Interview with David Autor

David AutorPhoto by Peter Tenzer

Trade liberalization and labor markets

Region: Ten or 15 years ago, there was a broad consensus among economists that trade liberalization was largely beneficial, at least in theory.

Autor: Yes, beneficial in aggregate—and just as important, there were no detectable adverse consequences in terms of income distribution.

Region: Right. But as China became a global economic power, the consensus that trade liberalization was benign grew less solid. Trade with China—and globalization more broadly—appeared to be contributing to unemployment and inequality in the United States and elsewhere.

You’ve looked at this very closely, often with David Dorn and Gordon Hanson, analyzing the impact of import competition, from China in particular. In a recent working paper, you suggest that economists must “more convincingly estimate the gains from trade,” and you argue that “developing effective tools for managing and mitigating the cost of trade adjustment should be high on the agenda for policymakers and applied economists.”

What empirical evidence led you to so strongly question the theoretical benefits? That paper looked at job churn and lifetime income, and your Journal of Labor Economics paper with Acemoglu and Price as well as Dorn and Hanson, estimated job losses as high as 2 million to 2.4 million.

Autor: That 2 million number is something of an upper bound, as we stress. Stepping back, I was raised on the consensus view that trade—and globalization more broadly—had not been an important disequalizing force for income distribution in developed-country labor markets. A large chunk of my thesis research analyzed the sources of rising inequality. If you were studying inequality at that time, you were very aware of the debate about whether this trend was primarily due to trade, to technology, to unions or to all kinds of other forces.

The trade and technology discussion was interesting because trade economists and labor economists came at the same problem from very different angles and reached the same conclusion: that globalization had not been a big factor in the rise of U.S. inequality. That was the right conclusion for that time period. The data that they were looking at was through the early ’90s. A lot of the final papers on that topic were published around 2000 in the Journal of International Economics by Krugman and by Leamer. So it was a consensus view.

This experience taught me that if you’re studying the impact of technological change, which is a general equilibrium phenomenon, you need to be thinking about all of the general equilibrium forces, trade being the most natural one. So I’ve been teaching trade in grad labor since I was hired by MIT in 1999.

Region: So you were teaching this as the transformation was taking place, in a sense.

Autor: Yes, but it was about five years ago that David Dorn, Gordon Hanson and I started working on this. And we started on it because, one, the facts were striking, in terms of how much globalization had progressed since the conversation ended around 2000, and two, we felt we had a methodology to get at it that we didn’t have before.

The first paper that Gordon, David and I wrote built on the work that Dorn and I had done in another paper on labor market polarization that appeared in the American Economic Review in 2013. That paper broke U.S. labor markets into commuting zones. The U.S. Department of Agriculture developed the CZ concept. Basically, they are contiguous counties where most of the people live and work within the same county cluster and where the centroid of the county is commutable from the edges of the CZs.

It was Dorn who discovered them in his doctoral thesis work, in the sense that they were out there, but no one outside the Department of Agriculture had used them for economic research. Following Dorn’s lead, we started using them because we thought they had a lot of attractive properties—they’re a sort of “revealed preference” measure of local labor markets. They cover the entire mainland United States, and their boundaries can be drawn consistently over time (unlike metropolitan statistical areas, which are regularly redefined as populations shift). Now a lot of scholars are using them, which is great.

That’s what got us started. We thought we had a convincing way to look at the local labor market consequences of rising trade with China. One, we had a lens, the CZs, to look at very small geographic areas. And two, we had a strategy for isolating what we thought of as the plausibly exogenous variation coming from China’s growing export capacity by looking at changes at the goods level in U.S. exports to all these other countries.

The basic idea is simple: We see China’s share of U.S. manufacturing goods consumption rising rapidly in the 1990s and even more so in the 2000s. Is this due to changes in China’s competitive position—lower prices, higher quality—or is it due to shifts in U.S. consumer tastes or even due to declines in U.S. production capacity (for example, our furniture manufacturers suddenly ran out of wood)? It’s hard to distinguish these explanations looking only at U.S. data.

Photo by Peter Tenzer

To make progress, we studied China’s exports to eight other high-income countries simultaneously in each of 392 goods categories (covering all of manufacturing). Our idea was that if Chinese exports to the United States are driven in part by falling costs and rising quality, then Chinese import penetration in other rich countries in precisely these same goods categories should rise in parallel. And this hypothesis is strongly confirmed by the data. The bivariate correlation between the rise in China’s market penetration at the product level in the United States and these eight other countries ranges from 0.55 to 0.96.

For our analysis of the impact of the China shock on U.S. labor markets, we use only the component of rising China—U.S. import penetration that is shared with these other countries—that is, we use the component that we can confidently attribute to China’s improved competitive position.¹

We take these predicted changes in import penetration into the United States by goods category, and then we project that down to these commuting zones, looking at the geographic structure of U.S. manufacturing employment. Manufacturing is always very geographically concentrated. When you’re talking about furniture, for example, you’re talking about Tennessee or the Carolinas; you’re not talking about 50 states making furniture. The same is true if you’re talking about toys or leather goods or textiles—they’re very localized.

