The inadequate quantity and quality of American jobs is one of the most fundamental economic challenges we face. It’s not the only challenge: Poverty, inequality, and stagnant mobility loom large, as well. But in a nation like ours, where wages and salaries are key to the living standards of working-age households, all these challenges flow from the labor market problem.
OK, but this is a supposed to be an article about technology. What’s the linkage between technology and this fundamental problem? As a D.C.-based economist who’s been working on the issue of jobs and earnings for almost 25 years, trust me when I tell you that most policy makers believe the following:
“Yes, there’s a problem of job quantity and quality, but it’s largely a skills problem. Because of recent technological advances, most notably computerization, an increasing share of the workforce lacks the skills to meet the demands of today’s workplaces.
What’s more, the pace at which technology is replacing the inadequately skilled is accelerating—think robotics and artificial intelligence. These dynamics explain growing wage stagnation, wage inequality, and the structural unemployment of those without college degrees.”
Problem is, most of that is wrong.
Technology and employers’ skill demands have played a critical role in our job market forever, but they turn out to be of limited use in explaining the depressed incomes of today, or of the past decade.
Consider: The demand for college-educated workers has actually slowed quite sharply since 2000 and their real wages have been flat. If that fails to surprise you, you may well be someone who’s recently graduated and looked for work. If that does surprise you, you may well be a high-level economic policy maker.
Before I go any further, allow me to assert the following, and not just to inoculate myself, but because I really believe it: Technology is a hugely important force in economies across the globe. Neither I nor any economist I know would question that we should want the most skilled workforce we can get, not to mention the best educated electorate. There’s no question that those with more education earn more than those with less—the college wage premium is as high as it’s ever been. No question that the upward mobility of far too many disadvantaged children is thwarted by unacceptably high barriers to attending and completing college. No question that way too many people lack the skills they need to make it in today’s job market.
But a number of important new studies show that it’s not technology-driven skill deficits that are depressing wage and job growth. It’s the weak economy, not yet recovered from the Great Recession, it’s persistently high unemployment robbing workers at almost every skill level of the bargaining power they need to claim their fair share of the growth, it’s terrible fiscal policy, it’s large and persistent trade deficits, it’s imbalanced sectoral growth as finance booms while manufacturing lags.
The policy implications that flow from these findings are profound. Improving workers’ skills is obviously insufficient. Supply doesn’t create demand. In fact, there’s evidence that as demand for college-educated workers has tailed off, they’ve been moving down the occupation scale, displacing workers with lower education levels.
If we want to improve the quantity of jobs, we’ll have to do more to promote labor demand. We’ll need to worry less about robots and more about austere fiscal policy, imbalanced trade, weak capital investment, and bubbles and busts. If we want the jobs we create to be of higher quality, we’ll have to do more to lift workers’ bargaining power, by enforcing labor standards, raising minimum wages, and leveling the playing field for collective bargaining. Supply-side solutions targeting workers’ skills may well help the targeted individuals, but they won’t help raise the number and quality of jobs.
Technology and jobs
The impact of technology on work and wages is and always has been profoundly important. The most popular economists’ theory about how this plays out is “skill-biased technological change,” or SBTC.
The conventional SBTC story is simple. As technology in the workplace evolves, it raises the cognitive demands that employers make of their workers. How that affects individual workers depends on the extent to which they productively interact with the new technologies.
In the econo-mese of labor economics, the relevant terms here are complements and substitutes. If a new machine does what you do but does it for less, you’re a substitute. Not good. But if the nature of your job and the skills you bring to it enable you to use the new machine to produce more or better output than the machine could produce without you, you’re a complement. Tech change is “biased” in your direction. Pass go and collect a hefty paycheck.
Is SBTC a useful model? Over the very long term, yes. Over the near term, it can be misleading. It’s a good telescope and a lousy microscope.
Taking the long view, the complementarity of skill demands and educational upgrading is why, despite the large increase in their share of the workforce, the most highly educated workers still have the lowest unemployment rates. For as long as we have data on occupations, we can observe technologically-induced “occupational upgrading,” that has enabled the increasing share of skilled workers—complements to the process—to be absorbed into the workplace.
Isn’t it possible for technology to displace so many workers that it creates more problems than it solves? What if the technological advances in the workplace flip too many of us from complements to substitutes? The greater presence and improved capacities of computers, robots and artificial intelligence in the workplace have led some to believe that the pace at which technology is displacing workers has accelerated.
Yet here again, evidence is lacking.
There are at least two ways to evaluate this concern about increasing technological unemployment: macro and micro. From a macro perspective, if technology were contributing to more output with fewer workers (more precisely, fewer hours of work), productivity growth would accelerate. In fact, as shown in the figure below, the rate of productivity growth appears to have decelerated in recent years (productivity changes are notoriously jumpy so the figure includes a smooth trend to help reveal the recent deceleration).
It’s worth remembering that much of what we hear about technology replacing workers is anecdote. Still, there are some pretty compelling anecdotes out there, and given that data is the plural of anecdote, I’m certainly open to the possibility that there’s more here than meets the eye. But as I recently wrote, “the robots-are-coming advocates need to explain why a phenomenon that should be associated with accelerating productivity is allegedly occurring at a time when the trend in output-per-hour is going the other way.”
The micro research on this question is particularly interesting. Economist David Autor, along with various co-authors, has taken the most granular look at how tech change is interacting with jobs through the lens of tasks. He breaks these tasks down into three bins: routine, manual, and abstract.
Routine tasks are repetitive and rules-based, like highly repetitive production work. If you stand in the same place and do the same thing at regular intervals on a production line and you haven’t yet been replaced by a robot, you soon will be.
