Current Population Survey
The Current Population Survey (CPS) is a monthly household survey conducted by the U.S. Census Bureau for the Bureau of Labor Statistics (BLS) to produce official estimates of employment, unemployment, labor force participation, earnings, and related demographic characteristics for the civilian noninstitutional population.[1][2] The survey targets a scientifically selected multistage probability sample of approximately 60,000 eligible housing units across all states and the District of Columbia, with households interviewed for four consecutive months, rotated out for eight months, and then re-interviewed for four more months to enable both current and repeated measurements.[3][4] Originating in the early 1940s as a response to the need for direct, sample-based unemployment measurement amid the Great Depression and wartime labor shifts, the CPS has served as the foundational source for U.S. government labor force statistics since its formal monthly implementation in 1942, evolving through methodological refinements to enhance accuracy and coverage while maintaining long-term data comparability.[5][6] Its outputs, including the widely tracked unemployment rate (U-3 measure), underpin economic indicators, policy decisions, and business cycle analysis, supplemented annually with modules on income, poverty, health insurance coverage, school enrollment, and voter turnout to broaden insights into social and economic trends.[7][8] Key defining features include its reliance on household respondents for self-reported data on work status, job search activities, and hours worked, which—while enabling detailed breakdowns by age, sex, race, education, and occupation—introduce challenges such as sampling variability, nonresponse bias, and definitional constraints that exclude certain marginally attached individuals from unemployment counts, prompting BLS to publish broader alternative measures like U-6 for a fuller picture of labor underutilization.[8][9] Ongoing modernization efforts, including sample redesigns and integration of administrative data, aim to address coverage gaps and improve precision amid demographic shifts and technological advances in surveying.[10][11] Despite these strengths, historical redesigns—such as those in the 1990s and 2000s—have periodically altered series comparability, necessitating adjustments for trend analysis and underscoring the survey's balance between innovation and continuity in empirical labor measurement.[6]History
Origins and Establishment (1940–1950s)
The Current Population Survey (CPS) originated as a monthly household sample survey initiated by the Works Progress Administration (WPA) in March 1940, marking the first national effort to directly measure unemployment and labor force status on a continuing basis.[12] This program addressed the limitations of decennial census data and administrative records, which failed to provide timely, granular insights into employment dynamics amid the Great Depression's lingering effects and the transition to wartime mobilization.[12] The survey employed an "activity concept" for classification—defining individuals as employed if working or with a job, unemployed if actively seeking work, or not in the labor force otherwise—and began with a small sample tested in December 1939 before expanding to approximately 8,000 housing units.[12] In August 1942, as the WPA dissolved amid full employment from World War II defense production, responsibility for the survey transferred to the U.S. Census Bureau under a Bureau of the Budget directive, with the name changing to Monthly Report on the Labor Force.[13][12] The Census Bureau promptly revised the methodology in 1943 to a full probability sample design, increasing the sample size to about 23,000 housing units by 1945 to enhance representativeness and reduce sampling error.[12] This shift ensured the survey's estimates aligned more closely with population benchmarks from the 1940 decennial census, establishing a foundation for reliable monthly labor force indicators.[14] By 1947, the Bureau of the Budget authorized expansion beyond basic unemployment metrics to include data on hours worked and earnings, reflecting growing demand for comprehensive labor market analysis in the postwar economy.[13] The survey was officially renamed the Current Population Survey in 1948 to encompass its broadened demographic, social, and economic scope.[13] In 1949, the Bureau of Labor Statistics (BLS) assumed oversight of core concepts, methods, and publication of unemployment data, initiating a collaborative framework with the Census Bureau that formalized their joint roles by 1959.[13] During the early 1950s, methodological refinements continued, including empirical studies culminating in the delineation of primary sampling units (PSUs) in late 1949 and early 1950, which stratified the sample for better geographic coverage.[15] A major redesign in 1954 expanded PSUs from 68 to 230 clusters—counties or metropolitan areas—without increasing costs, thereby improving estimate precision and variance reduction through self-weighting strata.[14] These enhancements solidified the CPS as the principal source of official U.S. labor force statistics, prioritizing empirical accuracy over administrative proxies.[12]Evolution and Key Expansions (1960s–1980s)
