Journal of Economic Literature, Vol. XL March 2002 ... · have technological advances been...

Journal of Economic LiteratureVol. XL (March 2002), pp. 7–72

Acemoglu: Technical Change, Inequality, and the Labor MarketJournal of Economic Literature, Vol. XL (March 2002)

Technical Change, Inequality,and the Labor Market

DARON ACEMOGLU1

1. Introduction

1.1 Motivation

What are the implications of techni-cal change for the labor market? Howdoes new technology affect the distri-bution of wages and income? Is tech-nology responsible for the changes inthe wage structure observed in manyadvanced economies since the 1970s?

The recent consensus is that techni-cal change favors more skilled workers,replaces tasks previously performed bythe unskilled, and exacerbates inequal-ity. This view is shaped largely by theexperience of the past several decades,which witnessed both major changes intechnology, including the rapid spreadof computers in workplaces and in ourlives, and a sharp increase in wage in-equality. In the United States, for ex-ample, the college premium—the wagesof college graduates relative to the wagesof high-school graduates—increased byover 25 percent between 1979 and1995. Overall earnings inequality alsoincreased sharply. In 1971, a worker at

the 90th percentile of the wage distri-bution earned 266 percent more than aworker at the 10th percentile. By 1995this number had risen to 366 percent(author’s calculations from March CPSdata). Many commentators see a directcausal relationship between technologi-cal changes and these radical shifts inthe distribution of wages taking place inthe U.S. economy. The title of AlanKrueger’s (1993) influential paper oncomputers and inequality summarizesthis view: “How Computers HaveChanged the Wage Structure.” JeremyGreenwood and Mehmet Yorukoglu(1997, p. 87) similarly give a succinctstatement:

Setting up, and operating, new technolo-gies often involves acquiring and processinginformation. Skill facilitates this adoptionprocess. Therefore, times of rapid technologi-cal advancement should be associated with arise in the return to skill.

They further argue that we are nowin the midst of a “Third Industrial Revo-lution,” fueled by advances in informa-tion technology, and that this revolu-tion is responsible for the increase ininequality (as does Francesco Caselli1999, in a paper entitled “TechnologicalRevolutions”).

The view that technological develop-ments favor skilled workers also receivessupport from accounts of earlier

7

1 Massachusetts Institute of Technology. I thankAlexis Leon for excellent research assistance andJoshua Angrist, David Autor, Olivier Blanchard,Lawrence Katz, Omer Moav, John McMillan,Lawrence Mishel, Donald Morrison, Lee Ohanian,Steve Pischke, Thomas Lemieux, Gianluca Vio-lante, Bas ter Weel, and two anonymous refereesfor comments.

episodes. For example, there were alreadysigns of significant technology-skillcomplementarity in the 1910s. ClaudiaGoldin and Lawrence Katz (1998) arguethat the spread of batch and continuous-process methods of production in-creased the demand for skills. Theyadd, “. . . the switch to electricity fromsteam and water-power energy sourceswas reinforcing because it reduced thedemand for unskilled manual workersin many hauling, conveying, and assem-bly tasks” (p. 695). Over this period,capital-intensive industries increasedthe demand for skills considerably (seeGoldin and Katz 1998, table 3), and thescope of these industries expanded withthe sharp fall in the price of electricity(see, for example, Arthur Woolf 1984,p. 178). The rapid increase in the im-portance of white-collar and clerical oc-cupations gave another boost to the de-mand for skills. Generalizing from theexperience of the 1920s, Harry Jerome(1934, p. 402) argued that “. . . in thefuture . . . there is considerable reasonto believe that the effect of further[mechanization] will be to raise theaverage skill required.”

The early twentieth-century evidencewas so powerful that Zvi Griliches(1969) suggested capital and skills areintrinsically complementary. RichardNelson and Edmund Phelps (1967),Finis Welch (1970), Theodore Schultz(1975), and Jan Tinbergen (1975) alsoargued that technological developmentsincrease the demand for skills. Eventssince then support this notion. Personalcomputers, computer-assisted produc-tion techniques, and robotics appear tocomplement skilled workers, replacingmany labor-intensive tasks. In this light,it is perhaps natural to view the in-crease in inequality over the past sev-eral decades as a direct consequence oftechnical change.

Although the consensus is now broad,

the idea that technological advances fa-vor more skilled workers is a twentieth-century phenomenon. In nineteenth-century Britain, skilled artisansdestroyed weaving, spinning, andthreshing machines during the Ludditeand Captain Swing riots, in the beliefthat the new machines would maketheir skills redundant. They were right:the artisan shop was replaced by thefactory and later by interchangeableparts and the assembly line (e.g., JohnJames and Jonathan Skinner 1985;Goldin and Katz 1998). Products pre-viously manufactured by skilled artisansstarted to be produced in factories byworkers with relatively few skills, andmany previously complex tasks weresimplified, reducing the demand forskilled workers.2 Joel Mokyr (1990, p.137) describes this process vividly:

First in firearms, then in clocks, pumps,locks, mechanical reapers, typewriters, sew-ing machines, and eventually in engines andbicycles, interchangeable parts technologyproved superior and replaced the skilledartisans working with chisel and file.

2 In the absence of detailed econometric evi-dence, it is difficult to generalize from the histori-cal examples and conclude that technical changewas definitely skill-replacing during the nine-teenth century. Still, the only econometric evi-dence that I am aware of supports this view. Usingthe 1850 Census of Manufacturers, James andSkinner (1985) find that there was more rapid sub-stitution of capital for skilled workers than un-skilled workers. It can also be argued that techni-cal change always increases the demand for“skills,” and the artisans who were hurt as a resultof new technology were not “skilled” since theylacked the flexibility to adapt to the requiredchanges. This argument is not totally convincing,since the artisans earned considerably more thanother laborers (for example, James and Skinner1985 report over 60 percent wage differentials forbuilding and printing workers relative to laborersin the 1850s). So the artisans possessed skills thatwere being rewarded by the market, and the stan-dardization of the production process destroyedthese rewards. On the other hand, it has to benoted that many of the skill-replacing technologiesof the nineteenth century may have also increasedthe demand for engineers and managers (see, e.g.,Goldin and Katz 1998).

8 Journal of Economic Literature, Vol. XL (March 2002)

Interchangeable parts were in fact verymuch designed to be skill-replacing(unskill-biased). Eli Whitney, a pioneerof interchangeable parts, described theobjective of this technology as:

. . . to substitute correct and effective op-erations of machinery for the skill of the art-ist which is acquired only by long practiceand experience; a species of skill which is notpossessed in this country to any considerableextent. . . . (quoted in H. J. Habakkuk 1962,p. 22)

The experience of the nineteenth andearly twentieth centuries led HarryBraverman (1974) and Stephen Marglin(1974) to argue that technical changewas “deskilling”—a major purpose oftechnical change was to expand the divi-sion of labor and simplify tasks pre-viously performed by artisans by break-ing them into smaller, less skill-requiringpieces. Braverman (1974, p. 113), for ex-ample, suggested that the first principleof management and production tech-niques of the period was “dissociation ofthe labor process from skills of the work-ers. The labor process is to be renderedindependent of craft, tradition, and theworkers’ knowledge.”

A longer view therefore suggests thattechnological advances may not havealways increased the demand for skills.In fact, most early nineteenth-centuryinnovations appear to have replacedskilled workers and expanded tasks per-formed by the unskilled. But then, whyhave technological advances been skill-biased in the twentieth century? And,are technological changes the major causeof the recent increase in inequality?

This essay attempts to answer thesequestions. It has two main theses:

• The behavior of wages and returns toschooling indicates that technicalchange has been skill-biased during thepast sixty years and probably for mostof the twentieth century. Furthermore,

an acceleration in skill bias during thepast few decades appears to be the maincause of the increase in inequality.

