Recalibrating the Jobs Thermometer

The U.S. Bureau of Labor Statistics is responsible for measuring labor market activity. Each month BLS releases some of the most up-to-date measures of economic health in The Employment Situation, often called “the jobs report.” We also release the Commissioner’s statement each month at the same time as the jobs report.

Most of the attention focuses on the headline numbers—how many jobs were added (or lost) that month and did the unemployment rate change? However, with the release of January data each February, we make some yearly updates to improve the accuracy of the numbers. In our survey of households, which is the source for the unemployment rate and other measures, we update the U.S. population totals to reflect the latest information about births, deaths, and international migration. In our survey of nonfarm establishments, which is the source of the jobs count, we make our annual benchmark revisions. Today I’m going to focus on the establishment data.

Each month, the establishment program surveys a sample of businesses and governments around the country. The survey asks how many people worked or received pay for the pay period that included the 12th of the month. While the establishment sample is large, covering about one-third of all nonfarm jobs, the employment changes reported each month are still subject to revisions. Monthly revisions result from more establishments reporting their numbers or correcting previous reports, and from updated information about seasonal employment patterns.

The establishment survey also benefits from another source of data, the Quarterly Census of Employment and Wages. That is a nearly complete count of all establishments, although it is available with a delay of about 6 months. A full count of employment helps us in several ways. We use the data to measure the error associated with the establishment survey. This way data users don’t have to guess how accurate the monthly employment data are. We have a stat for that! In case you are wondering, the data are very accurate. Annual benchmark revisions (which I will explain in a moment) have averaged only 0.3 percent in absolute terms over the past 10 years.

Besides measuring error, once a year we realign the sample-based estimates with the full count of employment. We call this “benchmarking.” This realignment makes sure the employment levels do not stray too far from the “truth” over time. (For several reasons that I won’t go into right now, the establishment survey employment totals will not exactly equal employment totals from the full counts. If you really want to know the details, you can read more about benchmarking, but remember I tried to spare you.)

During this annual benchmarking, we also introduce other changes to the survey. Sometimes we update the industry classification, like we will this year. We also use new information to update the statistical model that accounts for business births and deaths. We review the establishment sample for size, coverage, and response rates, and we may drop some series if the data quality doesn’t meet our standards. We also update the models and information used in seasonal adjustment.

As you can see, there’s a lot going on during this annual updating. All establishment data, including employment, hours, and earnings, are subject to adjustment. I hope this brief explanation and the material we have on our website help to make everything more transparent and easier to understand.

The same basic benchmarking that occurs nationally also happens for the state and local employment estimates. Want to know more? Visit their homepage. If you still have questions, call or email us. We are here to help.