formula.
The 2022 SAIPE release illustrated the geographic range of county-level poverty in the United States. The national child poverty rate stood at approximately 14.7 percent, representing roughly 10.7 million children. County child poverty rates ranged from under 5 percent in wealthy suburban counties in Virginia, Maryland, and New Jersey — concentrated around the DC and New York metro areas — to above 45 percent in a cluster of counties in the Mississippi Delta, the Rio Grande border region, and the Pine Ridge area of South Dakota. This more than nine-fold spread between low and high poverty counties underscores why geography-specific estimates matter: a national average obscures conditions that differ by an order of magnitude across counties.
Median household income by county follows a similar pattern. In 2022 county median household incomes ranged from under $30,000 in the lowest-income Appalachian and Delta counties to above $140,000 in high-income suburban counties around major metros. State median household incomes, also published by SAIPE, provide a model-based complement to the CPS ASEC state estimates; the two sources use different methodologies but generally track each other closely at the state level.
Title I-A Education Funding
The Elementary and Secondary Education Act Title I-A program is the largest source of federal funding flowing directly to local school districts, distributing approximately $17 billion per year. The allocation formula is built primarily on the SAIPE count of school-age children in poverty for each school district — specifically, children aged 5 to 17 whose family income falls below the poverty threshold. This count is the only official source of annual, district-level poverty data with the geographic specificity the formula requires.
The Title I-A formula is not a simple per-child allocation. Congress has built in holds-harmless provisions that limit how much a district's allocation can fall from one year to the next, regardless of how much its SAIPE poverty count changes. The formula also includes concentration grants that direct additional funds to districts where the count of poor children exceeds a threshold relative to total enrollment, recognizing that high-concentration poverty creates challenges beyond what a simple per-child formula captures. State minimum allocations protect small states from receiving tiny absolute dollar amounts even when their share of national poverty is small.
The practical consequence is that the SAIPE school district estimates carry enormous policy weight. If a district's estimated count of children in poverty increases by 1,000 in a given year, its Title I-A allocation typically increases by several hundred thousand to several million dollars, depending on its size and the national formula calculation for that year. State education agencies then sub-allocate state Title I-A funds to individual schools within districts based on school-level poverty indicators, typically free and reduced-price lunch eligibility or community eligibility provision enrollment. The SAIPE district estimate sets the total pot; school-level decisions happen below that level.
Because funding follows the SAIPE estimates, districts have a legitimate interest in the accuracy of those estimates for their boundaries. Districts that experience rapid demographic change — from new affordable housing development, economic shocks, or population mobility — may find that SAIPE estimates lag behind actual conditions. The Census Bureau accepts public comments on SAIPE methodology and periodically revises its models to address known biases, but the annual publication cycle means there is always some temporal gap between on-the-ground conditions and the estimates that determine funding.
Community Development Block Grants
The Department of Housing and Urban Development's Community Development Block Grant program distributes approximately $3.5 billion per year to more than 1,200 entitlement communities — cities with populations above 50,000 and urban counties that have opted into the entitlement program. The CDBG formula grants are calculated using a combination of Census-derived indicators, of which SAIPE county poverty counts are a central component.
The CDBG formula actually runs two separate calculations for each entitlement community and grants the higher of the two allocations. Formula A weights poverty count at 25 percent, population at 25 percent, and overcrowded housing units (persons per room above 1.0, from the ACS) at 50 percent. Formula B weights pre-1940 housing units (a proxy for housing stock age and deterioration) at 50 percent, poverty count growth lag (population growth relative to a national baseline) at 25 percent, and poverty count at 25 percent. The dual-formula structure reflects different theories of urban distress: Formula A emphasizes current poverty concentration while Formula B emphasizes older industrial cities with aging housing stock.
SAIPE enters both formulas through the poverty count variable. Because SAIPE provides annual single-year county estimates, the CDBG formula can be updated each year to reflect recent poverty conditions, rather than waiting for a new decennial census. This makes SAIPE's county-level accuracy directly consequential for how hundreds of millions of dollars in housing and community development funds are allocated across the country each fiscal year.
