Shane Haley and Caila Pinkleton are economists in the Division of Current Employment Statistics – National, Office of Employment and Unemployment Statistics, Bureau of Labor Statistics. Telephone: (202) 691‑6555; email: Contact CES
Summary of the revisions
With the release of January 2019 data on February 1, 2019, the Bureau of Labor Statistics (BLS) introduced its annual revision of national estimates of employment, hours, and earnings from the Current Employment Statistics (CES) monthly survey of nonfarm establishments.
The March 2018 benchmarked seasonally adjusted employment level for total nonfarm employment is 148,279,000. The not seasonally adjusted benchmarked employment level is 147,368,000.
Compared with the sample-based, published seasonally adjusted estimate for March 2018, total nonfarm employment had a revision of -1,000, or less than -0.05 percent. The not seasonally adjusted total nonfarm employment estimate was revised by -16,000, or less than -0.05 percent.
Table 1 presents revised total nonfarm employment data on a seasonally adjusted basis for January 2018 through December 2018. The revised data for April 2018 forward incorporate the effect of applying the rate of change measured by the sample to the new benchmark employment level, as well as updated net birth/death model forecasts and new seasonal adjustment factors. Revisions to November and December also reflect incorporation of additional sample receipts.
Table 1. Differences in seasonally adjusted levels and over-the-month changes, total nonfarm employment, January to December 2018 (in thousands)
IMPORTANT Special notice regarding seasonally adjusted historical data
Due to a major reconstruction effort, some seasonally adjusted series were revised further back than usual. Total nonfarm data was revised back to 1939. For more information, see the Reconstructions section of this article.
Establishment survey benchmarking is done on an annual basis to population 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 Unemployment Insurance agency four times per year. Approximately 97 percent of total nonfarm employment within the scope of the establishment survey is covered by UI. The UI data are obtained and edited by each state’s Labor Market Information (LMI) agency, and are tabulated and published through the BLS Quarterly Census of Wages and Employment (QCEW) program. Both the QCEW and CES are based on the North American Industry Classification System (NAICS). An employment count for the remaining 3 percent is constructed from alternate sources, primarily records from the Railroad Retirement Board (RRB), as well as County Business Patterns (CBP) and Annual Survey of Public Employment and Payroll (ASPEP) data from the Census Bureau. This 3 percent is referred to as noncovered employment (NCE). The combination of QCEW and NCE employment counts create the benchmark level and are used to replace CES sample-based estimates each March. The full benchmark level developed for March replaces the March sample-based estimate for each basic cell.
The total annual revision is the difference between the benchmark level for a given March and the published March sample-based employment estimate. Typically, the total revision is equal to the benchmark revision. The overall accuracy of the establishment survey is usually gauged by the size of the benchmark revision, which is often regarded as a proxy for total survey error. In years that are affected by historical reconstructions, however, affected CES series are re-estimated prior to benchmarking. The benchmark revision, in these cases, is the difference between the benchmark level and the newly reconstructed sample-based estimate. The benchmark revision can be more precisely interpreted as the difference between two independently derived employment counts, each subject to its own error sources.
In order 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, which assumes that the total estimation error accumulated at a steady rate since the last benchmark. For the 9 months following the March benchmark (also referred to as the post-benchmark period), CES 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 the Benchmarks section of the CES Technical Notes, as well as in the October 2017 Monthly Labor Review, "Benchmarking the Current Employment Statistics National Estimates."
Table 2 presents revised employment data on a seasonally adjusted basis for March 2018 by major industry sector. The revision to seasonally adjusted total nonfarm employment is -1,000.
Table 2. Seasonally adjusted employment revisions for major industry sectors, March 2018 (in thousands)
Typically, 5 years of seasonally adjusted data are revised with each CES annual benchmark. However, reconstructed series are seasonally adjusted over their revised time spans.
For technical information on how seasonal adjustment is performed in the CES program, refer to the Seasonal Adjustment section of the CES Technical Notes.
