Here’s one possible explanation for that really weird jobs report: trouble with seasonal adjustment.
That theory comes courtesy of Labor Secretary Tom Perez, who floated it in an interview Friday afternoon. I’m not totally convinced, and to be fair, neither is Perez — he was careful to frame it as a still-unproven hypothesis. But it’s at least a credible one. (Others have also raised this as a potential issue.)
As a quick reminder, Friday’s jobs report contained a conundrum: One part of the report, based on a survey of businesses, said the U.S. added 288,000 jobs in April. Another part, based on a survey of households, said 73,000 fewer people had jobs. It’s highly unlikely both are correct.
But those numbers are seasonally adjusted, meaning they’re tweaked to account for predictable patterns, such as the hiring of retail workers around Christmas and lifeguards in summer. The Bureau of Labor Statistics (BLS) uses an algorithm that essentially looks at the past few years of data to predict what hiring “should” do based on seasonal patterns. If the numbers are better than expected, the model will report a gain in jobs; if they’re worse, the model will treat that as a decline.
This time, the raw (not seasonally adjusted) household survey showed employment rising by 677,000 in April. That’s a lot of jobs, but not as many as the model was predicting: April is usually a good month for hiring, because seasonal businesses start staffing up for the warm weather. Last year, unadjusted employment rose by more than a million. So after applying the BLS algorithm, that 677,000-person gain turned into a 73,000-person loss.
That kind of swing isn’t unusual. But there’s some reason to be skeptical of the model this month. That’s because the Census Bureau, which carries out the household survey on behalf of BLS, effectively conducted its interviews a week earlier than normal. That could mean companies weren’t as far along in their spring-hiring process as usual, skewing the numbers down. Adding weight to this theory: The decline in employment was concentrated among the young and less educated, who are hired for many of the seasonal jobs.
But hold on a second. If something was wrong with seasonal adjustment, shouldn’t that have affected both surveys, not just one of them? Not necessarily. The household survey asks people about their employment status in whichever week contained the 12th of the month. In April, the 12th fell on a Saturday, so people were asked about their employment from Sunday, April 6, to Saturday, April 12. (This month, the so-called reference week will be May 11-17.)
The business survey, on the other hand, asks employers how many workers they had in the pay period that contains the 12th. That may sound like a small distinction, but it can make a big difference. For a business that pays workers on the 15th and 30th of every month, for example, it doesn’t matter where in the week the 12th falls. So in theory, a quirk in the calendar could have skewed the household survey but not the establishment survey.
That’s in theory. In practice, I’m a bit skeptical. For one thing, the reference week shifted by only one day this year; last year, April 12 fell on a Friday, with no apparent negative effect on hiring. Moreover, the month that really looks out of whack is not April but March, when the household survey showed an unadjusted gain of 956,000 workers (476,000 after adjustment). So it’s possible that the issue has less to do with seasonal adjustment and more to do with regression to the mean.
We’ll know more in a month. If next month’s jobs report shows a big surge in household-survey hiring, that will be a sign this month’s decline was a seasonal fluke. But the real lesson in all of this is that the jobs numbers, especially the ones from the household survey, are volatile. Treat them with caution.