The failure of automation and skill gaps to explain wage suppression or wage inequality

This was originally published as Appendix A in the EPI report Identifying the Policy Levers Generating Wage Suppression and Wage Inequality.

That U.S. workers have “skills deficits,” that is, lack the skills necessary to deal with technological change, including primarily automation, has been the predominant explanation offered by economists, pundits, policymakers, and the media to explain sluggish wage growth and inequality in the United States, at least until recently. This is the skill-biased technological change hypothesis, which points to the increased use of computer equipment in the workplace and the onset of the information age.

This narrative is sometimes presented as explaining the wage gaps between “skilled” and ‘‘unskilled” (meaning those without a college degree) earners and the disappointing wage growth for the vast majority. This report will show that the skills deficit/automation claim has always been a weak explanation for the trends since 1979, and, since the mid-1990s, all indications are that there is no basis at all for considering automation as a significant factor in wage suppression or the growth of wage inequality. For this reason, center-left economists have increasingly stopped highlighting these factors in discussions of the wage problems we face, though the narrative lingers among the punditry.

The conventional wisdom on automation and wage inequality

The general view of the last 30 years is that inequality and wage stagnation are the result of technological change in the workplace, meaning automation, and globalization driven by either technological advances or the political decisions of U.S. trading partners (China’s decision to join the world trading system, for example). These trends were seen as not only inevitable but desirable, in that the harm to workers is the byproduct of forces one would neither want to nor could change. The only appropriate remedy is to adapt, primarily by upgrading workers’ skills and education and perhaps by providing a more adequate safety net.

We examined in the body of the paper the impact of globalization for wage trends and drew two conclusions: Globalization played a nontrivial role in lowering the wages of non-college-educated workers, and this downward pressure has been strengthened by policy decisions creating selective and regressive shaping of the global rules. This report focuses on the automation/skills deficit dimension.

The automation narrative in the 1990s (the Clinton years)

In the 1990s center-left economists settled on skill deficits as the dominant explanation for growing wage inequality. As noted by President Clinton’s Council of Economic Advisers (1995):

The sluggish growth of incomes is due to dramatic changes in technology and in global competition that have affected industrialized economies around the world, reducing the relative demand for workers with less education and training…. [M]ost economists believe that a shift in the demand for labor in favor of more highly skilled, more highly educated workers has played a key role. Intensifying global competition is also cited as a factor in putting downward pressure on the wages of less educated workers. However, a number of studies have found that the easily measured direct effects of trade on the wage distribution were small….

An array of economists adopted the automation narrative in the 1990s. Conservatives drew on the work of Kevin Murphy, Gary Becker, and others to conclude that skills deficits explained wage gaps, pointing to higher returns to both education and other, unobservable (meaning not captured by specific variables in survey data) skills. The center-left drew on the work of Katz and Murphy (1992) and, later, an important book by Goldin and Katz (2008), The Race Between Education and Technology. While conservatives have produced little new empirical work on wage inequality in the last 20 years or so, the center-left has focused, until very recently, on developing a newer version of skill-biased technological change centered on the polarization of occupational employment patterns (Autor, Katz, and Kearney 2006; Autor 2010; Acemoglu and Autor 2011; Kearney, Hershbein, and Boddy 2015).

The automation narrative in the 2010s (the Obama years)

This narrative was the dominant one offered by the Obama administration, and President Obama (USA Today 2018) was still offering it in September 2018:

This change has happened fast, faster than any time in human history. And it created a new economy that has unleashed incredible prosperity. But it’s also upended people’s lives in profound ways. For those with unique skills or access to technology and capital, a global market has meant unprecedented wealth. For those not so lucky, for the factory worker, for the office worker, or even middle managers, those same forces may have wiped out your job, or at least put you in no position to ask for a raise. As wages slowed and inequality accelerated, those at the top of the economic pyramid have been able to influence government to skew things even more in their direction: cutting taxes on the wealthiest Americans, unwinding regulations and weakening worker protections, shrinking the safety net. So you have come of age during a time of growing inequality, of fracturing of economic opportunity.

