More than 40 million Americans have applied for jobless benefits over the past two months, thanks to the COVID-19 pandemic and shutdowns. Last week alone, 2.1 million new claims got filed. So how is it possible that ADP projects that the US economy lost only 2.76 million jobs in the whole month of May?

Did America go back to work while escaping everyone’s notice? Or, more accurately, did less of America leave work than everyone assumed?

Private sector employment decreased by 2,760,000 jobs from April to May according to the May ADP National Employment Report®. The report utilizes data through the 12th of the month. The NER uses the same time period the Bureau of Labor and Statistics uses for their survey. As such, the May NER does not reflect the full impact of COVID-19 on the overall employment situation.

Broadly distributed to the public each month, free of charge, the ADP National Employment Report is produced by the ADP Research Institute® in collaboration with Moody’s Analytics. The report, which is derived from ADP’s actual payroll data, measures the change in total nonfarm private employment each month on a seasonally-adjusted basis.

Let’s not get too happy about this. Losing 2.76 million jobs in a month is still pretty darned horrid, but it’s a far cry from the repeat of April’s 20-million-job meltdown everyone expected. In fact, most observers figured that the May benchmarks would be worse as the data was still too raw and ambiguous last month — which might still be the case with Friday’s official jobs report from the BLS.

The report shows that job losses hit almost ever sector and were as evenly distributed as one might imagine. Large businesses sustained the most losses (1.6 million), with medium-sized businesses losing 722,000 and small businesses 435,000. Manufacturing took a huge hit with 719,000 jobs lost, but transportation beat it at 826,000. Two sectors, however, showed upticks: Administrative/support services (40,000) and education (166,000), perhaps as a result of adjustments needed for remote work.

Recall too that ADP is a less-than-perfect bellwether to the official numbers. Their predictive value is more in broad strokes rather than specific numbers. Even so, economists expected this to be worse — whole lot worse:

The reported total was well below the 8.75 million estimate from economists surveyed by Dow Jones. The reason for the wide disparity was not immediately clear.

May’s count also marked a precipitous drop-off from the 19.6 million plunge in April, an estimate that was revised from the initially reported 20.2 million. The April loss was by far the worst in the history of the ADP survey.

“The impact of the Covid-19 crisis continues to weigh on businesses of all sizes,” said Ahu Yildirmaz, co-head of the ADP Research Institute. “While the labor market is still reeling from the effects of the pandemic, job loss likely peaked in April, as many states have begun a phased reopening of businesses.”

No further information was available on why the reported monthly change was so big or how it could have been so far off Wall Street estimates. Ian Shepherdson, chief economist at Pantheon Macroeconomics, cautioned investors in a note Tuesday that they should be “braced for surprises, in either direction” because of the model ADP uses to calculate the payroll total.

The truth is that there’s no entirely reliable way to measure the job market at the moment, the Wall Street Journal’s Eric Morath writes. All we know is that the economy has gotten battered by the shutdowns, but the measures we’d normally rely on can’t handle the rapid flux of changing policies:

Friday’s U.S. jobs report from the Labor Department is expected to show U.S. employers shed nearly 30 million positions from payrolls this spring as a result of the coronavirus pandemic and related shutdowns—but that is just one of several varying estimates of job destruction.

Other data suggest layoffs might have topped 40 million, while another count shows only about 20 million are tapping unemployment benefits. No matter the measure, job loss triggered by the pandemic is historically high and likely to leave a lasting mark on the U.S. economy.

Nonfarm payrolls fell by a combined 21.4 million in March and April, the Labor Department said. Economists surveyed by The Wall Street Journal expect the May employment report to show another 8 million jobs were lost last month—bringing the total decrease since the pandemic took hold in the U.S. to more than 29 million.

The nonfarm-payrolls figures historically have been the most closely watched measure of job creation and destruction. That is because they capture a net change in employment, balancing hiring and firing that are always occurring in both strong and weak economies. They also are based on a survey of 145,000 businesses and other employers, which is viewed as more accurate than asking individuals about their employment status, and can be benchmarked to tax records in the longer run.

True, but let’s also set some expectations for the BLS report, too. The Bureau of Labor Statistics calculates its jobs report (due on Friday) based on two surveys — one of households and the other of businesses. Those polls get conducted in the middle of each month, which means that the data we see reflects a period of time approximately two weeks old. Most of the time that doesn’t matter at all, but in this kind of crisis where emergency public policy greatly impacts economic activity and changes on a daily basis, it will matter, and matter greatly. On the other hand, ADP uses data from the same period as BLS (as it states in its report), so maybe we’ll get a pleasant surprise on Friday anyway.

Perhaps by July’s jobs report for June, we’ll get back to a more normal basis for these measurements. Until we reopen all of our public squares, however, we’d better get used to the ambiguity.