Census Current Population Survey: The Monthly Survey Behind Official US Poverty Rates and Income Inequality Measures

The Current Population Survey is the federal government's longest-running household survey and the primary source for three of the most-cited economic statistics in the United States: the monthly unemployment rate, the official annual poverty rate, and the Gini coefficient measuring income inequality. Conducted jointly by the Census Bureau and the Bureau of Labor Statistics since 1940, it interviews approximately 60,000 households every month and produces a body of longitudinal microdata that spans more than eight decades of American economic life.

Survey Design and History

The Current Population Survey began during the Depression-era crisis over unemployment measurement. Before 1940 no reliable national estimate of joblessness existed; the federal government estimated unemployment through indirect administrative proxies. President Roosevelt's Committee on Economic Security, working in parallel with the Works Progress Administration, commissioned a genuine household survey to count people directly. The first nationwide survey was fielded in 1940, and its fundamental architecture — asking individuals about their own employment status rather than relying on employer records — has remained intact ever since.

Responsibility for the CPS is formally shared. The Census Bureau designs the questionnaire, selects and maintains the sample, conducts the interviews, and processes the microdata. The Bureau of Labor Statistics funds the program, specifies the labor force definitions, and publishes the headline unemployment statistics. This joint stewardship means the CPS simultaneously serves two distinct policy-making communities: labor economists at BLS who need monthly labor market indicators and poverty researchers at Census who need annual income and benefit data.

The sample consists of approximately 60,000 occupied housing units each month. Addresses are drawn from the Census Bureau's Master Address File, which is built from decennial Census address canvassing and updated continuously through the American Community Survey and postal data partnerships. The sample is clustered into primary sampling units — counties or groups of counties — stratified to ensure geographic and demographic representativeness across all fifty states and the District of Columbia. Each responding household is interviewed by Computer-Assisted Telephone Interviewing (CATI) or, when telephone contact fails, by Computer-Assisted Personal Visit Interviewing (CAPI). Response rates have historically exceeded 85 percent, though they have declined somewhat in recent years, consistent with a broader trend across federal surveys.

The 4–8–4 Rotating Panel Design

The CPS uses a rotating panel structure that the Census Bureau describes as “4–8–4”: each housing unit is interviewed for four consecutive months, then left out of the sample for eight months, then returned for a final four consecutive months before permanently rotating out. The complete cycle spans sixteen months and means that at any given month roughly 75 percent of the sample overlaps with the previous month's sample and 50 percent overlaps with the sample from twelve months earlier.

This design is not arbitrary. The month-to-month overlap enables BLS to produce reliable estimates of month-to-month change in labor force status — including flows between employment, unemployment, and not-in-labor-force — which require matching individual records across interviews. The year-over-year overlap of exactly half the sample supports annual comparisons. Rotating households in and out rather than maintaining a permanent panel prevents respondent fatigue and the conditioning effects that distort longitudinal surveys when the same people are interviewed indefinitely.

A practical consequence of the rotation scheme is that CPS microdata users who want to construct matched panels — tracking the same individual across multiple months — must link records using household and person identifiers rather than a persistent survey identification number. IPUMS CPS provides matched-pair files for adjacent years that have already solved this linkage problem for researchers who need it.

The ASEC: March Supplement and Annual Data

The basic monthly CPS collects labor force data. Once a year, in February through April (with the bulk of fieldwork in March), the Census Bureau fields the Annual Social and Economic Supplement, formally designated the CPS ASEC. The supplement expands the sample to roughly 95,000 households and adds a detailed retrospective income module covering the prior calendar year. This is the supplement that produces the official poverty rate and median household income figure that the Census Bureau releases each September.

The ASEC collects fifteen separate income sources at the individual level: wages and salaries from each job held during the prior year; net self-employment income; farm income; Social Security and Railroad Retirement benefits; Supplemental Security Income; public assistance and welfare payments; interest, dividends, and net rental income; educational assistance; child support and alimony; financial assistance from outside the household; foster child payments; employer-provided pension and retirement income; disability payments; workers' compensation and veterans' benefits; and a residual other income category. This granular structure allows the Census Bureau to compute total money income by summing across all sources and to construct the poverty status determination that drives official poverty statistics.

The ASEC also collects data on health insurance coverage, the full range of government transfer programs received (including SNAP, housing subsidies, and Medicaid), and work experience during the prior year including weeks worked and usual hours per week. These additions make the ASEC the primary data source for the Supplemental Poverty Measure and for CBO's household income and federal tax burden analyses.