During the 1960s, the CPS underwent significant sample expansions to enhance national coverage and precision. In January 1960, the survey incorporated Alaska and Hawaii following their statehood, increasing primary sampling units (PSUs) from 330 to 333 and expanding the covered population by approximately 500,000 persons.[14] By 1966, the monthly sample size grew from 35,000 to 50,000 households to improve reliability of labor force estimates.[12] In January 1967, further expansion raised PSUs to 449 and monthly housing units to about 60,000, yielding a roughly 20% gain in estimate precision.[14] Methodological refinements in the late 1960s addressed conceptual gaps in labor force measurement, informed by the Gordon Committee recommendations. Starting January 1967, the survey adopted a four-week job search duration for unemployment classification, required evidence of specific search methods, and incorporated an availability test, while raising the civilian noninstitutional population age threshold from 14 to 16.[14][12] Population controls shifted to 1960 Census-based estimates in 1961–1962, with full implementation by 1962.[16] Into the 1970s, sample adjustments followed the 1970 Census, reducing households to 47,000 initially before a 1976 addition of 9,000 households targeted 23 smaller states for subnational reliability; PSUs reached 461 by 1973 via cluster enhancements.[12][14] In September 1975, a state supplementary sample of about 14,000 interviews was introduced across 26 states and the District of Columbia to bolster state-level estimates.[14] The 1980s featured further sample scaling and control updates amid fiscal constraints. In 1980, additions of roughly 9,000 households extended coverage to 40 states, pushing the total to 65,000, with PSUs rising to 629; an additional 6,000 metropolitan-focused households followed in 1981, though a 1982 cut of 11,000 stabilized monthly interviews at 60,000.[12][14] January 1979 introduced two-level ratio estimation to differentiate metropolitan and nonmetropolitan variances, while 1980 Census-based controls took effect in 1979 for population weights and fully in 1982, adjusting estimates upward by 2%.[14][16] Questionnaire expansions included October 1978 earnings queries for outgoing rotation groups and November 1982 additions for union membership alongside earnings.[14] For the Annual Social and Economic Supplement, 1979 expanded income sourcing to 51 categories with 27 response values, and 1983 added Hispanic population weighting.[16] These changes prioritized empirical accuracy in tracking labor market dynamics, with the Bureau of Labor Statistics assuming deepened analytical oversight since 1959.[12]Major Revisions and Modernization (1990s–Present)
A major redesign of the Current Population Survey (CPS) was implemented in January 1994, introducing a revised questionnaire to enhance the accuracy of labor force measurements, transitioning data collection from paper-and-pencil to computer-assisted personal interviewing (CAPI) and telephone interviewing (CATI), and adopting population controls derived from the 1990 decennial census adjusted for undercount via the Post-Enumeration Survey.[17] Key questionnaire modifications included rewording the primary employment question to "LAST WEEK, did you do ANY work for pay?" (excluding unpaid family work initially probed separately), adding a direct question on layoffs, and refining unemployment criteria to emphasize active job search methods such as contacting employers while de-emphasizing passive activities like reading advertisements.[17] These changes, informed by extensive testing since 1986, resulted in measurable shifts in estimates: the unemployment rate rose by approximately 0.5 percentage points (from 6.8% in 1993 to 7.3%), labor force participation increased by 0.4 points to 66.6%, and demographic variations emerged, such as higher female employment-population ratios.[17] Post-1994, the CPS underwent periodic methodological updates primarily through revisions to population controls and sample designs aligned with decennial censuses, without fundamental questionnaire overhauls. Population controls were benchmarked to the 2000 census starting in early 2001, increasing the civilian noninstitutional population estimates and prompting adjustments to labor force series; similar updates occurred with 2010 census-based controls implemented around January 2011, incorporating refined undercount estimates and housing unit frames.[18] A sample redesign effective in 2014, based on the 2010 census, incorporated new primary sampling units, updated stratification for metropolitan areas, and improved coverage of recent housing constructions to better represent population shifts, though these enhancements maintained continuity in core labor force concepts.[19] In response to declining response rates—from over 80% a decade prior to lower levels by the early 2020s, attributed to privacy concerns, rising cellphone-only households, and respondent burden—the U.S. Census Bureau and Bureau of Labor Statistics initiated comprehensive modernization efforts in 2023.[11] These include developing an Internet Self-Response (ISR) mode for initial data collection by 2027, adapting questions for self-administered formats, integrating multi-mode approaches (ISR alongside CAPI), and deploying new operational software with adaptive designs to prioritize high-response segments.[11] Field testing is scheduled for 2025 and 2026, with cognitive and usability evaluations ongoing to preserve data quality amid these shifts, alongside separate enhancements to the Annual Social and Economic Supplement using administrative records for validation.[11]Methodology
