Proposal to Revise Method for Estimation of Monthly Labor Force Statistics for Certain Subnational Areas; Request for Comments

AGENCY:

Bureau of Labor Statistics, Labor.

ACTION:

Request for comments on proposed action.

SUMMARY:

The Department of Labor, through the Bureau of Labor Statistics and, specifically, the Local Area Unemployment Statistics (LAUS) program, is responsible for the development and publication of State and local area labor force statistics. In the LAUS program, monthly estimates of the labor force, employment, unemployment, and the unemployment rate for more than 7,000 areas in the Nation are developed and issued. A major program redesign to improve the methodological basis of the LAUS estimates and update the geography and techniques to reflect 2000 Census data was initially funded in FY 2001. After completion of various long-term research projects, the BLS plans to implement improvements to the estimating methods with State and area LAUS estimates for January 2005, to be published in March 2005.

DATES:

Written comments must be submitted to the office listed in the Addresses section of this notice on or before December 10, 2004.

ADDRESSES:

Send comments to Sharon P. Brown, Chief, Division of Local Area Unemployment Statistics, Bureau of Labor Statistics, Room 4675, 2 Massachusetts Avenue NE., Washington DC 20212.

FOR FURTHER INFORMATION CONTACT:

Sharon P. Brown, Chief, Division of Local Area Unemployment Statistics, Bureau of Labor Statistics, telephone number 202-691-6390.

SUPPLEMENTARY INFORMATION:

I. Introduction

The Department of Labor, through the Bureau of Labor Statistics, is responsible for the development and publication of State and local area labor force statistics through the Local Area Unemployment Statistics (LAUS) program. Currently, monthly estimates of employment, unemployment, and the unemployment rate are prepared for more than 7,000 areas, including Census regions, Census divisions, all States and the District of Columbia, Puerto Rico, metropolitan and small labor market areas, counties, cities of 25,000 population or more, and all cities and towns in New England regardless of population. In a multi-year, multi-project initiative that began in FY 2001, the following improvements to State and area labor force estimation were identified:

• State time series estimating models with real-time benchmarking to the national monthly employment and unemployment levels that will address long-standing issues related to accuracy and end-of-year revision,
• the extension of model-based estimation to six additional substate areas and the respective balance-of-State areas, and
• two enhanced procedures for developing other substate areas that employ innovative and dynamic estimating methods.

II. Background

A hierarchy of estimation methods is used to produce the State and area labor force estimates, based in large part on the availability and quality of data from the Current Population Survey (CPS), the office measure of the labor force for the nation.

Improved Time Series Models with Real-time Benchmarking. The estimates for States, the District of Columbia, New York City, Los Angeles metropolitan area, and the balances of New York State and California are developed using signal-plus-noise models. These models rely heavily on monthly CPS data, as well as current wage and salary employment estimates and unemployment insurance statistics. The State CPS annual averages of employment and unemployment are used as benchmarks to the model-based estimates at the end of the year. In general, the current method of model estimation and annual benchmarking results in an overestimate of employment and an underestimate of unemployment and the unemployment rate in States as compared to the national CPS estimates. The annual benchmarking approach reintroduces sampling error into the series and results in significant end-of-year revisions in a large number of States, causes economic anomalies that are an artifact of the benchmarking approach, distorts seasonality in the previous year so that analysis is impaired, and often misses shocks to the economy.

To address these serious issues, the improved model-based approach to estimation will ensure that State estimates add to the national estimates of employment and unemployment each month, through real-time benchmarking. In doing so, the benchmark will change from annual State-level estimates of employment and unemployment to monthly national estimates of these measures. In this way, economic shocks will be reflected in the State estimates on a real-time basis, and end-of-year revisions will be significantly smaller.

The improved State models are signal-plus-noise models, where the signal is a bivariate model of the unemployment or the employment level. The same inputs used in the current models are used in the new models. Seasonal adjustment occurs within the new model structure, with the removal of the seasonal component. The proposed models with real-time benchmarking produce reliability measures for the seasonally adjusted and not seasonally adjusted series, and on over-the-month and over-the-year change.

Under real-time benchmarking, a tiered approach to estimation is used. Model-based estimates (using a univariate form) are developed for the nine Census divisions that geographically exhaust the nation. (Census division groupings are currently used to analyze and publish LAUS estimates.) These estimates are controlled to the national levels of employment and unemployment. State model-based estimates are then made and controlled to the Census Division estimates. In this manner, the monthly State employment and unemployment estimates will add to the national levels, precluding differences between the sum of States and the national estimates, and national shocks related to the business cycle or outliers like September 11 will be addressed.