• We can understand the behavior oftechnical change by recognizing thatthe development and use of technol-ogy is, at least in part, a response toprofit incentives.3 When developingskill-biased techniques is more profit-able, new technology will tend to beskill-biased. I suggest that the earlynineteenth century was characterizedby skill-replacing developments be-cause the increased supply of unskilledworkers in the English cities (resultingfrom migration from rural areas andfrom Ireland) made the introductionof these technologies profitable. Incontrast, the twentieth century hasbeen characterized by skill-biasedtechnical change because the rapid in-crease in the supply of skilled workershas induced the development of skill-complementary technologies. The re-cent acceleration in skill-biased tech-nical change is in turn likely to havebeen a response to the rapid increasein the supply of skills during the pastseveral decades. However, I also arguethat despite the acceleration in skillbias, we are most likely not in themidst of a “technological revolution”;what has changed is not necessarilythe overall rate of progress, but thetypes of technologies that are beingdeveloped.

3 Precedents of this approach include JacobSchmookler (1966), who emphasized demand pulland the extent of the market as key determinantsof innovations; the endogenous growth theory,e.g., Paul Romer (1990), Gene Grossman and El-hanan Helpman (1991), and Philippe Aghion andPeter Howitt (1992); the induced innovation the-ory, including Syed Ahmad (1965), Charles Ken-nedy (1964), Paul Samuelson (1970), YujiroHayami and Vernon Ruttan (1970), and PaulDavid (1975); and recent work including my own,Acemoglu (1998, 1999b, 2000), Acemoglu andFabrizio Zilibotti (1999), and Michael Kiley(1999).

Acemoglu: Technical Change, Inequality, and the Labor Market 9

Finally, I conjecture that recent tech-nological developments are likely to haveaffected the organization of the labormarket—including the way firms areorganized, labor market policies, andthe form of labor market “institutions”—and may have had a large effect on thestructure of wages through this channel.

In the process of developing this ar-gument, this essay sets out a simpletheoretical framework in which inequal-ity and returns to skills are determinedby supply and demand forces (technol-ogy).4 Using this framework as a unify-ing device, I critically survey many ofthe theories that explain the recent in-crease in inequality by technologicalfactors and discuss how various piecesof evidence can be interpreted withinthis framework.

1.2 Summary of the Argument

I begin with a roadmap of the argu-ment. Since the effect of new technol-ogy on the distribution of wages in therecent past is central to the focus here,I organize this essay around a numberof salient facts from the post-war U.S.economy.5 Briefly, these facts are:

1. The past sixty years have seen a largeincrease in the supply of more educatedworkers, while returns to educationhave risen.

2. Returns to education fell during the1970s, when there was a very sharpincrease in the supply of educatedworkers. Returns to education thenbegan a steep rise during the 1980s.

3. Overall wage inequality rose sharplybeginning in the early 1970s. In-

creases in within-group (residual) in-equality—i.e., increases in inequalityamong observationally equivalentworkers—account for much of thisrise.

4. Average wages have stagnated andwages of low-skill workers have fallenin real terms since 1970.

I argue that technical change over thepast sixty years, or even over the pastcentury, has been skill-biased. This con-clusion follows from fact 1 above: in theabsence of substantial skill bias in tech-nology, the large increase in the supplyof skilled workers would have depressedthe skill premium. In 1970, Welch (1970,p. 36) reached the same conclusion, andargued:

With the phenomenal rise in average edu-cation, why have rates of return failed todecline? . . .

It is obvious that changes have occurred toprevent the decline in returns to acquiringeducation that would normally accompany arise in average educational level. Presumably,these changes have resulted in growth in de-mand for . . . education . . . sufficient to ab-sorb the increased supply with constant orrising returns.

The thirty years after Welch wrote thesewords witnessed a much more rapid in-crease in the supply of education and asharp increase in the returns to moreskilled workers, suggesting that skill-biased changes in technology continuedthroughout the postwar period.

And yet, if technical change has beenskill-biased throughout the recent past,why did inequality increase during thepast thirty years, but not before? Thereare at least two possible answers to thisquestion. The first, which I call thesteady-demand hypothesis, maintains thatdemand for skills increases at a constantpace, so changes in inequality must beexplained by the pace of the increase inthe supply of skills. According to thishypothesis, inequality (returns to skills)

4 Precedents of the supply and demand ap-proach include, among others, Gary Becker(1964), Finis Welch (1970), and Jan Tinbergen(1975).

5 I limit the discussion of the major trends tothe U.S. economy because of space constraints andalso because there is notably more research tobuild upon.


was relatively stable before the 1970s,because the rate of skill accumulationin the U.S. economy was as rapidas the constant pace of skill-biasedtechnical change (e.g., Katz and KevinMurphy 1992; Murphy, Craig Riddell,and Paul Romer 1998; David Card andThomas Lemieux 2000). The recent in-crease in inequality is then explained,not by a major technological change,but by a decline in the growth rate ofthe supply of skills. The second possibleanswer comes from the acceleration hy-pothesis, which maintains that there hasbeen an acceleration in skill bias begin-ning in the 1970s or the 1980s. Accord-ing to this hypothesis, there is a dis-continuity in the growth rate of thedemand for skills. The most popularversion of this hypothesis claims thatthere has been a notable acceleration inthe skill bias of technology, driven byadvances in information technology, orperhaps a “Third Industrial Revolution.”

So was there an acceleration in skillbias? This question is difficult to an-swer, as we lack direct measures of thedegree of skill bias. To tackle this ques-tion, one therefore needs to look at avariety of evidence often pointing indifferent directions. I conclude belowthat skill-biased technical change islikely to have accelerated over the pastseveral decades. This conclusion isbased on the sharp increase in overallinequality starting in the 1970s and onthe fact that returns to schooling roseover the past thirty years despite theunusually rapid increase in the supplyof educated workers.

Why did the demand for skills accel-erate over this period? And why hasnew technology favored more skilledworkers throughout the twentieth cen-tury, but not during the nineteenth cen-tury? One approach would view tech-nology as exogenous, stemming fromadvances in science or from the behav-

ior of entrepreneurs driven by a varietyof nonprofit motives. Demand for skillsincreased faster during the past thirtyyears, this approach would maintain,because of a technological revolution ledby the microchip, personal computers,and the internet.6 New technologies ofthe early nineteenth century were skill-replacing (unskill-biased) because thetechnological frontier then only en-abled the invention of skill-replacingtechniques.

Nevertheless, there are several prob-lems with this approach. First, althougha number of papers, including Green-wood and Yorukoglu (1997), AndresHornstein and Krusell (1997), and Ga-lor and Moav (2000), show that rapidtechnical change may lead to slower to-tal factor productivity (TFP) growth,the slow rates of TFP and outputgrowth of the past several decades aredifficult to reconcile with a technologi-cal revolution during this time period.Second, demand for skills appears tohave accelerated starting in the late1970s, precisely when the supply ofskills increased very rapidly. Exogenoustechnology theories do not explain thetiming of this acceleration.7

6 See, among others, Krueger (1993), Eli Ber-man, John Bound, and Griliches (1994), and DavidAutor, Katz, and Krueger (1998) for evidence thatthe rapid spread of computers has increased thedemand for skills. See Per Krusell, Lee Ohanian,Victor Rios-Rull, and Giovanni Violante (2000),Oded Galor and Daniel Tsiddon (1997), Green-wood and Yorukoglu (1997), Aghion and Howitt(1998, ch. 9), Caselli (1999), Galor and OmerMoav (2000), Violante (2000), Yonna Rubinsteinand Tsiddon (1999), Aghion, Howitt, and Violante(2000), and Eric Gould, Moav, and Bruce Wein-berg (2000) for models in which rapid technicalchange increases the demand for skills and causesa rise in inequality.

