Economists predicted the official U.S. unemployment rate would hit 20 percent — or really close to it — in May. Instead, the world learned Friday morning that the official rate is actually 13.3 percent, an improvement from 14.7 percent in April.

It was, as economist Chris Rupkey emailed, the “biggest forecast miss of our life.”

What the heck happened?

In short, give some credit to the government relief efforts, especially the Paycheck Protection Program, for bringing back jobs. The program gave relief to small businesses (and a few larger ones, triggering public outcry) through loans that would not have to be paid back if most of the money went to rehire and pay employees. PPP money had to be used right away, and a lot of it started hitting small businesses’ bank accounts in late April and early May, which ended up triggering a net gain of 2.5 million jobs in May, the Labor Department reported. Many economists expected the PPP would be a big factor in June, but it turns out the impact was sizable in May.

A closer look at where those jobs gains occurred is telling. Over half (1.4 million) were in restaurants. Nearly 20 percent of the job gains came from construction. About 15 percent of the growth came from retail stores, and about 10 percent came from dental offices. Restaurants and retail were heavy beneficiaries of the PPP. On top of that, construction homes and commercial properties was deemed essential in many states, and projects ramped up as the weather improved.

While Wall Street is cheering the jobs news, it is important to remember that the unemployment rate in May is still a good bit higher than the worst days of the Great Recession. In short, it’s encouraging that 2.5 million got their jobs back in May, but 21 million are still unemployed.

Most economists look at these numbers and urge Congress and the White House to keep the PPP and other aid going. Most of the government relief money is slated to dry up by the end of July. But President Trump and many of his top advisers question whether more aid is needed, since the economy is showing several signs of improvement.

“13.3% unemployment is an economic and human disaster. Peak unemployment in the Great Recession was 10%,” tweeted Michael Strain, head director of economic policy studies at the right-leaning American Enterprise Institute. “Workers, families, and small businesses need Phase 4 [stimulus]. There’s no doubt about the need.”

The other reason economists got this so wrong is this is an unprecedented situation. Economists are very good at explaining what happened in the past, but it is difficult to forecast the future and even harder to do it in such an unusual situation. In normal times, economists warn the data could be off by as much as a hundred thousand. This time, it was an error rate in the millions.

Is there a mistake in the data?

The Bureau of Labor Statistics, which is part of the Labor Department, calculates this data. Some have questioned whether Trump might have manipulated it since it is from a government agency, but economists and former government staffers of both parties have firmly rejected that.

There’s been a lot of hubbub about the fact that the BLS admitted there was a “misclassification error.” In a special note at the bottom of the jobs report, the BLS said the May unemployment rate would be about 16.3 percent, down from an April unemployment rate of more than 19 percent, if this error hadn’t happened. So what was it? In short, some people who should have been classified as unemployed were instead classified as employed but “absent” from work for “other reasons.”

This “other reason” category is normally used for people on vacation, serving jury duty or taking leave to care for a child or relative. But in this unusual pandemic circumstance, the “other reason” category got applied to some people sitting at home and waiting to be called back.

The same thing happened in April, which would have added nearly another five percentage points to the 14.7 percent unemployment rate, the agency said. The unemployment rate comes from a survey where people are asked a) if they are working and b) if they are not working, why not? People basically told the surveyors they were just “absent” from work.

BLS does not like to change people’s answers, because they don’t want to tamper with the data. Whether you look at the official unemployment or this amended unemployment rate, the bottom line remains: Unemployment is still high, but it went down slightly in May.

Did Trump intervene?

“You can 100% discount the possibility that Trump got to the BLS. Not 98% discount, not 99.9% discount, but 100% discount,” tweeted Jason Furman, the former top economist for President Barack Obama. “BLS has 2,400 career staff of enormous integrity and one political appointee with no scope to change this number.”

The bureau used the same methodology that it uses every month: It looks at who was working and who wasn’t mid-month. Specifically, the agency looked at the week of May 10 to 16. Like always, the Census Bureau collected data for the BLS in two ways. In one survey, census workers literally go out and ask some people about their situation, and in the other, BLS workers look at company and government payrolls from 140,000 establishments to see how many people are employed and how many hours they are working.

Here’s why economists struggled: By May 16, more than 38 million unemployment claims had been filed, which easily implied a 20 percent unemployment rate. It was also several million more unemployment claims than at the end of April, which made economists think the May picture was getting worse. What economists missed is that some of those workers who had filed unemployment claims were starting to go back to work. There wasn’t good data on that.

Trump said Friday this is “probably, if you think of it, the greatest comeback in American history.”

Economists, however, pointed out that even the latest data shows 28 million people had their job cut or hours reduced during the pandemic. Plus, an alarming number — 2.3 million people — now say they have permanently lost their jobs. There’s a long way to go before the job market gets back to where it was before the pandemic.

Who isn’t counted in the unemployment rate?

The economic picture is further complicated by the fact that a lot of people stopped looking for work in May or had their hours reduced. These people’s jobs were probably affected by the pandemic, too, but they aren’t counted as part of the official unemployment rate. There were more than 6 million people who said they wanted full-time work but were instead working part time in May. And another 6 million who were out of a job but stopped looking actively, probably because of the health emergency. To be counted as unemployed, people needed to be sending out résumés in the past month.

In the end, perhaps the best figure to look at to gauge the health of the labor market — and how the economic rebound is going — is what share of the adult U.S. population is employed. Economists call this the “employment-to-population ratio.” (See the chart below.) It plunged massively in April and has come back only slightly.

Any improvement is encouraging. It makes a difference for those who did get their jobs back. But it’s clear there’s still a long way to go to get back to anything close to pre-pandemic levels.