Monthly Labor Force Data and the Unemployment Rate

BLS uses the basic monthly CPS to compute the unemployment rate, labor force participation rate, and employment–population ratio published on the first Friday of each month in the Employment Situation Summary. These statistics are derived entirely from the household survey, not from employer reports. That distinction is fundamental.

The CPS classifies each civilian noninstitutional person aged 16 or older into one of three labor force categories. Employed persons include anyone who worked at least one hour for pay or profit during the survey reference week, anyone who worked 15 or more hours as an unpaid worker in a family business, and anyone who had a job but did not work because of illness, vacation, labor dispute, or other reasons. Unemployedpersons are those who were not employed, were available for work, and had actively looked for a job at some point in the four weeks preceding the survey week. Not in the labor force covers everyone else: retirees, students, persons with disabilities who are not seeking work, caregivers, and the “marginally attached” — people who want work and are available but have not searched recently.

The headline U-3 unemployment rate is unemployed persons divided by the labor force (employed plus unemployed). BLS publishes six alternative measures of labor underutilization, labeled U-1 through U-6. U-1 counts only persons unemployed fifteen weeks or longer. U-2 includes job losers and persons who completed temporary jobs. U-3 is the official rate. U-4 adds discouraged workers — people who want work but have given up searching because they believe no jobs are available — to the unemployed count. U-5 adds all “marginally attached” workers who want work and are available but have not searched recently. U-6 further adds persons working part time for economic reasons who would prefer full-time employment. In periods of labor market stress U-6 typically runs six to eight percentage points above U-3.

A persistent source of confusion in public discussion is the relationship between the CPS household survey and the Current Employment Statistics program, often called the payroll survey or establishment survey. Both programs measure employment, but they measure different things. The payroll survey counts jobs filled at employer establishments; the CPS counts employed persons. A person holding two jobs appears once in the CPS and twice in the payroll count. The payroll survey covers only wage and salary workers on nonfarm payrolls; the CPS includes agricultural workers, the self-employed, and unpaid family workers. The reference periods differ. The payroll survey is administratively based and subject to much smaller sampling error; the household survey is the only source of the unemployment rate and labor force participation rate. Month-to-month divergence between the two measures is normal and expected.

The Official Poverty Measure

The official poverty measure has its origins in work done by Mollie Orshansky at the Social Security Administration in the early 1960s. Orshansky observed that the average American family in the late 1950s spent approximately one-third of its after-tax income on food. The Department of Agriculture had developed Economy Food Plan budgets for families of various sizes — the minimum cost of a nutritionally adequate diet under emergency conditions. Orshansky multiplied these food budgets by three to derive income thresholds at which families could just afford minimal nutrition, with two-thirds of income left for everything else.

The Johnson Administration adopted these thresholds as the official poverty line in 1969, with a retroactive determination that the thresholds as computed for 1963 would become the baseline. Since then the thresholds have been updated annually by the Consumer Price Index for All Urban Consumers (CPI-U), the same price index used to adjust Social Security benefits. No other structural update has occurred. The food budget that anchored the original calculation has never been revised to reflect changed consumption patterns; the multiplier of three — reflecting 1950s food expenditure shares — has never been updated to reflect the fact that housing and health care now constitute a far larger share of household budgets than food.

For 2023 the Census Bureau's poverty thresholds ranged from $14,580 for a single individual under age 65 to $30,900 for a family of four with two related children under 18. The thresholds vary by family size, number of children, and whether the family includes an elderly person, but they do not vary by geography: the same threshold applies in rural Mississippi as in San Francisco. Income is measured as total money income — the sum of all pre-tax cash income sources collected in the ASEC — and compared against the threshold for the family unit. A family is counted as poor if its total money income falls below its applicable threshold.

The official poverty rate for 2023 was approximately 11.1 percent, representing about 36.8 million people. Child poverty under the official measure was approximately 16 percent. These figures are released each September in the Census Bureau's annual Income and Poverty in the United States report, which draws entirely on ASEC microdata.

The Supplemental Poverty Measure

Recognition that the official poverty measure had fundamental methodological weaknesses led to decades of academic and policy debate. The National Academy of Sciences convened a panel in the early 1990s that published a comprehensive report in 1995 recommending a new approach. After further methodological development the Census Bureau released the first Supplemental Poverty Measure estimates in 2011, with the explicit understanding that the SPM supplements rather than replaces the official measure.

The SPM differs from the official measure in four principal ways. First, the threshold is based on observed consumer expenditures on food, clothing, shelter, and utilities (the FCSU bundle) for families in the second quintile of expenditure, using Consumer Expenditure Survey data, rather than on a fixed food budget multiplied by three. The threshold is updated as expenditure patterns change. Second, the resource measure is post-tax, post-transfer cash and near-cash income rather than pre-tax money income. Third, in-kind benefits — SNAP (food stamps), housing subsidies, the Low Income Home Energy Assistance Program, school lunch benefits — are added to resources because they affect families' actual consumption capacity. Fourth, necessary work expenses, child care costs, and medical out-of-pocket expenditures are subtracted from resources.