7 Naturally, supply and demand may havemoved together because supply responded to de-mand. I argue below that the large increase in thesupply of educated workers was not in anticipationof, or in response to, high returns, but driven by avariety of other factors. More generally, I oftenfocus on the effect of the supply of skills on tech-nology not because I view supply as exogenous,


An alternative theory maintains in-stead that new technologies are endoge-nous and respond to incentives. It wasthe large increase in the supply ofskilled workers, this approach claims,that induced the acceleration in the de-mand for skills. When skill-biased tech-niques are more profitable, firms willhave greater incentives to develop andadopt such techniques. A key determi-nant of the profitability of new tech-nologies is their market size; machinesthat can be sold in greater numbers willbe more profitable. Jacob Schmookler(1966), in his pioneering study, Inven-tion and Economic Growth, placedgreat emphasis on market size. He ar-gued that (p. 206) “invention is largelyan economic activity which, like othereconomic activities, is pursued for gain;. . . expected gain varies with expectedsales of goods embodying the inven-tion.” This reasoning implies that ma-chines complementary to skilled work-ers will be more profitable to developwhen there are more skilled workersto use them. New technologies havebecome more skill-biased throughoutmost of the twentieth century becausethe supply of skilled workers has grownsteadily. This perspective also suggeststhat a faster increase in the supply ofskills can lead to an acceleration in thedemand for skills (Acemoglu 1998). Sothe timing of the increases in supplyand demand is not a coincidence—instead, it reflects technology respond-ing to the supply of skills. Furthermore,rapid skill-biased technical change isnot necessarily associated with rapidoverall technical progress. In fact, an

acceleration in skill bias could cause aTFP slowdown because it creates animbalance in the composition of R&D.

This approach also provides a possi-ble explanation for the skill-replacingtechnical change of the early nine-teenth century. The emergence of themost skill-replacing technologies of thepast two hundred years, the factory sys-tem, coincided with a large change inrelative supplies. This time, there was alarge migration of unskilled workersfrom villages and Ireland to English cit-ies (see, for example, Habakkuk 1962;Paul Bairoch 1988; or Jeffrey William-son 1990). This increase in the “reservearmy of unskilled workers,” slightlyparaphrasing Karl Marx, created profitopportunities for firms to exploit by in-troducing technologies that could beused with unskilled workers. In fact,contemporary historians considered theincentive to replace skilled artisans byunskilled laborers to be a major objec-tive of technological improvements ofthe period. Ure, a historian in thefirst half of the nineteenth century,describes these incentives as follows:

It is, in fact, the constant aim and tendencyof every improvement in machinery to super-sede human labor altogether, or to diminishits costs, by substituting the industry ofwomen and children for that of men; of thatof ordinary labourers, for trained artisans.(quoted in Habakkuk 1962, p. 154)

These incentives for skill-replacing tech-nologies, I argue, were shaped by thelarge increase in the supply of unskilledworkers. So, it may be precisely the dif-ferential changes in the relative supplyof skilled and unskilled workers that ex-plain both the presence of skill-replacingtechnical change in the nineteenth cen-tury and skill-biased technical changeduring the twentieth century.

A major shortcoming of the “puretechnological” approaches—of both ex-ogenous and endogenous varieties—is

but simply because the effect of supply on tech-nology is more important in understanding thequestions posed above. I discuss below how supplymay respond to changes in skill premia, and howthis response may account for the joint behavior ofthe supply of, and demand for, skills over the pastcentury.


that they do not provide a natural expla-nation for the fall in the wages oflow-skill workers. Although a number ofpapers, including Caselli (1999), Green-wood and Yorukoglu (1997), and Galorand Moav (2000), show that technologi-cal revolutions may be associated with afall in the wages of low-skill workers, itis difficult to see how sustained techno-logical change can be associated withan extended period of falling wages oflow-skill workers and stagnant averagewages. This leads to the next questionfor this essay.

Why did the real wages of low-skillworkers fall over the past several de-cades? There are a number of possibleanswers. First, labor-market institu-tions, for example labor unions, under-went important changes over the pastthirty years, and these changes mayhave reduced the wages of many manu-facturing workers, causing an increasein inequality and a decline in the realwages of low-skill workers (e.g., Rich-ard Freeman 1991; John DiNardo, Ni-cole Fortin, and Lemieux 1995; DavidLee 1999). Second, international tradebetween skill-scarce less-developed coun-tries and skill-abundant rich economiesincreased over this period, and this mayhave put downward pressure on thewages of low-skill workers in the UnitedStates (e.g., Adrian Wood 1994; EdwardLeamer 1995). Third, there has been atransformation of the way in whichfirms are organized, or perhaps in theway that firms and workers match (see,for example, Acemoglu 1999a; MichaelKremer and Eric Maskin 1999; TimothyBresnahan 1997; Bresnahan, Erik Bryn-jolfsson, and Lorin Hitt 1999; andAutor, Frank Levy, and Richard Mur-nane 2000). Although each of these fac-tors could have been the cause of therecent changes in the wage structure, Iargue that their direct effect has beenlimited. Instead, I suggest that organi-

zational change, labor-market institu-tions, and international trade have in-teracted with technical change in a fun-damental way, amplifying the directeffect of technical change on inequality,and likely causing the decline in thewages of less-skilled workers.

Therefore, the overall picture thatemerges is not necessarily one in whichtechnology is the only factor affectingthe distribution of income. On the con-trary, the underlying thesis of this essayis that technology itself is no more thanan endogenous actor. To explain thechanges in the distribution of income,and to forecast what other changes mayhappen in the future, we need to under-stand the forces that shape technologi-cal progress, and how technology inter-acts with the overall organization of thelabor market.

There is considerable uncertainty onmany issues, and both more theoreticaland more empirical work are needed.Two areas deserve special attention.The first is the differential behavior ofresidual inequality and returns toschooling during the 1970s. Mosteconomists view changes in residual in-equality as related to changes in labor-market prices. It is therefore puzzlingthat during the 1970s, while returns toschooling fell, residual and overall in-equality increased. I argue below thatmodels based on a single skill index(one type of skill or many types of skillsthat are perfect substitutes) are unableto explain this pattern. Instead, we needmodels with multidimensional skills.Moreover, for this type of model to ex-plain the behavior of residual inequalityduring the 1970s and the 1980s, techno-logical progress needs to have changedthe demand for different types of skillsdifferentially. The endogenous technol-ogy models discussed above provide apossible explanation for why differentdimensions of skills may have been


affected differentially by technicalchange. Nevertheless, the reasons for thistype of behavior require much more re-search. More generally, we know rela-tively little about the determinants ofresidual inequality, and this topic is amajor research area for the future. Thesecond area is cross-country differencesin the behavior of wage inequality.While inequality increased sharply inthe United States, the United Kingdom,and Canada, it increased much lessin Germany and many Scandinavianeconomies. Although there are a num-ber of recent papers addressing thesequestions, much uncertainty still re-mains. I conjecture that cross-countrydifferences in wage inequality mayreflect, in part, technological choicesmade by these countries in responseto the different incentives created bytheir labor-market institutions, butmuch more research on this topic isrequired.

2. Empirical Trends

The objective of this section is to il-lustrate a number of major inequalitytrends from the past several decades.My aim is not to offer a comprehensivesurvey of the empirical literature, butsimply to highlight the most salienttrends to anchor the theoretical discus-sion (see, e.g., Peter Gottschalk 1997;George Johnson 1997; Katz and Autor2000, for recent surveys).