So that was the strategy; we thought it was a good idea, and we didn’t have a strong prior about what we would find. We thought we would find contraction of import-competing manufacturing employment, and we did. That was not surprising, but what happens after that was an unknown.

We were quite startled by how slow and incomplete the adjustment process was and the fact that you didn’t see offsetting gains in employment in other sectors. You see people entering unemployment or exiting the labor force, and wages falling modestly, but much adjustment was on the employment margin, not the earnings margin.

That is another thing that differentiated what we were doing. Historically, trade economists have relied upon full employment models, where people may lose jobs but they don’t lose work per se because they are quickly reallocated across sectors. According to such models, you should expect markets to clear by wages falling. But, in fact, what tends to happen is that people lose employment, and wages don’t really change for those who stay in employment.

Similarly, if we were in full general equilibrium all the time, the commuting zone would be irrelevant as an outcome measure because markets would clear nationally. The local shock is geographically dissipated because there’s effectively a law of one price of skill. And if that law did not hold in the very short run, workers would move to areas with higher wages until wages were again equalized. But it turns out that not only are the trade shocks locally concentrated—due to the concentrated geography of manufacturing—they also primarily play out locally. Much of the pain of adjustment is borne at the point of impact.

Region: And this is seen in the fact that market adjustment is so slow.

Autor: Yes. At the level of commuting zones, looking at workers initially in the impacted industry, you just don’t see the kind of diffusion or reallocation that general equilibrium models suggest. It’s not that it doesn’t happen eventually, but it happens slowly and painfully.

Anyone who has studied trade theory at even a basic level understands that trade is not Pareto improving: Trade causes prices to adjust, and so it creates winners and losers. It’s the Stolper-Samuelson theorem, that goods prices feed into factor prices, so if you’re an owner of a factor that suddenly becomes relatively abundant due to trade, you have a real fall in income.

And that’s the good scenario—which occurs in a world where there are no frictions! In a world where there are frictions, you have those same costs plus the adjustment costs.

Region: And that is something trade economists hadn’t looked at: the adjustment costs.

Autor: They hadn’t generally seen them as a first-order concern. To a labor economist, large worker adjustment costs are not altogether surprising. There’s an eminent labor literature starting with Jacobson, LaLonde and Sullivan in 1993 that shows when people are involuntarily displaced from long-term jobs, they experience deep enduring scars, in terms of lost earning, lost employment and even lost years of life. We know that the adjustment process is just as costly for career workers.

So at some level, it’s not surprising that we would find those same things when we looked at a trade shock. But we were startled by the magnitude and the robustness of the results. When we wrote that paper that you saw in February, that was a summary paper for the Annual Review of Economics. It encapsulated the last five years of work we have done, and we didn’t think it was going to get much attention, actually, because putting it in the Annual Review is basically saying, you know, graduate students will have to read this paper because it will be on their syllabus. It’s not exactly splashing across the CNN home page.

So we were startled when it got so much attention. One part of this is that we did not write it as a cut-and-dry literature review. We took the opportunity to say, “Here’s what we’ve done, here’s how we think about it, here’s how we think you should think about it.” It’s thematic much more than a typical journal paper. And then the timing with the presidential campaign and all of the contentious rhetoric about trade, and so on … it really got picked up.

Trade and family structure

Region: Let me ask you about one of the scars caused by trade shocks. In other recent work, again, with Dorn and Hanson, you studied the effects of rising import competition from China on marriage rates and child rearing in the United States and discovered that import shocks that affect male (not female) employment reduce marriage rates and fertility and increase the percentage of children living in poverty and/or single households. It’s not altogether shocking—it’s part of Gary Becker’s classic model of marriage—but the implications are really disturbing. Can you talk a bit about the study and the findings?

Autor: Sure. This is related to another line of work that I’ve done with my student, Melanie Wasserman (now an assistant professor at UCLA Anderson [School of Business]), David Figlio at Northwestern, Jeff Roth at University of Florida and Kris Karbownik, also at Northwestern.

Region: This is the Florida gender gap study? I’d like to ask you about that later.

Autor: Yes. Melanie and I started on this line of work a few years ago, trying to look at the interaction between the labor market and the marriage market. There has been this dramatic decline in marriage rates among low-education, noncollege-educated adults in the United States. And it’s very clear, too, as a stylized fact, that for groups of men who have experienced real declines in earnings, marriage rates have fallen dramatically for women of the same race and education groups; that is, the women who would be most likely to marry them.

Our operative hypothesis is that the declining [labor market] opportunities of low-skilled males have changed household structure, arguably to the detriment of kids, even more so for boys, in fact.

That was the idea we wanted to explore using the trade shock. Our thought was that the trade shock provides a very concentrated change in the opportunity set faced by workers, but particularly men; let’s see if that plays out in terms of reductions in marriage and changes in household structure.