Sampling Design and Population Controls
The Current Population Survey (CPS) utilizes a multistage probability sample of approximately 60,000 eligible housing units each month, targeting the civilian noninstitutional population aged 16 years and older residing in the United States and the District of Columbia.[20][3] The design selects primary sampling units (PSUs)—typically counties or groups of contiguous counties—within state-based strata, with selection probabilities proportional to population size; subsequent stages involve choosing enumeration districts or blocks within PSUs, followed by random selection of household addresses within those clusters.[21] This clustered approach balances cost efficiency with statistical reliability, allocating sample sizes across states to limit coefficients of variation for unemployment rate estimates to no more than 1.9% nationally and 8% at the state level, assuming a 6% unemployment rate.[20] Households follow a 4-8-4 rotation pattern: interviewed during four consecutive months, excluded for eight months, re-interviewed for four more consecutive months, and then permanently dropped from the sample.[19] This overlap across months enables composite estimation techniques that reduce variance by averaging data from incoming and outgoing rotation groups.[21] The sample excludes institutionalized populations, active-duty military personnel, and U.S. residents abroad, focusing fieldwork on occupied households via in-person and telephone interviews.[3] A redesign effective in 2025, based on the 2020 Census and incorporating a blended base of census data with administrative records, replaced the prior frame with 863 PSUs spanning 1,402 counties selected from an eligible pool of 1,983 PSUs and 3,144 counties; the transition occurs gradually through rotation groups from April 2025 to July 2026, maintaining the monthly housing unit target while yielding negligible shifts in published estimates.[10] Population controls consist of independent Census Bureau estimates of the civilian noninstitutional population, disaggregated by age, sex, race, Hispanic origin, nativity, and geographic divisions including states.[22] Derived from decennial census benchmarks adjusted annually for births, deaths, and net international migration using vital statistics and administrative data, these controls serve as weighting anchors to calibrate sample responses to actual population totals, mitigating sampling error and undercoverage biases.[22][18] Weights are computed via iterative proportional fitting across control categories, ensuring estimates reflect demographic distributions without over- or under-representing subgroups.[21] Updates to population controls are introduced each January, coinciding with the release of revised Census estimates; for instance, the January 2025 adjustment incorporated higher totals driven by updated immigration inflows, increasing labor force projections by about 0.6% compared to prior controls.[23][24] This annual process enhances temporal consistency but can introduce discontinuities if demographic shifts, such as migration surges, exceed prior projections, prompting BLS to publish bridged series for comparability.Data Collection Procedures
The Current Population Survey (CPS) data are collected monthly by the U.S. Census Bureau on behalf of the Bureau of Labor Statistics, targeting approximately 60,000 occupied households selected through a probability sample covering all 50 states and the District of Columbia.[26][4] Interviews occur during the calendar week that includes the 19th of the month, with responses referring to activities in the prior reference week, which encompasses the 12th of the month (adjusted for December to ensure the reference week falls within the survey month).[27][4] Field representatives conduct data collection using computerized questionnaires via computer-assisted personal interviewing (CAPI) for initial contacts or when telephone is unavailable, while telephone interviewing—facilitated by computer-assisted telephone interviewing (CATI) systems—is the preferred and predominant mode for subsequent interviews, accounting for about 85% of non-initial collections.[27][26] CATI operations are centralized in facilities located in Jeffersonville, Indiana, and Tucson, Arizona, handling roughly 12% of total interviews and achieving completion rates of 80-92% for assigned cases before recycling incomplete ones to field staff.[27] Households follow a sample rotation pattern wherein each is interviewed eight times over a 16-month span under a 4-8-4 scheme: four consecutive months in the sample, followed by eight months out, and then four more months back in.[26][4] The process begins with establishing a household roster during the first month-in-sample (MIS-1), typically via personal visit, capturing demographics for all members aged 15 and older (with labor force focus on those 16 and older).[27][26] Subsequent months employ dependent interviewing, importing prior responses to streamline updates and reduce respondent burden, while scripted questions probe employment status, job search activities, and supplemental topics in designated months (e.g., income in MIS-4 and MIS-8).[27] Noninterviews are categorized into types A (refusals or absences), B (temporary ineligibility), and C (permanent ineligibility), with supervisors conducting reviews and quality control reinterviews to validate data accuracy.[27] Collected data are transmitted daily to the Census Bureau's central processing system for editing and weighting.[26]
