Adam Curtis is probably the most interesting documentary filmmaker of our time. His latest, Can’t Get You Out of My Head (2021), further develops the theses he presented in The Century of the Self (2002), All Watched Over by Machines of Loving Grace (2011), and Hypernormalization (2016). He weaves together the great turn inward after the failed revolution of the Sixties, the degeneration of the counterculture into narcissism and consumerism, together with the intellectual history of psychology, the rise of information technology as a new mode of power, and the attendant erosion of Western democracy and the rise of the technocracy and antipolitics. It’s always a heady cocktail.
There are a number of striking theses in Can’t Get You Out. He notes the impossibility of collective action in the age of individualism. With the disappearance of unions and other working-class intermediaries, politicians no longer had the power of mass democracy behind them to confront the entrenched power of the elites. In a haunting sequence, he describes how Clinton was the first Western politician to realize that ‘the power had gone’. With politics evacuated of any vision of the future beyond money, the politicians spun a dreamworld.
The capitulation of social democracy across the West consisted of politicians giving away their power to unelected technocrats — not just central banks but a whole host of other ‘non-majoritarian institutions’ like regulatory bodies, special commissions and suchlike. The rise of the technocracy was an antipolitics that evacuated policymaking of political bargaining and replaced it with bureaucratic rationality and apolitical technocratic expertise. This development, reminicient of Bauman’s analysis of the Nazi regime, happened everywhere but was most pronounced in the European Union.
He also develops the intriging thesis that whereas the political economy of coal-powered industrial societies admitted powerful working-class institutions — manifest in dramatic industrial actions by miners — the political economy of oil-powered industrialism empowered the elites instead.
The rise of individualism after the smashing of the midcentury world made the earlier mechanisms of power unworkable. But the isolated, freely-floating elementary particles of the middle-class could be manipulated in other ways; above all, through the weaponization of psychology. The old structures of power had not gone away. They had mutated. Control was now to be exercised discreetly, via the panopticon of survellience capitalism, together with constantly-refined techniques of microtargeting and subliminal manipulation.
The isolated elementary particles of the neoliberal world were also lost. In another haunting sequence, Curtis describes the suburban middle-class housewives at midcentury as the vanguard of the new world. They were the first to discover the inner emptiness of the isolated individual, and not coincidently, the first to take valium en masse. The idea that was hit upon, after the capitulation, was that your feeling that ‘something was wrong with the world’ said less about the world and more about your brain chemistry. Before long, hundreds of millions were being prescribed medications to stabilize their feelings and mood cycles. Psychotropic drugs became the preferred means of pacification in the brave new world. Tellingly, valium was developed by the same firm that would later bring OxyContin to the market.
As Curtis is weaving these threads together the narrative moves between Britain, the United States, Germany, Russia, the Middle East, and for the first time, China. We are presented the fascinating story of Mao’s wife, Jiang Qing, who led the Cultural Revolution. What Curtis is attempting, although India is conspicuously missing from the story, is nothing short of a global history. Herein lies the problem. For Curtis himself falls into the trap of telling simplistic stories.
Curtis argues that China financed the American lending boom in the mid-2000s even as its rise as a manufacturing power deindustrialized the United States, devastated American working-class communities and triggered the rise of Donald Trump. Neat story. But one that does not survive scrutiny.
The housing-finance boom of 2003-2008 had nothing to do with China or other high-savings countries. It was driven by a logic endogenous to global financial intermediation — the financial boom was the great sucking sound of the wholesale funding market. If any foreigners are to be blamed, it was the European banks whose ‘round-trip finance’ was central to what Shin called the ‘global banking glut’.
It is true that some communities were hit hard by the China Shock, as Autor et al. have argued. But, as we shall see, China played a minor role in the story of American deindustrialization, and deindustrialization was itself only partly responsible for the decline of working-class communities in America and their turn to Trump. The unbundling of the value chains of American manufacturing firms and the automation of many value-added processes in said firms played a much larger role than Sino-American economic relations in the decline of American manufacturing employment. And large-scale transformations of the American occupational structure that gave us the hourglass economy, evacuated of medium-skilled jobs, played a considerably larger role in undermining working-class families and communities than deindustrialization. (At its peak, manufacturing employed fewer than 20m workers.) In turn, the transformation of the occupational structure had more to do with the Clinton betrayal than what the manufacturing firms were up to.
We have previously debunked the China Shock theory of Trump. But because this theory refuses to lay down and die, we’re gonna try to assassinate it again. We obtain the Autor-Hanson-Dorn measure of exposure to import competition from China for 722 commuting zone from Dorn’s website. We obtain overdose deaths from the USDA and electoral data from MIT election lab by county. We then aggregate the data by the commuting zones. Our response is the swing to Trump — the difference in GOP vote share between 2012 and 2016.