Data Access via the Census API
The Census Bureau publishes SAIPE data through its standard API at api.census.gov/data/timeseries/poverty/saipe. The endpoint supports queries by geography and by time year. The primary variables available are:
SAEPOVRTALL_PT— poverty rate for all ages (percent)SAEPOVALL_PT— estimated number of people of all ages in povertySAEPOVRT17_PT— poverty rate for children under 18 (percent)SAEPOV17_PT— estimated number of related children under 18 in povertySAEMHI_PT— estimated median household income in dollarsSAEPOV5_17_PT— estimated number of children aged 5 to 17 in poverty (the school-district level variable driving Title I-A)
Geographic levels are specified using the standard Census API geography codes. The for=state:* parameter returns all fifty states plus DC at summary level 040. The for=county:*&in=state:* combination returns all counties in all states at summary level 050. School district poverty data is at summary level 970 (unified school districts), 960 (secondary school districts), and 950 (elementary school districts), and is queried with for=school+district+(unified):*&in=state:*.
The time series coverage extends back to 1989 with selected years available before 2000 and full annual coverage from 2003 to the present. To query a specific year, pass the time=YYYY parameter. The Federal Reserve Economic Data database at fred.stlouisfed.org mirrors some SAIPE state-level series — including state-level median household income and poverty rates — under series identifiers following the pattern MHICA (median household income, California) and analogous codes for other states, making FRED a convenient starting point for state-level time series analysis without API key requirements.
Python Example: County Child Poverty from the SAIPE API
The following script fetches all county-level SAIPE estimates for a target year, identifies the fifty counties with the highest child poverty rates, summarizes total children in poverty by state, and computes the change in child poverty rate relative to a prior-year baseline. Replace YOUR_CENSUS_API_KEY with a key obtained at api.census.gov/data/key_signup.html. The script uses only requests and pandas.
import requests
import pandas as pd
# Fetch all county SAIPE estimates for the most recent year via the Census Bureau API.
# SAIPE endpoint: api.census.gov/data/timeseries/poverty/saipe
# Key variables:
# SAEPOVRTALL_PT - poverty rate, all ages
# SAEPOVALL_PT - poverty count, all ages
# SAEPOVRT17_PT - poverty rate, children under 18
# SAEPOV17_PT - poverty count, children under 18
# SAEMHI_PT - median household income
# NAME - county name
# STABREV - state abbreviation
# Geographic level 050 = county; use state=* to get all counties in all states.
API_KEY = "YOUR_CENSUS_API_KEY" # register at api.census.gov/data/key_signup.html
YEAR = "2023" # most recent SAIPE year at time of script
BASE = "https://api.census.gov/data/timeseries/poverty/saipe"
params = {
"get": "NAME,STABREV,SAEPOVRT17_PT,SAEPOV17_PT,SAEPOVRTALL_PT,SAEPOVALL_PT,SAEMHI_PT",
"for": "county:*",
"in": "state:*",
"time": YEAR,
"key": API_KEY,
}
resp = requests.get(BASE, params=params, timeout=90)
resp.raise_for_status()
raw = resp.json()
df = pd.DataFrame(raw[1:], columns=raw[0])
# Cast numeric columns
numeric_cols = [
"SAEPOVRT17_PT",
"SAEPOV17_PT",
"SAEPOVRTALL_PT",
"SAEPOVALL_PT",
"SAEMHI_PT",
]
for col in numeric_cols:
df[col] = pd.to_numeric(df[col], errors="coerce")
# Rename for readability
df = df.rename(columns={
"SAEPOVRT17_PT": "child_pov_rate",
"SAEPOV17_PT": "child_pov_count",
"SAEPOVRTALL_PT": "total_pov_rate",
"SAEPOVALL_PT": "total_pov_count",
"SAEMHI_PT": "median_hh_income",
})
# Drop any rows where child poverty rate is missing (territories, some jurisdictions)
df = df.dropna(subset=["child_pov_rate"])
# --- Bottom 50 counties by child poverty rate (highest child poverty) ---
bottom50 = (
df.sort_values("child_pov_rate", ascending=False)
.head(50)[["NAME", "STABREV", "child_pov_rate", "child_pov_count", "median_hh_income"]]
.reset_index(drop=True)
)
bottom50.index += 1
print("50 counties with highest child poverty rate (" + YEAR + " SAIPE):")
print(bottom50.to_string())
# --- State-level summary: mean child poverty rate weighted by child poverty count ---
state_summary = (
df.groupby("STABREV")
.agg(
total_child_poor=("child_pov_count", "sum"),
county_count=("child_pov_count", "count"),
median_child_pov_rate=("child_pov_rate", "median"),
)
.sort_values("total_child_poor", ascending=False)
.reset_index()
)
print("")