We’re still working on that hypothesis. We have a much more detailed set of results that are not yet in the paper about how the wage structure changes and specifically how it falls: Men’s wages fall sharply relative to women’s wages. We also find, also not yet in the paper, that the trade shock causes a fall in the virtual sex ratio—that is, the ratio of men to women in a given age bracket—within the affected commuting zone. We’re trying now to determine if that’s due to incarceration, mobility or military service (or mortality—though probably not enough to move the sex ratio needle).

And, yes, the trade shock leads to a decline in fertility, which we expected. It does not lead to a rise in nonmarital fertility, but it does lead to a rise in the share of births that are nonmarital births. And, of course, due to the reduction in marriage formation, it means more kids are going to be living in single-parent, impoverished households.

I want to emphasize that my [research] agenda is not to document the ill effects of trade on outcomes A through Z. One part of the agenda is to say, “Hey, here’s a remarkable historical trade event: China’s swift emergence as a technologically advanced, low-cost manufacturing leader. Let’s use that to study how a rapid shift in the trading landscape affects labor markets in competing countries.” Someone once said, “Never let a serious crisis go to waste.” As a labor economist, I’d say the same thing about a major exogenous shock: Don’t waste it; let’s understand how it plays out in labor markets.

And the second part of the agenda is to think about it more broadly; China’s rise provides a big economic shock that’s distinct and well measured at the industry and geographic level. We can use that shock to examine lots of outcomes—not just trade patterns—where we want to know, “What does change in the opportunity set faced by workers do, in that case, to household structure?”

This agenda is far from complete. We have another paper on the impact of the China shock on U.S. political outcomes. We have another studying patenting and innovative activity by firms that are in sectors facing sharp increases in competition.

The gender gap

Region: I’d like to return to some of that work, but let me first jump to a paper you just referred to, on the gap in achievement levels between boys and girls, finding that girls do better in terms of graduation rates and a variety of later life measures, and that gap is greater for African-American children than for whites.

Autor: That’s right. And for kids from low-income households, low-education households, and for single-headed versus married households.

Region: In May, with several co-authors, you released a study that looked at achievement by brothers and sisters in disadvantaged families in Florida and found that boys appear to be more sensitive to disadvantage than girls. And that fact accounts for some of the black/white difference in prevalence of the gender gap.

Autor: Right. Basically, disadvantage (broadly construed) is more prevalent among black than white households, and it matters more for boys than for girls. Therefore, black boys are on average more disadvantaged relative to their sisters, as compared to white boys relative to their sisters. The sibling gender gap is larger in black than white households in part because the average level of disadvantage is greater in black than white households.

Region: Tell me a bit about how you developed this data set on families, children and outcomes.

Autor: Sure. We’re drawing on a remarkable data set that David Figlio has pioneered with Jeff Roth over the course of many years, where he’s worked with the state of Florida to match birth records to schooling records to some juvenile detention records and eventually, we hope, to labor market outcomes and incarceration outcomes.

We start with the universe of Florida births between 1991 and 2001 and match them to public school records for the 80 percent of kids who end up in the Florida public schools; others move out of state or go to private schools. We can match kids to their moms. So for a woman with successive births, we know that it’s the same mother.

In many case, we can also determine if it’s the same father or a different father. There are three statuses for paternity—which turns out to be one of the crucial variables: One, the parents are married; two, paternity is claimed, but the parents are not married; and three, no paternity is claimed.

Photo by Peter Tenzer

You might think, “Oh, well, maybe they just didn’t get around to getting married.” But those are incredibly strong predictors of the household trajectory or environmental childhood experience. If we see a mother having a second kid and she was married on the prior birth, there’s an 80 percent probability she’s married on the next birth. If paternity was claimed, but they were not married on the prior birth, there’s a 60 percent probability that will be true on the next birth. If no paternity was claimed on the prior birth, 60 percent probability that that’s true on the current birth. In many nonmarital births, the father will differ across births.

Family structure was one of the key things we wanted to look at. Our initial hypothesis was that father absence itself was the big driver of the gender gap. But, of course, father absence is correlated with education, with income and with race. So it’s hazardous to simply say, “Oh, it’s father absence” or “It’s low education” because as soon as you start looking across those groups, you’re immediately also dividing on black versus Hispanic versus white, college versus noncollege, low versus high income.

So we look within families; we contrast siblings. We think of them as being exposed to the same environment. We account for the main effect of race. And then we’re asking, for example, among black households, do you see this gradient in the sibling gender gap as a function of household structure or as a function of education? That’s the basic approach.

Just to step back a bit, most of the research questions I have worked on have been motivated by a long-standing interest in the opportunities faced by less-educated and less-affluent households and their kids. And that’s sort of my background. I was a psychology student as an undergrad and was originally going to do a Ph.D. in psych. But I was dissatisfied with the methods; also I was interested in computer science.

