Estimation Methodology


Unemployment estimates have been developed for subnational areas for about 60 years. The program began during World War II under the War Manpower Commission to identify areas where labor market imbalance was created as a result of an inadequate labor supply, materials shortages, and transportation difficulties. After the war, emphasis was placed on identifying areas of labor surplus, and the program of classifying areas in accordance with severity of unemployment was established.

In 1950, the Department of Labor's Bureau of Employment Security (now Employment and Training Administration) published a handbook, Techniques for Estimating Unemployment, so that comparable estimates of the unemployment rate could be produced among the states. This led, during the late 1950s, to the formulation of the "Handbook" method, a series of computational steps designed to produce local employment and unemployment estimates without the expense of a large survey. This method relied heavily on data derived from the unemployment insurance (UI) system.

In 1972, the Bureau of Labor Statistics (BLS) began to develop the concepts and methods to be used by states to estimate labor force, employment, and unemployment. In 1973, a new system for developing labor force estimates at the state and substate level was introduced. This system combined the Handbook method with concepts, definitions, and estimation controls from the Current Population Survey (CPS), the Bureau of Census survey sponsored by BLS.

Since 1976, state samples of the CPS have been increased several times to improve the quality of state labor force estimates. In 1976, use of the CPS as an estimation control was extended to all states. Subsequently, BLS established a maximum expected coefficient of variation (CV) of 10 percent for unemployment, assuming an unemployment rate of 6 percent, as a criterion for using the monthly CPS data directly for official publication of labor force estimates. (The coefficient of variation of an estimate is defined as the standard error of the estimate divided by the estimate.) Based on this criterion, monthly CPS data were used, beginning in 1978, for official statewide labor force estimates for the 10 largest states (referred to as "direct-use" states)—California, Florida, Illinois, Massachusetts, Michigan, New Jersey, New York, Ohio, Pennsylvania, and Texas—and for 2 substate areas—the Los Angeles-Long Beach Primary Metropolitan Statistical Area (PMSA) and New York City.

Over the years, major improvements have been made to the UI database, an integral input to state and area estimation. The UI database project, conducted in 1976-78, standardized all UI claims data used in state and area labor force estimates, so that these data would be more consistent with the conceptual underpinnings of unemployment used in the CPS, as well as more comparable from state to state. The result of this project was the regular, automated development of data on UI claimants certifying to unemployment for the week including the 12th day of the month (the CPS reference week). These data are based on the claimants' state/county/city of residence and exclude those who had earnings from employment in the certification week.

In 1985, a state-based design for the CPS was fully implemented to incorporate the 1980 census information and to provide for improved reliability for each of the 50 states and the District of Columbia. North Carolina was added as another direct-use state, and the CV requirement for unemployment was reduced to 8 percent on monthly estimates for these 11 large states. For each of the 39 non-direct-use states and the District of Columbia, the reliability requirement was established at an 8 percent CV for annual average unemployment, assuming a 6-percent unemployment rate.

Until 1989, official monthly estimates for the non-direct-use states, the District of Columbia, and substate areas were based at least partially on the Handbook method. Beginning in 1989, estimates for these 39 states and the District of Columbia were based on time series models developed by BLS and tested by state employment security agencies, using standardized procedures. More advanced regression models for the 39 smaller states and the District of Columbia were introduced in 1994.

Seasonal adjustment of statewide estimates was introduced in 1992.

In 1994, major changes to the CPS were introduced, including a complete redesign of the questionnaire and the use of computer-assisted interviewing for the entire survey. A new sample based on a sampling frame which incorporated 1990 Census data was also introduced and phased in through mid-1995.

At the beginning of 1996, due to budgetary reductions, the CPS sample size was decreased from 56,000 to 50,000 households, and, therefore, direct-use estimation was discontinued. Since then, the estimates for the 11 large States, Los Angeles-Long Beach, New York City, and the balances of California and New York have been produced by the same time series models used in the smaller states. Seasonally adjusted data were introduced for these two large substate areas. In addition, the Handbook procedure, which continues to be used to create labor market area (LMA) estimates, was simplified to eliminate several nonessential steps.

In January 2005, a major program Redesign was implemented. Work on the Redesign began in Fiscal Year 2001, with a budget initiative to enhance the quality and quantity of LAUS program statistics. Major LAUS Redesign components include improvements to the current method of State and large area estimation including ‘real time benchmarking’, extending our ‘best estimating’ techniques to more areas, improving the methods used in all other areas through better techniques and input data, and updating the geography with 2000 Census-based areas.

