June 30, 2020 — The jobs report that comes out at the start of every month is the result of a rigorous, time-tested process that would be exceedingly difficult to corrupt, a former commissioner of the U.S. Bureau of Labor Statistics said.

But she knows it’s often difficult to convince partisans of that fact.

In a briefing for the National Press Foundation, former BLS Commissioner Erica Groshen explained how the federal agency compiles the jobs report – the number of people employed in the nation, and the corresponding unemployment rate. She also detailed the BLS’ recent “misclassification error” that made the April and May unemployment rates look lower than they actually were. That problem, she said, is being corrected.

The jobs report is generally released on the first Friday of the month, covering employment the previous month. It is real-time and often front-page news. During times of economic strain – the Great Recession of 2007-2009, the COVID recession of 2020 – the jobs report released at 8:30 a.m. Eastern is reported within seconds and drives the day’s stock market and news cycle.

If the unemployment rate is good, a president will typically tout it and his opponents will say it could be better. If it’s bad, the president usually blames outside factors and his opponents say, “We told you so.”

And in that political tug of war, the integrity of the numbers will sometimes be called into question. That’s what happened in 2016, when then-candidate Donald Trump said the relatively good employment numbers under President Barack Obama were “phony” and “totally fiction.”

But even if the bureau – staffed by career economists and statisticians – would want to put out phony numbers, doing so would be pretty tough, Groshen said. That’s because the monthly jobs report comes not from one but two separate surveys:

__The payroll survey measures employment, hours and earnings in the nonfarm sector. It comes from a representative sample of 600,000 work sites in industries around the country.

__The household survey measures whether somebody is in the labor force. It comes from a representative sample of 60,000 households, and it produces those big headlines about the monthly unemployment rate – an economic indicator closely watched in an election year. The household survey asks whether people are employed or unemployed; if they’re unemployed, they are asked if they are looking for work

“If you were mucking with the data, you’d have to muck with both” the household and payroll surveys, Groshen said. The two surveys come from different, independent sources, and their methods have been refined over the years.

Right now, the country is in one of its deepest recessions in decades and economists are unsure how quickly it will snap back. The unemployment rate shoots up over a few months at the start of a recession and then gradually – over several years – eases back down. The COVID-fueled unemployed rate, now in the mid-teens, is higher than at any time since the Great Depression in the 1930s.

The Bureau of Labor Statistics found itself in the news in June not because of the unemployment rate itself but because of a “misclassification error” that substantially altered that rate. The result: Unemployment looked better than it actually was. That lead critics of the president to float the possibility that the numbers had been monkeyed with.

Nobel Prize-winning economist Paul Krugman tweeted himself into the political dustup: “This being the Trump era, you can't completely discount the possibility that they've gotten to the BLS, but it's much more likely that the models used to produce these numbers — they aren't really raw data — have gone haywire in a time of pandemic.”

He quickly backed off his tweet. And in fact, the numbers had gone haywire.

Groshen explained what happened. When businesses closed because of the coronavirus, workers were sent home. In the household survey, they were classified as being “employed but not at work for other reasons” – a category that usually includes vacation, sick leave, jury duty and the like.

But in fact, they were laid off and unemployed and should have been counted as such. The impact on the unemployment rate was significant: In April, the official rate was 14.7%. If those misclassified workers had been added in, the rate would have been 19.5%.

The U.S. Census Bureau, which conducts the household survey, has since refined its instructions to interviewers to avoid the problem.

“It was neither incompetence nor manipulation” that led to the misclassification, Groshen said. They reality, she added, was that COVID-19 changed things rapidly – and the household survey didn’t adapt quickly enough. (She explains the issue in more detail here.)

Groshen is a senior extension faculty member at the Cornell University School of Industrial and Labor Relations, a research fellow at the W.E. Upjohn Institute for Employment Research and a fellow of the American Statistical Association.