I ended up spending several years directing a computer learning center for the poor in San Francisco at a black Methodist church, and then I did similar volunteer work in South Africa. This was all before starting at Harvard’s [John F.] Kennedy School [of Government]. So a lot of the questions I focused on were questions that really gripped me when I was doing social service work.

At the computer learning center—it was actually sponsored by Silicon Valley—the idea was to diffuse computers to people who couldn’t afford them. I was very interested in how technological change was changing skills and opportunities. I saw the household structures. I was working in the middle of a crack epidemic, in a very poor neighborhood.

Region: When was this?

Autor: This was in 1989 through ’92. For every kid, you knew who the mom was and you knew who the dad was, and often it was a different dad for each kid. So I became very aware of how the labor market and family structure were interacting in that setting.

For me, these questions just sort of keep bubbling up. And these research agendas—the question about household structure and trade, for example—they merge because they are both driven by the question of how is the opportunity set changing for less-educated workers? And what does that do, not just to career, but to income, and not just income, but also to family, and also indirectly to the investments they make in their kids?

Region: Just as a side note, when I first read your CV I thought, this is an unusual background. A lot of physics or math majors become economists, but unless you’re Daniel Kahneman, psychology is an unusual college major.

Autor: Yes, it all happened quite by accident. I didn’t even know what economics was, literally. I never had taken a class in it because I just thought it was kind of about money, and I always felt like, “Well, I like money, but I don’t want to study it.”

After I’d done this work in San Francisco for several years, I knew that I wanted to go on for something else, but I didn’t know what. I knew I would look like a really good candidate for a public policy master’s program, so I applied to Kennedy School. Once there, I thought I’d really like to study this question about technology and inequality, or something.

So I applied for the Ph.D. program. For that, of course, I was required to take the upper-level economics classes, and it was only there that I was like, “Oh, wow. This is actually sort of the combination of the questions I want to answer and the tools I want to use to answer them.” So that’s how I ended up in economics. I had to do major remedial education. I mean, I was taking Calc 1B with Harvard undergraduates when I was 30 years old! [Laughs]

So anyway, I consider myself very, very lucky to have fallen into economics, quite backwards and by accident, and then I got my Ph.D. obviously from the Kennedy School, so I was surprised, very surprised, to find myself in an economics department and at MIT.

Trade and polarization

Region: Well, it goes without saying that economics is lucky to have you!

Let’s talk about one of the other things you’ve looked at relative to trade: political polarization. I found this work fascinating. You analyze congressional elections in 2002 and 2010 relative to varying levels of exposure to import substitution, and you find significant political shifts when districts are affected by international trade.

Autor: On average, import substitution benefits Republicans more than Democrats; but that’s a fairly weak effect because there’s a divergence going on. In initially Republican districts, you see turnover of incumbents, but they’re replaced by other conservatives who are more conservative. So a lot of the movement is within party, across ideological space, toward more ideologically strident candidates.

Region: They become darker red.

Autor: Yes, and then in Democratic areas, you see some movement left, but not that much on average. If you look, however, at the 16 percent of electoral districts where the minority is white non-Hispanic—in other words, less than 50 percent is white non-Hispanic—in those districts, you see strong movements to the left. You see, in other words, replacement of moderate Democrats with liberal Democrats. But that’s just in that 16 percent of districts.

Gordon, David, Kaveh [Majlesi] and I have actually been working on this project for several years. It was empirically very challenging to bring this one into focus for two reasons. One is that congressional districts don’t correspond to standard measures of geography.

You have to get vote counts by congressional districts, and then you have to link them to the economic outcomes of the counties in which they reside and the trade shocks of the CZs that encompass those counties. So, that was the first challenge.

The second one was, if you simply looked at party as the outcome or vote share, you would not find anything interesting because most of the movement is ideology within parties. The only way you can see that is by looking at the actual behavior of elected officials, so you look at their NOMINATE scores.² The operative assumption is that the candidate communicates what she or he is going to do and the voters perceive that and they vote that person into office. And once they’re in office, we see their voting record.

Region: I can see why this was challenging empirically.

Autor: We worked on it for a long time; we came very close to giving up. Dorn was the one who saved the day. You know, I think Gordon and I were ready to write it off.

Region: Your persistence paid.

Autor: That paper was presented this summer at the Princeton International Economics Workshop, which Gene Grossman organizes, and Elhanan Helpman, you know, descended from Mount Sinai to discuss the paper and so that was great (in the sense that Elhanan is something of the Moses of this field).

Region: Impressive!

Autor: Yes! So, we’re excited about it. But there’s a bunch of questions still to answer. What’s really fascinating—something we were not aware of—is that while the polarization of Congress has been going on for 30 years, basically since Ronald Reagan took office, the polarization of the U.S. electorate is much more recent. It’s really just been in the last 10 years.

By “polarization,” I mean the clustering of beliefs along party lines. So, for example, it’s increasingly the case that your view on global warming is highly predictive of your view on Mexicans, is highly predictive of your view on how high or low taxes should be and whether people should have a right to open-carry weapons.