Estimates for States

For all states and the District of Columbia, the Los Angeles-Long Beach-Glendale, metropolitan division, New York City, and the respective balances of state, models based on a "signal-plus-noise" approach (Scott, 1974, and Bell, 1990) are used to develop employment and unemployment estimates. The model of the signal is a time series model of the true labor force which consists of three components: A variable coefficient regression, a flexible trend, and a flexible seasonal component. The regression techniques are based on historical and current relationships found within each state's economy as reflected in the different sources of data that are available for each state—the CPS, the Current Employment Statistics (CES) survey, and the UI system. The noise component of the models explicitly accounts for autocorrelation in the CPS sampling error and changes in the average magnitude of the error (Pfeffermann and Tiller, 2002). In addition, the models can identify and remove the effects of outliers in the historical CPS series. While all the state models have important components in common, they differ somewhat from one another to better reflect individual state labor force characteristics.

Seasonal adjustment occurs within the model structure through the removal of the seasonal component. The models also produce reliability measures on the adjusted and unadjusted series, and on over-the-month change.

The Redesign bivariate models incorporate a major change in the approach to benchmarking and the benchmarking process. Rather than continue with an annual average State benchmark applied retrospectively that reintroduces sampling error to the historical monthly estimates, the Redesign approach uses a reliable real-time monthly national benchmark for controlling current State model estimates of employment and unemployment. In this process, benchmarking is part of the monthly State model estimation process.

Under real-time benchmarking, a tiered approach to estimation is used. Model-based estimates are developed for the nine Census divisions that geographically exhaust the nation using univariate signal-plus-noise models. The Division models are similar to the State models, but do not use unemployment insurance claims or nonfarm payroll employment as variables. The division estimates are benchmarked to the national levels of employment and unemployment on a monthly basis. The benchmarked division model estimate is then used as the benchmark for the States within the division. The distribution of the monthly benchmark adjustment to the States is based on each State’s monthly model estimate. In this manner, the monthly State employment and unemployment estimates will add to the national levels.

Estimates for Substate Labor Market Areas

As noted, monthly labor force estimates for two large substate areas—New York City and the Los Angeles-Long Beach-Glendale, CA metropolitan division and the respective balances of New York and California—are developed using bivariate signal-plus-noise models. We have also developed signal-plus-noise models for six additional substate areas and their State balances. The areas are: the Chicago-Naperville-Joliet, IL metropolitan division; the Cleveland-Elyria-Mentor, OH metropolitan area; the Detroit-Warren-Livonia, MI metropolitan area; the Miami-Miami Beach-Kendall, FL metropolitan division; the New Orleans-Metairie-Kenner, LA metropolitan area; and the Seattle-Bellevue-Everett, WA metropolitan division. As with the Redesign State and Division models, these area models are based on the classical decomposition of a time series into trend, seasonal, and irregular components. A component to identify and remove the CPS sampling error is also included. Area models, like the Division models, are univariate in design in that only the historical relationship of the inputs is considered—UI claims and CES inputs are not used each month in the estimation process. Area and balance of State models are controlled directly to the State totals, which are themselves controlled to the national CPS via the Census division models.

The LAUS Handbook method is an effort to estimate unemployment for an area, using available information without the expense of expanding a labor force survey like the CPS. The Handbook presents a series of estimating "building blocks," in which categories of unemployed workers are classified by their previous status. Two broad categories of unemployed persons are: (1) Those who were last employed in industries covered by state UI laws, and (2) those who either entered the labor force for the first time or reentered after a period of separation. Handbook inputs were updated using the 2000 Census and other improvements to Handbook estimation were implemented with January 2005 estimates.

Employment. The total employment estimate is based on data from several sources. The primary source for most metropolitan areas (MAs) is the Federal-state CES survey. The CES is designed to produce estimates of the total number of employees on payrolls in nonfarm industries for the particular area. In small labor market areas and the remainder of the MAs, the establishment employment data come from the Quarterly Report of Quarterly Census of Employment and Wages (ES-202 Report).

These "place-of-work" employment estimates must be adjusted to a place-of-residence basis, as in the CPS. Estimated adjustment factors have been developed using employment relationships which existed at the time of the most recent decennial census. The adjustment approach implemented in January 2005 is more dynamic than the previous one and incorporates commuting to nearby labor market areas. These factors are applied to the place-of-work employment estimates for the current period to obtain adjusted employment estimates, to which are added synthetically developed estimates for employment not represented in the establishment series—agricultural workers, nonfarm self-employed and unpaid family workers, and private household workers.