Table 3 presents the employment benchmarks for March 2018, not seasonally adjusted, by major industry sector. The revision to not seasonally adjusted total nonfarm employment is -16,000.
Table 3. Not seasonally adjusted employment benchmarks for major industry sectors, March 2018 (in thousands)
(1) With the 2018 benchmark, CES reconstructed several series. The effects of these reconstructions are included in this table. For more information, see the Reconstructions section in the 2018 CES Benchmark Article.
Table 4 below shows the recent history of not seasonally adjusted total nonfarm percent and level benchmark revisions. Over the prior ten years, the annual benchmark revision at the total nonfarm level has averaged 0.2 percent (in absolute terms), with a range of −0.7 percent to 0.3 percent.
The differences listed in table 4 reflect the error due to normal benchmarking procedures after the incorporation of reconstructions. Those years are footnoted.
Table 4. Percent and level differences between nonfarm employment benchmarks and estimates by industry supersector, March 2008 – 2018 (in thousands)
(1) With the 2010 benchmark, CES reconstructed historical national employment levels of all employees for other federal government (91-999900) to reflect corrections to initial counts for Census temporary and intermittent workers for the 2010 Census. The reconstructions resulted in about 4,000 in employment moved out of other federal government (91-999900). For more information, see the Reconstructions section in the 2010 CES Benchmark Article.
(2) A review of industries for the possible presence of noncovered employment yielded 13 additional industries. As a result of including these industries, employment in the amount of 95,000 was added to the benchmark nonfarm level. For more information, see the Changes to noncovered employment section in the 2011 CES Benchmark Article.
(3) The percent and level differences in this column reflect reconstructions to series within CES supersectors financial activities and education and healthcare services. Each first quarter, the Quarterly Census of Employment and Wages (QCEW) program, whose data account for approximately 97 percent of the CES universe scope (see The Sample section of the CES Technical Notes), incorporates updated industry assignments. In 2013, these updates included two substantial groups of nonrandom, noneconomic code changes, one to funds, trusts, and other financial vehicles (NAICS 525), and the other, a reclassification of approximately 466,000 in employment from private households (NAICS 814), which is out of scope for CES, to services for the elderly and persons with disabilities (NAICS 62412), which is in scope. These changes also had an impact, beyond what would be considered typical for a given benchmark year, on corresponding CES series. For more information about the changes to these industries, see the QCEW First Quarter 2013 News Release or the Special notice regarding reconstructed data section in the 2013 CES Benchmark Article.
(4) With the 2015 benchmark, CES reconstructed the national employment series 65-624120, services for the elderly and persons with disabilities back to January 2000. CES previously reconstructed this series with the 2013 benchmark; however, between the 2013 and 2015 benchmark, a better source of information for the employment within NAICS 62412 for the state of California was found. The inclusion of the reconstructed series resulted in total nonfarm and total private employment that was 27,000 less than the originally published March 2015 estimate level. The difference between the benchmarked and originally published March 2015 estimate level is -199,000 or -0.1 percent. This table displays March 2015 data after accounting for the decrease of 27,000 from the reconstructed series. Similarly, for the education and health services supersector, this table displays March 2015 data after incorporating the reconstructed series. For more information, see the Reconstructions section in the 2015 CES Benchmark Article.
(5) With the 2017 benchmark, CES reconstructed the national employment series 60-561613, security guards and patrols and armored car services back to October 2016 to correct a microdata error. The inclusion of the reconstructed series resulted in total nonfarm and total private employment that was 3,000 more than the originally published March 2017 estimate level. The difference between the benchmarked and originally published March 2017 estimate level is 138,000 or 0.1 percent. This table displays March 2017 data after accounting for the increase of 3,000 from the reconstructed series. Similarly, for the professional and business services supersector, this table displays March 2017 data after incorporating the reconstructed series. For more information, see the Reconstructions section in the 2017 CES Benchmark Article.
