How economists blew the analysis of the manufacturing jobs shock

How economists blew the analysis of the manufacturing jobs shock

I came across this article yesterday, posted by – to his credit – Brad DeLong, whose argument it eviscerates. Entitled “The Epic MIstake about Manufacturing That’s Cost Americans Millions of Jobs,” it deserves widespread attention. So I am summarizing it here. But by all means go and read the entire piece.

Just to give you the frame of reference, here is the historical graph of manufacturing jobs in the US for the past 50 years:

After peaking in 1979, the number more or less gradually declined in the 1980s, and then stabilized in the 1990s, before plummeting right after 2000.


As written by Gwynn Guilford, the consensus of economists’ opinion was that while

the US had hemorrhaged manufacturing jobs, losing close to 5 million of them since  2000. Trade may have been a factor—but it clearly wasn’t the main culprit. Automation was.
For a decade or so, this phenomenon had been put forth by Ivy League economists, former US secretaries of treasurytransportation, and laborCongressional Research Services, vice president Joe Biden, president Barack Obama—and by Quartz too, for that matter. In a 2016 New York Times articletitled “The Long-Term Jobs Killer is Not China. It’s Automation,” Harvard economist Lawrence Katz laid out the general consensus: “Over the long haul, clearly automation’s been much more important—it’s not even close.”

Susan Houseman, an economist at the Upjohn Institute, and her colleagues, examined microdata available from the Federal Reserve, and discovered that this entire rationalization was based on technological improvements in only one industry — computers. As the article states, once they

 strip[ped] away the computers industry output from the rest of the data[, t]hat revealed just how the rest of manufacturing was doing—and it was much worse than what Houseman and her colleagues expected.

“It was staggering—it was actually staggering—how much that was contributing to growth in real [meaning, inflation-adjusted] manufacturing productivity and output,” says Houseman.

The culprit was “quality improvement.” The computing power of your smartphone is probably millions of times better than the Univac computer of the 1950s. In fact, it’s much better than its ancestor of only 10 years ago. So your new smartphone is clearly worth “more” than an equivalently priced smartphone of 10 years ago.

The problem is, this quality improvement was assumed to be general across all of American industry. It wasn’t. Strip away computer quality improvement, and, well:

according to Houseman’s data, without computers, manufacturing’s real output expanded at an average rate of only about 0.2% a year in the 2000s. By 2016, real manufacturing output, sans computers, was lower than it was in 2007.

Here’s the nut graph (if the formatting doesn’t show properly, the important thing to know is that the light blue line well below the others is real manufacturing output less computers):

To put this in more dinosaurian terms, imagine if American factories were churning out exactly as much of exactly the same stuff in 2019 as they had been in 1969. Obviously we would say that there was no growth at all. But now make one change: where in 1969 automakers were churning out 10 million 1969 Chevy Impalas, in 2019 they are churning out 10 million 2019 Honda Accords. OK, modern Accords are light-years better than the Impalas of long ago,  but we wouldn’t say that American industry across the board had improved.

That’s the mistake that economists made. Take away improvements in the computer industry, and American manufacturing really hasn’t made any progress at all – and still there are 4.5 million fewer jobs in manufacturing than there were at the end of 1999. And all of those jobs outside of the computer industry were due not to automation, but to trade.

Since the 2016 election, I’ve been leery of choosing sides between the “economic anxiety” trope and the “they’re all racists” theory, because I don’t see them as necessarily inconsistent. A white racist who is content with how things are going might vote for, e.g., Barack Obama, while that same racist, if stressed, will look for a scapegoat – like Mexicans and Muslims – to blame. Obama himself in a “Kinsey gaffe” referred to people “who cling to their guns and their religion.” That’s why in 2016 the national election result tracked so well with the economic voting models, while among the hardest-hit places in the 2016 “shallow industrial recession” were those of the upper Midwest that put Trump over the top.

(Dan here… Go here for a cogent video argument  start arround 17 minutes in to hear Susan Houseman explain.)