In this story, rapid technological change has led to substantial growth overall and for those at the very top—the ones with “unique skills” and “access to capital”—while those without unique skills and access to capital experience diminished demand and are unable to push for higher pay.

Similarly, a leading conservative labor economist and chairman of the George W. Bush Council of Economic Advisers from 2006 to 2009, Ed Lazear, offered the automation story in the Wall Street Journal in 2019:

How are American workers doing? Neither the middle class nor the poor have fared well in recent decades—but don’t blame tax cuts, a too-low minimum wage or the greed of the 1%. In rich countries around the world, the top half of the income distribution has been pulling away from the bottom half. Productivity growth among high-wage workers, driven by technological change, is the reason…. The likely explanation is that changes in trade and technology have raised the productivity of highly trained, highly educated workers relative to the less skilled. Wages tend to move with productivity, so that if differences in worker productivity grow, wage differences will also grow. (Lazear 2019)

The skills deficit story sounds logical, but it’s not true. It fails to explain wage patterns over the last four decades, and it is a prima facie implausible explanation for at least the last 20 or more years (since 1995 or 1999). Contrary to President Obama’s contention, technological change (or at least automation) has not been especially rapid in the last dozen years or so (there’s actually been a substantial deceleration). Further, the contention that there has been a shift in the demand for labor in favor of more highly skilled, highly educated workers was not true in the late 1990s and has happened far more slowly in the 2000s than in earlier periods.

Automation: A flawed explanation

There have been two versions of the automation narrative, one based on education wage differentials, used to explain the trend in 1980s, and one based on “polarization” in occupational employment, used to explain the 1990s and beyond. Both versions were recently offered by economist David Autor as being among four explanations of wage inequality (Greenhouse 2020). Autor was asked, “[D]espite productivity gains of around 75% since the 1970s, the average American worker saw their wages increased by only 20% since then. There are numerous theories as to what has caused this…. How do you think this happened?” He responded:

There’s no single answer to that question. I would say there are four really important forces. One of them is educational attainment—the rate of growth of educational attainment in the United States actually slowed in the early 1980s. So the rate at which people were completing college, but the demand was growing for college-educated workers. And that led to a lot of rising inequality, just because the wages of the more educated rose relative to others….

A second force is the direction of technological change, which has increased the value of abstract reasoning of creativity, of expertise, of judgment, and devalued a lot of skilled work that people did that followed well-understood rules and procedures. So that would be many clerical jobs, phone answering jobs, calculating, accounting, bookkeeping, copying and filing, but also many production jobs, which often involved skilled, repetitive tasks. But increasingly once we understand the rule book for that type of work, it’s feasible to encode it in software and have it executed by machines or by computers.

This section will briefly review the failures of both versions of the automation narrative to explain the key wage patterns identified in the paper. Our discussion focuses on automation and not “technology” or “technological change,” since the latter terms imply a more general dynamic than one in which the implementation of new technologies in workplaces substitutes software and equipment for labor.

The education narrative

The automation story based on educational differentials sees wage inequality as being driven by increasing education wage gaps. The reasoning is that workplace automation has had a “skill bias” in recent decades, meaning that automation has largely just reduced the demand for a subset of workers—those largely without four-year college degrees. In most discussions skill is equated with what people obtain with four-year college degrees. If the supply of college-educated workers keeps up with the demand for college graduates driven by ongoing automation, then the wage premium for having a college degree will be flat and there will be no increase in wage inequality. If supply fails to keep up with automation-driven relative demand—leading to “skill deficits”—then the relative price of college graduates will rise and drive up wage inequality. Some have argued (as mentioned earlier) that the supply of college graduates faltered around 1980 and failed to keep up with growing demand from ongoing automation.

But for automation to cause a change in wage patterns, technology/automation has to outrace skills.1 It is not enough for automation to be occurring in workplaces or to continue in the same manner as before. Automation, after all, has been a force in workplaces for over 200 years, while education levels have also grown rapidly. Moreover, it is possible for automation to be a large and ongoing force in shaping the pattern of jobs/occupations (rising white-collar and declining blue-collar employment shares) without it generating wage inequality. For instance, automation was ongoing in the 1950s and 1960s when real median wages rose and wage inequality did not increase.

