Are robots stealing workers' jobs? At last week's Democratic presidential debate, CNN moderator Erin Burnett dove into the thorny issue.
"According to a recent study, about a quarter of American jobs could be lost to automation in just the next 10 years," she said, asking candidates how they would respond to this problem.
The question ultimately pitted two divergent worldviews against each other. One came from entrepreneur Andrew Yang, who has centered his campaign on what he perceives as a massive employment threat from automation. His prescription is a universal basic income program.
The other point of view came from Massachusetts Sen. Elizabeth Warren, who took issue with Burnett's premise.
"So the data show that we have had a lot of problems with losing jobs, but the principal reason has been bad trade policy," she said. "The principal reason has been a bunch of corporations, giant multinational corporations, who've been calling the shots on trade, giant multinational corporations that have no loyalty to America."
The question of whether trade or automation affects employment more not only divides economists, but appears set to drive major questions in the Democratic primary.
Fact-checking Burnett's question
There are two fact checks here. One is of Burnett's question and one is of Warren's contention that trade has driven job losses.
In her question about what candidates would do to combat automation-related job losses, Burnett seemed to be referring to a January 2019 report from Brookings Institution (NPR has reached out to CNN to confirm this; it has not responded). The authors of the paper noted this in their own post-debate blog post.
One author says Burnett was misstating their conclusions.
"The question distorted what we and other and many analysts say about how this will play out," said Mark Muro, senior fellow and policy director of the Metropolitan Policy Program at the Brookings Institution.
The report found that about one-quarter of jobs have a "high susceptibility" to automation — meaning that 70% or more of the tasks in those jobs could potentially be automated using existing technology. But that doesn't necessarily mean those jobs will disappear.
"If you play that out, that's very different from saying 25% of jobs are going to be liquidated," Muro said.
Fact-checking Warren and Yang
Even if Burnett was overstating the report's findings, it's nevertheless true that automation takes a lot of blame for U.S. job losses, particularly in manufacturing. Warren said that blame is misplaced.
"So the data show that we have had a lot of problems with losing jobs, but the principal reason has been bad trade policy," Warren said.
There is good evidence for what Warren said, but there isn't total consensus among economists, and it's also true that trade and automation aren't entirely separable phenomena.
Moreover, how correct Warren is may depend upon which time period she's talking about.
"Automation and productivity growth has played a huge role in the decline of manufacturing as a share of employment in the postwar period," said David Autor, professor of economics at MIT.
The evidence often cited to show this is that manufacturing employment as a whole has declined massively in recent decades, while productivity has generally climbed.
But more recently, Autor contends, trade played a bigger role.
"When people think about the implosion of U.S. manufacturing post-1999, trade was more to blame for that," he said.
What happened post-1999 is that the U.S. normalized trade relations with China, and China entered the World Trade Organization. That gave China better access to global trade, so it could sell more of its goods worldwide (and, in addition, allow other countries to sell more of their goods to Chinese consumers). Autor and his co-authors have found that the resulting economic shocks of this increased trade with China accounted for up to 40% of the manufacturing job losses between 2000 and 2007.
And there is some evidence to support that those drops were larger than those caused by automation, as Warren said. One of the most-cited studies on the topic is a 2018 analysis from the Upjohn Institute for Employment Research, a Michigan-based think tank.
In the study, highlighted by multiple analyses of the debate (like this one from Vox), economist Susan Houseman argues that the computer and electronics industry drove much of that productivity growth, and that for other areas, automation's effects have been overstated. Their conclusion: "trade significantly contributed to the collapse of manufacturing employment in the 2000s," but there's "little evidence of a causal link to automation."
In addition, a 2018 study from economists at the University of Maryland (and cited post-debate by former Treasury economist Ernie Tedeschi) found that trade with China led to a decline in employment levels twice as big as the decline caused by the "adoption of industrial robots."
But then, they still found a sizable impact from automation: "our estimate is that robot adoption between 1999 and 2018 reduced employment by about 1.1 million jobs." That's smaller than their trade estimate, but certainly not nothing.
Yang's team, meanwhile, told NPR that he believes that automation is a bigger factor than trade. As evidence, they pointed to a 2017 analysis from researchers at Ball State University. That study found that trade accounted for about 13% of manufacturing job losses between 2000 and 2010, while productivity gains accounted for nearly 88%.
But again, the underlying truth is more complicated. Michael Hicks, a professor of economics at Ball State and co-author of that study, pointed out in an email to NPR that "productivity" does not equal "automation" — it can also refer to things like factory workers' increasing education levels, other types of technology and more efficient processes.
And while Hicks believes productivity is a much bigger factor, he also said that trade should not be discounted: "We have always said that in the 2000s, trade-related job losses were important, just not as important as productivity-related job losses."
In other words, while there's evidence that trade in recent years has driven job losses, there's certainly no consensus on which has disrupted labor markets more.
On top of that, trade and automation aren't mutually exclusive forces; in some ways, one can propel the other.
"Offshoring depends on all kinds of technologies, and we viewed software as a form of automation," Muro said of his team's study. "And so absent strong Internet, all kinds of software and so on, there's no globalization. So these things are deeply entwined."
What the candidates are really trying to say
It's easy to get lost in the which-phenomenon-mattered-more weeds. But what also matters is which points the candidates are trying to make.
In bringing up job losses from trade, Warren was warning against "bad trade policy," arguing that "bad policy" tends to disproportionately benefit massive corporations.
In looking to the future, though, it's not clear which kinds of shocks will result from even more global trade, because there are no more behemoths like China to rattle the U.S. job market.
"The China shock is largely kind of behind us in the sense of, it's not going to happen again in that form," Autor said. "There's not an equally populous country that I don't know about that's about to become part of the world trading system."
That doesn't invalidate Warren's ultimate argument about corporations' involvement in trade deals, but it does complicate this debate, because presidential candidates are trying to determine how to shape future policy. While trade may have led to more job losses than automation in the past, Autor also added, "That doesn't mean that in the long run automation won't be as or more important."
Yang, meanwhile, is trying to argue that automation is an existential threat to many American jobs. But even if that's true, it's an entirely separate question as to whether his universal basic income is the right prescription for that problem.
Ultimately, one important takeaway from this debate is that it's easy to oversimplify the effects of both trade and automation on the U.S. job market.
Trade can lead to offshoring of some jobs, but it also often leads to shifts in employment, from one sector to another; it doesn't necessarily lead to lower total employment.
Similarly, a new technology can replace a worker, but it can also simply alter a job's tasks. One common example is that after the introduction of ATMs, bank tellers remained employed (indeed, their numbers grew); they just handled less cash and ended up doing more work marketing for their banks.
This doesn't mean that trade and technology don't matter. They will continue to affect and even displace some workers. And a worker who loses a job likely doesn't care whether a robot or offshoring is to blame; the worker just wants a new job. And to Muro, it's important to figure out a way to support that person, regardless of cause.
"The U.S. does a terrible job of supporting worker transition and adjustment and helping people when things break down," he said.