Bureau of Labor Statistics
Technical information: Revisions to CES data for late sample reports, annual benchmarking, and other factorsBackground The Current Employment Statistics (CES) program (also known as the payroll survey or the establishment survey) is designed to measure trends in employment, hours and earnings by industry. Each month BLS surveys approximately 150,000 businesses and government agencies representing approximately 390,000 worksites throughout the United States. The monthly Employment Situation news release provides national CES data on employment, hours, and earnings, as well as labor force and unemployment estimates from the Current Population Survey (CPS), also known as the household survey. The Employment Situation is typically released on the first Friday of the month following the reference month. For example, the February 2, 2007, Employment Situation published CES first preliminary employment estimates for January 2007. CES estimates represent information reported by survey respondents for their pay periods that include the 12th of the month. Not all sampled firms are able to report their data in time to be used in the first preliminary estimates. Therefore, BLS continues to collect sample responses after the release of first preliminary estimates for incorporation into second preliminary and final sample-based estimates. Second preliminary estimates for a reference month are published the month following the initial release, and final sample-based estimates are published 2 months after the initial release. Sample-based estimates remain final until employment levels are reset to universe employment counts or “benchmarks” for March of each year; the benchmarks are primarily derived from Unemployment Insurance tax records. The annual benchmarking process results in revised data back to the last annual benchmark for not seasonally adjusted series and back 5 years for seasonally adjusted series. CES data are principal economic indicators and serve as input to many other economic series. Each time the CES employment estimates are revised, additional information that was not previously available is incorporated into the estimates. Below are discussions of how CES data revise: > Monthly, to include late sample reports; > Annually, due to benchmarking; and > Irregularly, to avoid discontinuities while incorporating methodological changes Monthly Revisions to Include Late Sample Reports CES data users typically are most concerned with revisions to over-the-month changes. This section profiles these monthly revisions of CES seasonally adjusted over-the-month changes and the sample collection rates that underlie the revisions. Revisions to CES over-the-month changes are calculated by comparing each month’s second preliminary over-the-month change to the first preliminary over-the-month change, the final sample-based over-the month change with the second preliminary over-the-month change, and the final sample-based over-the-month change to the first preliminary over-the-month change. See www.bls.gov/web/cesnaicsrev.htm for a table of revisions to seasonally adjusted total nonfarm over-the-month changes from January 1979 forward. The monthly employment change figures shown in the table do not reflect subsequent changes due to the introduction of benchmark revisions, seasonal adjustment, or other updates. Mean revisions and mean absolute revisions for each calendar year are included in the table. Mean absolute revisions indicate the overall magnitude of change to the estimates, while the mean revisions are a measure of whether there is a bias in direction of the revisions. The closer the mean revision is to zero, the less indication that revisions are predominantly either upward or downward. For example, if in a given year there were six upward revisions of 50,000 and six downward revisions of 50,000, the mean revision would be zero; however, the mean absolute revision would be 50,000. BLS begins collecting sample reports for a reference month as soon as the reference period, the establishment’s pay period that includes the 12th of the month, is complete. Collection time available for first preliminary estimates ranges from 9 to 15 days, depending on the scheduled date for the Employment Situation news release. The Employment Situation is scheduled for the third Friday following the week including the 12th of the prior month, with an exception for January. (For January, the news release is delayed a week if the third Friday following the week of the 12th occurs on January 1st, 2nd, or 3rd.) Given this short collection cycle for the first preliminary estimates, many establishments are not able to provide their payroll information in time to be included in these estimates. Therefore, CES sample responses for the reference month continue to be collected for two more months and are incorporated into the second preliminary and final sample-based estimates published in subsequent months. Additional sample receipts are the primary source of the monthly CES employment revisions. Prior to 1991, most of the CES sample was collected by mail in a decentralized environment, by each State Workforce Agency. BLS has gradually centralized collection and adopted automated sample collection methods with the result that collection rates have gradually risen over time. BLS uses a variety of collection techniques, tailored to individual firm preferences to encourage participation in this voluntary survey. Computer Assisted Telephone Interviewing (CATI) is used for initial enrollment and collection of sample. Sample units are often transferred later to an automated method of self-reporting, such as through Fax, Internet, or Touchtone Data Entry (TDE). Many large, multi-establishment firms report through Electronic Data Interchange (EDI); the firms provide electronic files to BLS that include all of their worksites. A small percentage of sample units still report via mail. Collection rates are defined as the percent of reports received for a monthly estimate compared to the total number of actively-reporting sample units on the sample registry. CES collection rates back to 1979 can be found on www.bls.gov/web/cesregrec.htm. Much of the month to month variation in the first preliminary collection rates is a function of the number of collection days in the individual months. The overall upward trend over time is attributable to replacing decentralized mail collection with automated techniques. Benchmark Revisions Annual CES benchmark revisions are published along with January first preliminary estimates in February of each year. Benchmark revisions reflect a re-anchoring of CES sample-based estimates to incorporate near universe counts of employment. These comprehensive counts of employment, or benchmarks, are derived primarily from employment counts reported on unemployment insurance (UI) tax reports that nearly all employers are required to file with State Workforce Agencies. The benchmark revision is the difference between the universe count of employment for March and its corresponding sample-based estimate. A table of benchmark revisions from 1979 forward:
* less than 0.05 percent Each annual benchmark revision affects 21 months of data for not seasonally adjusted series and 5 years of data for seasonally adjusted series as described below. BLS revises employment from the current March benchmark month back to the previous year’s March benchmark using a simple linear wedge procedure. Revised estimates for post-benchmark months are derived by applying the previously calculated over-the-month sample changes to the revised March levels for April through October. Revised business birth/death estimates also are incorporated into the post benchmark months. Additionally, seasonal adjustment models are rerun, and seasonally adjusted estimates are replaced for 5 years back. November and December estimates revise due to both impacts of benchmarking and additional sample. Additionally new sample units are rotated into the survey starting with November. As an example of benchmark effects, the March 2006 benchmark revisions (published in February 2007) resulted in revised series from April 2005 through December 2006 on a not seasonally adjusted basis and revised series from January 2002 through December 2006 on a seasonally adjusted basis. See www.bls.gov/web/cesbmart.htm for more details on the benchmarking process. Historical time series reconstructions to avoid discontinuities when incorporating methodological changes Beyond the monthly revisions and the benchmark revisions, CES employment estimates have been reconstructed several times in order to avoid series breaks and provide users with continuous, comparable employment time series suitable for economic analysis. The major reconstruction efforts are briefly described below. Improvement to seasonal adjustment methodology - With the release of the 1995 benchmark revision (in June 1996), BLS refined its seasonal adjustment procedures to control for survey interval variations, sometimes referred to as the 4- versus 5-week effect. This improvement mitigated the effects that a variable number of weeks between surveys had on the measurement of employment change, thus improving the measurement of true economic trends. At that time, data for 1988 forward were revised to incorporate this new methodology. CES Sample redesign Over a 4-year period, BLS introduced a new probability-based sample design; it replaced an outmoded and less scientific quota sample-based design. The new design was phased in by major industry division with the June 2000 through June 2003 benchmark releases. As each industry was phased in, the post-benchmark estimates for that year were affected by the new sample composition.
Industry reclassification Along with the last phase of the probability sample conversion, BLS also made the switch from producing estimates based on the Standard Industrial Classification (SIC) system to the North American Industry Classification System (NAICS). All historical employment estimates were reconstructed at that time. All NAICS-based estimates were derived from the sample from 2002 forward, while earlier history was reconstructed from SIC-based estimates using a ratio technique. See www.bls.gov/ces/cesnaics.htm and www.bls.gov/opub/mlr/2003/06/art1full.pdf for more information on the conversion to NAICS. Prior to NAICS, the SIC structure had been updated periodically. CES data were updated from the 1972 SIC classification to the 1987 SIC classification in August 1990. Other factors contributing to revisions Over the time period covered by the revision and collection rate tables referenced above, CES has introduced many program improvements; some of these affect the revision patterns observed over time. Monthly Revisions – As noted above, the overall magnitude of these revisions has trended down over time mainly due to automated and improved data collection techniques which raised the collection rates for the first and second preliminary estimates. Other factors of note include: Timing of benchmark revisions -Between 1980 and 2003, annual benchmark revision updates were introduced in June of each year, concurrent with the March final sample-based estimates and the April second preliminary estimates. The monthly revisions for March and April for these years were often larger than for other months, because the March final and April second preliminary estimates were incorporating not only additional sample but also other benchmark-related changes updates. - Beginning with the 2004 benchmark revision (published in 2005), BLS reduced the time required to produce the annual revisions by four months and thus began publishing benchmark revisions in February rather than June. Therefore from 2004 forward, the November final and December second preliminary estimates are affected by benchmark revision updates, rather than the March final and April second preliminary estimates. Timing of seasonal adjustment updates - Between 1980 and June 1996 seasonal factors were updated on an annual basis along with the benchmark revisions. Thus March final and April second preliminary were affected by the re-computation of seasonal factors as well as other benchmarking procedures and additional sample receipts. - Between November 1996 and November 2002, BLS updated seasonal factors on a semi-annual basis, meaning that September final and October second preliminary estimates as well as March final and April second preliminary revisions were affected by seasonal factor updates. - Since June 2003 the CES program has used a concurrent seasonal adjustment procedure, meaning that seasonal adjustment is rerun every month using all available months of estimates including the month currently being estimated for first preliminary. This technique yields the best possible seasonal adjustment for the current month and reduces benchmark revisions to over-the-month changes. In the application of the concurrent procedure, the previous two months are revised to incorporate not only additional sample receipts but also new seasonal factors. Thus there are no longer individual months that are more affected than others by seasonal factor updates. However this practice does mean that revisions from second preliminary to final sample-based estimates for each month are affected by the CES replacement policy. Because CES revises only two months of estimates each month, the fourth month back from the current first preliminary estimate is adjusted using a different set of seasonal factors than the third month back. For example, with the release of October first preliminary data, factors are revised for September and August, but not July. Change in estimation technique/sample re-sizing Up through March 2002 benchmark published in June 2003, CES estimating cells were stratified by industry and then by employment size. All sample units were refit to size classes based on their number of employees with each benchmark revision, a practice known as resizing. Resizing often resulted in large revisions between estimates between the March final and April second preliminary estimates. New benchmarks were introduced at the same time, making it difficult to separate out the effects of re-sizing, new seasonal adjustment factors, and newly benchmarked data. Beginning with the March 2003 benchmark, CES changed to an improved sample design which no longer required the use of size class estimation cells or annual re-sizing.
Last Modified Date: March 28, 2008 |
Tools |
Calculators |
Help |
Info |