For a hypothesis to have resonance it should connect the major observations within its purview, and the theory that skill-biased technological change explains growing wage inequality fails across many dimensions.2

Top 1% wage growth. The leading research promoting education wage differentials as the driver of overall wage inequality (Katz and Murphy 1992; Goldin and Katz 2008) does not address the redistribution to the very top. Some might attempt to explain this as increased returns to the skills of executives and professionals in finance corresponding to the rise in the college wage premium. But there is no persuasive evidence to support a skill explanation for rising top 1% wages and income or, specifically, the superlative executive compensation growth and expansion in the financial sector that lays behind it (Bivens and Mishel 2013).

“The more you learn the more you earn.” The education story, at least as applied to the 1980s in Katz and Murphy (1992), offers to explain both the rising 90/50 and 50/10 wage gaps as reflecting the rising relative wage differentials between every level of education: college over some college, some college over high school, high school over less-than-high-school. Such an explanation, however, cannot explain why automation did not generate rising education wage differentials between dropouts, high school graduates, and associate college graduates after the 1980s. The flat or declining 50/10 wage gap in the 30 years after 1987 is inconsistent with the skills-gap narrative, since middle-wage workers who have more education than low-wage workers have not reaped a growing advantage since then. Acemoglu and Autor (2012) identified this as a major failure of the education narrative in their review of Goldin and Katz (2008). Mishel, Bernstein, and Schmitt (1997a) made that same point years earlier. (The changed behavior of the 50/10 wage gap in the 1990s—stable or flat rather than growing, as in the 1980s—has been cited by proponents of the occupational employment polarization story as a motivation for adopting this new framework; see Mishel, Shierholz, and Schmitt 2013 for the history.)

The sharp deceleration in automation-driven relative demand for college graduates in the mid-1990s. The college wage premium flattened in the early and mid-1990s. Several studies, all by proponents of the automation explanation, found that the impact of automation on the relative demand for college graduates substantially declined after the mid-1990s relative to earlier decades. Autor (2017) updated the Katz and Murphy (1992) model and showed that the automation-driven relative demand for college graduates decelerated by a third in the early to mid-1990s. Goldin and Katz (2007, Table 1) also showed a large deceleration in 1990–2005 relative to earlier decades going back to 1950, noting “a slowdown in demand growth beginning in the early 1990s” (p. 6).

Autor, Goldin, and Katz (2020), updating the Goldin and Katz (2008) metrics from The Race Between Education and Technology, confirmed the dramatic deceleration of automation’s impact:

[T]he model’s results…divulge a puzzling slowdown in the trend demand growth for college equivalents starting in the early 1990s. Rapid and disruptive technological change from computerization, robots, and artificial intelligence is not to be found though the impact of these technologies may not be well captured by this two-factor setup.

Their results (based on Autor, Goldin, and Katz 2020, Table A2) show a deceleration in growth of relative demand for college graduates in the 1999–2017 period relative to earlier periods: a 45.8% deceleration relative to the 1979–1999 period and a 41.8% deceleration relative to the longer 1959–1999 period. The period since 1999, therefore, has been one featuring a historically small impact of automation on (relative) demand for college graduates.

If automation’s impact has been far less in the last 25 years than in earlier decades, then it cannot explain the ongoing strong or even faster growth of wage inequality in the top half, illustrated by the growth of the 95/50 and 90/50 wage gaps.

It is ironic that just as the education narrative was becoming dominant in the mid-1990s the actual automation-driven relative demand for college graduates became markedly slower, negating the story that automation’s impact was accelerating and causing inequality.

Growing within-group wage inequality. The rise of education wage differentials is, at best, only a partial explanation of rising wage inequality because roughly 60% of the increase is due to greater inequality within education groups (Mishel et al. 2012, Table 4.20; Autor, Goldin, and Katz 2020). Autor, Goldin, and Katz (2020) acknowledge that growing within-group wage inequality is a challenge to the automation narrative.3

Stagnation of wages for college graduates. The story that automation-induced unmet demand for college graduates is lifting the wages of those with more education while punishing the wages of those with less is belied by the actual labor market experience of college graduates. For one, the inflation-adjusted wages of college graduates did not rise between 2000 and 2014, making the “lifting of the most educated” story not very convincing. The widespread use of unpaid internships for college students and graduates provides further evidence that employers do not have “unmet needs.” There is also ample evidence that the wages of entry-level college graduates slumped in the 2000s (Gould 2020) and that many young college graduates filled jobs that did not require a college degree (Abel and Dietz 2014). The median annual wage of recent college graduates, according to the New York Federal Reserve Board, rose by only 1% from 2000 to 2019, hardly a sign of winning in a race between education and technology/automation.