Figure 1 plots a measure of the sup-ply of college skills between 1949 and1995, constructed along the lines ofAutor, Katz, and Krueger (1998), as theratio of college equivalents (those withat least college + 0.5 × those with somecollege) to noncollege equivalents(those with high school or less + 0.5 ×those with some college).8 It also plotsreturns to college. This picture summa-

rizes many of the salient trends I wantto emphasize. In particular,

1. There has been a remarkable in-crease in the supply of skills in the U.S.economy over the past sixty years. In1939, just over 6 percent of Americanworkers were college graduates. By1996 this number had increased to over28 percent. In 1939, almost 68 percentof all workers did not have a highschool degree. In 1996, this numberhad fallen to less than 10 percent (see,for example, Autor, Katz, and Krueger1998, table 1). The relative supply ofskills plotted in figure 1 provides asummary of these changes.

2. There has been no tendency forthe returns to college to fall in the faceof this large increase in supply—on thecontrary, there is an increase in the col-lege premium over this time period. Animportant issue is whether changes inthe returns to college (or more gener-ally other measures of wage inequality)correspond to true changes in the re-turns to skills. As is well-known in thelabor literature, observed schooling pre-mia may reflect returns to ability. Thisraises the possibility that changes in thereturns to schooling may be driven bycomposition effects (changes in thecomposition of ability across schoolinggroups). In the appendix, I elaboratehow changes in the distribution of un-observed skills across groups can createcomposition effects, and show thatthese composition effects cannot be re-sponsible for the changes in the wagestructure. Therefore, here I interpretthese changes in the observed returnsto schooling as changes in the trueprice of skills.

3. Following an acceleration in thesupply of skills, returns to college fellsharply during the 1970s, leadingRichard Freeman to conclude that“Americans are over-educated” (Free-man 1976). Returns to college then rose8 See the appendix for data details.


very sharply during the 1980s. This in-crease in the returns to schooling hasbeen one of the major motivating factsfor the empirical inequality literature(e.g., Bound and Johnson 1992; Katzand Murphy 1992).

There have also been importantchanges in the overall distribution ofwages. Figure 2 plots the 90th, 50thand 10th percentiles of the overallwage (weekly earnings) distributionfor white male workers between 1963and 1997 (with the 1963 values forall series indexed to 100).9 This figure

illustrates two more important patterns:1. Overall wage inequality started to

increase sharply in the early 1970s aftera period of relative stability—prior tothe 1970s, the 90th, 50th, and 10th per-centiles of the wage distribution fol-lowed each other closely, but cameapart sharply in the 1970s.

2. Median wages stagnated from 1975onwards, while workers at the 10thpercentile of the wage distribution (i.e.,

9 Sample constructed as described in the appen-dix. I focus here on wage inequality for white mensince labor market participation of women in-creased substantially over the sample period, and

this would likely contribute to the composition ef-fects. Moreover, male-female wage difference nar-rowed substantially over the same time period aswell. School quality for black men also under-went significant transformation (e.g., Welch 1973;Card and Krueger 1992), and this could createsignificant composition effects.

.6

96

Relative Supply of College Skills and College Premiumyear

College wage premium Rel. supply of college skillsC

olle

ge w

age

prem

ium

Rel

. sup

ply

of c

olle

ge s

kills

Figure 1. The Behavior of the (log) College Premium and Relative Supply of College Skills (weeks worked bycollege equivalents divided by weeks worked of noncollege equivalents) between 1939 and 1996. Data fromMarch CPSs and 1940, 1950, and 1960 censuses.

897969594939

.5

.4

.3 0

.2

.4

.6

.8


“low-skill workers”) saw their earningsfall in real terms to levels even belowthose in 1963.10

Figure 3 turns to another measure: re-sidual (within-group) inequality, whichshows inequality among observationallyequivalent workers. This figure displaysthree measures of residual inequalityamong white male workers between1963 and 1997: 50–10, 90–50 and 0.5times 90–10 log wage residual differen-tials (I plot 0.5 times 90–10 wage differ-entials in order to fit this on the samescale as the other measures).

10 Average wages, like median wages, have stag-nated. For example, white men aged 30–49 earned409 a week in 1999 dollars in 1949, and 793 in1969, which corresponds approximately to a 3.4percent a year increase in real wages between1949 and 1969. In contrast, the same age groupearned 909 in 1989, or experienced only a 0.6 per-cent a year increase between 1969 and 1989 (allnumbers author’s calculation from census data).The behavior of the median and average wagegrowth depends on the consumption deflator. Ihave followed the literature in using the personalconsumption expenditure deflator. It has been ar-gued that this deflator overstates inflation becauseof difficulties in measuring quality change (e.g.,Michael Boskin et al. 1995). Even in the presenceof such measurement problems, a large gap re-mains between the rate of increase of real wagesbefore and after the 1970s, unless there is an “ac-celeration” in this bias exactly around the 1970s.Part of this gap is due to the increased importanceof nonwage income and benefits. In fact, thanks to

the increase in benefits, the share of labor in na-tional income has not fallen over this period (see,e.g., Krueger 1999). So whether average wageshave stagnated or continued to increase in linewith output growth depends on how benefits arevalued relative to earnings. It is also important tonote that if these non-wage benefits are taken intoaccount, inequality appears to have increased evenmore than the numbers here indicate (BrooksPierce 2000). This is because high-wage workersare the primary recipients of such benefits.

150

95

Indexed Wages for White Males 1963–97year

Figure 2. Changes in the Indexed Value of the 90th, 50th, and 10th Percentiles of the Wage Distribution forWhite Males (1963 values normalized to 100). Data from March CPSs.

908580757065

index 10th pctile wagesindex 90th pctile wages

140

130

120

110

100

90

index 50th pctile wages


To calculate these measures, I look atthe residuals from a standard Mincerianwage regression of the form

ln wit = X′itβt + vit, (1)

where wit is weekly earnings for individ-ual i observed in year t, and Xit is a set ofcontrols which include nine educationdummies, a quartic in experience, andregion controls (constructed from theMarch CPSs; see the appendix for detailsof the sample). The fact that βt is in-dexed by t indicates that returns to theseobserved characteristics are allowed tovary from year to year. The measures ofresidual inequality are calculated as thedifference between the 90th and the10th (or the 50th and the 10th, etc.)percentile values of the residual distri-bution from this regression, vit. Residualinequality appears to have increased very

much in tandem with overall inequality—it shows a sharp increase starting in theearly 1970s.11 Remarkably, all three mea-sures of residual inequality behave verysimilarly, suggesting that forces affectingthe top of the male wage distribution(90–50) are also affecting the bottom ofthe wage distribution (50–10). Finally, notean important contrast between figure 1and figures 2 and 3. While returns toschooling fell during the 1970s, overall

11 DiNardo, Fortin, and Lemieux (1995) provideevidence from the May Current Population Survey(CPS) data that residual and overall wage inequal-ity started to increase in the 1980s. Katz and Autor(2000), on the other hand, find that residual wageinequality started to increase in the 1970s fromMarch and May CPS data, and from census data.See the appendix. Recent work by Thomas Pikettyand Emanuel Saez (2001) also finds an increase ininequality during the 1970s using data from theInternal Revenue Service tax returns.

.65

95

Residual Inequality Measures for White Males 1963–97year

Figure 3. 90–50, 50–10 and 0.5 × 90–10 Differentials from Log Weekly Wage Regressions for White Males Aged 18–65.

908580757065

90–50 residual differences0.5 times 90–10 residual diffs

.6

.55

.5

.45

.4

50–10 residual differences


and residual inequality increased. Ireturn to this issue later in the essay.

