What this covers

Susan Houseman of the Upjohn Institute sits down with Russ Roberts to challenge the standard economic narrative about US manufacturing decline. Houseman argues that economists have systematically misread the data over the past two decades, mistaking a narrow statistical illusion for broad-based prosperity. The conversation revolves around a central claim: nearly all measured output and productivity growth in manufacturing traces to a single small sector—computers and electronics—while the rest of the industry has been stagnant or contracting. Rather than automation displacing workers, Houseman contends that trade has been the dominant force behind the loss of roughly one-third of manufacturing employment since 2000.

The discussion examines three layers of measurement error that distort the standard picture. First, quality adjustment in the computer sector—improvements in product performance captured through hedonic adjustment methods—accounts for the apparent productivity miracles, not manufacturing automation. Second, price index construction fails to capture price drops when sourcing shifts to cheaper foreign suppliers, systematically overstating domestic value added and understating import competition. Third, roughly half the workers producing manufactured goods are reclassified as services employees through offshoring to temporary agencies and contractors, distorting employment tallies. Houseman emphasizes that rapid, concentrated shocks to regional economies differ fundamentally from gradual technological change, with long-term consequences that trade policy should address. The pair also explore whether policy should focus on easing worker transition or reassessing trade relationships, and whether loss of manufacturing capacity might threaten future R&D capacity and national security.

Sharpest takeaway

Houseman argues that the standard narrative attributing US manufacturing job losses to automation and high productivity is a misinterpretation of the data: nearly all apparent manufacturing output and productivity growth is driven by the computer/electronics sector's quality-adjusted price declines, while the rest of manufacturing has been stagnant and trade, not automation, is the dominant cause of employment losses.

  • Real output and productivity growth in manufacturing is almost entirely attributable to one small sector—computer and electronic products—via R&D-driven quality adjustment, not automation of production.
  • Measurement conventions fail to capture price drops when sourcing shifts to cheaper foreign suppliers, overstating US value added and productivity and understating real imports.
  • Roughly half of jobs needed to produce manufactured goods are counted outside manufacturing, distorting employment figures.

The argument · threads5 threads · 25 claims
0.80

Quality adjustment methodology inflates apparent manufacturing output and productivity growth.

5 pointscentrality 5/5
  • The apparent robust real output growth and productivity growth in US manufacturing—going back to the late 1970s—is largely attributable to one relatively small industry, computer and electronic products, which has accounted for only about 15% of manufacturing output for decades yet grew at 20-25% per year (1997-2007) once quality-adjusted, while most other industries were stagnant or declining; remove computers and 1979-2000 output growth falls to about 45% of aggregate GDP growth and 2000s output is essentially stagnant or down about 12%.

    “it turns out that the apparent robust, real output growth and fantastic productivity growth in the manufacturing sector in recent years, as well as going back at least to the late 1970s was largely attributable to one relatively small industry--computer and electronic products.”

  • Adjusting prices for changing product quality (e.g., a bigger or better TV at the same price representing a real price drop) requires hedonic adjustment models that demand extensive data on how product attributes change and on how much consumers value those changes, making accurate quality adjustment extremely expensive and inevitably lagging.

    “The best way arguably to do this is to run what's called hedonic adjustment models... it just requires a lot of data.”

  • Manufacturing's share of GDP had been falling almost commensurate with its falling employment share, yet its real (inflation-adjusted) output growth had largely kept pace with the aggregate economy for many decades, creating an apparent puzzle that economists resolved by inferring lower manufacturing inflation and higher productivity.

    “manufacturing as a share of our Gross Domestic Product had been falling almost commensurate with the fall of employment share.”

  • Although computer output grew rapidly in quantity terms, falling prices in the sector mean its share of nominal GDP has remained roughly the same, which resolves the apparent paradox of a fast-growing sector staying a constant fraction of manufacturing.

    “in quantity terms, but not in--and remember, their prices are falling, so as a share of nominal GDP, the share has remained roughly the same.”

  • For autos specifically, the BLS has long run a survey of auto manufacturers that tries to determine how much of a price increase is due to quality changes, showing that quality adjustment is done for some high-value, frequently-changing products even though it cannot be done for everything.

    “for autos, it has run a survey of auto manufacturers for a long time that tries to get at how much of the price increase is due to quality changes.”

0.80

Trade, not technology automation, caused the unprecedented manufacturing job collapse of the 2000s.

5 pointscentrality 5/5
  • The strong narrative that decades of manufacturing job loss—including in the 2000s—was just about automation stifled an intelligent debate about trade and trade policy that should have happened long ago, and acknowledging trade's effects on communities earlier might have allowed more to be done and avoided the current trade-war extreme.

    “The problem with that is that it kind of stifled an intelligent debate about trade and trade policy that we needed to have long ago.”

  • It is misleading to conclude from the manufacturing statistics alone—as most economists did—that because productivity growth was high, automation and the march of technology were what displaced manufacturing workers, since that high measured productivity was driven by computer-sector quality gains rather than automation of production.

    “it would be misleading from these statistics alone, which most economists were doing, to jump to the conclusion that because productivity growth was so high in the manufacturing sector, that reflected automation.”

  • The output and productivity growth in the computer sector is driven by quality improvements coming from research and development—not from automation of the production process—which is fundamentally different from the automation story used to explain manufacturing job loss, and it occurred even as the US lost market share and production of computers and semiconductors shifted to Asia.

    “it would be misleading from these statistics alone, which most economists were doing, to jump to the conclusion that because productivity growth was so high in the manufacturing sector, that reflected automation.”