The slow growth of the college wage premium since 1995 or 2000. This fact makes the education wage-gap narrative a prima facie implausible explanation for the growing wage gap in the top half. While the college wage premium grew minimally since 1995 and especially since 2000, the 95/50 wage gap continued to grow strongly. The log 95/50 wage gap rose 0.76 points per year in 1979–2000 and rose even faster, by 1.00 log points per year, over 2000–2019 (see Figure A). In sharp contrast, the key education wage gap, the college–high school wage premium, grew far more slowly in the latter 2000–2019 period—hardly growing at all, just 0.13 log points per year, 87% slower than in the 1979–2000 period.

Figure A

The college wage premium cannot explain growing wage inequality since 2000: Average annual percentage-point changes in wage gaps, 1979–2000 and 2000–2019

Log 95/50 ratio College wage premium
1979–2000 0.76 1.00
2000–2019 1.00 0.13
ChartData Download data

The data below can be saved or copied directly into Excel.

Notes: The college wage premium is the percent by which hourly wages of four-year college graduates exceed those of otherwise-equivalent high school graduates. This regression-based gap is based on average wages and controls for gender, race and ethnicity, education, age, and geographic division; the log of the hourly wage is the dependent variable. The 95/50 wage ratio is a representation of the level of inequality within the hourly wage distribution. It is logged for comparability with the college wage premium.

Source: Author’s analysis of EPI Current Population Survey Extracts, Version 1.0 (2020), https://microdata.epi.org.

Copy the code below to embed this chart on your website.

As noted above, the growth of the wage gap in the top half, illustrated by the 95/50 or 90/50 gap, is a key wage pattern that needs to be explained, as it was the only source of growing wage inequality other than that of the top 1% in the last three decades. These data show that any explanation of wage inequality based on education wage gaps is implausible for the period since 2000 (and probably since the mid-1990s). That is, it is implausible that a sharply decelerating growth of the college wage premium can help explain an accelerated growth of the 95/50 wage gap since 2000.

The slowdown of college completion. The education wage-gap narrative sometimes focuses on the slower growth of college completion post-1979 as the cause of wage inequality. This makes sense within the conventional framework that ignores the role of other factors—globalization, weaker unions, lower minimum wages, and so on—besides supply and automation-driven relative demand. Nevertheless, the slowdown of college completion cannot explain wage inequality since the early to mid-1990s, the era of slight growth in the college wage premium. After all, slow college completion is said to increase wage inequality as the relative supply of graduates falls short of relative demand. The growth of the relative supply of college graduates, according to Autor, Goldin, and Katz (2020, Table 3), was very slow in the most recent two decades (1999–2017), yet the college wage premium barely grew.

The framework put forth by Katz and Murphy (1992) and by Goldin and Katz in The Race Between Education and Technology (2008) relies on competitive labor markets driven solely by relative supply of education and automation-driven relative demand for education. The notion that the relative demand for education (as a proxy for automation) can be deduced from education wage premiums and supply trends presumes that factors other than automation (unions, globalization, minimum wages, corporate structure changes, and others) have no impact. The evidence presented in this paper indicates otherwise. Simply put, the college–high school wage premium goes up and down for lots of reasons besides automation and supply factors, and we cannot readily deduce automation’s impact from data on wages and education supply.

There have been many critiques of the skills deficit/education wage-gap narrative in the past, and the analyses have stood the test of time; these include Mishel and Bernstein 1994; Mishel, Bernstein, and Schmitt 1997a, 1997b; Mishel and Bernstein 1998; Howell and Wieler 1998; Galbraith 1998; Howell 2001; Card and DiNardo 2002; Mishel, Shierholz, and Schmitt 2013; and Howell and Kalleberg 2019.

