Political Machinery: Did Robots Swing the 2016 U.S. Presidential Election?



Carl Benedikt, Thor Berger, and ChinChih Chen

Was the outcome of the 2016 U.S. Presidential Election shaped by workers losing out to automation? According to a recent poll of unemployed Americans who were able to work, 37 percent stated automation as a prime reason for their misfortunes (Hamel et al., 2014). Moreover, a staggering 72 percent of surveyed American’s fear a future in which computers and robots can do more human jobs, while 85 percent favor policies to restrict the use of machines to hazardous jobs (Pew Research Center, 2017). Even though the causes of the populist backlash in America and Europe are far from conclusive, parallels have been drawn with the machinery riots of the British Industrial Revolution, when “Luddites” smashed power looms in fear of losing their jobs. A post-election article in The Wall Street Journal featuring the headline “Trump’s focus on jobs, globalization and immigration tapped anxiety about technological change,” speaks to the frequent belief that automation was a real cause of voter concern. Despite such beliefs, empirical efforts to examine the extent to which automation shaped the outcome of the U.S. Presidential Election have remained scant. What is clear is that the vote for Donald Trump was a vote against the status quo: according to the Exit Polls, 82 percent of voters believed that the Republican candidate would perform best in bringing about change, while the corresponding figure for Hillary Clinton was a meager 14 percent.

This paper examines the link between workers exposure to automation and voting patterns in the 2016 U.S. Presidential Election through the lens of economic history. Our analysis builds on two sets of observations. First, technological change is rarely a Pareto improvement: as automation has made inroads into a wider set of industries and occupations, it has left a sizable fraction of the workforce worse off. In particular, the sharp reduction in middle-income jobs in the U.S. economy cannot be explained without reference to the disappearance of “routine jobs”—i.e., occupations mainly consisting of tasks following well-defined procedures that can easily be automated (Autor et al., 2003; Acemoglu and Autor, 2011). As traditional middle-income jobs have dried up, many workers have shifted into low-income service occupations (Autor and Dorn, 2013), while others have dropped out of the workforce altogether (Cortes et al., 2016a). According to Eberstadt’s (2016) timely book Men Without Work, 24 percent of primeaged men in the U.S. will be out of work by 2050 at current trend. A prime explanation is the robot revolution, which has contributed to both joblessness and wage reductions especially among American men (Acemoglu and Restrepo, 2017).

Second, the economics of automation cannot be separated from its politics. For ordinary workers, their skills constitute their capital; it is from their human capital that they derive their subsistence. Because automation is accompanied by creative destruction in employment, which often comes with social costs—including vanishing incomes, forced migration, skill obsolescence, and episodes of unemployment—it threatens not only the incomes of incumbent producers but also the power of incumbent political leaders (Acemoglu and Robinson, 2013). The reason is simple: if workers that have lost out to automation do not accept labor market outcomes, they will resist the force of technology through nonmarket mechanisms, such as political activism (Mokyr, 1990, 1998; Mokyr et al., 2015). The British Industrial Revolution provides a case in point. The downfall of the domestic system—which was gradually displaced by the mechanized factory—inflicted substantial social costs on workers, leading them to rage against the machines that pioneers of industry marveled about. The 1779 riots in Lancashire and the Luddite risings of 1811 to 1813, are only two of many attempts to bring the spread of machines to halt (Mantoux, 2013). Although the Industrial Revolution began with the arrival of the factory, it ended not just with the construction of the railroads but also with the publication of the Communist Manifesto. While the accelerating pace of technological progress paved the way to modernity, it also bred many political revolutionaries along the way.

Against this background, we examine if robots caused American voters to opt for radical political change. As shown by the Exit Polls, the election of President Trump was unquestionably a vote against the status quo. Building on a recent study by Acemoglu and Restrepo (2017), showing that adoption of robots has caused nonemployment and wage reductions, we explore if robots shaped the outcome of the 2016 U.S. Presidential Election. Figure 1 provides a first glance of our key finding, documenting the positive relationship between differences in in votes cast in the 2016 and 2012 elections for the Republican candidate and changes in the exposure to automation across electoral districts. We show that this relationship holds also when controlling for a range of other demographic and economic factors, as well as differences in exposure within states serves to show that similar patterns are also evident when factoring out historical state-level divisions along party lines. To account for the potential endogeneity of robot adoption, we present additional instrumental variable estimates that exploit historical differences in industrial specialization across local labor markets and the adoption of robots in countries other than the U.S. to show that this relationship is presumably causal. As a final empirical exercise, we estimate how the outcome of the 2016 election would have changed under different counterfactual levels of robot 3 adoption. Had the adoption of robots been 2 percent lower between 2011 and 2015, all else equal, the electoral college would have been won by the Democratic candidate. Thus, our findings bolsters the view that that an increased level of automation in the U.S. labor market in recent years tilted the electorate into opting for radical political change.

View the full pdf below:
Carl Benedikt is Co-Director of the Oxford Martin Programme on Technology and Employment at the Oxford Martin School, and Economics Associate of Nuffield College, both University of Oxford. He is also a Senior Fellow of the Programme on Employment, Equity and Growth at the Institute for New Economic Thinking in Oxford, and the Department of Economic History at Lund University. His research focuses the transition of industrial nations to digital economies, and subsequent challenges for economic growth, labour markets and urban development.
Thor Berger is a Post Doctoral Fellow at the Department of Economic History, Lund University and an Associate Fellow at the Oxford Martin Programme on Technology and Employment, University of Oxford. His research interests straddles the intersection of economic geography, economic history, and urban economics and focuses on how technological advances—spanning the potato’s introduction in the 18th century, to the 19th-century railroads, to advances in Machine Learning in the 21st century—affects employment, inequality, and regional or urban growth. 
Chinchih Chen is a Postdoctoral Researcher of the Oxford Martin Programme on Technology and Employment at the Oxford Martin School. She obtained her PhD at London School of Economics and Political Science in 2014. Her primary areas of specialisation are economic geography and applied microeconometrics. Secondary fields are international trade and regional economics.

Copyright © Oxford Martin School, University of Oxford 2018. All rights reserved.

Original report was published here.