3. Introduction to the Theoryof Skill Premia

While, undoubtedly, many factors af-fect the distribution of wages, a naturalstarting point for an economic analysisis that of supply and demand. In the in-troduction to his pioneering study ofincome distribution, Tinbergen (1975,p. 15) wrote

. . . what matters is the difference betweenqualities available and qualities required bythe demand side, that is by the organizationof production. (italics in original)

This is where I begin as well. I intro-duce a simple framework that linkswages to supply of skills and to demandgenerated by the technology possibili-ties frontier of the economy. There aretwo types of workers, skilled and un-skilled (high and low-education work-ers), who are imperfect substitutes. Im-perfect substitution between the twotypes of workers is important in under-standing how changes in relative sup-plies affect skill premia. For now Ithink of the unskilled workers as thosewith a high school diploma, and theskilled workers as those with a collegedegree.12 So the focus in this section ison returns to schooling (or between-group inequality), and I use the terms“skill” and education interchangeably.In practice, education and skills areonly imperfectly correlated, so it is use-ful to bear in mind that since there areskilled and unskilled workers within thesame education group, an increase in

the returns to skills will also lead to anincrease in within-group inequality.

Suppose that there are L(t) unskilled(low-education) workers and H(t) skilled(high-education) workers, supplying la-bor inelastically at time t. All workersare risk neutral, and maximize (the pres-ent value of) labor income. Also sup-pose that labor markets are competitive(I return below to the role played bynoncompetitive elements, especially inthe European context).

The production function for the ag-gregate economy takes the constantelasticity of substitution (CES) form

Y(t) = [(Al(t)L(t))ρ + (Ah(t)H(t))ρ]1 ⁄ ρ, (2)

where ρ ≤ 1, and Al(t) and Ah(t) are factor-augmenting technology terms. Althoughall the results of interest hold for a generalconstant returns to scale F(.) function, Ifocus on the CES production function tosimplify the discussion. I drop the timeargument when this causes no confusion.

The elasticity of substitution betweenskilled and unskilled workers in thisproduction function is σ ≡ 1 ⁄ (1 − ρ). Irefer to skilled and unskilled workers asgross substitutes when the elasticity ofsubstitution σ > 1 (or ρ > 0), and grosscomplements when σ < 1 (or ρ < 0).Three noteworthy special cases are: (i)σ → 0 (or ρ → −∞) when skilled and un-skilled workers will be Leontieff, andoutput can be produced only by usingskilled and unskilled workers in fixedproportions; (ii) σ → ∞ when skilled andunskilled workers are perfect substi-tutes, and (iii) σ → 1, when the pro-duction function tends to the CobbDouglas case. The value of the elasticityof substitution will play a crucial role inthe interpretation of the results that fol-low. In particular, in this framework,technologies either increase the pro-ductivity of skilled or unskilled workers,that is, there are no explicitly skill-replacing or unskilled-labor-replacing

12 This classification of workers into skilled andunskilled groups is obviously appropriate only forthe twentieth-century United States. In manyother countries or periods, high school graduatescould be considered to be “skilled” workers. Moregenerally, the two-skill model is a convenientsimplification, less realistic than a world with acontinuum of imperfectly substitutable skills.


technologies.13 But, as we will see be-low, depending on the value of the elas-ticity of substitution, an increase in Ah

can act either to complement or to“replace” skilled workers.

The production function (2) admitsthree different interpretations:

1. There is only one good, andskilled and unskilled workers are imper-fect substitutes in the production of thisgood.

2. The production function (2) is alsoequivalent to an economy where con-sumers have utility function [Yl

ρ + Yhρ]1 ⁄ ρ

defined over two goods. Good Yh is pro-duced using only skilled workers, and Yl

is produced using only unskilled work-ers, with production functions Yh = AhH,and Yl = AlL. In this interpretation, it isimportant that the economy is closed.We will see in section 6.3 that differentresults obtain when international tradein these two goods is allowed.

3. A mixture of the above twowhereby different sectors producegoods that are imperfect substitutes,and high- and low-education workers areemployed in both sectors.

Although the third interpretation ismore realistic, I generally use one of thefirst two, as they are easier to discuss.Since labor markets are competitive,the unskilled wage is

wL =∂Y∂L

= Alρ[Al

ρ + Ahρ(H ⁄ L)ρ](1 − ρ) ⁄ ρ. (3)

This equation implies ∂wL ⁄ ∂H ⁄ L > 0: asthe fraction of skilled workers in thelabor force increases, the unskilled wageshould increase. Similarly, the skilledwage is

wH =∂Y∂H

= Ahρ[Al

ρ(H ⁄ L) − ρ + Ahρ](1 − ρ) ⁄ ρ,

which yields ∂wH ⁄ ∂H ⁄ L < 0; everythingelse equal, as skilled workers becomemore abundant, their wages should fall.Combining these two equations, the skillpremium—the wage of skilled work-ers divided by the wage of unskilledworkers—is14

ω =wH

wL=

Ah

Al

ρ

HL

−(1 − ρ)

=

Ah

Al

(σ − 1) ⁄ σ

HL

−1 ⁄ σ

.

(4)

Equation (4) can be rewritten in a moreconvenient form by taking logs,

ln ω =σ − 1

σln

Ah

Al

−1σ

ln

HL

. (5)

Naturally, the skill premium in-creases when skilled workers becomemore scarce, i.e.,

∂ ln ω∂ ln H ⁄ L

= −1σ

< 0. (6)

This is the usual substitution effect, andshows that for given skill bias of technol-ogy, as captured by Ah ⁄ Al, the relativedemand curve for skill is downward slop-ing with elasticity 1 ⁄ σ = (1 − ρ). Intui-tively, an increase in H/L creates twodifferent types of substitution. First, if

13 A more general formulation would replaceequation (2) with the production function

Y(t) = [(1 − bt)(Al(t)L(t) + Bl(t))ρ

+ bt(Ah(t)H(t) + Bh(t))ρ]1 ⁄ ρ,

where Bl and Bh would be directly unskilled-labor-replacing and skill-replacing technologies, and anincrease in bt would correspond to some of thetasks previously performed by the unskilled beingtaken over by the skilled (see, e.g., Johnson andStafford 1999 on this). For most of the analysishere, there is little to be gained from this moregeneral production function (but see section 5.3).

14 For some parameter values, skilled workersmay have lower wages than the unskilled, i.e.,ω ≤ 1. One may want to impose

Ah

Al

σ − 1

>HL

,

to avoid this. Alternatively, one could assume thatskilled workers can use the technologies normallyused by the unskilled, Al, and be more (or equally)productive at this than the unskilled.


skilled and unskilled workers are produc-ing the same good, but performing dif-ferent functions, an increase in the num-ber of skilled workers will necessitate asubstitution of skilled workers for taskspreviously performed by the unskilled.Second, if skilled and unskilled work-ers are producing different goods, thegreater number of skilled workers willlead to a substitution of the consumptionof the unskilled good by the skilled good.In both cases, this substitution hurts therelative earnings of skilled workers.

Figure 4 draws the relative demandfor skills, equation (5), against the rela-tive supply of skills, H/L, which is takento be given for the purposes of this ex-ercise. An increase in the relative sup-ply, from H/L to H′/L′, moves the equi-librium point along the downwardsloping relative demand curve, andreduces the skill premium from ω to ω′.

