The jobs report, explained: payrolls, the unemployment rate and revisions
The monthly US jobs report comes from two different surveys. Here is what each measures, why they can disagree, and why the first number is often revised.
On the first Friday of most months, the Bureau of Labor Statistics publishes the Employment Situation report, better known as the jobs report. Markets move on it within seconds, and it is quoted for weeks. Yet the two numbers people remember, "jobs added" and "the unemployment rate", come from two separate surveys that measure different things. Knowing how each works makes the headlines far easier to judge.
Two surveys, two questions
The establishment survey, formally the Current Employment Statistics (CES) program, asks employers how many people are on their payrolls. It covers a large sample of businesses and government agencies, representing hundreds of thousands of individual worksites. From it come the headline change in nonfarm payroll employment, average hourly earnings and the average workweek.
The household survey, the Current Population Survey (CPS), is run for the BLS by the Census Bureau. It interviews people in roughly 60,000 eligible households about whether they worked, looked for work or did neither during a reference week. From it come the unemployment rate, the labor force participation rate and many demographic breakdowns.
The establishment survey counts jobs; the household survey counts people. A person with two part-time jobs counts twice in the first and once in the second. The household survey includes the self-employed, farm workers and people working in private households; the payroll survey does not. That is why the two can point in different directions in a given month.
What "unemployed" means
To be counted as unemployed in the household survey, a person must have had no job during the reference week, have been available for work, and have actively looked for work in the previous four weeks (or be waiting to be recalled from a temporary layoff). Someone who wants a job but has stopped looking is not counted as unemployed. They are outside the labor force.
The official unemployment rate, known as U-3, is the number of unemployed divided by the labor force (employed plus unemployed). The BLS also publishes broader measures. The most cited, U-6, adds people who are "marginally attached" to the labor force, including discouraged workers, and people working part time for economic reasons who want full-time work. U-6 is always higher than U-3, and the gap between them says something about slack in the job market that the headline rate misses.
Why the first estimate changes
The payroll number published on jobs Friday is a first estimate based on the survey responses received by then. Not every employer replies in time. As more responses arrive, the BLS revises each month's figure in the following two monthly reports. The revisions are published in the same release, and they can be large enough to change the story: a month first reported as weak can later look solid, and the reverse.
Once a year, the BLS also carries out a benchmark revision. Payroll estimates are re-anchored to counts drawn largely from state unemployment insurance tax records, which cover nearly all employers. A preliminary estimate of the benchmark is published in late summer, and the final revision is incorporated with the January report the following February. Benchmark revisions have occasionally shifted the level of employment by several hundred thousand jobs.
The household survey is not revised month to month in the same way, but its population controls are updated each January, which can cause breaks in the level of some series.
Seasonal adjustment
Hiring has strong seasonal patterns: retailers add staff before the holidays, schools shed staff in summer, construction slows in winter. The headline figures are seasonally adjusted to remove these regular patterns so that one month can be compared with the next. Unadjusted data are also published, but most news coverage uses the adjusted series. Unusual weather, strikes and natural disasters are not removed by seasonal adjustment, and the BLS often notes when such events affected a month's data.
How to read a single report
- Look at the three-month average of payroll gains rather than one month. Individual months are noisy and will be revised.
- Read the revisions line. The net revision to the two prior months is often as informative as the new number.
- Check participation. The unemployment rate can fall because people stopped looking for work, not because they found jobs. The labor force participation rate and the employment-population ratio help separate the two.
- Look at wages and hours together. Average hourly earnings can be pushed up if lower-paid workers lose jobs, which changes the mix rather than pay. A shorter average workweek can signal that employers are cutting hours before cutting staff.
- See which industries moved. Gains concentrated in one or two sectors, such as health care and government, say something different from broad gains across goods-producing and service industries.
The margin of error
Both surveys are samples, so both have sampling error. The BLS publishes confidence intervals: a monthly change in payrolls needs to be well over 100,000 jobs to be statistically significant, and a change in the unemployment rate of about 0.2 percentage points is needed for the same. A headline difference of a few tens of thousands of jobs between the estimate and forecasts is, statistically, not much of a surprise at all.
The bottom line
The jobs report is one of the most carefully produced economic statistics in the world, but it is an estimate that gets better with time. Treat the first print as a draft, look at trends and revisions, and read the payroll and household numbers as two different views of the same labor market.