Supply, Demand, and the Skill Premium

Supply, Demand, and the Skill Premium

There is no question that the economic returns to skill have increased dramatically since the end of the Great Compression.  The college wage premium, for example, has grown from 56 percent for men and 45 percent for women in 1980 to 133 percent for men and 114 percent for women by 2010.[1]  Meanwhile, the wage premia enjoyed by super-elite college graduates have exploded alongside exploding super-elite wages—with the economic returns to a post-BA degree relative to a high school degree rising from 70 percent for men and 90 percent for women in 1980 to 237 percent for men and 204 percent for women in 2010.[2]  Indeed, at some point, the story of super-elite skill premia and rising top wages converges on the recent history of the labor incomes of the 1 percent.

Like every good in a market economy, skilled labor commands a price—or wage—set in the first instance by the balance of its supply and demand. Skilled workers command increasing wages and wage premia when the supply of their skills does not increase by enough to keep up with rising demand.  The labor market’s skill fetish therefore has—in theory—two potential sources:  an insufficient supply of skilled workers and an excess demand for skills.  The Meritocracy Trap emphasizes one of these sources, arguing, in Chapter Eight, that the enormous labor incomes that superordinate workers enjoy have their roots in accelerating technological innovations that dramatically increase the demand for the exceptional skills that they distinctively possess.[3] But an endnote in the book recognizes the alternative explanation of the rising skill premium—namely that the demand for elite skills has grown more or less steadily and that the wage premium increased principally because the growth in the supply of college skills began to slow.

This supply-side view is most powerfully developed by Claudia Goldin and Lawrence Katz in their book The Race Between Education and Technology.  Goldin and Katz acknowledge that skill-biased technical change (and its attendant effects on the demand for skill) is important to the wage structure, writing that “educational wage differentials would have declined substantially from 1950 to 2005 in the absence of rapid skill-biased technological change.”[4]  But they claim that accelerating demand-side effects are not needed to explain the contrast between the relatively compressed wage structure of the Great Compression and the distinctive acceleration of the skill premium, and the explosion of elite labor income, beginning in the late 1970s.  Instead, they propose that “[a] unified explanation for the two halves of the twentieth-century inequality experience (at least with respect to educational wage differentials) does not require much (if any) acceleration in the skill-bias to technological change.  Rather, the difference between the two halves came about primarily because of a change in the supply of educated and skilled labor.”[5]  Goldin and Katz thus insist that “educational supply changes have been the tail wagging the wage-premium dog.   Supply variations were far more important in changing relative wages than were differential demand changes across periods,”[6] and indeed that the “higher-frequency movements in the college premium” are consistent with “stable demand growth.”[7]  (“[B]y and large,” Goldin and Katz add, “the growth rate of the demand for more educated relative to less educated labor was fairly constant over the 1915 to 2005 period.”[8]) This leads Goldin and Katz to conclude that “[t]he most important conclusion to be drawn from [our] analysis is that the driving force behind the explosion of the college wage premium of the last 25 years was a sharp decrease in the growth rate of the relative supply of educated workers and not an increase in the growth rate of the relative demand for skill.”[9]

This supply-side view requires less speculative inference than the demand-side view that The Meritocracy Trap adopts.  Trends in the demand for skill are not straightforwardly observable but must instead be derived from close analyses of task intensity or indirectly inferred from the interaction between trends in the supply of skill and trends in the skill premium.  The supply of skill, by contrast, may be observed by looking to enrollments at colleges and universities—the institutions that produce skill—and counting up the number of graduates, or skilled workers.  This is, as arguments below make plain, not quite so simple a matter as Goldin and Katz suppose, but their approach nevertheless retains an intuitive appeal.  Moreover, The Meritocracy Trap makes substantial use of “raw data” that Goldin and Katz deploy, only to “classify and interpret” the data differently.[10]  Both considerations make it important to say directly why The Meritocracy Trap departs from Goldin’s and Katz’s conclusions.