Unemployment. The estimate of unemployment is an aggregate of the estimates for each of the two building-block categories. The "covered" category further consists of two unemployed worker groups: (1) Those who are currently receiving UI benefits and (2) those who have exhausted their benefits. Only the number of those currently collecting benefits is obtained directly from an actual count of UI claimants for the reference week. The estimate of persons who have exhausted their benefits is based upon the number actually exhausting benefits in previous periods "survived" using a conditional probability approach based on CPS data.

The second category, "new entrants and reentrants into the labor force," cannot be estimated directly from UI statistics, because unemployment for these persons is not immediately preceded by the period of employment required to receive UI benefits. In addition, there is no uniform source of new entrants and reentrants data for States available at the LMA level; the only existing source available is from the CPS at the State level. Separate estimates for new entrants and for reentrants are derived from econometric models based on current and historical state entrants data from the CPS. These model estimates are then allocated to all Labor Market Areas (LMAs) based on the age population distribution of each LMA. For new entrants, the area’s proportion of 16-19 years population group to the State total of 16-19 years old population is used, and for reentrants, the handbook area’s proportion of 20 years and older population to the State total of 20 years and older population is used.

Substate adjustment for consistency and additivity. Each month, Handbook estimates are prepared for labor market areas that exhaust the entire state area. To obtain a labor force estimate for a given area, a "Handbook share" is computed for that area which is defined as the ratio of that area's Handbook estimates of employment and unemployment to the sum of the Handbook estimates of employment and unemployment for all LMAs in the state. These ratios are then multiplied by the current, statewide estimate for employment and unemployment to produce the final adjusted LMA estimates.

Estimates for Parts of LMAs

Current labor force estimates at the sub-LMA level are required by several Federal programs. Disaggregation techniques are used to obtain current estimates of employment and unemployment for counties within multi-county LMAs and cities, towns, and townships within counties. Two alternative methods are used to disaggregate the LMA estimates.

The population-claims method is the preferred technique. If residence-based UI claims data are available for the subareas within the labor market area, the ratio of claims in the subarea to the total number of claims within the LMA is used to disaggregate the estimate of experienced unemployed to the subarea level. To ensure the quality of the claims data used in this technique, claimant records are processed through a residency assignment system that verifies and/or corrects residence addresses and assigns the associated residency codes. This provides a more accurate count of claims by city. The estimates of unemployed entrants are allocated based on the latest available census distribution of adult and teenage population groups. Employment is disaggregated using decennial census employment-population ratios updated by current population estimates. Estimates for all disaggregated counties and New England cities and towns are developed using this method.

If the necessary UI claims data are not available, the census-share method is used. This method uses each subarea's decennial census share of total LMA employment and unemployment, respectively, in order to disaggregate employment and unemployment. Very few states will be using this method for data after 2004.

Annual Activities

Once each year, labor force estimates are revised to reflect updated input data and new Census Bureau population controls. As part of this procedure, all of the state and substate models are reviewed, revised as necessary, and then reestimated; this reestimation is called "smoothing."

When new population controls are available from the Bureau of the Census, typically in January, CPS estimates for all states, the District of Columbia, New York City; the Chicago-Naperville-Joliet, IL metropolitan division; Cleveland-Elyria-Mentor, OH metropolitan area; Detroit-Warren-Livonia, MI metropolitan area; Los Angeles-Long Beach-Glendale, CA metropolitan division; Miami-Miami Beach-Kendall, FL metropolitan division; New Orleans-Metairie-Kenner, LA metropolitan area; and, the Seattle-Bellevue-Everett, WA metropolitan division are adjusted to these controls. Additionally, the time series regression models for the states and model-based areas are reestimated based on the latest input data.

Other substate estimates for previous years are also revised on an annual basis. The updates incorporate any changes in the inputs, such as revisions to establishment-based employment estimates or claims data and updated historical relationships. The revised estimates are then readjusted to the latest statewide estimates of employment and unemployment.

Bell, W.R. and Hillmer, S.C. (1990), "The Time Series Approach to Estimation for Repeated Surveys," Survey Methodology, 16, 195-215.
Scott, J.J., and Smith, T.M.F. (1974), "Analysis of Repeated Surveys Using Time Series Methods," Journal of the American Statistical Association, 69, 674-678.
Pfefferman, D. and Tiller, R. (2002), "State Space Modelling with Correlated Measurements with Application to Small Area Estimation Under Benchmark Constraints," State Space and Unobserved Components Models in Honour of Professor J. Durbin, Amsterdam.


Last Modified Date: March 10, 2005