An interesting case study of the re-sponse of the returns to schooling to anincrease in the supply of skills is pro-vided by the experience in the WestBank and Gaza Strip during the 1980s.As Joshua Angrist (1995) illustrates,there was a very large increase in thesupply of skilled Palestinian labor asPalestinian institutions of higher educa-tion, totally absent before 1972, beganto open. Angrist shows that premia to

college graduate workers (relative tohigh school graduates) that were as highas 40 percent quickly fell to less than 20percent. The extent of substitution wasalso clear. First, many college graduateworkers could not find employment inskilled jobs: Angrist (1995) shows asharp increase in the unemploymentrate of college graduates, and Ze’evSchiff and Ehud Ya’ari (1989) reportthat only one in eight Palestinian gradu-ates could find work in his profession,with the rest working as unskilled labor-ers, mainly in the construction industry.Second, premia for tasks usually per-formed by more educated workers fellsharply. Between 1984 and 1987, thepremium for administrative and mana-gerial jobs (relative to manual laborers)fell from .32 to .12, while the premiumfor clerical workers fell from .02 to –.08(see Angrist 1995 for details).

As equation (6) shows, the elasticityof substitution, σ, is important for thebehavior of the skill premium whensupply changes. The elasticity of substi-tution is also crucial for the response ofthe skill premium to changes in tech-nology. Unfortunately, this parameter israther difficult to estimate, since it re-fers to an elasticity of substitution thatcombines substitution both within andacross industries. Nevertheless, thereare a number of estimates using aggre-gate data that give a range of plausiblevalues. The majority of these estimatesare between σ = 1 and 2 (see for exam-ple Freeman 1986). The response ofcollege premium for Palestinian laborreported in Angrist (1995), for example,implies an elasticity of substitution be-tween workers with sixteen years ofschooling and those with less thantwelve years of schooling of approxi-mately σ = 2.

Given the focus of this essay, it is alsoimportant to know how the skill premiumresponds to technology. Differentiation

ω′′

Figure 4. The Relative Demand for Skills

H/L

Relative supplyof skills

Relative demandfor skills

Skill-biased tech.change

Skill premium

H′/L′

ω

ω′


of (5) shows that the result depends onthe elasticity of substitution:

∂ ln ω∂ ln (Ah ⁄ Al)

=σ − 1σ

.

Therefore, if σ > 1 (i.e., ρ ∈ (0, 1]), thenimprovements in the skill-complementarytechnology increase the skill premium.This can be seen in figure 4 as a shift outof the relative demand curve, moving theskill premium from ω to ω′′. The con-verse is obtained when σ < 1: that is,when σ < 1, an improvement in the pro-ductivity of skilled workers, Ah, relativeto the productivity of unskilled workers,Al, shifts the relative demand curve inand reduces the skill premium. This caseappears paradoxical at first, but is, infact, quite intuitive. Consider, for ex-ample, a Leontieff (fixed proportions)production function. In this case, whenAh increases and skilled workers becomemore productive, the demand for un-skilled workers increases by more thanthe demand for skilled workers. In somesense, in this case, the increase in Ah iscreating an “excess supply” of skilledworkers given the number of unskilledworkers. This excess supply increases theunskilled wage relative to the skilledwage. This observation raises an impor-tant caveat. It is tempting to interpretimprovements in technologies used byskilled workers, Ah, as “skill-biased.”However, when the elasticity of substitu-tion is less than 1, it will be advances intechnologies used with unskilled work-ers, Al, that increase the relative produc-tivity and wages of skilled workers, andan increase in Ah relative to Al will be“skill-replacing.”

Nevertheless, the conventional wis-dom is that the skill premium increaseswhen skilled workers become relativelymore—not relatively less—productive,which is consistent with σ > 1. In fact,as noted above, most estimates show an

elasticity of substitution between skilledand unskilled workers greater than 1.

It is also useful to compute the aver-age wage in this economy. Without con-trolling for changes in the educationalcomposition of the labor force, theaverage wage is

w =LwL + HwH

L + H=

[(AlL)ρ + (AhH)ρ]1 ⁄ ρ

1 + H ⁄ L, (7)

which is also increasing in H/L as long asthe skill premium is positive (i.e., ω > 1or Ah

ρ(H ⁄ L)ρ − Alρ > 0). Intuitively, as the

skill composition of the labor forceimproves, wages will increase.

Therefore, the results so far implythat in response to an increase in H/L:

1. The relative wages of skilled workers,the skill premium ω = wH ⁄ wL, decreases.

2. Wages of unskilled workers increase.3. Wages of skilled workers decrease.4. The average wage (without control-

ling for education) rises.

It is also useful to highlight the impli-cations of an increase in Ah on wagelevels. First, an increase in Ah, with Al

constant, corresponds to an increase inAh ⁄ Al; the implications of this changeon the skill premium were discussedabove. Moreover if Ah increases, every-thing else being equal, we expect boththe wages of unskilled and skilled work-ers (and therefore the average wage) toincrease: in this framework, technologicalimprovements always increase all wages.

The most central result for our pur-poses is that as H/L increases, the skillpremium, ω, should fall. In terms offigure 4, the increase in supply cor-responds to a rightward shift in the ver-tical line from H/L to H′ ⁄ L′, whichwould move the economy along thedownward sloping demand curve forskills. But this tendency of the skill pre-mium to fall could be counteracted bychanges in technology, as captured by


σ − 1

σln (Ah ⁄ Al). Therefore, this simple for-

mulation encapsulates the essence ofthe two forces that Tinbergen (1975)emphasized:

The two preponderant forces at work aretechnological development, which made fora relative increase in demand and hence inthe income ratio . . . and increased accessto schooling, which made for a relativedecrease. (p. 35, italics in original)

As discussed in the empirical trendssection, the past sixty years, and par-ticularly the past thirty years, have wit-nessed a rapid increase in the supply ofskills, H/L, but no corresponding fall inthe skill premium. This implies that thedemand for skills must have increased—as a result of Tinbergen’s “technologicaldevelopment”—to prevent the relativewages of skilled workers from declining.Although in richer models there couldbe other factors leading to such a steadyincrease in the demand for skills, thecause highlighted by this simple frame-work, skill-biased technical change, is anatural candidate. More explicitly, therelative productivity of skilled workers,(Ah ⁄ Al)(σ − 1) ⁄ σ, must have increased.

The increase in (Ah ⁄ Al)(σ − 1) ⁄ σ can beinterpreted in a number of differentways. In a two-good economy, such skill-biased technical change corresponds toan increase in Ah ⁄ Al and ρ > 0 (σ > 1)—i.e., skilled workers becoming moreproductive. Skill-biased technical changecould also take the form of a decreasein Ah ⁄ Al and ρ < 0 (σ < 1). In this casethe “physical” productivity of unskilledworkers would increase, but their rela-tive wages would fall due to relativeprice effects. Alternatively, with theone-good interpretation, skill-biasedtechnical change simply corresponds toa change in the production functionthat increases (Ah ⁄ Al)(σ − 1) ⁄ σ.

Some back-of-the-envelope calcula-tions provide a sense of the rise in Ah ⁄ Al

implied by the changes in the structure

of wages and employment. Autor, Katz,and Krueger (1998) report employmentand wage-bill shares for differentgroups of workers in their appendix ta-ble A1. If we assume a specific value forσ, we can translate these numbers intochanges in Ah ⁄ Al. In particular, noticethat the relative wage bill of skilledworkers is given by

SH =wHHwLL

=

Ah

Al

(σ − 1) ⁄ σ

HL

(σ − 1) ⁄ σ

. (8)

Hence, we haveAh

Al=

SHσ ⁄ (σ − 1)

H ⁄ L. (9)

In table 1, I calculate the impliedAh ⁄ Al values for σ = 1.4 and for σ = 2using workers with some college, col-lege graduates, and college equivalentsdefinitions of Autor, Katz, and Krueger(1998)—see their paper for a more de-tailed analysis that controls for potentialcomposition effects. In all cases, thereis a large implied increase in Ah ⁄ Al and(Ah ⁄ Al)(σ − 1) ⁄ σ. For example, the num-bers indicate that, assuming an elastic-ity of substitution of 1.4, the relativeproductivity of college graduates, Ah ⁄ Al,which was approximately 0.030 in 1960,increased to 0.069 in 1970, and to 0.157in 1980. Between 1980 and 1990, it in-creased by a factor of almost three toreach 0.470. As equation (5) shows,changes in the demand indexD = (Ah ⁄ Al)

σ − 1

σ may be more informativethan changes in Ah ⁄ Al, so table 1 alsogives the evolution of D.