Certainly, Goldin and Katz are not wrong to observe that the share of Americans to get college degrees grew rapidly during the Great Compression and much more slowly in the years since.  Between 1940 and 1980—that is, over the course of the Great Compression—America’s university sector served a rapidly growing share of the population:  the proportion of Americans between 25 and 29 who had obtained a BA or higher nearly quadrupled, from 5.9 percent in 1940 to 22.5 percent in 1980.[11]  (The share of 30-year-olds to have attended at least some college also grew rapidly, more than doubling from around 20 percent in 1955 to roughly 50 percent in 1980.[12])  By contrast, the rate of growth in the share of 25- to 29-year-olds with college degrees or greater has slowed since 1980, rising only incrementally from just under a quarter in 1980 to slightly over a third in 2015.[13]  (The growth rate in the share of 30-year-olds to have attended at least some college also slowed, from roughly a half in 1980 to below two-thirds today.[14])

The supply of the elite graduates whose incomes dominate rising economic inequality, moreover, has stagnated more completely still.  Such recent growth among college and university enrollments as exists has been concentrated in the least-elite parts of the sector:  two-year colleges, for example, enrolled only about a quarter of public-sector college students in the 1960s but nearly half by 2005;[15] and the fraction of college students in public rather than private schools was roughly three times as great in 2000 as it had been in 1900,[16] with the relative sizes of public versus private colleges also roughly tripling over the same period.[17]  At the same time, the most elite colleges and universities have hardly grown at all, nor have new schools joined the ranks of the super-elite.  Among the 35 private institutions to make the top 50 in the US News & World Report 2006 college rankings, “only 4 began college level instruction in the twentieth century and just one was founded after 1900.”[18]

It is therefore only a small exaggeration to say that the share (relative to the size of the workforce) of college-educated workers has been effectively fixed since the end of the Great Compression and that the share of elite college graduates has in fact declined.  This of course supports Goldin’s and Katz’s instinct to find the cause of the rising skill premium on the supply side of the labor market.

Nevertheless, four considerations cut against the supply-side account that Goldin and Katz prefer and in favor of The Meritocracy Trap’s demand-side view.  All four arise internally to the logic of Goldin’s and Katz’s argument.  They amount to reasons (of increasing importance) why the data that Goldin and Katz deploy do not support their conclusions.

First, and most narrowly, the purely supply-side account that Goldin and Katz prefer cannot explain the path of the skill premium during the 1970s and 1980s.  Goldin and Katz admit this[19] and propose an ad hoc explanation that emphasizes the 1973 oil price and inflation shocks and the relatively greater wage-stickiness enjoyed by unionized non-college-educated workers in their wakes.[20] This explanation is opportunistic, however, and also fits badly with the fact that rising elite wages rather than falling middle class wages drove the college wage premium steeply up in the 1980s.  The demand-side account presented in The Meritocracy Trap handles this period more elegantly and more accurately.

Second, and more broadly, the purely supply-side explanation that Goldin and Katz prefer may be an artifact of inaccuracies in the ways in which they measure the demand for college skills.  Goldin and Katz derive demand from the shares of college-educated and non-college-educated workers in the workforce at each moment, combined with the college wage premium and the elasticity of substitution between college and non-college labor.[21] This approach assumes that the labor market is in equilibrium at every moment in time.  A disequilibrium in the labor market might produce a downward bias in Goldin’s and Katz’s estimates of the demand for skilled labor.

Third, although Goldin and Katz observe the supply of college graduates directly, they do not observe skill directly but rather infer skill in a fashion that imposes a downward bias on their estimates of the supply of skill, especially in the years following the Great Compression, when the college wage premium has increased most dramatically.

