The government agency that compiles the official U.S. unemployment rate — the Bureau of Labor Statistics — said it is still working to fix an unprecedented data-collection error that has badly skewed the rate, dismissing suggestions that data had been manipulated.

The May jobs report included an unusual note saying that the unemployment rate would have been 16.3 percent, not the official rate of 13.3 percent, if not for errors made during the data’s collection, a problem that also plagued monthly reports in March and April. Both the official and corrected May unemployment rates showed improvement compared with April figures. The errors have sowed doubt among some parts of the public about the integrity of the figures, as the economy has cratered during the pandemic.

William W. Beach, an appointee of President Trump who became Bureau of Labor Statistics (BLS) commissioner in 2019, said that the agency is working to improve its data-collection process during the pandemic. Any questions about whether the unemployment rate had been tampered with stem from an “enormous ignorance” of how his agency works, he said.

“It’s just really absurd,” Beach said in an interview with The Washington Post. "There’s no evidence for that allegation. And I can tell you, no one from any political source has contacted me, since I’ve been basically commissioner, on any of this stuff except to get information. It’s just bad business to try to undermine your infrastructure. And the BLS is absolutely essential to our economic infrastructure.”

Prominent Democrats, including Sen. Kamala D. Harris (D-Calif.), began to wonder publicly Friday whether there had been interference from the Trump administration, as the president has made unemployment figures a central part of his campaign pitch in recent years. In the past, Trump has disregarded norms that presidents have traditionally followed, such as waiting at least an hour to tweet after the jobs data is released.

The errors could complicate the country’s efforts to address the economic fallout from the novel coronavirus, coming in such a tense moment with the economy mired in a deep recession and social unrest on the rise. Economists said they could not think of another statistical issue of comparable magnitude in the country’s modern era. And the errors come at a time when missteps at public agencies have combined with an administration that has spread falsehoods and misinformation, fueling mistrust with public institutions.

BLS associate commissioner Julie Hatch Maxfield said that the errors came from mistakes made by some of the roughly 3,000 Census Bureau workers who go out and survey nearly 60,000 households for the monthly survey.

Interviewers wrongly classified some workers as employed but “absent from work" due to “other reasons,” when those people probably should have been counted as temporarily laid off, and therefore unemployed, while their workplaces were shut down. In the May report, this misclassification affected nearly a million workers.

“Essentially what happened is people ended up in the wrong bucket," said Maxfield, who has been at the agency since 1999 and now oversees the monthly jobs reports.

Beach and Maxfield told The Post that the White House has no influence on the jobs report and that the BLS is working with the Census Bureau to retrain field surveyors. The agencies plan to put a notice on the laptops that surveyors use this month to try to eliminate the error for the June report. If successful, it could result in a slight increase in the official June unemployment rate, some economists predict.

“I have no contact with president, and I’ve never met him,” Beach said, when asked about Trump.

Both the BLS and the Census Bureau said they were investigating the errors and why they had persisted, despite months of training and guidance given to staff.

Beach and Maxfield reiterated what former officials from both sides of the aisle have said: The data is immune from political interference because of how it is collected, but the pandemic has posed a huge challenge to data collection. Census workers normally conduct a lot of interviews in person for the monthly jobs report, but they have had to shift to phone interviews, which has driven down the response rate.

The final jobs data is presented to Beach by career staff only in aggregate, in its nearly final form. Trump, like his predecessors, is not briefed on the data until the afternoon or evening before the report is released to the public to prevent it leaking to a wider audience, Beach said.

“He would not be briefed until after he had finished all of his engagements. That then prevents the president from inadvertently signaling by an expression or by a term the direction in which the numbers are going the next day,” Beach said. “The numbers are also shared with the secretary of the Treasury and with the Federal Reserve chairman. And that’s it.”

The labor secretary gets to see the numbers at about 8 a.m. on the Friday they are released, he said, which is a half-hour before they go public.

Other federal labor experts agreed that tampering is implausible, noting that the agency has only one political appointee, in this case Beach.

“The numbers are produced in a very automated way, with not that many people touching them until they are processed, and the staff are some of the most dedicated professionals that I’ve ever worked with,” said Erica Groshen, who was in charge of the BLS for four years in the Obama administration.