Annual historical benchmarking will still continue for State estimates but would be greatly altered. The updating of model inputs, model reestimation, and incorporation of updated population controls would be performed each year. However, the impact on the historical series of these benchmark activities is considered to be fairly small.

Extending Model-based Estimation to Additional Areas. Currently, monthly labor force estimates for New York City, the balance of New York State, the Los Angeles metropolitan area, and the balance of California are developed using model-based methods. (These models will be updated to the form used for States and described above.) As part of the LAUS improvement efforts, model-based estimation will be extended to the following areas and the respective balance-of-State areas: Chicago metropolitan division, Cleveland metropolitan area, Detroit metropolitan division, Miami metropolitan division, New Orleans metropolitan area, and Seattle-Everett metropolitan division. This will improve the statistical basis of the estimation for these areas, and provide important tools for analysis such as measures of error and seasonally adjusted series.

These area models will follow the form of the Census divisions (univariate), and will be benchmarked to the State employment and unemployment estimates on a real-time basis. As with the State models, seasonally adjusted series will be produced, along with measures of error for the seasonally adjusted and not seasonally adjusted series, and on over-the-month and over-the-year change.

New and Reentrant Unemployment. There has been a long-standing concern in the LAUS program regarding the estimation of unemployment at the substate level (for areas other than New York City, Los Angeles, and the balances of New York State and California). Of specific concern is the measurement of unemployed new and reentrants to the labor market. The difficulty in estimating new and reentrants led to the use of a proportionate adjustment of area estimates to the State total unemployed as a way of controlling for the underestimate at the area level. The current research has led to a proposal for an improved methodology.

The new methodology incorporates the CPS new and reentrants State data and utilizes improved econometric modeling techniques. The new model follows the basic form of the model created in 1983 and used today, but has been updated and improved. The proposed model uses a stochastic nonlinear estimation process rather than the global linear procedure used currently. A stochastic, or random, coefficient is one whose value is allowed to change over time. In this model, the values of the model coefficients change from month to month as the models are updated with information from current observations.

The model estimates are distributed to each labor market area in the State based on the area's share of the State population. New entrants are distributed based on the area's share of the State 16-19 year old population, and reentrants are distributed based on the area's share of the State 20 years and older population.

The new method of estimation successfully addresses the issue of underestimation and eliminates the need for significant proportionate adjustment of area estimates to the monthly State levels of unemployment.

Residency Adjustment. The underlying concepts and definitions of all labor force data developed by the LAUS program are consistent with those of the Current Population Survey (CPS), including the requirement that measures relate to the place of residence of the labor force participant. Establishment-based data on the number of nonagricultural wage and salary jobs by place of work from the Current Employment Statistics (CES) or the Quarterly Census of Employment and Wages (QCEW) programs is the only current, geographically comprehensive source of information on employment at the substate level, and are a significant input to LAUS estimation. The establishment series differs from the CPS in that the CPS counts employed persons where they reside rather than jobs by place of work. Thus, the establishment-based data must be adjusted to account for multiple-job holding and residency prior to use in LAUS estimation.

The current procedure utilizes a single adjustment ratio for each estimating area, using Decennial census data and March-April average establishment-based data. The Census estimate of all employed residents in an area is divided by the job count. This ratio is then applied each month to the nonfarm wage and salary estimate for the area to produce the resident nonfarm wage and salary employed estimate for the area.

A basic problem with the current Census-based procedure of adjusting for residency was the limited geographic scope for influencing the area's estimate of resident employed and static nature of the approach. Recognizing that labor market areas often are not defined to the point where commutation is zero, and that, in the intercensal period, job growth can and does occur in the areas surrounding the estimating area, a new approach to developing resident employment was considered.

The proposed method postulates that resident employment in an area is a function not only of the relationship between employed residents and jobs in that area, but in other areas within commuting distance. The procedure is more dynamic than the current method insofar as job count changes in commuting areas can affect resident employment. As in the current procedure, however, the commuting ratios themselves are fixed for the intercensal period.

Detailed descriptions of the current and Redesign approaches are available at the above address and at the BLS LAUS Web site http://www.bls.gov/lau/home.htm.

Comments submitted in response to this notice will be summarized and included in the Notice of Decision on this proposal.

This notice is a general solicitation of comments from the public.