United States Department of Labor
If things were good enough yesterday, why would we change them today? Good enough is OK for folding laundry, cleaning the junk drawer, and raking leaves, but not for official statistics from BLS. We do our best to provide a timely look at the labor market and economy, but we often learn more after we publish those initial data. As a result, we sometimes revise our statistics. That’s mostly a good thing, but there is a fine line between the frequency of revisions and introducing noise and possibly confusion.
I recently wrote about the importance of maintaining and sometimes changing official historical records, using baseball as an example. Today I want to highlight two of our statistical programs: the Job Openings and Labor Turnover Survey (JOLTS) and the Local Area Unemployment Statistics (LAUS) data. We publish monthly statistics from these programs and revise them the following month as more information comes in. In addition to the monthly revisions, we incorporate more information once a year.
The COVID-19 pandemic continues to have a huge impact on our lives. Check out our summary of how the pandemic affected the labor market and economy in 2020. The magnitude of the labor market changes stress tested the JOLTS and LAUS programs. Based on what we observed in real time, and what we know now, we realized we needed to respond to this unusual economic environment. We change our estimating techniques infrequently, but even the best techniques need adjustments to respond to such significant shocks. These adjustments reflect our commitment to continuous improvement.
Changes in Job Openings and Labor Turnover Estimates
The economic conditions caused by the pandemic led us to make two changes to JOLTS procedures. First, we changed the way we handled unusual reports, which we call outliers. In normal times, these outliers may be businesses with unusually large numbers of job separations. This process mutes the outlier impact on the estimates because those outliers are unlikely to represent other businesses. At the start of the pandemic, however, very large increases in separations were followed by very large increases in hires in many businesses. During this period, we adjusted the JOLTS outlier-detection techniques to accept as normal those extreme changes. Under these circumstances, these “outlier” reports did in fact represent many other businesses.
Second, JOLTS uses data from the much larger Current Employment Statistics (CES) sample to adjust estimates of hires and separations to stay in sync with the monthly employment changes. This procedure assumes that, over the long term, the difference between JOLTS hires and separations is close to the CES employment change. This assumption, however, was not appropriate in late March 2020 as people, businesses, and governments tried to contain the spread of COVID-19. The two surveys have different reference periods. The CES reference period is the pay period that includes the 12th of the month, whereas JOLTS estimates of hires and separations cover the entire month. Hires and separations during the latter half of March 2020 were not included in the CES employment change for March but were included in the JOLTS estimates for the month. To accurately capture the timing of this unprecedented event, we stopped aligning the JOLTS estimates of hires and separations with the CES employment change from March through November 2020.
More changes to JOLTS estimates came with the publication of the January 2021 news release. As we do every year, we revised the past 5 years of historical JOLTS data using updated CES employment estimates. We also updated the seasonal adjustment factors and applied them over the past 5 years. In addition, because we stopped using the alignment procedure for most of 2020, the difference between CES and JOLTS estimates had become quite large by December. To preserve the true economic differences between CES and JOLTS but reduce the divergence by the end of 2020, we adjusted estimates of hires and separations for the months in which the alignment procedure was turned off. These adjustments ensure that we report the highest quality data as quickly as we can, while improving accuracy as we learn more information.
Changes in State Labor Force and Unemployment Estimates
We also made real-time changes during the pandemic to the models we use to produce state labor force and unemployment estimates. The primary inputs to the models are from the Current Population Survey (CPS), the source of the monthly national unemployment rate and other labor market measures. Because the CPS sample is not large enough to support state estimates on a monthly basis, we also use CES employment data and counts of continued claims for unemployment insurance to help inform the models. All of these model inputs experienced extreme movements, especially in the early part of the pandemic.
Starting with March 2020, we introduced two monthly adjustments we usually perform only once a year. These adjustments involved closer review and adjustment of outliers from all model inputs and level shifts. We discussed these changes in notes that appeared in the State Employment and Unemployment news releases for March 2020, April 2020, and May 2020.
These changes in 2020 provided a short-term solution for the state models. For the longer term, we respecified the relationships of the model inputs to provide more flexibility when unusual disruptions occur in the labor market. We explain these changes in our “Questions and Answers.”
We implemented the new estimation procedures for model-based areas in early 2021. They were reflected in the estimates published in the Regional and State Unemployment – 2020 Annual Averages news release. We replaced all previously published state data using the new procedures to ensure historically comparable estimates. The recent data revisions also reflect the best available inputs for model estimation. If you are interested in the details, you can read all about them at the LAUS Estimation Methodology page.
The speed with which the JOLTS and LAUS staff researched and implemented these improvements reflects the high quality of the BLS staff and their commitment to producing gold standard data. They make me proud to lead this great agency.