General Discussion

This In Focus provides an overview of BLS annual benchmark revisions, which have been part of regular BLS employment estimation processes since 1935. Benchmark revisions incorporate less timely but more comprehensive data into employment estimates.

Survey

The CES survey, also known as the payroll survey or the establishment survey, is a monthly survey of approximately 119,000 businesses and government agencies. The employers surveyed are randomly selected to represent all nonfarm employers by state, industry, and employer size categories. After selection, employers are surveyed for at least two years. Each month, BLS asks these employers how many civilians were on their payroll during the pay period that included the 12th day of the month.

Employment reports received by the last Friday of the month are incorporated into "first preliminary estimates," which BLS releases on the first Friday of the next month. Additional employment reports received in time for the following month's estimates are incorporated into revised, "second preliminary estimates." Employment reports received from more employers responding to the survey in time for a third month of estimation are incorporated into "third and final sample-based estimates" for each month.

Business Birth and Death Modeling

Some businesses cannot be chosen for the CES survey because they are too new to be included. Other businesses do not answer the survey because they are no longer operating. Patterns of business openings and closings over the year vary by industry in predictable ways. BLS uses historical data to model employment changes due to business "births" and "deaths" and updates this model with new data every year. This modeled component is part of all national-level CES employment estimates (except during the COVID-19 pandemic, when use of the model was modified because business births and deaths were not following usual patterns).

Seasonal Adjustment

Some month-to-month changes in employment are part of seasonal patterns, such as the opening and closing of ski resorts as the seasons change. Seasonal employment patterns differ by industry. BLS uses historical data to model these patterns and produces both seasonally adjusted and unadjusted estimates of month-to-month changes.

Benchmarking

The CES program produces estimates of month-to-month changes in the number of jobs soon after the end of each month. However, BLS revises these estimates of monthly employment changes based on additional data on overall civilian employment levels.

Additional Data

BLS receives additional data on employment levels after publishing the survey-based CES estimates. Employers are required to report the number of workers in jobs covered by Unemployment Insurance (UI) on their payrolls to states every quarter. UI covers about 97% of nonfarm payroll employment in the United States. These payroll reports are then forwarded to BLS, which receives them several months after CES survey responses. BLS also receives periodic employment data for employers not covered by UI, such as railroads.

These additional data sources provide information on the actual employment of all covered employers. They are not affected by which businesses respond to a survey or by the accuracy of models. BLS uses these additional data to further revise, or benchmark, estimates of employment. The same data are also used to produce the model of employment changes due to business births and deaths.

Benchmark Process

BLS uses the additional data to make a more accurate estimate of employment levels by industry for the previous March. The difference between the previous estimate of employment and the new, more accurate estimate is called the "benchmark revision."

At the time of the benchmark revision, estimates of employment changes in the 11 months before and 9 months after March are adjusted proportionately to match the new, more accurate March employment levels by industry.

Benchmark Timing

Benchmark employment estimates for the previous March are released in early February. However, in August BLS releases a "preliminary estimate" of the overall size of the upcoming benchmark revision.

Although BLS benchmarks employment levels only once per year, it publishes quarterly employment totals from the UI data by industry and geography about six months after the quarter ends. Federal Reserve district banks in Dallas, Philadelphia, and New York use these employment totals to estimate and publish their own estimates of benchmarked employment more frequently than BLS.

BLS continues to benchmark national level employment estimates annually because quarterly revisions can be larger than annual revisions and because the seasonal pattern of employment changes in the CES is different than the seasonal pattern of employment changes in UI data.

Establishment survey benchmarking is done each year to align employment estimates from the survey with employment counts derived primarily from the administrative file of employees covered by Unemployment Insurance (UI). All employers covered by UI laws are required to report employment and wage information to the appropriate state UI agency four times per year. The UI data are obtained and edited by each state's Labor Market Information agency. They are tabulated and published through the BLS Quarterly Census of Employment and Wages (QCEW) program. Both the QCEW and CES categorize their data using the North American Industry Classification System (NAICS). About 97 percent of total nonfarm employment within the scope of the establishment survey is covered by UI and is available to the CES program via QCEW records.

To create a continuous time series between the new March benchmark level and historical sample-based data from the prior March benchmark level, employment estimates for the months between the most recent March benchmark and the previous year's benchmark are adjusted using a linear "wedge-back" procedure. This procedure assumes that the total estimation error accumulated at a steady rate since the last benchmark. For the 9 months following the March benchmark (also called the post-benchmark period), BLS applies previously derived over-the-month sample changes to the revised March level to get the revised estimates. New net birth-death model forecasts are also calculated and applied during post-benchmark estimation. More information on benchmarks in the CES program is available in Benchmark in the Calculation section of the CES Handbook of Methods.

For the CES-N program, annual benchmarks are constructed to realign the sample-based employment estimates for March of each year with the universe employment counts for that month. Population counts are much less timely than sample-based estimates, and they are used to provide an annual point-in-time census of employment. Only the not seasonally adjusted March estimates are replaced with population counts.

Population counts are derived primarily from administrative files of employees covered by unemployment insurance (UI). All employers covered by UI laws are required to report employment and wage information to their state’s labor market information agency (LMI) four times a year. Approximately 97 percent of employment falling within the scope of the establishment survey is covered by UI. A benchmark for the remaining 3 percent is constructed from alternate sources, primarily records from the Railroad Retirement Board (RRB) and County Business Patterns (CBP). This 3 percent is collectively referred to as noncovered employment. The full benchmark employment level developed for March of a given year replaces that March’s sample-based estimate for each basic cell. Each annual benchmark revision affects 21 months of not seasonally adjusted data from April of the prior year through the following year’s December. The CES-N program also updates seasonal adjustment models with each year’s benchmark revision, and 5 years of seasonally adjusted data are revised with the benchmark release.

Monthly estimates for the year preceding the March benchmark are readjusted using a "wedge-back" procedure. The difference between the final benchmark level and the previously published March sample-based estimate is calculated and spread back across the previous 11 months. The wedge is linear; eleven-twelfths of the March difference is added to the February estimate, ten-twelfths to the January estimate, and so on, back to the previous April estimate, which receives one-twelfth of the March difference. This procedure assumes that the total estimation error since the prior benchmark accumulated at a steady rate.

Estimates for the 7 months following the March benchmark (April through October) are also recalculated. Beginning with the March benchmark as base employment, the post-benchmark estimates are calculated by multiplying the sample-based link, or relative over the month change, for each month to the prior month’s employment estimate and then adding an updated net birth–death forecast.

All data revised as a result of benchmarking are published in February of each year with the first release of January estimates. Additional history may be revised if corrections or series reconstructions are necessary.