The SPM geographic adjustment is significant: thresholds vary by housing tenure and by state, reflecting differences in the cost of housing across the country. A family renting in high-cost metropolitan areas faces a higher SPM threshold than the same family in a low-cost rural county.

Because the SPM counts SNAP benefits and the Earned Income Tax Credit as resources, it is far more sensitive to policy changes than the official measure. When the American Rescue Plan expanded the Child Tax Credit in 2021, the SPM child poverty rate fell to a historic low of 5.2 percent. When the expanded credit expired at the end of 2021, the 2022 SPM child poverty rate rose to 12.4 percent — the largest single-year increase on record. The official measure showed no comparable spike because refundable tax credits are not counted as income. The 2023 overall SPM poverty rate was approximately 12.9 percent, higher than the 11.1 percent official rate primarily because the high cost of medical care creates out-of-pocket burdens that reduce measured SPM resources.

Income and Inequality Measurement

The ASEC is the source for the Census Bureau's annual measures of household income and income inequality. The headline figure — real median household income — was approximately $80,610 in 2023, measured in 2023 dollars. Median household income is the income level that divides households into equal upper and lower halves; it is preferred over mean income for distributional analysis because it is not distorted by extreme values at the top of the distribution.

The Census Bureau publishes income quintile shares: the share of total aggregate household income received by each fifth of households ranked by income. In 2023 the bottom quintile received approximately 3.4 percent of aggregate income and the top quintile received approximately 51.5 percent. The top 5 percent alone received about 23 percent of aggregate income. These quintile shares are based on money income before taxes and do not include capital gains, which primarily accrue to high-income households; the Congressional Budget Office produces distributional analyses that incorporate capital gains using the CPS as an underlying source.

The Gini coefficient is the Census Bureau's summary measure of income inequality. The Gini ranges from 0 (perfect equality, every household has the same income) to 1 (perfect inequality, one household has all income). The United States Gini for 2023 was approximately 0.482 — higher than most peer wealthy nations but consistent with the trend that has prevailed since the early 1970s. The Gini computed from pre-tax money income systematically understates redistribution; distributional analyses using post-tax post-transfer income (the SPM resource measure) produce a meaningfully lower Gini.

The P90/P10 ratio — the income at the 90th percentile divided by the income at the 10th percentile — and analogous ratios such as P90/P50 and P50/P10 provide complementary perspectives on inequality that are less sensitive to the top tail of the distribution than the Gini. These ratios are commonly computed directly from IPUMS CPS microdata using the INCTOT or FTOTVAL variables.

Effects of EITC and SNAP on Poverty

The EITC and SNAP are the two largest anti-poverty programs in the United States measured by their effect on the poverty rate. Because the official poverty measure does not count SNAP benefits or refundable tax credits as income, the official measure cannot capture these programs' poverty-reducing effects. The SPM is designed precisely to make these effects visible.

Census Bureau analyses of SPM data consistently find that SNAP lifts roughly 4 million people above the SPM poverty threshold annually. Refundable tax credits — the EITC and the refundable portion of the Child Tax Credit — lift approximately 5 to 6 million people above the threshold. Social Security remains the largest single program in terms of people lifted from poverty: its removal would raise the poverty rate for persons 65 and older from roughly 14 percent to over 50 percent under SPM calculations. Medicaid and CHIP reduce medical out-of-pocket expenses and thus increase SPM resources, though estimating their poverty effect requires imputing a dollar value to insurance coverage, which the Census Bureau does using a Health Insurance Value approach.

IPUMS CPS: Harmonized Historical Microdata

The Integrated Public Use Microdata Series CPS project, housed at the Minnesota Population Center at the University of Minnesota, is the primary scholarly infrastructure for longitudinal CPS analysis. IPUMS CPS harmonizes all CPS basic monthly and ASEC waves back to 1962 into a single integrated database with consistent variable coding, uniform labeling, and a common documentation system. As of 2025 IPUMS CPS provides over 200 variables covering labor force status, income, poverty, demographics, and health insurance across more than sixty years of survey data.

The harmonization problem is substantial. Variable names, coding schemes, question wording, and the precise definitions of labor force concepts have changed dozens of times since 1962. IPUMS CPS resolves these changes by recoding historical data to consistent schemes where feasible and by documenting inconsistencies where recoding would create false comparability. The result is a dataset that allows researchers to trace trends in median income, poverty rates, labor force participation, and industry composition from the Kennedy era to the present using a single consistent extract.