The view that the post-war period ischaracterized by skill-biased technicalchange also receives support from thewithin-industry changes in employmentpatterns. With constant technology, anincrease in the relative price of a factorshould depress its usage in all sectors.Since the college premium increasedafter 1979, with constant technology,there should be fewer college graduates


employed in all sectors—and the sec-toral composition should adjust in orderto clear the market. The evidence is verymuch the opposite. Berman, Bound,and Griliches (1994) and Murphy andWelch (1993) show a steady increase inthe share of college labor in all sectors.

This discussion leads to my first con-clusion, which I highlight for futurereference.

Conclusion 1. The past sixty years musthave been characterized by skill-biasedtechnical change.

Furthermore, Goldin and Katz (1998)provide evidence of technology-skill

complementarity during the 1910s and1920s. In light of this evidence, onemight consider the bulk of the twenti-eth century to have been characterizedby skill-biased technical change, thoughwhether technical change during theearly twentieth century was or was notskill-biased is not central for the focusof this paper.

4. Steady-Demand and AccelerationHypotheses

The previous section highlighted theimportance of skill-biased technicalchange over the past several decades. Afirst hypothesis is that skill-biased

TABLE 1EMPLOYMENT SHARES AND SKILL-BIASED TECHNICAL CHANGE, 1940–1990

Employment Share Wage Bill Share

Somecollege

Collegegraduate

Collegeequivalent

Somecollege

Collegegraduate

Collegeequivalent

1940 6.4 6.1 9.3 8.9 12.3 16.71950 9.5 7.7 12.4 11.0 11.9 17.41960 12.5 10.1 16.4 14.1 16.4 23.41970 16.4 13.4 21.5 16.5 21.5 29.71980 23.6 19.2 31.0 22.4 28.1 39.31990 30.8 24.0 39.3 28.5 36.7 51.0

σ = 1.4 σ = 2

Somecollege

Collegegraduate

Collegeequivalent

Somecollege

Collegegraduate

Collegeequivalent

Ah

AlD Ah

AlD Ah

AlD Ah

AlD Ah

AlD Ah

AlD

1940 0.004 0.21 0.016 0.31 0.035 0.38 0.140 0.37 0.303 .055 0.392 0.631950 0.006 0.24 0.011 0.28 0.030 0.37 0.146 0.38 0.219 0.47 0.313 0.561960 0.013 0.29 0.030 0.37 0.080 0.48 0.189 0.43 0.343 0.59 0.476 0.691970 0.017 0.32 0.069 0.47 0.179 0.61 0.199 0.45 0.485 0.70 0.652 0.811980 0.042 0.40 0.157 0.59 0.486 0.81 0.270 0.52 0.643 0.80 0.933 0.971990 0.090 0.50 0.470 0.81 1.777 1.18 0.357 0.60 1.064 1.03 1.673 1.29

Note: The first panel gives the ratio of the employment of skilled relative to unskilled workers, and the wage bill ofskilled to unskilled workers for the corresponding skill categories. These data are taken from Autor, Katz, andKrueger (1998). Some college refers to those with more than high school (hence the measure is those with morethan high school divided by those with high school or less). College graduate refers to all those with a collegedegree, and college equivalent is defined, as in Autor et al., as those with a college degree + 0.5 × those with somecollege (correspondingly, the unskilled are defined as those with high school and less + 0.5 × those with somecollege). The bottom panel gives the implied technology shifts using equation (9) above for different values of theelasticity of substitution. The demand index D is defined as (Ah ⁄ Al)

σ − 1σ .


technical change takes place steadily—at a constant pace—over time. Alfred Mar-shall begins The Principles of Econom-ics by reiterating the eighteenth-centurytaxonomist Carl Linneaus’ motto that“Nature does not make leaps.” It is alsonatural in this context to begin with ahypothesis in which skill-biased techni-cal change does not make jumps, butprogresses steadily. The alternativewould be a process that is at times moreskill-biased than others—or even unskill-biased during some episodes. In thissection, I contrast the steady-demandhypothesis, which maintains that skill-biased technical change has progressedat a constant pace over the post-war pe-riod, against the acceleration hypothe-sis, which sees a break with past trendsin recent decades.

4.1 Steady-Demand Hypothesis

According to this hypothesis, therehas been no major change in the struc-ture of demand for skills. Versions ofthis story have been suggested by Free-man (1976), and it has been proposedas an explanation for the changes in thewage structure during the 1970s andthe 1980s by Katz and Murphy (1992),and more recently by Murphy, Riddell,and Romer (1998) and Card andLemieux (2000).

In a simple form, this hypothesis canbe captured by writing

ln

Ah(t)Al(t)

= γ0 + γ1t, (10)

where t is calendar time. Substitutingthis equation into (5), we obtain

ln ω =σ − 1

σγ0 +

σ − 1σ

γ1t −1σ

ln

HL

. (11)

It is useful to link this equation to thetwo forces discussed above, and empha-sized by Tinbergen (1975). According

to equation (11), “technological devel-opments” take place at a constant rate,but the supply of skilled workers couldgrow at different rates. Therefore,changes in the returns to skills arecaused by uneven growth in the sup-ply of skills. When H/L grows fasterthan the rate of skill-biased technicalchange, (σ − 1)γ1, the skill premium willfall. And when the supply growth fallsshort of this rate, the skill premium willincrease. The story has obvious appealbecause the 1970s, when returns toschooling fell sharply, were a period offaster than usual increase in the supplyof college graduate workers as figure 1and table 1 show. In contrast, the 1980swere a period of slow increase in thesupply of skills relative to the 1970s.Katz and Murphy (1992) estimate a ver-sion of equation (11) above using ag-gregate data between 1963–87.15 Theyfind

ln ω = 0.033 ⋅ t(0.01)

−0.71 ⋅ ln

HL

(0.15)

This approach does fairly well in captur-ing the broad features of the changes inthe college premium between 1963 and1987. The predicted values from theabove equation are quite close to the ob-served movements in the college pre-mium, suggesting that the U.S. labormarket since 1963 can be characterizedby an elasticity of substitution betweencollege graduate workers and noncollegeworkers of about σ = 1 ⁄ 0.71 ≈ 1.4, and anannual increase in the demand for skillsat the rate of about 3.3 percent (orγ1 ≈ σ

σ − 1⋅ 0.033 ≈ 0.115). The increase

in the college premium during the 1980sis then explained by the slowdown in the

15 They use the relative supply of collegeequivalent workers. This is defined as collegegraduates + 0.29 × some college – 0.05 × highschool dropouts divided by high school gradu-ates + 0.69 × some college + 0.93 × high schooldropouts.


growth rate of the supply of collegegraduate workers.

