Important Information on Revisions to Data for Model-Based Areas in 2022
The Local Area Unemployment Statistics (LAUS) program will be issuing revised estimates for model-based areas from January 2017 to December 2021 and publishing official annual averages for 2021 in conjunction with the Regional and State Unemployment—2021 Annual Averages news release on March 2, 2022. (The model-based areas consist of the census regions and divisions, all states and the District of Columbia, and the following seven substate areas, along with their corresponding balance-of-state areas: the Los Angeles-Long Beach-Glendale, CA Metropolitan Division; the Miami-Miami Beach-Kendall, FL Metropolitan Division; the Chicago-Naperville-Arlington Heights, IL Metropolitan Division; the Detroit-Warren-Dearborn, MI Metropolitan Statistical Area; New York city, NY; the Cleveland-Elyria, OH Metropolitan Statistical Area; and the Seattle-Bellevue-Everett, WA Metropolitan Division.) Revisions to model-based data for at least the most recent five years typically are issued by the LAUS program in early March. Some notable features of the upcoming model revision cycle are as follows:
Implementation of "blended base" population controls from April 2020 forward
Each year, the Census Bureau provides BLS with updated monthly series of civilian noninstitutional population ages 16 and older (CNP16+) back to the decennial estimates base period. These series are used to recontrol the primary Current Population Survey (CPS) inputs to the time-series models for state estimation of employed and unemployed people. Annual recontrolling involves simple ratio adjustment, applying the ratio of the new CNP16+ estimate to the prior CNP16+ estimate to both the employment and unemployment levels from the CPS for the given month prior to model re-estimation. From 2012 to 2021, the monthly CNP16+ recontrol series provided by the Census Bureau were extrapolated from the 2010 Census, with the estimates base period having been April 2010. In 2022, the recontrol series reflect a "blended base," using population totals from the 2020 Census but demographic distributions still based on the 2010 Census. These blended-base series begin in April 2020. (Information on the underlying methodology for the blended base is available from the Census Bureau at www.census.gov/data/academy/webinars/2021/methodology-updates-for-the-vintage-2021-estimates.html.
Use of wedged population recontrols for the 2010s
Changes in the decennial census basis for the CNP16+ recontrol series often entail level breaks due to "closing error," which is defined as the difference between the final, post-censal estimates for April of the new decennial base year as extrapolated 10 years out from the prior census and the enumerated data from the latest census. The rates of closing error for the final 2010-extrapolated CNP16+ series relative to the new blended-base series range from -3.5 percent to +6.1 percent across model-based areas. These would manifest as some notable discontinuities in CNP16+ levels between March and April of 2020, which then would be transmitted proportionally to the CPS inputs to the state employment and unemployment models. This period coincides with the peak impact of the coronavirus pandemic (COVID-19) on the labor market. In the absence of intervention, BLS would have introduced non-economic discontinuities stemming from the closing error of the population recontrol series on top of economic discontinuities stemming from COVID-19. In some cases, the ones stemming from the closing error would have been substantially offsetting with respect to the ones stemming from the pandemic. This would have been untenable for state model re-estimation in 2022.
To mitigate the impact of the closing error on model re-estimation, BLS implemented a wedged CNP16+ series for the 2010s. This involved walking the April 2020 closing error forward cumulatively in even increments starting from the April 2010 base period of the final 2010 Census-extrapolated series that was implemented in early 2021. As the scope of model revision for the current annual processing cycle is 2017–21, some breaks in the CNP16+, civilian labor force, employed, and unemployed levels will be evident between December 2016 and January 2017. (The relatively larger the April 2020 closing error, the more conspicuous the 2016–17 level breaks.) However, unemployment rate, employment-population ratio, and labor force participation rate series should remain consistent through the point of splicing. The temporary level breaks between December 2016 and January 2017 are preferable to allowing non-economic changes in the underlying population control series to complicate model re-estimation at the onset of the pandemic.
Note that this use of wedged population recontrols for the 2010s through March 2020 is intended as a temporary measure. They will remain in use until such time as the Census Bureau can provide BLS with an intercensal state recontrol series for the decade, which is anticipated sometime in 2024.
Incorporation of outlier effects into benchmarking
With the implementation of the fifth generation of LAUS time-series models in early 2021, real-time hierarchical benchmarking to the national estimates of employment and unemployment reverted from a model-based approach to a pro-rata approach. However, effects of outliers continued to be incorporated outside of the benchmarking process. This has been determined to have produced some distorting effects in census divisions where there are substantial differences across states in the extent of the recovery from the pandemic through a given point in time. To reduce these distortions, outlier effects will be incorporated into the real-time benchmarking process with the annual processing revisions for 2017–21 to be issued on March 2, 2022.
Updates to pandemic outlier specifications
New pandemic outlier specifications were developed, covering the period from March 2020 to December 2021. Preference was given to new outlier models that replaced level shifts with temporary change regressions. Any additive outliers that had been identified in 2021 were excluded from the seasonally adjusted series.
Discontinuation of real-time outlier detection in 2022
Effective with the release of the January 2022 preliminary estimates for states on March 14, 2022, real-time implementation of level-shift outliers in the employment and/or unemployment inputs to the models will be discontinued. These real-time level shifts were introduced in 2020 in response to the pandemic to preserve movements in the published estimates that the models otherwise would have discounted. BLS has determined that these interventions are no longer necessary on a real-time basis. Outlier detection and treatment will revert to retrospective activities conducted during annual processing.
Last Modified Date: February 18, 2022