Employment and Labor Force Classification Criteria
The Current Population Survey (CPS) classifies individuals in the civilian noninstitutional population aged 16 and older into employed, unemployed, or not in the labor force based on their activities during a designated reference week, defined as the calendar week that includes the 12th of the month (with adjustments for November and December if the 12th falls on a Friday or Saturday).[8][28] This classification ensures each person is counted in only one category, with the civilian noninstitutional population excluding active-duty military personnel and those in institutions such as prisons or nursing homes.[9][28] Employed persons are those who, during the reference week, worked at least one hour for pay or profit (including self-employment), worked 15 or more hours unpaid in a family-owned business or farm, or were temporarily absent from a job due to reasons such as illness, vacation, strike, or childcare, regardless of whether they were paid for the absence.[8][9] Multiple jobholders are counted only once as employed, even if holding more than one position.[9] The CPS collects this status through direct questions about work performed or job attachment during the reference week.[28] Unemployed persons are defined as those who did not work during the reference week, were available to start work (except for temporary illness), and either made at least one specific active effort to find a job in the prior four weeks or were on temporary layoff expecting recall without needing to search.[8][9] Active job search efforts must involve direct actions with potential to result in a job offer, such as contacting employers, submitting applications, or interviewing, but exclude passive activities like reviewing ads or job listings without follow-up.[8][28] Persons waiting to start a new job are classified as unemployed only if they actively searched within the last four weeks; otherwise, they are considered not in the labor force.[8] The labor force comprises all employed and unemployed persons aged 16 and older in the civilian noninstitutional population.[9][8] Those not in the labor force include individuals who are neither employed nor unemployed, such as retirees, students, homemakers, or the disabled who do not meet unemployment criteria; this group may encompass marginally attached workers (who want and are available for work but did not search in the last four weeks) and discouraged workers (a subset who believe no suitable jobs are available).[9][8] Survey responses are weighted using Census Bureau population controls to represent the national totals.[28]Data Products
Core Monthly Labor Force Data
The core monthly labor force data from the Current Population Survey (CPS) comprise estimates of employment status for the civilian noninstitutional population aged 16 years and older, excluding active-duty military personnel and those in institutions such as prisons or nursing homes. Conducted monthly by the U.S. Census Bureau on behalf of the Bureau of Labor Statistics (BLS), the survey targets approximately 60,000 eligible households and yields key indicators on the size and composition of the labor force, including levels of employment and unemployment as well as derived rates.[29] These data form the basis for the household survey component of the BLS's monthly Employment Situation report, providing timely insights into labor market dynamics distinct from payroll-based establishment surveys.[1] Individuals are classified into employed, unemployed, or not in the labor force based on their activities during a designated reference week that includes the 12th of the month. Employed persons include those who worked for pay or profit, or who were temporarily absent from a job; unemployed persons are jobless, available for work, and have actively sought employment in the prior four weeks; the labor force sums the employed and unemployed, while those not in the labor force meet none of these criteria.[8] Derived metrics include the unemployment rate, computed as the unemployed divided by the labor force multiplied by 100; the labor force participation rate, the labor force divided by the civilian noninstitutional population multiplied by 100; and the employment-population ratio, the employed divided by the population multiplied by 100.[1] These series are published in both seasonally adjusted and unadjusted forms, with standard errors reflecting sampling variability.[29]| Key Series | Description |
|---|---|
| Civilian Noninstitutional Population | Total U.S. population aged 16+, excluding armed forces and institutionalized persons.[8] |
| Labor Force Level | Sum of employed and unemployed persons.[1] |
| Employment Level | Number of persons who worked or had jobs during the reference week.[8] |
| Unemployment Level | Number of persons jobless, available, and seeking work.[8] |
| Unemployment Rate | Percentage of labor force that is unemployed ((unemployed / labor force) × 100).[1] |
| Labor Force Participation Rate | Percentage of population in the labor force ((labor force / population) × 100).[1] |
| Employment-Population Ratio | Percentage of population employed ((employed / population) × 100).[1] |