print("State summary: total children in poverty by state (" + YEAR + " SAIPE):")
print(state_summary.head(15).to_string(index=False))
# --- Compare to a prior year: compute change in child poverty rate by county ---
PRIOR_YEAR = "2019" # pre-pandemic baseline
params_prior = dict(params)
params_prior["time"] = PRIOR_YEAR
resp2 = requests.get(BASE, params=params_prior, timeout=90)
resp2.raise_for_status()
raw2 = resp2.json()
df2 = pd.DataFrame(raw2[1:], columns=raw2[0])
df2["SAEPOVRT17_PT"] = pd.to_numeric(df2["SAEPOVRT17_PT"], errors="coerce")
df2 = df2.rename(columns={"SAEPOVRT17_PT": "child_pov_rate_prior"})
# Merge on state + county FIPS
merged = pd.merge(
df[["state", "county", "NAME", "STABREV", "child_pov_rate"]],
df2[["state", "county", "child_pov_rate_prior"]],
on=["state", "county"],
how="inner",
)
merged["rate_change"] = (merged["child_pov_rate"] - merged["child_pov_rate_prior"]).round(2)
largest_increase = merged.nlargest(10, "rate_change")[
["NAME", "STABREV", "child_pov_rate_prior", "child_pov_rate", "rate_change"]
]
largest_decrease = merged.nsmallest(10, "rate_change")[
["NAME", "STABREV", "child_pov_rate_prior", "child_pov_rate", "rate_change"]
]
print("")
print("Counties with largest increase in child poverty rate (" + PRIOR_YEAR + " to " + YEAR + "):")
print(largest_increase.to_string(index=False))
print("")
print("Counties with largest decrease in child poverty rate:")
print(largest_decrease.to_string(index=False))
The script illustrates three common SAIPE workflows: a simple extract of the most recent year for all counties, an aggregation to the state level to compare total poverty burdens, and a year-over-year change analysis using the pre-pandemic baseline. Because SAIPE estimates carry model-based standard errors, production analyses should fetch the corresponding margin-of-error variables and account for uncertainty before drawing conclusions from small differences between counties or years.
Relationship to ACS and CPS
SAIPE occupies a distinct position in the federal poverty measurement landscape alongside the American Community Survey and the Current Population Survey. The three programs serve different purposes and should not be treated as interchangeable.
The CPS ASEC is the source for the official national poverty rate and is designed for national and state comparisons over time. It produces the most methodologically defensible poverty estimates for the nation as a whole and for large states, but its sample is too small to support reliable county-level estimates for most of the country. State-level CPS estimates require multi-year averaging for small-population states.
The ACS five-year estimates provide poverty data at the census-tract level and below, enabling neighborhood-scale analysis that neither SAIPE nor the CPS can support. However, the five-year averaging period means ACS estimates reflect conditions across a rolling window rather than a single calendar year. ACS and SAIPE poverty estimates for the same county and year often differ because they use different methodologies, income definitions, and reference populations; neither is simply more accurate than the other, and users should choose based on the analytic task at hand.
SAIPE is the right choice when the application requires annual single-year county or school district estimates with consistent methodology across all geographies — most importantly when working with Title I-A funding, CDBG formulas, or any federal program that uses SAIPE as its official data source. The program's model-based approach is a limitation for users who need pure survey estimates, but for federal funding applications it is the authoritative source by statute.
For national and state poverty measurement methodology, the official poverty thresholds, and the Supplemental Poverty Measure, see Census Current Population Survey: The Monthly Survey Behind Official US Poverty Rates and Income Inequality Measures, which covers the CPS ASEC, the Orshansky thresholds, and the EITC and SNAP poverty reduction estimates that SAIPE cannot capture at the sub-state level.
For neighborhood-level poverty data at the census-tract scale and the five-year rolling estimates that underpin local planning decisions, see Census ACS: The American Community Survey and the Federal Demographic Dataset Behind Every Policy Decision, which covers the ACS's tract-level access, the differences in poverty methodology between the ACS and SAIPE, and how the two datasets complement each other for community development analysis.