(6) With the 2018 benchmark, CES reconstructed several national employment series. A recoding effort in the QCEW resulted in about 336,000 in employment in wholesale trade agents and brokers (41-425120) moving into other series within the wholesale trade, retail trade, transportation and warehousing, and professional and business services major industry sectors. Affected basic series were reconstructed for their entire history, generally back to January 1990. Additionally, a reclassification of a state employer to private ownership caused a shift of about 17,000 in employment from the CES series other state government (90-922999) into services for the elderly and persons with disabilities (65-624120). Affected basic series were reconstructed from March 2018 back to January 2018. For more information, see the Reconstructions section in the 2018 CES Benchmark Article.
(7) Absolute revision is less than 0.05 percent.
(8) Indented series are part of trade, transportation, and utilities.
The routine benchmarking process results in revisions to the series for production and nonsupervisory employees (PE) and women employees (WE). There are no benchmark employment levels for these series; they are revised by preserving ratios of employment for the particular data type to the all employee (AE) level prior to benchmarking, and then applying these ratios to the revised all employee level. These figures are calculated at the basic cell level and then aggregated to produce the summary estimates. Average weekly hours (AWH), average hourly earnings (AHE), and in manufacturing industries, average weekly overtime hours (AWOH) are not benchmarked; they are estimated solely from reports supplied by survey respondents at the basic estimating cell level. New employment benchmarks can additionally affect indirectly estimated data types. For more information on indirectly estimated data types, see the Available Data section in the CES Technical Notes.
Table 5 lists directly estimated data types and their common abbreviations. Directly estimated data types except for AE are collectively called non-AE data types.
Table 5. Directly estimated data types
Data Type
Abbreviation
All employees
AE
Production and nonsupervisory employees
PE
Women employees
WE
Average weekly hours of all employees
AE AWH
Average hourly earnings of all employees
AE AWE
Average weekly overtime hours of all employees
AE AWOH
Average weekly hours of production and nonsupervisory employees
PE AWH
Average hourly earnings of production and nonsupervisory employees
PE AWE
Average weekly overtime hours of production and nonsupervisory employees
The aggregate industry levels of the hours and earnings series are derived as a weighted average. AE and PE estimates for basic cells act as weights for their respective hours and earnings estimates for broader industry groupings. Adjustments of AE estimates to new benchmarks may alter the implicit weights used for both AE and PE hours and earnings, which, in turn, may change the estimates for both AE and PE hours and earnings at higher levels of aggregation.
Generally, new employment benchmarks have little effect on hours and earnings estimates for major industry groupings. To influence the hours and earnings estimates of a broader industry group, employment revisions have to be relatively large and must affect industries that have hours or earnings averages that are substantially different from those of other industries in their broader group.
Table 6 and table 7 provide information on the not seasonally adjusted levels of major industry sector hours and earnings series resulting from the March 2018 benchmark. At the total private level, there was no change in average weekly hours estimates for both AE and PE from the previously published level. Total private average hourly earnings increased by 4 cents for AE and increased by 3 cents for PE from the previously published level.
(1) With the 2018 benchmark, CES reconstructed several series. The effects of these reconstructions are included in this table. For more information, see the Reconstructions section in the 2018 CES Benchmark Article.
Table 7. Effect of March 2018 benchmark revisions to production employees average weekly hours and average hourly earnings estimates, major industry sectors
(1) With the 2018 benchmark, CES reconstructed several series. The effects of these reconstructions are included in this table. For more information, see the Reconstructions section in the 2018 CES Benchmark Article.
The CES sample alone is not sufficient for estimating the total nonfarm employment level because each month new firms generate employment that cannot be captured through the sample. There is an unavoidable lag between a firm opening for business and its appearance on the CES sample frame. The sample frame is built from Unemployment Insurance (UI) quarterly tax records. These records cover virtually all U.S. employers and include business births, but they only become available for updating the CES sampling frame 7 to 9 months after the reference month. After the births appear on the frame, there is also time required for sampling, contacting, and soliciting cooperation from the firm, and verifying the initial data provided. In practice, BLS cannot sample and begin to collect data from new firms until they are at least a year old.