The job polarization automation narrative

A new automation-causes-wage-inequality narrative emerged around 2005 to replace or supplement the education wage-gap narrative and to overcome one of the latter’s key weaknesses—the inability to explain why the 50/10 wage gap grew in the 1980s but flattened or declined in the 1990s. The job polarization story claims that the nature of automation changed starting in the 1990s such that automation replaced middle-wage occupations/jobs more than jobs in low-wage and high-wage occupations. This is demonstrated by mapping the changes in occupational employment patterns by wage level. That is, this story relies heavily on the expansion of low-wage occupations characterized by routine manual tasks, the expansion of high-wage occupations requiring abstract, nonroutine tasks, and a corresponding shrinkage of occupations in the “middle,” which perform routine manual tasks.

Autor, Katz, and Kearney (2006) articulated this narrative:

[T]hese models also imply a puzzling deceleration in relative demand growth for college workers in the early 1990s, also visible in a recent “polarization” of skill demands in which employment has expanded in high-wage and low-wage work at the expense of middle-wage jobs. These patterns are potentially reconciled by a modified version of the skill-biased technical change hypothesis that emphasizes the role of information technology in complementing abstract (high-education) tasks and substituting for routine (middle-education) tasks.

This occupational employment polarization narrative was introduced into the policy world in a Center for American Progress/Hamilton Project paper by David Autor (2010) and articulated as the necessary replacement for the education narrative in Acemoglu and Autor (2012).

But the job or occupational polarization story fails to explain key trends and wage patterns:

Growth in top 1% wages. The job polarization narrative, as with the education wage-gap narrative, fails to address the superlative wage growth of the top 1% and top 0.1%, a major dynamic driving wage inequality.

Lack of evidence of job polarization since 1999 or in the 2000s. This was demonstrated in Mishel, Shierholz, and Schmitt (2013) and Beaudry, Green, and Sand (2013, 2014) and further confirmed in Hunt and Nunn (2019). The key chart in Autor (2010, Figure 1) showed there was no job polarization in the 1999–2007 period, but that finding was not mentioned in the text. Autor’s paper (2015, 149–50) for the Kansas City Federal Reserve Board’s annual Jackson Hole conference acknowledged the lack of polarization in the post-1999 period through 2012, the latest data available at that time:

Although the polarization hypothesis can explain some key features of the U.S. and cross-national data, reality invariably proves more complicated than the theory anticipates. The clearest evidence for this general dictum is the unexplained deceleration of employment growth in abstract task-intensive occupations after 2000, which is discussed by Beaudry, Green, and Sand (2013, 2014) and Mishel, Shierholz, and Schmitt (2013)…. The final empirical regularity highlighted by Chart 7 is that growth of high-skill, high-wage occupations (those associated with abstract work) decelerated markedly in the 2000s, with no relative growth in the top two deciles of the occupational skill distribution during 1999 through 2007, and only a modest recovery between 2007 and 2012. Stated plainly, the U-shaped growth of occupational employ­ment came increasingly to resemble a downward ramp in the 2000s.

A “downward ramp” and the absence of the “U-shaped growth of occupational employ­ment” are acknowledgments, though expressed in a not-very-straightforward manner, that job polarization was not present between 1999 and 2012.

Obviously, a narrative based on the changing composition of employment by occupation (expanding low-wage and high-wage occupations and shrinking middle-wage occupations) cannot be relevant to explaining post-1999 wage patterns if those occupational employment patterns have not been evident. In fact, the Autor (2014) data show that nearly all of the employment growth was in low-wage occupations, a pattern that does not readily explain why the wage gap between the top and the middle kept expanding and did so at an accelerated pace. For occupational employment patterns to explain top-half wage-gap growth would require finding a rapidly expanding need for high-wage workers able to carry out abstract, nonroutine tasks, a pattern not present since 1999.