Specifically, Goldin and Katz treat a college degree (more precisely, what they call a “college equivalent,” namely either one BA holder or two workers with some college training but no degree[22]) as containing the same quantity of labor skill in every year of their study, and in particular in every year between 1975 and the present.  This approach neglects that although growth in the shares of the population to attend and to graduate college (as measured by head count) slowed sharply after 1975, growth in investment in college training did not slow.  The share of GDP devoted to post-secondary education grew from 2.2 percent in 1970 to 3.1 percent in 2014.[23]  And real expenditures per college student grew by nearly 60 percent during precisely the period in which the relative supply of college graduates stagnated.[24] Per-student expenditures at elite colleges grew more rapidly still:  in the Ivy league, for example, by 80 percent since just the turn of the millennium.[25]

The rapid growth in real investment per college graduate means that, as the years progress, a single hour of college-graduate labor contains, one might say, much more skill. In addition, the dramatic increase in elite work hours means that the effective supply of labor per college graduate has also increased.  These effects cause Goldin’s and Katz’s measure of the supply of college skills to underestimate the true supply of skill, by increasing amounts over time.[26]  (If spending on college education had taken its actual course while spending per student had remained constant at 1970 levels, then there would be many more skilled workers today than there are in fact—60 percent more college graduates, and over 80 percent more Ivy League graduates.)  And if the supply of skill has grown more rapidly than Goldin and Katz presume, then the wage premia that they observe must reflect an increased—and, given the data, an accelerating—rate of growth in the demand and a rising—indeed, accelerating—skill bias in technological change.  (The fact that Goldin’s and Katz’s estimates of the demand for skill themselves depend, as they must, on the authors’ (too-small) estimates of the supply of skill means that claims about a steady rise in demand cannot provide any independent warrant for Goldin’s and Katz’s conclusions.)

Finally, and most importantly, Goldin and Katz place the dividing line between skilled and unskilled workers at the same place—with all college degrees counted as skilled, and any time spent in college counted as half-skilled—for every year of their series.  In fact, however, the most consequential dividing line between unskilled and skilled workers, for purposes of setting wages, has in recent decades moved dramatically up the skill distribution.   This causes Goldin’s and Katz’s measures of both the relative supply of skill and the skill premium increasingly to dilute truly elite skilled workers with nominally college-educated but functionally un- or at least only mid-skilled workers.  For the most recent years, the dilution is enormous:  on Goldin’s and Katz’s accounting, roughly half of the US workforce (the roughly one-third with college degrees plus half of the remaining roughly one-third with some college training but no degree) is skilled.  But in fact, labor incomes have been rising for only the very most elaborately skilled workers—the top 5 or even top 1 percent.

Goldin and Katz make no separate estimate of the trends in supply of or demand for these truly skilled workers, and their broader estimates provide no insight into whether rising wage premia at the very top of the skill distribution are best explained by slowed growth in supply or increased growth in demand.  As Goldin and Katz admit, demand-side explanations of the rising super-skill premium are compelling: “Computerization prior to the 1990s largely substituted for non-college clerical and production tasks.  More recent advances in information technology have increasingly led to organizational changes that eliminate many lower- and middle-paid college jobs but greatly complement top-end managers and those with strong problem-solving skills.  Demand for those who graduated from more selective institutions as well as those with post-BA degrees is still soaring and they are doing spectacularly well.  But demand for many other college workers is less strong and their earnings have not risen as much relative to non-college workers since 1990.”[27] By lumping all college graduates together in the same way throughout every year of the past 50, Goldin’s and Katz’s demand estimate disguises accelerating demand growth for super-skills by dilution through decelerating demand growth for ordinary college skills.  They thus neglect demand-side explanations for precisely the super-skilled wage premia whose explosion drives the more modest (because diluted) increases in the college premium overall.

These considerations cut against Goldin’s and Katz’s claim that a deceleration in the growth of the supply of skills, rather than an acceleration in the growth of the demand for skills, best explains the rising skill premium and therefore also the rising incomes of superordinate workers.  In this way, they reinforce the account of snowball inequality laid out in Chapter Eight of The Meritocracy Trap.

