Implementation of Outlier Treatment Improvement
Since the onset of the coronavirus pandemic, BLS has been examining the inputs to the employment and unemployment models for outliers and implementing level shifts in real time, where appropriate, based on statistical evaluation of movements in the inputs. These level shifts help to preserve movements in the published estimates that the models otherwise would have discounted. The implementation of a level shift due to an outlier identified in the Current Population Survey input to the Detroit-Warren-Dearborn, MI Metropolitan Statistical Area unemployment model for January 2021 produced some distortions in the pro-rata benchmarking factors across the modeled areas within the East North Central census division that have increased over subsequent months. The modeled areas within the East North Central census division include the states of Illinois, Indiana, Michigan, Ohio, and Wisconsin, as well as the Chicago-Naperville-Arlington Heights, IL Metropolitan Division and the balance of Illinois, the Detroit-Warren-Dearborn, MI Metropolitan Statistical Area and the balance of Michigan, and the Cleveland-Elyria, OH Metropolitan Statistical Area and the balance of Ohio.
BLS has modified the outlier intervention effective the October production cycle to reduce these distortions for the September revised and October preliminary estimates. This modification resulted in estimates that better reflect the model inputs for these most recent two months. Distortions to data for earlier months in the year will be addressed during the upcoming annual revision cycle for 2021. BLS expects to issue the 2021 revisions for model-based areas in early March of 2022. Until that time, data users are encouraged to view the year-to-date changes through September revised or later as more accurately reflective of recent labor market behavior (when seasonally-adjusted data are used) or over-the-year changes through September revised or later (when not-seasonally-adjusted data are used).
Both seasonally-adjusted and not-seasonally-adjusted unemployment data for modeled areas were impacted. Data for non-modeled substate areas are controlled, or forced to sum, to the not-seasonally-adjusted totals for their respective model-based areas. Hence, distortions across the modeled areas within the East North Central census division attributable to the January level shift were transmitted proportionally to the non-modeled substate areas through unemployment additivity adjustment up through the September preliminary estimates. By extension, the modification to the outlier intervention effective the October production cycle reduced distortions for the September revised and October preliminary estimates for non-modeled substate areas in proportion to the mitigation for their respective model-based areas’ not-seasonally-adjusted unemployment estimates. Similar to the plan for model-based areas noted above, distortions to data for non-modeled substate areas for prior months in the year will be addressed during the upcoming annual revision cycle, with the 2021 revisions for non-modeled substate areas expected to be issued by BLS in mid-April of 2022.
BLS is scheduled to issue its State Employment and Unemployment news release for October 2021 on Friday, November 19, 2021, at 10:00 AM (Eastern Time). Through the federal-state cooperative partnership, state workforce agencies have flexibility to schedule data releases for their own areas at times different from the corresponding BLS issuance. For the affected states of the East North Central census division, the state workforce agency release dates and times for the October 2021 preliminary and September 2021 revised statewide estimates are scheduled as follows:
Each year, historical estimates from the Local Area Unemployment Statistics (LAUS) program are revised to reflect new population controls from the Census Bureau, updated input data, and reestimation. The data for model-based areas also incorporate new seasonal adjustment, and the unadjusted estimates are controlled to new census division and U.S. totals. Substate area data subsequently are revised to incorporate updated inputs, reestimation, and controlling to new statewide totals.
On March 3, 2021, the LAUS program issued 2020 annual-average civilian labor force and unemployment estimates for census regions and divisions; all states and the District of Columbia; 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, as well as the seven corresponding balance-of-state areas. At the same time, historical data for these areas were replaced back to their series beginnings based on a new generation of time-series models. The revisions to model-based data for states and substate areas in 2019 and 2020 also incorporated updated estimation inputs, while the revisions to data for all model-based areas from April 2010 forward also reflected new population controls from the U.S. Census Bureau. Both not-seasonally-adjusted and smoothed-seasonally-adjusted (SSA) monthly data for these areas were affected.
Statewide data for Puerto Rico (which are not model-based, but rather are tabulated from a household survey similar to the Current Population Survey for the United States) were revised from January 2018 forward. The not-seasonally-adjusted data reflect the incorporation of new population controls, while the SSA data reflect seasonal adjustment and smoothing of the recontrolled not-seasonally-adjusted data.
On March 19, 2021, data for geographic areas below the state level (other than the model-based areas noted above) were revised from their series beginnings through 2009. These revisions consisted exclusively of controlling to the new model-based totals described above.
On April 16, 2021, revisions were made to data from 2010 through 2020 for geographic areas below the state level (other than the model-based areas noted above), and official annual averages for 2020 were issued. For substate areas, estimation inputs generally were revised back to 2019, while the revisions for 2010–18 largely consisted of controlling to the new model-based totals described above. However, some states had estimation input revisions as far back as January 2010 for their substate areas.
Puerto Rico was not able to conduct normal data collection for its household survey in March or April 2020. Hence, neither statewide nor local area data were published for those two months. Annual-average data for Puerto Rico and its local areas in 2020 cannot be calculated in the absence of the March and April data.
On March 19, 2021, model-based data for Michigan; Detroit-Warren-Dearborn, MI Metropolitan Statistical Area; and Balance of Michigan, state less Detroit-Warren-Dearborn MSA were corrected to reflect outlier effects in both the employment and unemployment inputs for July 1998. Both seasonally-adjusted and not-seasonally-adjusted data for these three areas were affected, and the 1998 annual averages were re-calculated to reflect the corrections to the not-seasonally-adjusted estimates for July.
Last Modified Date: November 16, 2021