It was shaping up to be another hot summer.
On Aug. 4, at 8:30 a.m. Eastern time, the Bureau of Labor Statistics released monthly hiring data for July. A pack of economists and journalists pounced, excited to spread the numbers across social media. Good news! The United States had added 209,000 jobs, an above-average result that exceeded the consensus prediction of Wall Street analysts.
“America’s economy has beaten expectations,” the Guardian reported. Stock markets jumped when trading opened an hour later; the Dow Jones industrial average climbed to a record high. President Trump was moved to open his Twitter account. “Excellent Jobs Numbers just released – and I have only just begun,” he said. “Many job stifling regulations continue to fall. Movement back to USA!”
Here’s the thing: The U.S. didn’t add 209,000 jobs in July. Not even close.
We’ve broken down the latest figures in the jobs report and explained the context you need to understand them. Explore our jobs report interactive »
One month after the announcement, the BLS revised its original estimate. The new number: 189,000 — decent, but unimpressive. And then in October, the BLS revised the July figure a third and final time: 138,000 — mediocre at best.
Don’t be too hard on yourself if you were still under the impression that the U.S. economy had had a summer for the ages. The shift from “excellent” to so-so did not receive much attention from the media. The mainstream press went bananas over the BLS’s headline employment number, as it usually does when the first estimate of the data comes out, and then barely acknowledged the revisions, also standard practice. This flawed approach affects how people view the economy.
“There is a lot misunderstanding around the mechanics of the jobs report,” said Matt McDonald, who is a partner at the Washington-based consulting firm Hamilton Place Strategies and writes a monthly analysis of the jobs report. “It can take on a life of its own.”
McDonald said there is no reason to believe the BLS’s results are manipulated for political purposes. Less-than-perfect figures are inevitable when attempting to make a monthly estimate of hiring and firing among a population of 326 million people. (The headline number that gets most of the attention each month is based on a survey of about 150,000 companies and government agencies; it changes because responses continue to come in after the initial report is published. The unemployment rate, another popular indicator, comes from a separate survey of about 60,000 households, adding another layer of complication.)
Even so, conspiracy theories about how the numbers are rigged have continued. Until recently, Trump himself was among the champions of those notions. In March, The Washington Post recounted 19 examples of the president dismissing the jobs data as fake before he took office. Example: “The unemployment number, as you know, is totally fiction,” Trump told an audience in Des Moines, Iowa, on Dec. 8, 2016.said: “I talked to the president prior to this, and he said, to quote him very clearly, ‘They may have been phony in the past, but it’s very real now.’”">1
FiveThirtyEight: How accurate are those job report numbers?
Innocent or otherwise, the overemphasis of politicians, journalists and traders on a preliminary attempt by the BLS to measure the labor market is problematic, and not only because it lends validity to the charge that the data is “fake.” The traders who buy and sell assets in the seconds after the latest figures are released and the politicians who make partisan statements in the immediate aftermath are basing those decisions on incomplete information.
Does that matter? Not in the way that increasing tension between the U.S. and North Korea does. But the willingness of decision-makers and people who can influence those decision-makers to jump on the initial BLS estimates is arguably an abuse of public trust. For example, in October, when the July hiring figures were reduced a second time, Trump didn’t go on Twitter to correct the record.
A little secret of the economics trade is that the jobs data is the statistical equivalent of a best guess. The 150,000 non-farm companies that are surveyed is an acceptable sample size from a statistical perspective, but it’s capable of providing only an approximation of reality. According to the BLS, the actual monthly change in the number of jobs likely falls somewhere in the range of 120,000 more or 120,000 less than its estimate. That means you can be confident that job numbers grew in a given month only if the change in monthly hiring is 120,000 or greater. That was the case in January, when the BLS reported an increase of 216,000. But consider March. The BLS estimated that 50,000 jobs were added then, meaning that the actual change in the number of jobs could be anywhere between a gain of 170,000 and a loss of 70,000.
To be extra clear: None of this is meant to suggest that the BLS is guilty of bad statistics, only that there are limits to a methodology designed to deliver monthly snapshots of a massive, diverse economy. “They are the best numbers we have,” McDonald said.
Rather, the point is that most journalists and many analysts are guilty of bad, or at least short-attention-span, reporting. And if you pay attention only to the initial hiring estimate that most mainstream media outlets blast out every month, you are operating with a skewed impression of what’s really going on in the economy.
Since 2003, the average change between the first estimate of a month’s job gains or losses and the third and final estimate is an increase of 11,000 jobs, according to the BLS’s Current Employment Statistics program, which conducts the survey of companies. The revisions in monthly hiring have been different this year, however; through August, the average change from initial to final estimate was a decrease of 11,250 jobs.
The BLS itself has long tried to make sure that the public understands the limits of its methodology. As Michael Calvillo and Tyler Downing, a couple of BLS economists, wrote in October 2016: “The CES program often discourages users from placing too much emphasis on data from any single month and encourages them instead to analyze data series over a longer term.”