Revisiting the China Shock (with Joseph Francis)

Russ Roberts frames the discussion around the influential work of David Autor, David Dorn, and Gordon Hanson (ADH). U.S. manufacturing employment fell by about 6 million jobs—roughly one-third—between 2000 and 2010, while nominal imports from China rose from about $100 billion to $365 billion, increasing from roughly 1% to 2.5% of GDP. ADH argued that Chinese imports caused a substantial share of the decline and later associated the “China shock” with lower wages, unemployment, reduced household income, welfare transfers, mortality, marriage-market outcomes, and Republican voting. Joseph Francis explains how they tried to identify causation rather than mere correlation.

Francis says ADH did not simply use U.S. imports from China, because those imports respond to U.S. demand and are therefore endogenous. They wanted to isolate China’s supply shock: the transition from communism toward capitalism, rapid productivity growth, and China’s increased ability to produce and export. Their instrumental variable used Chinese imports into seven or eight other OECD countries, mainly in Europe and Japan, as an supposedly exogenous measure of China’s expanding supply. Roberts notes that technological change, rising U.S. demand, and other forces were occurring simultaneously, making the central challenge separating China’s effect from general manufacturing productivity growth.

Francis describes a 2022 Review of Economic Studies article by Kirill Borusyak, Peter Hull, and Xavier Jaravel. They argued that ADH had not adequately distinguished the long-running decline of U.S. manufacturing towns from the portion specifically caused by Chinese imports. ADH’s design effectively compared areas with manufacturing against places with little or none, so any general decline in manufacturing could be attributed to China. The correction added controls comparing commuting zones with similar manufacturing levels but different exposure to Chinese imports. The estimated effect on manufacturing’s employment share fell by about half, from approximately 0.6 to 0.3.

Roberts recalls ADH’s claim that China explained about 25% of the manufacturing-employment decline from 1990 to 2007, and about 45% of the decline from 2000 to 2007—roughly 1.5 to 2 million jobs. Applying the later correction could reduce that estimate to approximately 700,000 to 1 million jobs. Francis cautions that the 25% and 45% figures were not directly produced by the raw regression coefficient corrected by Borusyak and colleagues; they involved an additional calculation. Even after the reduction, the localized losses could matter politically and humanly, particularly where many displaced workers struggled to find comparable employment.

Francis says he initially focused on manufacturing employment, as most discussion had, but then checked what the Borusyak-Hull-Jaravel corrections did to ADH’s other outcomes. On page 45 of the online statistical appendix, the corrected regressions showed that the previously reported negative effects on unemployment, wages, household income, and other living-standard measures disappeared. Yet ADH’s published article did not discuss this result prominently; it referred readers to an appendix table in a footnote. Francis finds this especially important because the living-standard effects, rather than the mechanical possibility that imports displace competing factory jobs, were the most consequential part of the China-shock narrative.

Roberts argues that cheaper Chinese imports could free U.S. resources for other goods and services, eventually raising living standards even though adjustment costs are real. Francis speculates that academic incentives may explain why the corrected living-standard results received little attention: young economists need publications in top journals and may avoid antagonizing senior scholars. The ADH work was cited thousands of times and became a political shorthand, regardless of whether readers understood its qualifications. Francis recalls a 2005 New York Times argument by Nicholas Kristof that elites used China as a scapegoat; he says Democrats initially promoted that framing and Donald Trump later helped make it a bipartisan consensus.

Roberts emphasizes that U.S. manufacturing employment had been declining steadily since at least 1945, even while manufacturing output remained strong. Producing more goods with fewer workers through technology and innovation is not evidence that the economy was being hollowed out. Francis’s economic-historian prior is that countries industrialize and then deindustrialize as rising wealth makes manufacturing work less attractive; globalization facilitates the process, and China may have accelerated it without causing it. Japan, Malaysia, or India might have supplied U.S. consumers had China remained a communist “basket case.” Both agree that identifying China as the sole cause risks misguided tariffs and industrial policy.

Neither speaker denies that trade creates localized losers. Roberts notes that workers may be unable to move because housing is expensive, while people with limited education may not find equally attractive jobs after manufacturing declines. Those adjustment problems can produce anger toward elites, even if national employment remains healthy and new jobs appear elsewhere. Francis says economists have long known, since work associated with Eli Heckscher in the 1930s, that international trade has different regional effects. He favors debating regional assistance or other ways to compensate losers rather than simply blaming China. His study of Argentina also makes him skeptical that protectionism and industrial policy are easy solutions.

Roberts recounts asking a seminar presenter how many regressions he had run while searching for a dramatic result. The presenter could not answer, illustrating Roberts’s concern that researchers can vary specifications, controls, and instruments until they find a publishable conclusion. Francis links this to the “credibility revolution,” when economists adopted sophisticated empirical methods after being criticized as overly theoretical. He thinks younger econometricians are now pushing back by demanding robustness, citing the corrections to ADH as an example. His title, “Whimsy and Fragility,” pays tribute to Edward Leamer, who argued in “Taking the Con Out of Econometrics” that results often depend on arbitrary assumptions and specification choices.

Francis favors a more old-fashioned economic history: time-series analysis, smaller econometric questions assembled into a broader account, theory-informed interpretation, and detailed description of what happened. He argues that deindustrialization involves multiple causes that a simple X-causes-Y regression cannot capture. Both speakers discuss the replication crisis and the file-drawer problem, in which null or inconvenient results remain unpublished. Francis calls economics peer review a largely opaque process: editors can override reviewers, relationships may matter, and published work is often treated as settled. He contrasts this with scientific fields such as Nature, where review reports and post-publication criticism are more visible. He says peer review is necessary but needs a vigorous, open post-publication process.