BLS has researched both sample-based and model-based approaches to measuring birth/death units that have not yet appeared on the UI universe frame. The research demonstrated that sampling for births was not feasible in the very short CES production timeframes, so the Bureau uses a model-based approach to account for this employment. This model incorporates two components; the first component is an indirect imputation for business deaths and the second component is an Autoregressive Integrated Moving Average (ARIMA) time series model designed to estimate the net birth/death employment not accounted for by the imputation from the first component. More information on the CES birth/death model is available in the Birth/Death Model section of the CES Technical Notes.
Forecasted vs. actual net birth/death
Only error from the second component is directly measurable. Error from this component is measured by comparing the actual net of births and deaths with the model-based forecast that was used in the CES sample-based estimates. Most recently, the data from April 2017 to March 2018 can be measured. As table 8 shows, the actual net birth/death for April 2017 to March 2018 was approximately 65,000 below the forecasted amount used in the CES monthly estimates for the same period.
Table 8. Differences between forecasted and actual net birth/death, total private employment, April 2017 – March 2018 (in thousands)
Net birth/death adjustments to the post-benchmark period
From April 2018 to December 2018, also known as the post-benchmark period, estimates were recalculated for each month based primarily on new benchmark levels and new net birth/death forecasts. Net birth/death forecasts were revised to incorporate information from the most recent year of universe employment counts. Table 9 shows the net birth/death values for the supersectors over the post-benchmark period. From April 2018 to December 2018, the net birth/death model cumulatively added 1,055,000, compared with 963,000 in the previously published April 2018 to December 2018 employment estimates.
Table 9. Net birth/death forecasts by industry supersector, April to December 2018 (in thousands)
In addition to the regular benchmark revisions, CES employment, hours, and earnings estimates are sometimes reconstructed in order to avoid series breaks and to provide users with continuous, comparable employment time series suitable for economic analysis. Reconstructions may be necessary to correct processing errors, reporting errors, changes in scope, or to address other issues that may cause breaks in CES time series that are not economic in nature. The reconstructions that were incorporated with this year’s annual benchmark are described below.
Wholesale trade recoding
During processing for the 2018 benchmark, a QCEW recoding effort for the first quarter of 2018 moved a significant amount of employment out of NAICS 425120. For CES purposes, this resulted in about 336,000 in employment in wholesale trade agents and brokers (41-425120) moving into other series within the wholesale trade, retail trade, transportation and warehousing, and professional and business services major industry sectors. This amounted to approximately 38.8 percent of the previously published March 2018 employment level of wholesale trade agents and brokers. Affected basic series were reconstructed for their entire histories, generally back to January 1990. Series for AE, PE, WE, PE AWH, and PE AHE were reconstructed back to January 1990 for all affected basic and independent series. AE AWH and AE AHE were reconstructed back to the start of those data types, March of 2006. A complete list of affected basic and independent series is available in exhibit 1. Basic series are those at the most detailed industry level that get aggregated into other series levels; independent series are detailed series estimated independently of aggregation. For more information about CES aggregation procedures, see the Aggregation Procedures section in the CES Technical Notes.
Some of the changes associated with the recoding of NAICS 425120 affected levels of NAICS detail not published by CES; therefore, only cases where CES industries were affected are discussed in detail here. Reconstructions were applied to not seasonally adjusted CES data.
Basic-level reconstruction methods
The reconstruction of data from wholesale trade agents and brokers redistributes a single NAICS (425120) that made up the entire CES industry wholesale trade agents and brokers (41-425120) to many NAICS that span industries in the CES major industry sectors wholesale trade, retail trade, transportation and warehousing, and professional and business services. Employment in wholesale trade agents and brokers was split into 69 basic series, including the portion that remained in the original series, and 6 independent series.