Finally, these observations and the account developed by Goldin and Katz unite to support the idea that meritocratic inequality feeds on itself.  Growth in the supply of college graduates has slowed for a very particular reason and in a very particular way:  elites remade education, including college education, to grow expenditures more rapidly than enrollments, concentrating relatively more intensive training in relatively fewer students.  Even purely supply-side accounts of the rising skill premium therefore identify a feedback loop between prior and subsequent inequalities.  Once again, meritocratic inequality is snowball inequality and therefore vulnerable to the moral attacks that The Meritocracy Trap mounts, in Chapter Nine, on this account.

[1] These premia, for prime-aged men and women, are reported by Michael Simkovic, in Michael Simkovic, “The Knowledge Tax,” University of Chicago Law Review 82, no. 4 (2015): 2036–37, tables 1–2.  Simkovic uses data on average earnings for 30- to 54-year-old labor force participants by education level, collected by the Minnesota Population Center, University of Minnesota, Integrated Public Use Microdata Series, Current Population Survey, archived at

[2] These premia, for prime-aged men and women, are reported in Simkovic, 2036–37, tables 1–2.  Simkovic uses data on average earnings for 30- to 54-year-old labor force participants by education level, collected by the Minnesota Population Center, University of Minnesota, Integrated Public Use Microdata Series, Current Population Survey, archived at

[3] Prominent statements of the connection among technical change, the rising demand for skill, and the rising skill premium include:  David Autor, Lawrence F. Katz, and Melissa S. Kearney, “The Polarization of the U.S. Labor Market” (working paper no. 11986, National Bureau of Economic Research, 2006),; Daron Acemoglu and David Autor, “What Does Human Capital Do? A Review of Goldin and Katz’s The Race Between Education and Technology” (working paper no. 17820, National Bureau of Economic Research, 2012),; Daron Acemoglu and David Autor, “Skill, Tasks and Technologies: Implications for Employment and Earnings,” in Handbook of Labor Economics, ed. Orley Ashenfelter and David Card, vol. 4b (Amsterdam: Elsevier, 2011),; David Autor, The Polarization of Job Opportunities in the U.S. Labor Market: Implications for Employment and Earnings (Center for American Progress and The Hamilton Project: 2010),; David H. Autor, Lawrence F. Katz, and Melissa S. Kearney, “Trends in U.S. Wage Inequality:  Revising the Revisionists,” Review of Economics and Statistics 90, vol. 2 (May 2008),; Mark Doms, Timothy Dunne, and Kenneth R. Troske, “Workers, Wages, and Technology,” Quarterly Journal of Economics 112, no. 1 (February 1997); David H. Autor, Lawrence F. Katz, and Melissa S. Kearney, “The Polarization of the U.S. Labor Market,” American Economic Review 96, no. 2 (May 2006); Frank Levy and Richard J. Murnane, The New Division of Labor: How Computers Are Creating the Next Job Market (Princeton, New Jersey: Princeton University Press, 2004); David H. Autor, Frank Levy, and Richard J. Murnane, “The Skill Content of Recent Technological Change:  An Empirical Exploration,” Quarterly Journal of Economics 118, no. 4 (November 2003); Daron Acemoglu, “Technical Change, Inequality, and the Labor Market,” Journal of Economic Literature 40, no. 1 (March 2002); Steven G. Allen, “Technology and the Wage Structure,” Journal of Labor Economics 19, no. 2 (April 2001); David H. Autor, Lawrence F. Katz, and Alan B. Krueger, “Computing Inequality: Have Computers Changed the Labor Market?” Quarterly Journal of Economics 113, no. 4 (November 1998); Stephen Machin and John van Reenen, “Technology and Changes in Skill Structure:  Evidence from Seven OECD Countries,” Quarterly Journal of Economics 113, no. 4 (November 1998); Daron Acemoglu, “Why Do New Technologies Complement Skills? Directed Technical Change and Wage Inequality,” Quarterly Journal of Economics 113, no. 4 (November 1998).

