April 5 brings us another "jobs day," the first Friday in each month, when the latest employment numbers come out. With much business-media fanfare, the numbers will be parsed for what they tell us about the economy and where it is headed. The jobs numbers do contain some valuable information, but they mean much less than is often assumed—because, as the Department of Labor cautions, they're estimates. They are subject to significant revision, they are volatile, and they tell us very little about the direction of the labor market.
The "jobs day" information comes from two different sources of data provided by the Bureau of Labor Statistics. Household data are used to estimate the size of the labor force and the unemployment rate, and the widely reported "job creation" numbers come from data supplied by employers. It is possible, however, to obtain job-creation numbers by analyzing the household data.
There is often a disparity between the employer-based numbers and the job-creation numbers that come from household data. In over half of the months since 1996 to the present, the household and establishment job-creation figures differed by more than 50%. For example, in a month when the employer-based number is 200,000 additional jobs, it would be typical to find the household numbers reporting only 100,000 new jobs.
Over longer periods of time, however, there is significant agreement between the two data sources. The annual employment levels from the household survey are very closely related to annual employment levels from the employer-establishment survey.
It is widely accepted that the employer-based establishment data are better for the purposes of estimating job creation than are the household data. Unfortunately, even the establishment data are less than accurate, at least when they are first announced.
Each month, the Bureau of Labor Statistics reports employment and the change in employment since the previous month. There are subsequent revisions, one and two months after the first announcement, until the number becomes final, sometimes up to two years later. The error in any given month tends to be very large, which means that its reliability is low.
For instance, the average number of jobs created per month during the 1996-2012 period was 78,000. But in the typical month, the initial estimate missed the final number by 73,000 in one direction or the other. This means that the average error in the initial report is almost as large as average job creation itself.
The quarterly numbers are somewhat better, but not by much. The picture improves somewhat for annual job growth, with the typical error being about half as large as the average job change itself.
In some months, the error is enormous. For example, the initial report released in January 2011 stated that 130,712,000 Americans were working in December 2010. Two months later, that number was revised downward to 130,260,000—a difference of 452,000. When the final numbers came in, the actual overstatement turned out to be 366,000. That's worth keeping in mind at a time when much excitement can be stirred up by a monthly jobs fluctuation in the tens of thousands.
The December 2010 example is by no means the worst case. In late 2009, the difference between initial reports and revisions was well over one million. For the entire period studied, the initial estimates of job creation deviated from final estimates by at least 50% in almost one in every four months.
Over the very long run, the reported numbers look better. Although there is significant error in any given month, the overstatements of job growth in some months tend to offset understatements of job growth in other months. The preliminary data would have yielded job growth of about 17 million between 1996 and 2012. The actual job growth was 16 million, an error of about 6.5%.
Accuracy goes up considerably during more normal times when job growth is positive (about two thirds of the time). Then, the average error is only about one-third as large as the average number of jobs created during this period. This still implies that true job growth is missed by 68,000 in one direction or the other during the typical positive-job-growth month.
For example, during the past year, 164,000 jobs were added each month on average. But consider that the typical error range would put job growth during that period somewhere between 96,000 and 232,000 in the average month.
Trying to deduce anything about the direction of the jobs market by referring to such wobbly numbers is essentially guesswork. Indeed, in any given month, about 70% of what happens to job growth in the following year has nothing to do with changes that occurred in that particular month. In almost one-fourth of cases, the job growth in any given month does not even move in the same direction as the job change in the 12 months that follow. Using even the entire previous quarter's job growth provides no better signal of where the labor market is headed.
Employment numbers have value, especially when considered over long periods, like a full year. "Jobs day" chatter is irresistible but almost without content. Monthly jobs numbers provide imperfect portraits of the recent past, and they are very poor predictors of the labor market's future.
Mr. Lazear, who was chairman of the President's Council of Economic Advisers from 2006-09, is a Hoover Institution fellow and a professor at Stanford University's Graduate School of Business.