  • Manufacturing employment, after being relatively stable through the 1990s, fell by over three million jobs (about 20%) between 2000 and 2007, then by another two to three million in the Great Recession, dropping by roughly a third in a single decade—an unprecedented decline with devastating effects on regional economies, with employment still down about 25% from 2000.

    “Between 2000 and 2007 manufacturing employment declined by over three million jobs, or about 20% just in seven years.”

  • Both trade and technology affect manufacturing jobs and they often work together interactively, so simple decompositions like 'technology accounted for 70% and trade 30%' of job loss are not valid or possible; the strong narrative that job loss was almost all about technology is not supported by the data going back to the 1980s.

    “the simple decompositions that researchers and policymakers often want to do such as, 'Well, technology accounted for 70% ; and trade 30%,' is not a valid--it's not possible. You can't do it. They're interactive.”

0.71

Large sectoral shocks destroy regional economies in ways gradual change cannot and warrant policy attention.

6 pointscentrality 4/5
  • Bounding the size of the input-substitution mismeasurement effect from purchased materials inputs in their 2011 Journal of Economic Perspectives paper, the authors found it lowered measured figures by a few tenths of a percentage point—not huge, but big enough to care about—while noting this captured only purchased materials and not growing services offshoring.

    “it did seem to lower things by a few tenths of a percentage point. Not huge, but big enough to care about.”

  • Very large, fast shocks to a sector—like manufacturing's loss of a third of employment and roughly 20% of factories in a few years—are fundamentally different from gradual marginal change because they destroy both physical and human capital and can send regional economies into a downward spiral that takes a generation or more to recover from, a robust result in labor economics that should inform trade policy.

    “When you have very, very large shocks to a sector, like the manufacturing sector experienced in the 2000s, those are fundamentally different than the kinds of changes over longer periods of time, on the margin”

  • Because computer-sector value added comes from R&D and design done in the US while physical production (like phone screens) is offshored, less-educated workers who would have held production jobs lose opportunities, which is one of the factors driving the growth of inequality.

    “that is, in part, one of the factors--not the only one, but one of the factors that is likely driving the growth of inequality. That, some of these formerly good-paying jobs in the United States are now offshored.”

  • The United States no longer does consumer production of computers; what remains is largely defense and government computers, and the semiconductor industry, while still strong, has lost share across all chip types to Asia.

    “Computers, what are we manufacture in computers anymore? Not much. What we do is a lot of defense and some government computers.”

  • Once the measurement problem is pointed out, there is little debate among economists who understand statistics that the standard narrative is a misinterpretation of the data—that one industry drove apparent robust growth and productivity, that it was not about automation, and that the statistics masked great weakness in many underlying manufacturing industries.

    “it would be misleading from these statistics alone, which most economists were doing, to jump to the conclusion that because productivity growth was so high in the manufacturing sector, that reflected automation.”

  • An MIT university-wide study raised concern that losing manufacturing and its associated skills could adversely affect the US ability to do R&D, because research and development often (though not always) needs to be co-located with manufacturing, raising concerns about the country's innovative vibrancy and national security if it lacks a strong manufacturing sector.

    “oftentimes, not always, but oftentimes, R&D needs to be co-located with manufacturing. So, that raises big concerns about the vibrancy of our ability to innovate in this country if we don't have a strong manufacturing sector.”

0.65

Trade's true impact on domestic production is understated because import price switching goes unmeasured.

2 pointscentrality 5/5
  • Because price drops from switching to cheaper foreign sources are not captured, a given dollar increase in imports from a low-cost country (e.g., $100 million) can displace a substantially larger dollar amount of domestic production (e.g., $150 million) than the trade statistics show, so price-adjusting imports does not fix this and the impact of imports on US production is understated.

    “picking up $100 million worth of increased imports from a lower cost country, what kind of impact that has in the United States, because maybe it displaces $150 million worth of production here because of those price differences.”

  • Because price indexes track the price of a specific good at a specific outlet period to period, the one-time price drop that occurs when a buyer switches to a lower-cost supplier (e.g., a foreign producer or a discount retailer) is intrinsically not captured by the way indexes are constructed, regardless of how quickly statisticians add the new outlet.

    “When there is a substitution from one source to a cheaper source, you miss that price drop.”

0.56

Manufacturing employment data undercounts actual workforce by excluding temporary and contracted workers.

3 pointscentrality 4/5
  • Manufacturing employment statistics count only payroll employees of businesses classified as manufacturers, so many workers needed for or actually working in factories—such as temporary help agency staff and independent contractors—are counted in services or as self-employed and excluded from the manufacturing workforce.

    “you have to be a payroll employee of a business that's a manufacturer to be counted in the manufacturing sector”

  • On the order of half of all jobs required to produce manufactured goods are counted somewhere outside the manufacturing industry, and the practice of outsourcing to temporary help agencies and independent contractors appears to be growing over time, leading to undercounting of people involved in producing manufactured goods.

    “Our estimates suggest that on the order of half of all jobs required to produce goods--a good, manufactured goods--are counted somewhere outside of the manufacturing industry.”

  • Manufacturing output is measured as value added—the value of what a producer's labor and capital produce, less purchased inputs—on a dollar basis, then adjusted for price changes via extensive BLS price surveys to separate quantity changes from price changes; value added and GDP are the same concept.

    “We think of the value added of that steel producer, what its labor and its machines--capital--produce, less its purchased inputs. So, that's value added.”