Lack of a relationship between occupational employment patterns and wage inequality. Remarkably, the job polarization narrative relies on mapping occupational employment patterns to explain wage inequality but has never presented evidence that these occupational employment shifts affect wages. In fact, Mishel, Shierholz, and Schmitt (2013) show that changes in occupational employment shares (whether an occupation expands or contracts employment relative to other occupations) are not related to changes in relative wages by occupation (whether wages in that occupation rose or fell relative to wages in other occupations). That is, one would expect that occupations that expand (contract) would have rising (falling) wages relative to other occupations. Looking at the relation between changes in occupational employment shares and the corresponding relative wages of occupations, Mishel, Shierholz, and Schmitt (2013) found no relationship in each of the decades of the 1980s, 1990s, and 2000s. It is also worth noting that middle-wage occupations have shrunk and higher-wage occupations have expanded since the 1950s, but median wages and wage inequality have risen and fallen over this time with no apparent correspondence to job polarization trends.

If occupational employment patterns do not affect occupational relative wages, then they certainly bear no relationship to changes in wage inequality, since presumably the mechanism for automation to cause changes in wage gaps is for automation-induced changes in occupational employment patterns to alter the relative wages of occupations. The effort to track occupational employment patterns has no implications for understanding wage patterns.

Footprints of automation

The discussion so far has relied on economic analyses that derive the pace and skill bias of automation from patterns of occupational employment growth or from wage and education supply trends. It is worthwhile to examine other perhaps more direct footprints of automation to discern the pace of automation in recent years compared to earlier periods. These data are illustrated in Figure B, drawn from Mishel and Bivens (2017) and Mishel and Shierholz (2017).

Figure B

Average annual growth rates of labor productivity, capital, and IT hardware and software, 1973–2016

Growth in labor productivity and the capital stock has decreased in recent periods

Null column 1947-1973, The Post-War Boom Null column 1973-1995, The Great Productivity Slowdown  Null column 1995-2002, IT-led resurgence  Null column 2002-2007, pre-recession deceleration Null column 2007-2016* (2016* = 2015Q4/2016Q3), Great Recession and its aftermath 
Labor productivity 3.3% 0 1.50% 3.30%  2.20% 1.30%
Capital investment 3.90% 3.80% 5.10% 2.90% 1.70%

Copy the code below to embed this chart on your website.

Capital investment in information technology has also slowed

1947–1973  1973–1995  1995–2002  2002–2007  2007–2016
Hardware 10.57%  12.91%  15.64%  7.97% 4.82% 
Software 0.00% 16.43% 16.31% 6.46%  4.18% 

Copy the code below to embed this chart on your website.

Note: Using latest available data, 2016 measure includes data from 2015Q4–2016Q3.

Source: EPI analysis of data (xls) compiled by John Fernald of the Federal Reserve Bank of San Francisco.

Copy the code below to embed this chart on your website.

Automation is what occurs as new technologies are incorporated along with new capital equipment or software to replace human labor in the workplace. Labor productivity and capital investment are both measures of automation in that they necessarily accompany the substitution of capital for labor. Thus, if there were a recent surge of robots or automation, we would expect to see the footprints in trends in productivity, capital investment, and software investment. The trends shown in the figure suggest that automation has been far slower since 2002 than in the three earlier periods: the early postwar years from 1973 to 1973; from 1973 to 1995; and over the high-tech boom years of 1995–2002. There is certainly no evidence of automation having accelerated. These data affirm the findings above that automation, given its slow pace in recent years, is unlikely to have been a major factor driving wage stagnation or wage inequality in the last two decades.

Automation (along with shifts in consumer demand and trade) would be expected to be a major factor in why employment in some occupations expands and employment in other occupations declines. Therefore, we can also examine the pace of change in the occupational composition of employment to deduce trends in automation.

Using the data in Atkinson and Wu (2017), Mishel and Bivens (2017) computed a metric to examine the pace of occupational employment shifts in each decade. Specifically, Mishel and Bivens examined the shares of total employment for each of the 250 occupations in the data for the beginning and end years of each decade and computed the changes in these shares. The sum of the occupational share gains will automatically be equal to the occupational share losses, so the metric of change in each decade is half the sum of the absolute change in employment shares. This metric adjusts for differences in the rate of employment growth in each decade and the absolute employment size of individual occupations. This metric of shifts in occupational employment measures the shares of total employment exchanged between occupations that gain and occupations that lose employment shares each decade.

