Local Area Unemployment Statistics
Notes on Using Current Population Survey (CPS) Subnational Data
- Sampling error of estimates. CPS estimates are based on a sample of the population, rather than a complete count. Therefore, they may differ from the figures that would have been obtained if it had been possible to take a complete census using the same questionnaire and procedures. This is particularly true for estimates of small groups or in states with small sub-samples. (The national sample consists of about 60,000 eligible households, but state sub-samples, the sizes of which depend on a number of factors, including state population, range from approximately 500 to 4,600 households.) Tables in appendix B of the Geographic Profile of Employment and Unemployment publications provide approximations of sampling errors at the 90-percent confidence level for employment and unemployment estimates for census regions and divisions and states for the corresponding years. The sampling errors can be used to construct confidence intervals around the estimates and to determine if estimates across areas or over time are significantly different. In addition, the first table in each section displays 90-percent confidence intervals for unemployment rates. The CPS sampling interval is roughly 1 in 2,500 nationally, and varies by state. Thus, on average, an estimate of 5,000 reflects only two sample cases with that set of characteristics.
- Differences between CPS and LAUS estimates. Due to differing methodologies, CPS estimates differ from annual-average employment and unemployment estimates produced by the Local Area Unemployment Statistics (LAUS) program. However, both the CPS and the LAUS program use the same underlying concepts and definitions, including measuring people by place of residence, rather than jobs by place of work. The CPS sample is designed to support reliable annual-average estimation for the total population in each of the 50 states and the District of Columbia. The official LAUS estimates for states are derived from signal-plus-noise models that use the monthly employment and unemployment measures tabulated from the CPS as the primary inputs. Payroll employment estimates from the Current Employment Statistics survey of establishments and unemployment insurance claims counts from the state workforce agencies also are used as LAUS model inputs to help mitigate volatility in the monthly state-level CPS estimates. The LAUS models are controlled, or forced to sum, to the national not-seasonally-adjusted employment and unemployment estimates from the CPS. These model estimates furthermore serve as controls for LAUS substate area estimation, so that the monthly estimates are additive and comparable across geographic levels for over 7,500 areas nationwide. These areas include all counties and cities with populations of at least 25,000, in addition to census regions and divisions and the 50 states and the District of Columbia. The LAUS data, which are used by various federal programs to allocate funds or determine program eligibility, constitute a large set of information that allows comparisons across thousands of diverse areas throughout the nation on a monthly basis.
- Lack of subnational population controls for demographic groups. Apparent differences in estimates between and among demographic groups—particularly in estimates of levels (that is, numbers of persons)—may not reflect real differences, especially prior to 2003. Instead, they may result solely from differences or changes in the demographic composition of the sample due to chance, combined with the interaction of the national and state population controls used in the estimation procedure. Population controlling for subnational areas incorporates several sets of independent demographic population estimates at the national and state levels. This approach can result in a possible source of error in subnational demographic components beyond the sampling error. On balance, the introduction in 2003 of independent population estimates for some demographic groups at the state level, though more limited and generally less reliable than those available at the national level, improved the reliability and comparability of many demographic population and labor force estimates relative to earlier years, when statewide controls were available for the total population only.
- Updating of population controls. Due to differing population controls, level (that is, number-of-person) estimates are not fully comparable between years. Level estimates for CPS data are controlled to independent estimates of population from the Census Bureau’s Population Estimates Program. Each year, the Census Bureau revises its post-censal population series back to the latest decennial estimates base, currently April 1, 2010. (CPS annual average subnational estimates through 2010 are on April 1, 2000, controls.) The CPS subnational data appearing in the Geographic Profile of Employment and Unemployment generally are re-controlled to population estimates that become available shortly after the end of the year to which the data pertain. This differs from the controlling of the national CPS data, for which the latest population controls from the Census Bureau are implemented on a forward-basis at the beginning of each year, without re-controlling of data for the previous year. It also differs from the updating of population controls in the LAUS program, in which estimates for at least the latest five years are re-controlled at the beginning of each year. The effect of population re-controlling on the statewide CPS estimates is generally less than one percent. Furthermore, re-controlling tends to have a minimal impact on rates, ratios, or percent distributions, because all levels for a given state are scaled by the same factor, representing the revision to the state’s population between the Census Bureau's initial and 1-year-later estimates, as applied to CPS levels.
- Impact of CPS redesign. Due to a major questionnaire redesign and a fundamental change in the data collection mode introduced in January 1994, estimates for years prior to 1994 may not be fully comparable with data for later years. See www.bls.gov/osmr/research-papers/1995/pdf/ec950090.pdf for an analysis of these differences, including which data items were most affected by the redesign.
- Regional CPS data vs. sum-of-state CPS data. For 1999 and earlier years, summing the CPS estimates for the states that comprise a census region or division may not yield exactly the published estimates for the region or division, due to differential weighting and rounding. CPS region and division estimates were created separately, and the weights for those large areas were not precisely the sum of the component state weights. Differences in estimates between the larger geographic areas and the summed component states, if present, were quite small—generally no more than a few thousand persons.
- Publication standards for subnational CPS data. To achieve comparability of the data across geographic areas for publication purposes, a unique requirement for minimum levels of employment and unemployment is developed for each area. This requirement is based on the known differences in sampling rates across areas. Before estimates are published for a specific category (such as Asian unemployment in a particular state), a predetermined critical cell must satisfy a 50-percent coefficient of variation requirement. Minimum bases of employment and unemployment are developed for each area based on this requirement. These are listed in appendix B, table 1, of the Geographic Profile of Employment and Unemployment publication for the given year. As a result, a small estimate for a particular category may be published in one state, but a larger estimate may be suppressed in another state. As noted above, the sampling ratio is, on average, about 1 in 2,500, but varies considerably by state.
- Discontinuation of CPS data for substate areas. The estimates for substate areas (i.e., large metropolitan areas, metropolitan divisions, and principal cities) that had been tabulated from the CPS and published in section 3 of the Geographic Profile of Employment and Unemployment through 2014 are characterized by (a) an absence of level (that is, number-of-person) estimates due to the unavailability of independent population controls, (b) limited geographic coverage that (c) lags Office of Management and Budget (OMB) updates to the federal statistical area system considerably, (d) a lack of demographic coverage despite (e) a relatively weak publication standard (corresponding to a maximum expected coefficient of variation of 50 percent), and (f) a high degree of sampling error on the data that do meet the relatively weak publication standard. Even though BLS believes that the CPS is the superior survey instrument for measuring unemployment in the U.S. and states, we have concluded that data users often are better served by substate area data from the Census Bureau’s American Community Survey (ACS). Data from the ACS include level estimates, provide more extensive geographic and demographic coverage, and have smaller sampling errors. Furthermore, the ACS sample is flexible enough to accommodate boundary changes from OMB the year that they become effective. General information on differences between labor force estimates from BLS and the ACS is available at www.bls.gov/lau/acsqa.htm.
- Additional information. Users of subnational CPS data are further encouraged to refer to the appendixes of the Geographic Profile of Employment and Unemployment publication.
Last Modified Date: March 20, 2020