Ratios of employment moving out of wholesale trade agents and brokers and into other basic-level CES series were created from QCEW data for March 2018 and used to reconstruct each series. Ratios were held constant and applied to each monthly CES employment estimate for the entire history of the series, usually back to 1990, to determine the amount of employment moving from wholesale trade agents and brokers to a different CES industry. Average weekly hours and average hourly earnings were recalculated by applying the same ratios to the amount of old total hours and old total payroll to move the correct amount of hours or earnings into the new CES industry. Total hours and total earnings summed for the new industries were then averaged using the new employment levels. More information about the calculation of CES employment, hours, and earnings estimates is available in the CES Technical Notes under Estimation Methods.
Exhibit 1 shows the ratios that were applied to 41-425120 to calculate proportions of employment, total hours, and total earnings moving to other series. AE series are sometimes published at a more detailed level than non-AE series. Furthermore, aggregate-level series for AE may be basic series for non-AE datatypes, as noted in exhibit 1.
Exhibit 1. Ratios used to move employment out of wholesale trade agents and brokers and into other industries
Reconstructed basic series were summed to aggregate levels using the normal CES aggregation methods (see the Aggregation Procedures section in the CES Technical Notes), and, consequently, unadjusted aggregate employment series are subject to change from the detailed aggregate series to the major industry sector levels for trade, transportation, and utilities and professional and business services. Hours and earnings series are subject to change from detailed aggregate series to aggregate major industry sectors for private service-providing, service-providing, total private, and total nonfarm. Not seasonally adjusted employment for private service-providing, service-providing, total private, and total nonfarm are unaffected.
Reconstruction methods for series that go back further than 1990
For industries that have series with earlier data than January 1990, ratios at the major industry level were created from summed reconstructed data and the originally published estimates. These pre-1990 ratios were calculated for employment and total earnings using January 1990 data for all series in exhibit 2 and applied to the pre-1990 histories for these series. Pre-1990 PE AWH did not change, but the historical PE AWH data was used to calculate PE AHE changes for pre-1990 data, which had minor changes between 1 and -2 cents for some months.
Exhibit 2. Start years of pre-1990 industries affected by recoding wholesale trade agents and brokers
For all series affected by the redistribution of employment, hours, and earnings from wholesale trade agents and brokers into other industries, the corresponding seasonally adjusted series were also subject to change for the entirety of their histories. Aggregate industries as well as component series used as indirect seasonal adjustment are also subject to change, up to and including total private and total nonfarm. More information about seasonal adjustment of CES series is available in the CES Technical Notes under Seasonal Adjustment.
A reclassification of some employers from state to private ownership caused a shift in employment of about 17,000 from the CES series other state government (90-922999) into services for the elderly and persons with disabilities (65-624120). This shift amounted to approximately 3.5 percent of the previously published March 2018 employment level of other state government. The change occurred in January 2018, necessitating a reconstruction of 3 months of CES data for January, February, and March of 2018. Exhibit 3 lists data types affected by the change in ownership. Changes include AE, PE, and WE in services for the elderly and persons with disabilities and AE and WE in other state government. Hours and earnings are not applicable for government series, so the hours and earnings for services for the elderly and persons with disabilities were not affected by the change at the most detailed industry level. Changes to employment at the most detailed level, however, can affect the hours and earnings at aggregate levels.
Exhibit 3. CES series involved in state government reconstructions
CES Industry Code
CES Industry Title
Data Type(s) Involved
Time Span
65-624120
Services for the elderly and persons with disabilities
A ratio of QCEW employment using the new ownership total to QCEW employment using the old ownership total for other state government for March 2018 was calculated. The ratio was applied to CES employment for other state government for January, February, and March of 2018 for AE and WE to determine the amount remaining in those series. The difference between the old employment and new employment for each month was added to the corresponding month for services for the elderly and persons with disabilities. For PE, the original sample ratios used to calculate PE in services for the elderly and persons with disabilities were applied to the new totals for AE.