Those beliefs didn’t used to be as correlated. So people have more strongly held, more divergent views. They have much more negative views of the other party as well. There is rising disapproval of cross-party marriage: The idea that you would be upset if your kid married someone of the other party has risen. Pew has documented this phenomenon. This has coincided with the rise of the Tea Party, and it’s most pronounced after 2006.

We’re seeing this same phenomenon in the data that we’re using to look at the trade exposure. We don’t want to say that all political polarization is due to trade exposure; I’m sure it’s not. But the localized adverse economic impact of the China shock in the 2000s does appear to be a kind of unnoticed contributor. In fact, you can see the antecedents of the current divisive presidential race playing out in the House in the 2000s—and specifically, in the locations where manufacturing was most hard hit.

Technology and labor polarization

Region: You’ve also worked a great deal on technology and labor, and I know you’ll be discussing Jim Bessen’s paper on job loss, inequality and computer automation tomorrow at the NBER Summer Institute. Just as with trade liberalization, economists tend to view technological progress as a nearly unmitigated good. True, there are winners and losers, but overall it provides net benefits to society.

You’ve looked closely at that relationship and suggested that technological automation—unlike import substitution—isn’t responsible for a lot of job loss, but it has caused significant polarization in labor markets.

I wonder if you would review that research a bit, and also comment on something I sensed from your 2014 paper, “Polanyi’s Paradox”: that your outlook on this subject seems more optimistic than that of a lot of pundits and the general public—the fear that we’re all going to lose our jobs to robots. Am I right about that?

Autor: Yes, that’s true. I am.

Region: And a third question: What is your sense of the likely trajectory of occupational change in light of current trends in automation?

Autor: Well, just to give you a bit more of my personal history on this: I came out of a technology background in the sense that I had been teaching computer skills and I had worked for several years as a programmer as well. So I was very, very interested in treating this question of how are computers affecting what people did.

When I was a grad student, the state of the art was to look at people who use a computer and ask about their wage history. Alan Kreuger had written a very well-known paper on that topic, and I was very impressed by the work. But I didn’t think it was using quite the right lens. First of all, you don’t have to be using a computer to be affected by a computer; and second, it’s not necessarily computer skills that are the relevant thing—it’s what they complement and what they substitute for.

My administrative assistant, for example, is not an ace programmer, but her work has been completely transformed by computers because so much of the work she used to do is now done electronically; her job is now organizing, coordinating and dealing with problems, as opposed to typing and filing and so on. And that work is more intellectually demanding than the stuff she did 15 years ago. But using the computer is the easy part!

Photo by Peter Tenzer

That interest led to the 2003 paper with Levy and Murnane, “The Skill Content of Recent Technological Change.” It laid out what we thought was a useful way of thinking about this. The quotation from Michael Polanyi is in that paper as well. His observation was, “We know more than we can tell.” Polanyi was also influenced by Herb Simon, the economist and computer scientist who also wrote about the question of what technological change could and couldn’t do.

It’s funny because we were somewhat in the vanguard of those arguing that computers are having a big impact on the labor market and explaining how and why they’re changing things. Many people now accept that computers have changed the labor market, but they’ve also adopted a very gloomy view of where that’s going.

I do not share that gloomy view, though I can understand why people would think that. It’s a very natural thing to think that if computers do more work, people do less work. But I think the answer is much more nuanced—and, of course, economists have recognized these nuances for centuries.

Computerization changes what type of work people do—that’s very clear; we see the occupational change going on. But the part that people miss is that displacement of a set of tasks or even entire job categories does not augur the end of work. In the last 200 years, technology has totally changed the work that we do. Most of the jobs we have didn’t really exist in any significant number 200 years ago. As a result, work is much better. It’s more interesting, it’s more productive, it’s safer and more rewarding.

My optimism on this topic comes in part from the fact that we’ve already gone through incredibly dramatic adjustments and have been largely made enormously better off for it. It’s not just because we’ve increased aggregate wealth, but more of us work in paid jobs now than 100 years ago. At the turn of the 20th century, most women worked in grueling unpaid employment in the household. Now the majority work in better jobs, for pay, in the formal labor market.

We’ve adjusted to the displacement of human labor by automation along at least three margins. One is that we’ve just created many new and interesting things to do. Think about software development or tourism or all kinds of travel and food and restaurants. We do all kinds of creative and interesting things we didn’t do before.

Two, more of us work, but we work fewer hours. People don’t work until the day they die. They work 40- and 50-hour weeks instead of 80-hour weeks. They work five days a week instead of seven. They take vacations. So they’ve spread the work in a way that’s constructive and leads to a better quality of life.

And the other thing, of course, is that as we get wealthier, our consumption demands rise, so we create a lot of work because we choose to consume rather than just taking it all in leisure. If a worker in 2015 wanted to have a 1915 level of income, he or she could work about 17 weeks a year. But most of us choose not to. We’d rather have a bigger house and a couple of cars and whatever else.

