The jobs number is always wrong

The 2004 edition of the Economic Report of the president included a creative proposal it hoped the statistical agencies would consider: reclassifying fast-food cooks as manufacturing workers.

''When a fast-food restaurant sells a hamburger,” the report asked, “is it providing a 'service' or is it combining inputs to 'manufacture' a product?'' 

I hear you snickering, but it did make some fair points.

The report noted, for example, that “mixing water and concentrate to produce soft drinks is classified as manufacturing” — so why shouldn’t assembling a hamburger count, too?

The answer lies in the Census Bureau’s definition of manufacturing, which the job counters at the BLS follow: “the mechanical, physical, or chemical transformation of materials, substances, or components into new products.”

Heating a frozen hamburger patty does indeed create a “chemical transformation” — heat causes a burger’s proteins to unfold and reconfigure in ways that irreversibly change it.

(You can freeze and melt a soda as many times as you like and still drink it, but try that with a burger and you’ll regret it.)

It would be a stretch, however, to argue that heating a burger transforms it into a “new product,” so it’s no surprise that the BLS continued to categorize burger flippers as service workers. 

If the BLS rejected the White House’s suggestion on its merits, everyone else rejected it on its politics — a transparent attempt by the White House to make the manufacturing sector look healthier than it was.

It wasn’t the first time the seemingly mundane process of counting jobs became a political flashpoint. 

In 1971, the Nixon White House shut down BLS press briefings after the agency unenthusiastically described a 0.2% drop in unemployment as only “marginally significant” (the Secretary of Labor described the same data as “of great significance”).

A month later, a statistical error caused the BLS to overstate a further drop in unemployment, this time raising fears that the White House was manipulating the data to make the economy seem better than it was.

Investigations found no evidence of political influence on the jobs data, but the OMB nevertheless responded by issuing a directive that tightly restricted early access to the data for political appointees.

More surprisingly, there have also been accusations that the White House manipulated jobs data to make the economy look worse than it was.

In 1961, Reader’s Digest published an article accusing the Kennedy White House of using data techniques to “magnify the unemployment problem” as a pretext for more government spending and regulation.

Again, an investigation found no basis for the claim.

A similar investigation in 1944 dismissed similar claims that the BLS had "obsequiously acquiesced" to White House demands to understate inflation, with the goal of keeping wages down too (while the government had war-time powers to set wages).

All of these unfortunate episodes are recounted on the BLS website, which highlights just how much precedent there is behind President Trump’s new accusations of political bias at the non-partisan agency.

In fact, his shock decision to fire Erika McEntarfer wasn’t even the first time a BLS commissioner lost their job for political reasons.

In 1932, Ethelbert Stewart was “involuntarily retired” as head of the BLS for publicly disagreeing with the Hoover administration’s rosy portrayal of the Depression-era labor market.

In response, The San Francisco News opined that "in the city named for George Washington, it seems they fire people for telling the truth."

Now, by contrast, McEntarfer has been fired for the gravest form of not telling the truth: statistics.

On Friday, President Trump accused the BLS commissioner of “miscalculations” that he is sure were politically motivated. 

But every jobs report is a miscalculation — by design.

When the BLS reported on Friday that the US economy had added 73,000 jobs in July, McEntafer and everyone involved with the number knew it was wrong.

Like every month, July’s report was based on incomplete data: The BLS doesn’t wait for all 121,000 surveyed employers to respond. 

Instead, it goes with what it’s got at the end of the month — typically just 60% or so of what it’d like to have — and then updates its models as additional responses trickle in afterwards.

But even with all the data in, it’s still just an estimate based on a lot of assumptions.

Without adjusting for seasonality, for example, the BLS would have reported that the US economy lost 1,066,000 jobs in July.

The difference between 1,066,000 and the 73,000 that everyone thinks of as “the” number of jobs created in July is just one measure of how “wrong" the BLS’ model is.   

“All models are wrong,” as statistician George Box famously said, “but some are useful."

The BLS model that comes up with a monthly jobs number is one of the useful ones — an early warning system that allows businesses, investors and the Federal Reserve to adjust to the direction of the job market. 

The true size of the job market won’t be known until everyone reports their taxes in a year or so, as I’m sure the president is aware. 

But the president also seems acutely aware of the power of numbers to affect our perception of reality.

A hamburger, for example, is far more than its calorie count: It’s protein, a night out with friends and a piece of culture, too. 

But the moment you read "1,600 calories" on the menu, a bacon cheeseburger becomes something else entirely: judgment. Liability. Guilt.

Numbers often wield more power than the reality they attempt to represent, so we should of course try to get them correct.

That’s especially true with economic data, where feedback loops can cause the perception of a slowing economy to become the reality of a recession.

But with high-frequency data like non-farm payrolls, correct is the enemy of useful.

And always has been.

— Byron Gilliam

_{Brought to you by:}

Arkham is a crypto exchange and a blockchain analytics platform that lets you look inside the wallets of the best crypto traders — and then act on that information.

Arkham’s Intel Platform has a suite of features including real-time alerts, customizable dashboards, a transaction visualization tool, and advanced transaction filtering — all of which is accessible on all major blockchain networks, and completely free.

Arkham’s main product is the exchange, where users can express their trade ideas against the market.

_{Brought to you by:}

Katana is a DeFi chain built for real sustainable yield and deep liquidity. It concentrates liquidity into core applications and channels the chain’s revenue back to the users.

Creating a better DeFi experience that benefits the active users on the chain.

Earn boosted yield and KAT tokens: Deposit directly into vaults on the katana app and start earning on your ETH, BTC, USDC, and more.