Aggregate-level reconstruction methods
The previous employment was maintained at total nonfarm (00-000000), but aggregate levels were affected by the change in ownership. Exhibit 4 lists the affected aggregate industries. As with usual processing, 5 years of data are subject to seasonal adjustment during benchmark processing.
Exhibit 4. Aggregate series affected by the change in ownership in other government services
Aggregation processing error for production and nonsupervisory employees hours and earnings estimates
During the 2018 benchmark, a processing error was found that caused both PE AWH and PE AHE to aggregate incorrectly, specifically between 1947 and 1989 for the total private, goods-producing, and private service-providing sectors. The maximum absolute differences in aggregation totals were 0.1 hours for PE AWH and 2 cents for PE AHE. These were corrected by re-aggregating all hours and earnings estimates between 1947 and 1989.
With the release of the January 2019 first preliminary estimates on February 1, 2019, CES incorporated series changes related to annual sample adequacy and disclosure review.
Series changes due to annual sample review
All CES series are evaluated annually for sample size, coverage, and response rates. The following changes result from a re-evaluation of the sample and universe coverage for CES industries, which are based on the 2017 North American Industry Classification System (NAICS). Some industries no longer have sufficient sample to be estimated and published separately and were discontinued or combined with other similar industries for estimation and publication purposes. This information is also available at www.bls.gov/web/empsit/cesnewseries.htm.
Some series have new CES industry codes or titles as a result of the series changes (see table 10). These CES industry code or title changes have been applied to all data types published for the designated series. Historical data for these series with new CES industry codes or CES industry titles are affected as noted on the remainder of this page; historical data are available under the new CES industry codes or new CES industry titles.
Table 10. Series with CES Industry Code or Title Changes
NAICS Code
Previous
New
CES Industry Code
CES Industry Title
CES Industry Code
CES Industry Title
335312,3,4
31-335314
Switchgears, switchboard apparatus, relays, and industrial controls
31-335314
Motors and generators, switchgears and switchboard apparatus, and relay and industrial controls
Table 11 through table 16 show the new CES industry codes and titles, not the previous CES industry codes and titles, as noted in table 10.
In order to more easily identify affected series and because AE series are published at a more detailed industry level than non-AE series, series changes are shown for AE and non-AE data types. The first three tables reference the AE data type and the second three tables reference all non-AE data types. The three tables in each group display the discontinued, collapsed, and new series. Discontinued series tables (table 11 and table 14) display series for which the data types noted are no longer published. Collapsed series tables (table 12 and table 15) display series for which the data types noted are no longer published because the industry no longer has sufficient sample to be estimated and published separately. Affected industries have been combined with other similar industries for estimation and publication purposes. Historical data for these series were reconstructed to provide consistent time series. Series that no longer publish AE will also no longer publish other directly estimated datatypes or derivative series. New series tables (table 13 and table 16) display series for which the data types noted are now published.
Table 11. Discontinued all employees series
NAICS Code
CES Industry Code
CES Industry Title
Next Highest Published Industry
32311
32-323110
Printing
Printing and related support activities (32-323000)
323113
32-323113
Commercial screen printing
Printing and related support activities (32-323000)
32312
32-323120
Support activities for printing
Printing and related support activities (32-323000)
LABSTAT, the BLS public database, contains all historical employment, hours, and earnings data revised as a result of this benchmark, including both not seasonally adjusted and seasonally adjusted data. The data can be accessed at www.bls.gov/ces/data, the CES National Database page.
Previously published data are available on both a not seasonally adjusted and seasonally adjusted basis for all CES industries down to the 3-digit level from the CES Vintage Data. CES vintage data are typically updated in late February following the annual benchmark revision.