Mishel and Bivens found that occupational employment changes were “fairly uniform over the 1940–1980 period and far more rapid than for any period since 1980. The period since 2000 has seen the lowest rate of change—half the rate of change of the 1940–1980 period.” These findings indicate that the pace of occupational employment shifts was extremely slow in the 2000s, affirming the deceleration found in the slower growth in productivity, software, and capital investments identified in Figure B.

Notes

1. Mishel and Bernstein (1994) referred to this as the need to show “acceleration of technology/automation.”

2. The term “college graduate” in this discussion is explicitly restricted to those with a four-year bachelor’s degree and excludes those with “some college” or an associate degree and those with an advanced degree beyond a bachelor’s. Some discussions lump the college and advanced degree returns or supply together. That can be misleading when the evidence is used to suggest we increase the number of college graduates if, in fact, the evidence may suggest we need more post-college graduates.

3. Some researchers have asserted, without any empirical backup, that within-group inequality reflects the returns to unobservable (not captured by any metric in our regular data) skills. That is, of course, the only way to preserve the skill-biased technological change story without having to look beyond pure supply-and-demand factors. It is not, however, persuasive without further evidence or even some conjecture about what patterns we would expect returns to unobserved skills to display.

References

Abel, Jaison R., and Richard Deitz. 2014. “Are the Job Prospects of Recent College Graduates Improving?” Liberty Street Economics (Federal Reserve Bank of New York blog), September 4.

Acemoglu, Daron, and David Autor. 2011. “Skills, Tasks and Technologies: Implications for Employment and Earnings.” In Handbook of Labor Economics, edited by Orley Ashenfelter and David Card, 4, part B: 1043–171. Elsevier.

Acemoglu, Daron, and David Autor. 2012. “What Does Human Capital Do? A Review of Goldin and Katz’s The Race between Education and Technology.” Journal of Economic Literature 50, no. 2: 426–63.

Atkinson, Robert D., and John Wu. 2017. “False Alarmism: Technological Disruption and the U.S. Labor Market, 1850–2015.” Information Technology & Innovation Foundation.

Autor, David. 2010. “The Polarization of Job Opportunities in the U.S. Labor Market: Implications for Employment and Earnings.” Center for American Progress and Brookings Institution: The Hamilton Project.

Autor, David. 2014. “Polanyi’s Paradox and the Shape of Employment Growth.” Working Paper 17040. National Bureau of Economic Research. https://doi.org/10.3386/w20485.

Autor, David. 2015. “Polanyi’s Paradox and the Shape of Employment Growth.” Reevaluating Labor Market Dynamics (Federal Reserve Bank of Kansas City, Economic Policy Proceedings), 129–77.

Autor, David. 2016. “Will Automation Take Away All Our Jobs?” TEDxCambridge.

Autor, David. 2017. “How Long Has This Been Going On? A Discussion of ‘Recent Flattening in the Higher Education Wage Premium: Polarization, Skill Downgrading, or Both?’ by Robert G. Valletta.” Massachusetts Institute of Technology.

Autor, David, Claudia Goldin, and Lawrence F. Katz. 2020. “Extending the Race Between Education and Technology.” AEA Papers and Proceedings 110: 347–51.

Autor, David H., Lawrence F. Katz, and Melissa S. Kearney. 2006. “The Polarization of the U.S. Labor Market.American Economic Review 96, no. 2: 189–94.

Beaudry, Paul, David A. Green, and Benjamin M. Sand. 2013. “The Great Reversal in the Demand for Skill and Cognitive Tasks.” Working Paper 18901. National Bureau of Economic Research.

Beaudry, Paul, David A. Green and Benjamin M. Sand. 2014. “The Declining Fortunes of the Young Since 2000.” Papers and Proceedings of the 126th Annual Meeting of the American Economic Association 104, no. 5: 381–6.

Bivens, Josh, and Lawrence Mishel. 2013. “The Pay of Corporate Executives and Financial Professionals as Evidence of Rents in Top 1 Percent Incomes.” Journal of Economic Perspectives 27, no. 3: 57–78.