So I guess I’d say a couple of things:

One, technology does result in enormous opportunity. That we should think of our own productivity increases as a threat to us is sort of alien, if you think about it. Rising productivity—an expanding production possibility frontier—is a good problem to have.

And two, it’s not technologically determined how this gets sorted out. There are multiple possible outcomes that are within societal control. I like to give the analogy of comparing Saudi Arabia to Norway, two countries that have vast opportunity that comes from their oil resources. That oil is manna from heaven—like the productivity gains that come from automation.

These countries have managed that opportunity completely differently. In Norway, you have an engaged society where people are happy. They work. They play. It functions pretty well, people feel pretty “bought in.” Labor force participation is above 80 percent for both sexes. In Saudi Arabia, 90 percent of the private sector work force is foreign nationals. Most Saudi nationals who work, work for the government, many in low productivity jobs. There’s a lot of dissatisfaction.

Region: And the analogy with tech is we can manage that as well.

Autor: Or mismanage it, but it’s not deterministic. It’s within societal control, and the analogy is that in both cases, it’s like discovering a resource that allows you to accomplish your goals without your own labor. If you’re in Saudi Arabia, you can have money without any labor. You own the oil; just let other people pay to come extract it and pay them in oil, right?

Technological change is like that in a way too. My argument is not, “Don’t worry; it will take care of itself.” Our success in adapting to technological change in the past has come from deliberate decisions to invest in our human capital, for example. The high school movement is the best example, as documented in Claudia Goldin and Larry Katz’s wonderful book The Race Between Education and Technology. It’s just that expanding the set of tasks that can be done by machines rather than people is not bad news; it’s an opportunity that we have to rise to.

The “gig” economy

Region: On a very different tack, I’d like to ask for your thoughts on the so-called gig economy. The growth of Uber and on-demand services has gotten a lot of attention from the media. I’m not sure how many economists have looked at it.

Autor: Yes, in fact, there are economists who are working hard on this.

Region: Do you view it as a significant economic phenomenon? That is to say, do you see the gig economy as something that will employ growing numbers of workers for increasing hours? And what impact is that likely to have on inequality?

Autor: Not everything can be Uberized, clearly. There have been all these attempts to Uberize everything, and many of them fail. Like the company that will park your car for you and then bring it back to you when you want. There’s another that just makes Asian food for you and delivers it to your house. But many of them turn out not to be economical, though some of them will be, of course.

Substantively, I think the gig economy is a bit Janus-faced in the following sense.

Suppose you’re a “talent,” meaning someone who has a unique capability that is not completely interchangeable with someone else. If you’re a good computer programmer, for example, or a performer or an architect or if you’re a skilled medical expert, anything like that, having thick talent markets is really good for you. There was a time when the best programmers were trapped inside IBM for a 40- or 50-year career, and they were probably going to get paid 10 percent more than the lousy programmer who worked right next to them. There just wasn’t much differentiation. Now they can be a superstar. The gig economy works like a talent agency for these differentiated talents.

But at the bottom, at the other end of the labor market, it acts like a commodifier. For example, if you go to Amazon right now and buy a faucet, they will offer to have it installed in your house for $60. Well, previously, you dealt with plumbers whom you knew. You had a relationship with them, and they benefited from that in many ways, once they were established. They liked to be known to their customers.

But if they get Uberized in that way, which is what Amazon does, they just become a commodity, just a thing that gets this installation done. You don’t really think about who the person is; I don’t remember all the Uber drivers I’ve met. So I think the gig economy can actually work in countervailing directions for people at different places in the labor market hierarchy.

Region: It might polarize jobs, as technology does.

Autor: A little bit, yes. I mean, it’s by no means all negative. In the case of Uber, it’s replacing the traditional taxi business, and taxis were just a rigged game that yielded terrible service, high prices for riders and low wages for drivers—a lousy setup for everybody except taxi medallion owners, of course. Clearly, many Uber drivers are better off than they would have been driving taxis, and many passengers are better off than if they were riding taxis. It’s just hard to see where there’s a negative in that.

But I do think it will lead, or can lead, to more insecure, irregular employment. On the one hand, it’s good if you can do a little work when you need it. On the other hand, can you count on it? I think our labor market institutions are going to have to adapt to this change in the stability of employment arrangements. They’re really structured around the idea of full-time, full-year, stable employment. That’s the way unemployment insurance works. That’s the way our health care was originally designed when it was bound to employment during the Second World War.

So I think that the gig model will create challenges, but hopefully it will also create gains. You could argue that an even more important example than Uber is something like Airbnb. You have all these nonexistent markets for resources that are sitting idle most of the time—not people, but capital: empty vacation houses, parked cars, recreational equipment that gets used once a year. That’s a huge waste. Airbnb addresses part of it. There ought to be—and I believe there is—an Airbnb for car sharing, except that I wouldn’t call it “sharing.” You’re paying for something, so it’s not sharing, right? It’s the person-to-person rental economy.