The misclassification issue first came across the BLS radar in March. The clue that something was awry was that the number of people who were classified as absent from work for “other reasons” had spiked over typical levels, Maxfield said.

“It was obvious in March we needed to do more," she said.

Interviewers typically put in a note about what the “other” reason is, and BLS staff could see many mentions of coronavirus, covid and pandemic — signaling those people probably should have been classified as unemployed, Maxfield said.

The global pandemic was declared March 11, just before interviewers went into the field for a week to ask the 60,000 households the jobs questions. The BLS and Census Bureau, anticipating confusion, sent out instructions via email on how to classify people laid off or furloughed during the pandemic, Maxfield said, though there was little time to act.

For the next month, the agencies increased their training to try to educate field workers about the different classification specifications. Despite these efforts, the errors continued in the April report, adding up to a whopping five-point difference between the official rate — 14.7 percent — and the one that corrected for the error, 19.7 percent.

After that report, which became public May 8, field supervisors held calls with their staff, giving detailed instructions about the data-collection process and hosting question-and-answer sessions.

But the problems persisted in the May jobs report that came out last week.

This month, interviewers’ laptops will flash a special notice when they try to use the “employed but absent” category to advise the interviewer the category should not be used for anyone who is not working because of the effects of covid-19, the disease caused by the coronavirus. Census officials will continue to do additional staff training, the bureau said in an email.

While the debate rages over which employment figure is more meaningful, the overall improvement in jobs has been consequential.

Friday’s job report helped fuel a stock market rally, as markets have trended toward pre-pandemic highs. Trump held a news conference Friday to hail the drop in the unemployment rate, even if though it remains the highest since the Great Depression, touting it as a sign that the economy’s recovery was well underway. Some Republicans in Congress have pointed to the falling unemployment rate as a sign that the economy doesn’t need billions of dollars more in federal aid.

In contrast, some Democrats, such as Harris, have used the revised numbers to argue that the economic picture being painted by the White House is rosier than it actually is.

“I think these headline figures are really important for shaping people’s impressions of how serious the recession is,” said Barry Eichengreen, a professor of economics and political science at the University of California at Berkeley. “These numbers shape the impression of policymakers and of people who put pressure on them to do more or do less.”

Eichengreen said that he couldn’t think of a more significant issue with a key economic measurement since the data-collection system was built in the aftermath of the Depression.

“I don’t think in the modern era of statistical reporting that we’ve had a possible error in a major economic statistic this large,” he said. “Because we’ve never seen a movement of the economy within a month or two, as what’s happened this time.”

To some economists, there are other causes for concern in the monthly report. Only 67 percent of households responded to the survey in May, down from over 80 percent before the pandemic. As fewer and fewer people take the survey, it can become less representative of the adult U.S. population.

“The fall in the response rate is a concern because you are trying to come up with an estimate of labor market outcomes for over 200 million people from a survey of what is ideally 60,000 households,” said Ernie Tedeschi, a former Treasury chief economist who is now at Evercore ISI.

Heidi Shierholz, former chief economist in President Barack Obama’s Labor Department, was one of many former federal economists who defended the BLS and Census Bureau, and tried to put out the conspiracy theories Friday.

She said that the BLS should be lauded for its transparency about the problem but could understand why people had doubted the numbers.

“This administration has shown over and over they’re comfortable with data manipulation, so I understand people questioning this, but I do not believe that this is where it’s happening,” she said. “If there’s an undermining of trust in these numbers broadly, that’s just bad for all of us.”

She said that she thought perhaps the BLS would have been better served by including the disclosure about the erroneous data at the top of the report, not on the sixth page of a 42-page report.

Former staff at the Census Bureau also defended the integrity of that agency, pointing out how difficult it is to conduct surveys during a pandemic.

“Even Census runs into unprecedented conditions, none quite so unprecedented as a nationwide pandemic that itself is not yet fully understood,” said Kenneth Prewitt, who was director of the bureau during the 2000 Census.

Beach, the BLS commissioner, echoed those considerations, saying that the pandemic had created an “extraordinary” set of circumstances for government statisticians to sort through. But he said revising the March, April and May unemployment data was “off the table.”

“Our systems are created in more normal times,” he said. “We are doing everything we can short of revising the unemployment rate, which we don’t do."