Tyler Cowen has popularized a similar view.  Tyler Cowen, Average is Over: Powering America Beyond the Age of the Great Stagnation (New York: Penguin Group, 2013).

[4] Claudia Goldin and Lawrence F. Katz, The Race Between Education and Technology (Cambridge, MA: Belknap Press, 2008), 101–02.

[5] Goldin and Katz, 93.

[6] Goldin and Katz, 303.

[7] Goldin and Katz, 298.

[8] Goldin and Katz, 292.

[9] Goldin and Katz, 101.

[10] Daniel Markovits, The Meritocracy Trap: How America’s Foundational Myth Feeds Inequality, Dismantles the Middle Class, and Devours the Elite (New York: Penguin Press, 2019), 393n252.

[11] National Center for Education Statistics, Digest of Education Statistics (2015 Table 104.20),  Goldin and Katz report a similar trend for the share of the US-born population to have completed at least 4 years of college by age 30.  Goldin and Katz, The Race Between Education and Technology, 249, fig. 7.1.  Goldin and Katz use data from the 1940 through the 2000 Census IPUMS, and the 2005 Current Population Survey MORG databases.

[12] Goldin and Katz, 250, fig. 7.2.

[13] The precise share for 2015 is 35.6 percent.  National Center for Education Statistics, Digest of Education Statistics (2015 Table 104.20),  For the trend through 2005, see Goldin and Katz, The Race Between Education and Technology, 249, fig. 7.1.

[14] In 2015, 65 percent of Americans aged 25 to 34 had at least some college education.  Camille L. Ryan and Kurt Bauman, “Educational Attainment in the United States: 2015,” Current Population Reports, 2, table 1,; Goldin and Katz, The Race Between Education and Technology, 250, fig. 7.2.

[15] Goldin and Katz, 280.

[16] Goldin and Katz, 267, fig. 7.7.

[17] Goldin and Katz, 261–62.  In 1900, private colleges were often larger than public colleges, and Yale and Harvard were bigger than the University of Michigan and the University of Minnesota, the two largest public universities at the time.  Overall, ratios of median public college to median private college enrollments have grown steadily over the past century: from 1.8 in 1900, to 3.4 in 1923, to 4.1 in 1933, to 5.2 by the 1990s. Goldin and Katz, 261–62.

[18] Goldin and Katz, 256.

[19] Goldin and Katz, 300–02.

[20] Goldin and Katz, 301–02.

[21] Goldin and Katz, 294, equation 3.

[22] Goldin and Katz, 297, notes to table 8.1.

[23] Markovits, The Meritocracy Trap, 357n138, citing National Center for Education Statistics, “Expenditures of Educational Institutions Related to the Gross Domestic Product, by Level of Institution: Selected Years, 1929– 0 through 2014– 5,” Table 106.10, https:// pro grams/ digest/ d15/ tables/ dt15_ 106.10.asp? referrer= report.

[24] Markovits, The Meritocracy Trap, 357n138.

[25] Markovits, 357n138.

[26] Goldin’s and Katz’s account of how they compute the supply of skills, and in particular their use of “college efficiency units,” touches on this issue but does not directly resolve it. Goldin and Katz, The Race Between Education and Technology, 295, 431–32n15. Moreover, their principal test of the reliability of their method of estimation emphasizes a paper—Lawrence F. Katz and Kevin M. Murphy, “Changes in Relative Wages 1963–1987:  Supply and Demand Factors,” Quarterly Journal of Economics 107, no. 1 (February 1992): 35–78—that employs data from the period for which their model fits badly and that does not (because, given when it was written, it cannot) consider the bulk of the recent rise in the college wage premium or the bulk of the period in which growth in per-graduate investments in college outstripped growth in relative graduation rates.

[27] Goldin and Katz, The Race Between Education and Technology, 302.

How America’s foundational myth feeds inequality, dismantles the middle class, and devours the elite. 

Share this post