Labor market intermediaries

Region: You’ve done a lot of work on labor market intermediaries.³ In the context of job loss and occupational change, I think, such work is especially germane and important. In a 2010 article with Susan Houseman, you look at one such entity, Work First, Detroit’s welfare-to-work program.

Autor: That’s right, where people were being placed in temporary help agencies.

Region: Yes, and you found that temp-help job placements don’t improve subsequent earnings and employment outcomes, whereas direct-hire job placements do. Your sense was that job stability was the key factor there, that temp jobs simply foster job churn.

Autor: Especially at the very low end. The population we studied is Work First clients, meaning that they were welfare recipients. This is not a population that suffers from excess job stability.

Region: In a more recent paper with Susan Houseman, you return to the Detroit data and find quite interesting nuances in the 2010 findings. Can you describe that research?

Autor: The context here is that Work First was a response to the belief that there had not been high returns on public training investments. The big idea of Work First, which originated in Riverside, California, is, “We’ve just got to keep them activated.” And there’s something to that. Many countries spend a lot of public money on so-called labor market activation: keeping people engaged.

The Work First program is basically, “Get your resume together. We’ll give you a day of job preparedness training and then send you out the door to find a job.” An even more extreme manifestation of that is, “We don’t care what job it is.” Temporary help agencies find that variant of Work First attractive because many agencies are basically looking for day laborers.

Ostensibly, they would make promises to Work First offices: “Oh, these are stable jobs; they’ll last at least 90 days.” But those are promises that temporary help firms are really not set up to fulfill. So in reality, the level of “stickiness” of those job placements is incredibly low. And remember that Work First clients are having trouble finding regular employment already; employers don’t have high demand for them. So temporary help placements just don’t tend to lead anywhere. In fact, Sue Houseman and I find that for more-skilled workers, temporary help placements just crowd out better direct-hire jobs that they would have gotten.

Specifically, we show that if you look at the people who do OK in temporary help, they would have done better in a direct-hire job. The losses from temporary help placements are actually greater for them because the opportunity cost is greater. Why is that? Our hunch is that direct-hire jobs involve the employer making some form of commitment. They’re actually paying a fixed cost to hire you; you’re not just on tap. They must expect that they’re going to keep you for a while or they wouldn’t go to the effort. If you can get that job, you can have a greater expectation of some ongoing employment unless you’re doing something wrong.

Disability compensation

Region: You’ve also studied the labor market effects of government disability insurance for a number of years, including SSI (Supplemental Security Income) and SSDI (Social Security Disability Insurance), and more recently the Veteran Disability Compensation program. Could you elaborate a bit on that recent research with Mark Duggan, Kyle Greenberg and David Lyle on the Veterans Administration disability program?

Autor: Sure. The VA system is very understudied relative to the others. It’s growing extremely rapidly, not because of growth in the number of veterans, but because of growth in a fraction of veterans who are using it and the severity of the claims that they’re making.

The funny thing is that if you look at the Veteran Disability Compensation program like an economist, it looks as if they got the design just right. But when you study it closely, you realize that it works exactly the opposite of what you would have anticipated.

Our regular federal SSDI program and SSI are categorical disability programs: You either are disabled or you’re not. If you are deemed disabled, you get benefits, but you’re not allowed to work to any substantial level. The VA program, by contrast, is not categorical. You can have a rating between 0 and 100 in increments of 10; your benefits rise, and there are no explicit work restrictions. In theory, you’re just being compensated for the loss of health resulting from your service. So you would think this has all the right incentive properties: Veterans receive a carefully calibrated income transfer, and they have no incentive to distort their labor supply to increase their benefits. This is opposite to SSI and SSDI, where adults in ill health but with some remaining work capacity face a categorical choice: work or disability benefits.

Well, what happens in reality? The graduated system creates what people who know it well call an “escalator.” As soon as you get onto the first step, you just start rising, and there is an incentive to want to get higher because the benefits are steep and nonlinear. If you rate 100 percent disabled, you can get $3,000 a month, which is not subject to federal income tax.

It’s really a lot of money. $36,000 a year is something like $50,000 to $55,000 of pretax income. In addition to that, there’s something called the Individual Unemployability benefit, where they can deem you unable to work even if you don’t have a 100 percent disability rating; the combination of disabilities you have makes you unable to work. And in that case, you also get a 100 percent benefit. That creates an incentive for people to document that they can’t work. To make it all worse, you can receive Individual Unemployability when you’re 90 years old, when you never would have been working. It doesn’t stop at retirement; it continues to death. This program has grown explosively.

We were interested in trying to understand how access to this program affected people’s participation. We looked at a change in policy in 2001 that retroactively made people who served in theater in Vietnam or Cambodia eligible for Veteran Disability Compensation if they had type 2 diabetes.