Manual tasks are relatively simple for people but hard for computers. The person who comes into my office to empty the wastebasket appears to be doing a pretty straightforward bit of work, but because I may move the wastebasket around, she ends up doing something—noiselessly and effortlessly identifying the trashcan—that’s easy for her but hard to automate.
Autor provides a nice example of what differentiates these two different kinds of jobs/tasks:
Modern automobile plants…employ industrial robots to install windshields on new vehicles as they move through the assembly line. But aftermarket windshield replacement companies employ technicians, not robots, to install replacement windshields. Why not robots? Because removing a broken windshield, preparing the windshield frame to accept a replacement, and fitting a replacement into that frame demand far more real-time adaptability than any contemporary robot can approach.
Abstract tasks involve high levels of reasoning, mastery of complicated stuff like statistical analysis, and nuanced communication.
Armed with this typology, Autor tackles both ends of the “technological unemployment” thesis and finds it wanting. He finds more techno-complements than substitutes at the high end of the wage/skill spectrum and fewer opportunities for automation at the low end. Repetitive production jobs will continue to go missing, but that’s a very long-term trend. Higher end jobs will continue to complement technology while the inability of computers to replace workers in manual jobs that require “flexibility, judgment, and common sense remain[s] immense.”
That’s jobs. What about wages?
Technology and wages
When it comes to wage trends, SBTC is even a less useful tool. Generally speaking, SBTC would predict a fanning out of wages by skill levels, with complementarity, or the skill bias, rising as you climb up the pay scale, and ever-increasing substitutability, or a negative bias, as you go down the scale.
The problem, as Larry Mishel and colleagues at the Economic Policy Institute have shown, is that this has not been the actual pattern of wages over the past few decades. Both in real and relative terms, actual wage trends have not moved the way SBTC says they should.
- While wages fanned out as SBTC would predict in the 1980s, they’ve generally failed to do so since. For example, mid-level wages fell relative to low wages throughout much of the 1990s.
- SBTC would predict that the wage premium of more highly educated workers would continue to rise, yet starting around the mid-1990s, that differential slowed, if not plateaued. This flat trend is inconsistent with the oft-made claim that SBTC is driving post-2000 wage inequality.
- SBTC would predict productivity gains would be reflected in the wages of workers with complementary skills to the new technologies. But productivity increases have diverged not just from the real pay of the lowest paid workers, but from those of middle and upper-middle wage workers as well.
David Autor in particular has tried to save SBTC by coming up with interesting ways in which it could still be tapped to explain these wage patterns. Most prominently, he and colleagues argued that based on the tasks-framework discussed above, SBTC in the 1990s was leading to a polarization of both jobs and wages (as opposed to the more linear impact of traditional SBTC: bad for the bottom, less bad for the middle, good for the top). Routine production jobs, which tend to pay solid, middle-class wages, got whacked because they’re easily mechanized (so they’re substitutes); manual, non-routine jobs at the low end of the wage scale did better because they’re much harder to automate; and high-end, high-paying jobs did well because they’re complementary to the new tech.
But things moved around again in the 2000s in ways that forced yet a new morphing of SBTC. Polarization was gone, and Autor wrote that the formerly “U‑shaped growth of occupational employment came increasingly to resemble a downward ramp in the 2000s,” with pretty strong growth at the bottom and not much in either the middle or top.
This turns out to be an important analytic problem. Researchers have made interesting and compelling linkages, at least at the micro level, between technology and tasks at work. But they’ve failed to tie these occupational employment trends to wage inequality. Mishel, et al, put not too fine a point on it (my bold): SBTC, polarization, and all that task analysis simply “fail to explain the key wage patterns in the 1990s [they] intended to explain, and provide no insights into wage patterns in the 2000s. We conclude that there is no currently available technology-based story that can adequately explain the wage trends of the last three decades.”
If you have to bend a theory like SBTC that much to explain the changing reality of wage trends over the past few decades, it’s probably not the right one. It clearly helps explain the long-term absorption of so many more college-educated workers, and that still secures the theory a place on the economics mantelpiece. But it doesn’t explain much about wages.
So, what can economics tell us about the impact of technology on jobs and wages?
While I still think that the long-term positive correlation between technology and skill demands will eventually reassert itself, right now we’re going through one of the rare periods when the evidence for it is especially elusive. Indeed, in the 1990s, it appears that employers over-estimated the complementarities between computer technology and high-skilled workers. We were still working off a tech-hiring bubble when the other bubble—in housing—burst in late 2007. Such an explanation tells you a lot more about bubbles and busts and their impact on jobs and wages than technology stories do.
Second, based on the macro and micro evidence presented above, I’m a lot less worried about the threat of automation than I am about the threat of bad policy that fails to offset the imbalances in demand, trade, income and opportunity. The automation threat is always a possibility and it demands attention. But the absence of a progressive economic policy agenda is a reality, not a threat.
Finally, EPI is right. The technology stories economists tell us fail to fit the wage data—which means we’ve got to get beyond skill solutions as our favored response to inequality and stagnation.
That’s not to disparage more and better education, which remains critical, especially to those who face ever-higher access barriers.
But with great respect for the very smart phone in my pocket and the computer I’ve been staring at for the last few hours, if we fail to deal with the lack of bargaining power suffered by workers across the spectrum of wages and skills, the bubble and bust cycles that have defined the macroeconomy in recent decades, the erosion of labor standards, the austere fiscal policy, the sectoral imbalances, and the absence of full employment, too many of us will continue to experience eroding living standards—regardless of technological progress.