Key IPUMS CPS variables for income and poverty research include: FTOTVAL (total pre-tax family income from all sources), INCTOT (individual total pre-tax income), POVERTY (the ratio of family income to the applicable poverty threshold, expressed as a percentage, where values below 100 indicate poverty), and OFFPOV (the official poverty status flag). The POVERTY ratio variable is particularly useful for near-poverty analysis — identifying families between 100 and 200 percent of the poverty threshold who do not count as officially poor but remain economically precarious.

IPUMS CPS data is accessed through the IPUMS website at ipums.org, where users submit extract requests specifying variables, samples, and case selections. The ipumsr R package provides a full API for working with IPUMS extracts in R; a Python equivalent is available as the ipumspy package. Both packages handle the fixed-width format, variable labels, and value labels that IPUMS provides and that are absent from raw Census Bureau microdata files.

Data Access

The Census Bureau provides multiple access points for CPS data depending on the user's technical sophistication and analytical goals. The Census API at api.census.gov serves pre-tabulated CPS ASEC statistics — poverty rates, income medians, and employment counts by state, age group, race, and sex — without requiring users to work with microdata. The endpoint structure for CPS ASEC data follows the patternhttps://api.census.gov/data/<year>/cps/asec/mar, and available variables and their codes can be explored through the Census API variables documentation page for each year.

The Census Bureau's CPS Tables tool at census.gov provides a point-and-click interface for generating standard poverty and income tables in Excel or CSV format. DataFerrett, a browser-based tool, allows custom cross-tabulations of CPS microdata without downloading raw files, though its interface has not kept pace with modern data tools and most researchers prefer direct microdata access.

For basic monthly CPS data, the Census Bureau releases public-use microdata files in fixed-width format each month. These files contain the individual-level labor force records needed to reproduce BLS's unemployment estimates and to conduct custom labor market analyses. BLS publishes a data dictionary and technical documentation (the CPS Technical Paper) that specifies every field, its width, and its code values.

The Federal Reserve Economic Data (FRED) database at fred.stlouisfed.org hosts long historical series derived from CPS outputs, including UNRATE (seasonally adjusted unemployment rate), MEHOINUSA672N (real median household income in 2023 dollars), and dozens of demographic-specific unemployment and participation rate series. FRED provides a clean JSON API and a Python library (fredapi) that make it the simplest access point for historical time series that do not require microdata.

Python Example: State Poverty Rates from the CPS ASEC

The following script queries the Census Bureau API for state-level poverty rates from two consecutive CPS ASEC releases, merges the results, computes year-over-year changes, and ranks states by both rate and change. Replace YOUR_CENSUS_API_KEY with a key obtained at api.census.gov/data/key_signup.html.

import requests
import pandas as pd

# Pull state-level poverty rates from the CPS ASEC via the Census Bureau API.
# The CPS ASEC variables are published under the "cps/asec/mar" endpoint.
# Variable S0901_C02_001E is the poverty rate for persons (all ages).
# We fetch the most recent available year and compute year-over-year change.

API_KEY = "YOUR_CENSUS_API_KEY"  # register at api.census.gov/data/key_signup.html

CURRENT_YEAR = "2023"
PRIOR_YEAR   = "2022"

BASE = "https://api.census.gov/data/{year}/cps/asec/mar"

# Variables:
#   POVRATE  - official poverty rate, percent
#   GESTFIPS - state FIPS code (geographic identifier)
#   NAME     - state name
# Note: CPS ASEC poverty data is also available via the "cps/asec/mar" profile tables.
# For direct microdata access use IPUMS CPS. For published state poverty rates
# the pre-tabulated endpoint below provides the official Census release figures.

# --- Fetch current year ---
params_current = {
    "get": "NAME,POVRATE",
    "for": "state:*",
    "key": API_KEY,
}
url_current = BASE.format(year=CURRENT_YEAR)
resp = requests.get(url_current, params=params_current, timeout=60)
resp.raise_for_status()
data_current = resp.json()
df_current = pd.DataFrame(data_current[1:], columns=data_current[0])
df_current = df_current.rename(columns={"POVRATE": "pov_rate_current"})
df_current["pov_rate_current"] = pd.to_numeric(df_current["pov_rate_current"], errors="coerce")