How would you get that during a war? Well, you might have been exposed to Agent Orange, which is dioxin, and there is only very limited evidence suggesting that that could increase your risk of precursors to diabetes.

This policy change caused a huge uptick in the number of Vietnam era veterans qualifying for Veteran Disability Compensation. Once they got on, they moved up the escalator very quickly. In the space of five years, something like 40 percent of them are judged 100 percent disabled, many of them with PTSD [post-traumatic stress disorder].

We looked at how that affected their employment and found pretty large effects. But the thing we can’t tell is how much of that is a pure “income” effect—that basically you have more money and so you stop working—and how much of it is an incentive effect, because of the escalator phenomenon. That question is left open. But we do think that Veteran Disability Compensation is understudied. It’s really politically a hot potato, even more than the civilian disability programs (SSDI and SSI), which also need design improvements.

I’m currently requesting permissions to conduct field experiments with the disability insurance program. I’m not trying to arrange big randomized controlled trials at present; simply modest proof-of-concept exercises that show that we can learn a lot while doing no harm. But it’s challenging; the SSA [Social Security Administration] is not used to working with outside researchers in this capacity, and they are trying to determine what scope they have available for novel arrangements.

Growth of the SSDI program has peaked now because the baby boom cohorts have started to retire. But the cost is extremely high, and the structure of incentives is such that it potentially makes people disabled. Instead of viewing disability and work as potentially complementary—that is, you have a work limitation, but can still work, possibly with some assistance from the SSA—work and disability are viewed as opposites. That’s how the laws were written; the SSA does not have a choice about this unless or until Congress modernizes the statutes.

This categorical view of disability also harms targeting efficiency. There are people who could benefit that don’t receive it. I have trained two Ph.D. students who faced significant disabilities, neither of whom could walk. But neither could get assistance from SSDI because that would prevent them from working. These are people who have significant health needs and very significant health costs, but cannot obtain assistance from SSDI unless they forfeit their careers. Simultaneously, a kind of an industry grew up just to get marginal cases allowed onto SSDI. There are law firms, some quite well known, that made their whole practice on disability cases. So, there’s a real need to modernize the design of the program to recognize that disability and work are not opposites, that work limitations are not all or nothing and that the goal of disability assistance should be to support self-sufficiency to the degree possible—though not more than that.

Raising the minimum wage

Region: One last question, about the minimum wage and inequality. The past couple of years have seen a widespread call in the United States for raising the minimum wage to about $15 an hour. The idea in part is that the existing minimum doesn’t provide a livable wage or income because the cost of living has increased faster than the minimum.

Economists have studied the relationship between the declining inflation-adjusted value of the federal minimum wage and trends in inequality, and many studies have concluded that the declining real minimum is indeed responsible for a substantial part of that rise.

In a 2016 paper with Alan Manning and Christopher Smith, you reassess the question with a very deep analysis and come up with a more nuanced understanding. Can you tell me about that?

Autor: Sure. But I first want to say, the most important reason to think about minimum wage is not about inequality per se; it’s about whether you can boost incomes and what this costs in terms of employment.

Our study is not about the employment consequences. It’s really just about the inequality consequences. I got involved in that question because, again, in this inequality debate that started in the ’90s, one hypothesis was that rising inequality was due to technological change, another hypothesis was globalization and another was unions and minimum wage. There is a very well-known paper by David Lee, who’s now the provost of Princeton. It said that almost all the rise in inequality in the 1980s was due to the falling real value of the minimum wage, and there was another paper by a Dutch economist named Coen Teulings, who reached a similar conclusion.

I had been teaching the Lee paper for a decade at MIT, but I just didn’t think the results seemed right. Even people like Larry Mishel, the head of the Economic Policy Institute, a left-leaning policy outfit, had said to me, “Yes, we know the minimum wage decline is important for inequality, but it can’t be as high as that paper says. How could it be that big?” Alan Manning was visiting MIT, and we had a conversation about it. He had the same feeling about it, and we said, “Well, we should work on this.” Christopher Smith was one of my students. We actually ended up taking a very long time because the problem was quite subtle. We didn’t come away saying that the declining real value of the minimum wage has no effect on inequality. We just said that its effects were smaller than leading papers had concluded.

That, in some sense, was a much more literature-driven paper than almost any other paper I’ve written. I’m satisfied we got the right answer but, again, it doesn’t change my view at all about the goodness or badness of the minimum wage. That has to be evaluated on how much it raises or lowers employment, how much it redistributes income effectively, relative to better-designed tools like the EITC [earned income tax credit].

There’s a lot of research now going on surrounding this topic. The Arnold Foundation is putting a lot of money into trying to do catalytic work on the minimum wage. There is a real research opportunity, given that many states are experimenting with the minimum wage. Part of the problem with studying the minimum wage in the United States over the last 30 years is that the levels have been so milquetoast that there’s just not that much to learn from them. Now, we’re actually seeing some aggressive moves that will hopefully at least provide more clarity, whether the news is good or bad.