On its face, it sounds like the Bureau of Labor Statistics has an easy task: Just count the jobs. A lot of people have jobs, some people don’t, they’re all out there, just count them up.
Of course, it can’t literally go and count every person each month, so it uses statistical methods to survey employees and employers. But people have been doing surveys forever. Just send them out and run it through a computer and write the report. Easy.
No. Not at all. Not even close.
Individual businesses don’t even know how many people they employ in a given month. That might sound hard to believe, but think through how it works.
The BLS surveys 121,000 businesses each month (that’s a lot) and asks them how many people they employed in the week of the month that included the 12th day. A business can probably have an easier time saying how many people it employed in a year, but a specific week in a specific month is really hard.
About 4 million people change jobs each month in the U.S. Let’s assume for sake of argument that means about 1 million people change jobs per week: That means there are about 1 million possibilities for double-counting where two employers could say they employ the same employee the week of the 12th day. Then, there’s the simple fact that sometimes people will quit or get hired in that month after the business sends the paperwork back.
And that’s assuming they send it back at all. Before Covid, the survey response rate was about 60 percent. Today, it’s just 43 percent. The BLS is working with less data on more businesses and a larger workforce.
But back to the business side. The difficulty of answering the question of how many people they employ is especially large for large businesses, where countless managers have hiring and firing authority throughout the company. And that’s especially a problem for counting the number of jobs because large businesses employ a lot of people.
When a big business announces in a press release that it’s hiring 1,000 people or laying off 1,000 people, it matters to the BLS what the exact number is. And the business probably doesn’t know. If the press release said 1,000, but after the fact it turns out the big business actually laid off 1,009 people, nobody would accuse the big business of lying. But that means a different nine-employee small business was just canceled out of the jobs report entirely. The BLS knows that businesses probably don’t know exactly how many employees they have in a given month, so it has statistical methods to correct for those errors.
But it gets worse: Businesses form and fail in the hundreds of thousands each month. The BLS has a decent chance of sending its surveys to companies that don’t exist anymore. And the new companies that do exist definitely employ people, but the BLS has no way of contacting them because they didn’t have an address or a phone number until this month. So the BLS has statistical methods to estimate that as well.
Covid threw off those methods because there was a sudden surge in new business formations, likely to take advantage of Covid relief programs administered to employers. BLS knew a lot of those were phony, but how many, and for how long? Really hard to say.
Then, there are seasonal patterns in employment. Everybody knows more teens work in the summer, more ski instructors work in the winter, and more agricultural laborers work during harvest season. But is it more more-than-normal or less more-than-normal? And what should we expect “normal” to be? The BLS seasonally adjusts the data, again with complicated statistical methods.
What about self-employed people, who won’t show up on the employer survey but will show up on the household survey? What if there’s a big storm and it throws off employment in one city for the week that the survey is out that month? What if a bunch of public employees in another city go on strike? There are so many possible anomalies that would throw off the data for a month that the BLS has to keep in mind and correct for.
There is a way to get very reliable information on all of this, and that’s to look at the data for unemployment insurance. But those data only come out once a year, for tax purposes. So the BLS uses that data to revise its monthly estimates after the fact, along with a different quarterly survey that is also of higher quality.
But people don’t want just one really good report on employment per year. They want to know every month. They want to be able to track trends. So the BLS does the monthly jobs report under the full knowledge that it isn’t as precise as other methods that are less frequent.
It can’t just put out a jobs report that says “good” or “bad” or “okay.” It has to report a number. All of those questions, all of that confusion, ultimately has to boil down to a number. Every month.
The original estimate for the number of jobs in the month of June was 159,724,000. Then, after the revision with better data, it was 159,466,000. That’s a 0.161 percent correction, based on higher-quality information that didn’t exist at the time of the original estimate.
So, no, the BLS is not incompetent, and it does not have an easy job at all. The reasons for revisions are incredibly boring and technical and have nothing to do with politics or ideology. Or at least they didn’t, until the president fired the BLS commissioner because he didn’t like the jobs numbers. Now, the incentives have changed, and there is a reason to cook the books out of self-preservation.