# --- Fetch prior year ---
url_prior = BASE.format(year=PRIOR_YEAR)
resp2 = requests.get(url_prior, params=params_current, timeout=60)
resp2.raise_for_status()
data_prior = resp2.json()
df_prior = pd.DataFrame(data_prior[1:], columns=data_prior[0])
df_prior = df_prior.rename(columns={"POVRATE": "pov_rate_prior"})
df_prior["pov_rate_prior"] = pd.to_numeric(df_prior["pov_rate_prior"], errors="coerce")

# --- Merge and compute year-over-year change ---
df = pd.merge(df_current, df_prior[["state", "pov_rate_prior"]], on="state")
df["yoy_change"] = (df["pov_rate_current"] - df["pov_rate_prior"]).round(2)

# --- Rank states by poverty rate ---
df = df.sort_values("pov_rate_current", ascending=False).reset_index(drop=True)
df.index += 1  # rank starts at 1

print("State poverty rates (" + CURRENT_YEAR + " CPS ASEC), ranked highest to lowest:")
print(df[["NAME", "pov_rate_current", "pov_rate_prior", "yoy_change"]].to_string())

# --- Identify largest improvements and deteriorations ---
improved = df.nsmallest(5, "yoy_change")[["NAME", "yoy_change"]]
worsened = df.nlargest(5, "yoy_change")[["NAME", "yoy_change"]]

print("")
print("Five states with largest poverty rate decrease (" + PRIOR_YEAR + " to " + CURRENT_YEAR + "):")
print(improved.to_string(index=False))

print("")
print("Five states with largest poverty rate increase:")
print(worsened.to_string(index=False))

The script demonstrates the core workflow for CPS ASEC API queries: specify the year in the URL path, request the POVRATE variable and the state geographic filter, then merge annual pulls to compute changes over time. State-level CPS poverty rates carry larger margins of error than the national estimate because the state-level sample is smaller; for trend analysis, three-year rolling averages are more reliable than single-year comparisons.

Relationship to Other Federal Data Programs

The CPS sits at the center of a network of complementary federal surveys that together constitute the infrastructure for measuring economic conditions. The American Community Survey produces income and poverty statistics at the census-tract level with far greater geographic granularity than the CPS, but the ACS is a different instrument with different methodology and the two programs' poverty estimates are not directly comparable. The primary structural difference is that the ACS uses a different poverty concept and income measure than the CPS ASEC, and the ACS does not support the Supplemental Poverty Measure calculation.

The Survey of Income and Program Participation (SIPP), also administered by the Census Bureau, provides a longitudinal panel that tracks income dynamics, program participation, and asset holdings at shorter intervals than the annual ASEC. SIPP is better suited to studying transitions into and out of poverty and program participation spells; the CPS ASEC is better suited to annual national and state estimates and to the long historical trend series enabled by IPUMS CPS.

The Panel Study of Income Dynamics at the University of Michigan, begun in 1968, provides a true longitudinal household panel tracking families across generations. Its smaller sample (about 18,000 individuals) limits geographic analysis, but its multigenerational linkages enable intergenerational mobility research that the CPS cannot support.

Limitations and Methodological Debates

The CPS ASEC's income measurement has well-documented limitations. High-income households are systematically underrepresented due to survey nonresponse, topcoding of income variables in the public use files (reported income is capped at a maximum value to prevent identification of respondents), and the fact that high-income individuals disproportionately decline to participate. Administrative data comparisons, most notably work by economists using IRS tax records linked to the CPS, consistently find that the CPS understates income in the top decile.

Underreporting of transfer income is a separate problem. The Urban Institute's research on “administrative undercount” finds that households receiving SNAP, TANF, and Medicaid report those benefits in the CPS at rates well below what administrative enrollment records indicate. Because the SPM relies on self-reported benefit receipt, SPM poverty rates are biased downward — the measure appears to show more poverty reduction from programs than actually occurs when administrative coverage is taken as the benchmark.

The 2014 redesign of the ASEC income questions introduced a split-sample test and then a full redesign intended to improve income reporting comprehensiveness. The redesign changed how multiple income sources within the same category are collected and introduced a minimum-income-level probe. The Census Bureau publishes bridging estimates for 2013–2014 that allow researchers to assess the redesign's effect, but the break is real: long-run income trend analysis should account for the discontinuity.

For neighborhood-level income and poverty data with census-tract granularity, see Census ACS: The American Community Survey and the Federal Demographic Dataset Behind Every Policy Decision, which covers the ACS's five-year estimates, tract-level access, and the differences in poverty methodology between the ACS and the CPS ASEC.

For the employer-side employment and payroll data that complements the CPS household survey, see BLS Current Employment Statistics: The Monthly Jobs Report Behind Every Payroll Number, which covers the Establishment Survey, seasonal adjustment, benchmark revisions, and average hourly earnings.