Friday at 8:30 the Bureau of Labor Statistics will be releasing its monthly Employment Situation Report, the so-called “jobs report.” These things inevitably get a lto of media attention, but now that we’re not in the middle of a presidential campaign it’s perhaps a good time to just write about what this report is and how it works.
For starters, the most important thing to understand is that the jobs report is really two different things. One, the establishment survey is based on a survey of workplaces. They ask, basically, “how many people are working here?” The other, the household survey is based on a survey of individuals. The two surveys are not only methodologically different, they measure slightly different things. The establishment survey measures jobs while the household survey measures people. Since a given person can hold multiple jobs, in theory the number of jobs could greatly exceed the number of employed people. On the other hand, self-employed people won’t show up in a survey of establishments. In practice the number of jobs measured in the household survey is invariably larger than the number of jobs found in the establishment survey. Reported self-employment, in other words, exceeds reported multiple jobs holdings. The establishment survey also ignores farmers, and is therefore provides a number known officially as non-farm payrolls (and, no, I don’t know why it leaves the farmers out but it does).
The two surveys are also used for two different things. The establishment survey is where we get our raw count of jobs. If you read that XXX,XXX jobs were created last month, that’s an establishment survey figure. Since it surveys establishments, this is also where we get our sector-by-sector breakdowns—jobs were added in retail establishments, or local government establishments. The establishment survey is generally regarded as more accurate.
So why do the household survey? Well primarily because if you want to know the unemployment rate you need a denominator. After all lots of people don’t have jobs without being unemployed. They might be in school or retired or taking care of kids. To find the denominator you need to survey the population, and to take consistent numbers you need both the numerator and the denominator to come from the same survey. In addition to the main unemployment rate, people often also like to point at the broader measures of unemployment that do things like count people who have part-time jobs and say they’re looking for full-time work but can’t find it, or count “discouraged workers” who were looking for a job but gave up because they couldn’t find one.
But the most important thing in any jobs report is the revisions. Each month’s employment data gets revised over the next two months as states report more complete information via their unemployment insurance programs and other such things. These revisions are always quite small relative to the total number of jobs in the country, but they’re often quite large relative to the change in the total number of jobs. Everyone has a hard time restraining themselves from focusing on the newest data, but the most accurate data is the revised data from a couple of months back. Note that the BLS can’t survey an establishment if it doesn’t know the establishment exists, so during a recovery when new firms launch at a faster-than-expected rate you tend to see systematic upward revisions. During a downturn the reverse happens as firms are created at a smaller-than-expected rate.
Last but by no means least, responsible people will be discussing seasonally adjusted data. The business cycle is driven by the seasons to a surprising extent. Each year, employment booms in December and collapses in January due to Christmas surge hiring. Then it steadily recovers from the January low point through the summer, and then there’s a mini-recession in the fall as the swimming pools and amusement parks close and restaurants shut down their outdoor seating. Then the Christmas boom again. This would make raw data counts totally useless if not for the seasonal adjustment formula.