If the China Shock caused the Trump swing, then the Autor-Hanson-Dorn measure of exposure to import competition from China should be correlated with the swing to Trump. However, the correlation could be confounded as follows. Exposure to China was greater in places with large concentrations of the industrial working class. If there was a working-class swing to Trump for an unrelated reason, that would generate a spurious correlation between exposure to China and the Trump swing in the cross-section of US commuting zones. We must therefore control for a proxy of working-class predominance.
We use the share of blue-collar manufacturing employment in 1990 — the proportion of working age individuals without a college degree employed in manufacturing — as our control. It captures the degree to which the industrial working-class predominated in a commuting zone. Ie, it captures how blue-collar the area is. If the China Shock theory of Trump is right, the correlation should be robust to this control. We use partial correlation coefficients with and without our control and with and without Winsorizing to test the theory. We Winsorize the data by throwing out 10 percent of the most extreme observations.
Figure 1 displays the scatter plots for the Trump swing against four features: overdose death rates, change in blue-collar mfg emp, level of blue-collar mfg emp in 1990, and log-transformed China shock (the correlation with the raw variable is even weaker). We can see that overdose death rates, the level of blue collar mfg emp, and the China Shock seem positively correlated with the Trump swing, while change in blue-collar mfg emp seems negatively correlated with the Trump swing.
Table 1 displays the partial correlation coefficients. The China Shock is weakly correlated with the Trump swing (r = 0.207, P < 0.001). The coefficient halves after we Winsorize the data (r = 0.103, P = 0.021). And it vanishes altogether and bears the wrong sign once we control for the level of manufacturing employment whether we use the full sample (r = -0.047, P = 0.208) or Winsorize (r = -0.061, P = 0.286). We can therefore rule out the China Shock theory of the Trump swing.
More interestingly, change in manufacturing employment is only weakly correlated with the Trump swing (r = -0.115, P = 0.002). Moreover, it vanishes after we Winsorize the data (r = 0.048, P =0.285). Furthermore, after controlling for the level of manufacturing employment it bears the wrong sign in the full sample (r = 0.126, P = 0.001) or Winsorized (r = 0.127, P = 0.010). This shows that decline in manufacturing employment does not predict a swing to Trump after controlling for working-class predominance.
Meanwhile, we find a robust correlation between the level of manufacturing employment and the swing to Trump (r = 0.357, P < 0.001). And although it weakens somewhat when we Winsorize the data (r = 0.225, P < 0.001), it remains virtually unchanged when we control for change in mfg emp in the full sample (r = 0.361, P < 0.001) or in the Winsorized subset (r = 0.219, P < 0.001). This result is congruent with our previous finding that any proxy for the working class predicts the swing to Donald Trump. Moreover, what happened in 2016 was the culmination of a long process of class-partisan realignment in the United States, as the following graph demonstrates. Note that class-partisan polarization intensified further in 2020 when Trump’s share of the vote actually increased in 72 percent of US counties.
We also look at predictors of the overdose death rate. We find that the China Shock is correlated with overdose death rates, but the correlation weakens after we Winsorize. It weakens even more if we control for the level of manufacturing employment. And vanishes altogether if we Winsorize and control for blue-collar predominance. Meanwhile, decline in manufacturing employment is correlated with overdose death rates. And this correlation survives Winsorization and controlling for working-class predominance. The same is true of our proxy for working-class predominance. In short, both working-class predominance and decline in manufacturing employment robustly predict overdose death rates while the China Shock does so only weakly after controlling for working-class predominance. This is congruent with Case and Deaton’s result that deaths of despair are more or less confined to whites without college degrees.
Finally, we also run a multiple regression of overdose death rates on our three features. We find that working-class predominance predicts overdose deaths with or without controls. So do decline in manufacturing employment and the China Shock, even after we control for working-class predominance. However, the slope coefficient of the China Shock vanishes after we also control for decline in manufacturing employment. This suggests that decline in manufacturing employment is an effective mediator between the China Shock and overdose deaths (since it “kills the coefficient”).
To sum up, we have documented a number of facts:
(1) The predominance of the working-class strongly and robustly predicts the swing to Trump — there was a generic blue-collar swing to Trump independent of the China Shock and deindustrialization.
(2) The correlation with the China Shock is confounded by (1) — areas exposed to Chinese competition did swing to Trump but no more than we would expect given the generic working-class swing to the GOP.
(3) Deindustrialization per se is uninformative about the partisan realignment — decline in manufacturing employment does not predict the Trump swing after controlling for the predominance of the working class.
(4) Overdose deaths are a working class phenomenon — working-class predominance is a robust predictor of overdose death rates.
(5) Deindustrialization is strongly associated with overdose deaths — decline in manufacturing strongly predicts overdose deaths even after controlling for working-class predominance.
(6) The China shock is also associated with overdose deaths. Moreover, the effect of the China Shock on overdose deaths was, reasonably enough, mediated by the decline in manufacturing employment.
The bottomline is that the China Shock did devastate many working-class communities in the United States. But it was not responsible for putting Trump in the White House. The logic governing the working-class revolt is endogenous to the United States and cannot be blamed on foreigners.