Card, David, and John DiNardo. 2002. “Skill-Biased Technological Change and Rising Wage Inequality: Some Problems and Puzzles.Journal of Labor Economics 20, no. 4: 733–83.

Council of Economic Advisers. 1995. Economic Report of the President (1995). U.S. Government Printing Office.

Galbraith, James K. 1998. Created Unequal: The Crisis in American Pay. Free Press.

Goldin, Claudia, and Lawrence F. Katz. 2007. “The Race Between Education and Technology: The Evolution of U.S. Educational Wage Differentials, 1890 to 2005.” Working Paper 12984. National Bureau of Economic Research.

Goldin, Claudia, and Lawrence F. Katz. 2008. The Race Between Education and Technology. Harvard Univ. Press.

Gould, Elise. 2020. State of Working America Wages 2019: A Story of Slow, Uneven, and Unequal Wage Growth over the Last 40 Years. Economic Policy Institute.

Greenhouse, Steven. 2020. “02: The Future of Work.” The World as You’ll Know It (podcast transcript), 32:38.

Howell, David R. 2001. “The Skills Myth.” American Prospect, December 19, 2001.

Howell, David R., and Arne L. Kalleberg. 2019. “Declining Job Quality in the United States: Explanations and Evidence.” Russell Sage Foundation Journal of the Social Sciences 5, no. 4: 1–53.

Howell, David R., and Susan S. Wieler. 1998. “Skill-Biased Demand Shifts and Wage Collapse in the United States: A Critical Perspective.” Eastern Economic Journal 24, no. 3: 343–66.

Hunt, Jennifer, and Ryan Nunn. 2019. “Is Employment Polarization Informative About Wage Inequality and Is Employment Really Polarizing?” Working Paper 26064. National Bureau of Economic Research.

Katz, Lawrence F., and Kevin M. Murphy. 1992. “Changes in Relative Wages, 1963–87: Supply and Demand Factors.” Quarterly Journal of Economics 107 (February): 35–78.

Kearney, Melissa S., Brad Hershbein, and David Boddy. 2015. “The Future of Work in the Age of the Machine.” Brookings Institution: Hamilton Project.

Lazear, Edward P. 2019. “Mind the Productivity Gap to Reduce Inequality: It Isn’t Only an American Problem, but the U.S. Has Lessons to Learn from Other Wealthy Countries.” Wall Street Journal, May 6.

Mishel, Lawrence, and Jared Bernstein. 1994. “Is the Technology Black Box Empty? An Empirical Examination of the Impact of Technology on Wage Inequality and the Employment Structure.” Technical Paper. Economic Policy Institute.

Mishel, Lawrence, and Jared Bernstein. 1998. “Technology and the Wage Structure: Has Technology’s Impact Accelerated Since the 1970s?” Research in Labor Economics 17: 305–56.

Mishel, Lawrence, Jared Bernstein, and John Schmitt. 1997a. The State of Working America 1996–1997.  An Economic Policy Institute book. M.E. Sharpe.

Mishel, Lawrence, Jared Bernstein, and John Schmitt. 1997b. “Did Technology Have Any Effect on the Growth of Wage Inequality in the 1980s and 1990s?” Economic Policy Institute.

Mishel, Lawrence, and Josh Bivens. 2017. “The Zombie Robot Argument Lurches On: There Is No Evidence That Automation Leads to Joblessness or Inequality.” Economic Policy Institute.

Mishel, Lawrence, Josh Bivens, Elise Gould, and Heidi Shierholz. 2012. The State of Working America, 12th Edition. An Economic Policy Institute book. Cornell Univ. Press.

Mishel, Lawrence, and Heidi Shierholz. 2017. “Robots, or Automation, Are Not the Problem: Too Little Worker Power Is.” Economic Policy Institute.

Mishel, Lawrence, Heidi Shierholz, and John Schmitt. 2013. “Don’t Blame the Robots: Assessing the Job Polarization Explanation of Growing Wage Inequality.” Economic Policy Institute.

USA Today. 2018. Read Transcript of Former President Obama’s Speech, Blasting President Trump.” September 7.

See more work by Lawrence Mishel and Josh Bivens