Nevertheless, there are a number ofreasons for preferring a cautious inter-pretation of this regression evidence.The regression uses only 25 aggregateobservations, and there is significant se-rial correlation in the college premium(as also noted by Katz and Murphy). Ifthe true data were generated by an ac-celeration in skill bias and a largervalue of the elasticity of substitution,this regression could estimate a smallerelasticity of substitution and no accel-eration in the demand for skills (see be-low on this). For example, Katz andMurphy show that if the true elasticityof substitution is σ = 4, a significant ac-celeration in the skill bias of technicalchange is required to explain the data.Moreover, from the wage-bill sharedata reported above, Autor, Katz, andKrueger (1998) conclude that even forthe range of the values for the elasticityof substitution between σ = 1 and 2,skill-biased technical change is likely tohave been more rapid during the 1980sthan the 1970s. This can also be seen intable 1 above, where, for most mea-sures, the increase in (Ah ⁄ Al)

σ − 1

σ appearsmuch larger between 1980 and 1990than in other decades. We thereforeneed to discuss more detailed evidenceon this issue.

4.2 Evidence on Steady-Demand versusAcceleration

The first piece of evidence often putforth in support of an acceleration re-lates to the role of computers in thelabor market. Krueger (1993) arguedthat computers changed the structure ofwages, and showed that workers usingcomputers are paid more, and this com-puter wage premium has increased overtime. Although this pattern is striking,it is not particularly informative aboutthe presence or acceleration of skill-

biased technical change. It is hard toknow whether the computer wage pre-mium is for computer skills, or whetherit is even related to the widespreaduse of computers in the labor market.For example, DiNardo and Jorn-SteffenPischke (1997), Horst Enthorf andFrancis Kramartz (1998) and Lex Bor-ghans and Bas Ter Weel (2000) showthat the computer wage premium islikely to be a premium for unobservedskills.16

The second set of evidence comesfrom the cross-industry studies of,among others, Berman, Bound, andGriliches (1994), Autor, Katz andKrueger (1998), and Stephen Machinand John Van Reenan (1998). These pa-pers document that almost all industriesbegan employing more educated work-ers during the 1970s and the 1980s.They also show that more computerizedindustries experienced more rapid skillupgrading, i.e., they increased their de-mand for college-educated workers morerapidly. For example, Autor, Katz, andKrueger run regressions of changes inthe college wage-bill share in three-digit industries on computer usebetween 1984 and 1993. They find:

∆Sc80 − 90 = .287(.108)

+ .147 ⋅ ∆cu84 − 93(.046)

∆Sc90 − 96 = −.171(.196)

+ .289 ⋅ ∆cu84 − 93(.081)

where ∆Sc denotes the annual change inthe wage-bill share of college graduatesin that industry (between the indicateddates), and ∆cu84 − 93 is the increase inthe fraction of workers using computersin that industry between 1984 and 1993.These regressions are informative since

16 Equally, however, it would be wrong to inter-pret the findings of DiNardo and Pischke (1997)and Enthorf and Kramartz (1998) as evidenceagainst an acceleration in skill-biased technicalchange, since, as argued below, such technicalchange would increase the market prices for avariety of skills, including unobserved skills.


the college wage-bill share is related tothe demand for skills as shown by equa-tion (9). The results indicate that in anindustry where computer use increasedby 10 percent, the college wage-billshare grew by about 0.15 percent a yearfaster between 1980 and 1990, and 0.3percent faster a year between 1990 and1996.

Although this evidence is suggestive,it does not establish that there has beena change in the trend growth of skill-biased technology. As pointed out inconclusion 1 above, the only way tomake sense of the post-war trends is toincorporate skill-biased technical changeover the whole period. Moreover,Goldin, and Katz (1998) present evi-dence suggesting that capital-skill com-plementarity may have been as highduring the 1910s as during the recentperiod because of increased demand forskills coming from the introduction ofelectricity in most manufacturing pro-cesses. Similarly, even though therewere few computers in workplacesbefore the 1970s, other technologicaldevelopments may have increased thedemand for skills as rapidly as—or morerapidly than—computers. Therefore, thequestion is whether computers and theassociated information technology ad-vances increased the demand for skillsmore than other technologies did dur-ing the 1950s and the 1960s, or evenearlier. This question cannot be an-swered by documenting that computer-ized industries demand more skilledworkers.17

Cross-industry studies also may notreveal the true impact of computers on

the demand for skills, since industriesthat are highly computerized may de-mand more skilled workers for otherreasons as well.18 In fact, when Autor,Katz, and Krueger (1998) run the aboveregressions for the 1960–70 collegewage-bill shares, they obtain

∆Sc60 − 70 = .085(.058)

+ .071 ⋅ ∆cu84 − 93.(.025)

Therefore, industries investing more incomputers during the 1980s were al-ready experiencing more skill upgradingduring the 1960s, before the spread ofcomputers (though perhaps slower thanafter the 1980s, since the coefficienthere is about half of that between 1980and 1990). This suggests that at leastpart of the increase in the demand forskills coming from highly computerizedindustries may not be the direct effect ofcomputers, but reflect a secular long-runshift toward more skilled workers. Sofaster skill upgrading by highly comput-erized industries is not inconsistent withthe steady-demand hypothesis.

The third, and probably most power-ful, piece of evidence in favor of an ac-celeration in skill bias also comes fromAutor, Katz, and Krueger (1998). Theydocument that the supply of skills grewfaster between 1970 and 1995 than be-tween 1940 and 1970—by 3.06 percenta year during the latter period com-pared to 2.36 percent a year during theearlier thirty years. In contrast, the col-lege premium increased between 1970and 1995 by about 0.39 percent a year,while it fell by about 0.11 percent ayear during the earlier period. If de-mand for skills had increased at a steadypace, the college premium should have

17 This argument is related to a point first raisedby Lawrence Mishel and Jared Bernstein (1994).They pointed out that much of the evidence pre-sented in favor of the impact of technology on in-equality shows that there has been skill-biasedtechnical change during the 1980s, not that skill-biased technical change has accelerated relative toearlier periods.

18 Mark Doms, Timothy Dunne, and KennethTroske (1997) show that new technologies (butnot computers) are adopted by plants that havemore skilled and more highly paid workers, andthese plants do not increase their wages or de-mand for skills after the implementation of thesetechnologies.


also fallen since 1970.19 Moreover,Autor, Katz, and Krueger (1998) docu-ment greater within-industry skill up-grading in the 1970s, 1980s, and 1990sthan in the 1960s, which is also con-sistent with more rapid skill-biasedtechnical change during these laterdecades.

A simple regression analysis also con-firms this point. I combined the datafrom the March CPSs and decennialcensuses used in figure 1 above. Usingthese data, a regression similar to thatof Katz and Murphy for the period1939–96 yields similar results:

ln ω = 0.025 ⋅ t(0.01)

− 0.56 ⋅ ln

HL

,(0.20)

with an R2 of 0.63 and an implied elastic-ity of substitution of 1.8, which is some-what larger than the estimate of Katzand Murphy. However, adding higherorder terms in time (i.e., time squared,time cubed, etc.) improves the fit of themodel considerably, and these higher-order terms are significant. In figure 5, Iplot the implied time trends from regres-sions with higher-order terms as well as thelinear trend (all numbers were rescaledto fit in one graph). All three of these

more flexible time trends show an accel-eration in the relative demand for skillsduring the 1970s or the 1980s (the cubicand quartic time trends are almost iden-tical, hence practically indistinguishablein the figure).

A fourth piece of evidence comesfrom Greenwood and Yorukoglu (1997)and Per Krusell et al. (2000). Theseauthors argue, based on the work ofGriliches (1969), that equipment capitalis more complementary to skilled work-ers than unskilled workers. This prem-ise may be reasonable since advances inequipment often appear to substitutemachines for tasks previously per-formed by unskilled workers. Followingthe work by Robert Gordon (1990) andGreenwood, Zvi Hercowitz, and Krusell(1997), these papers document that thepost-war period has witnessed a seculardecline in the relative price of equip-ment capital, and argue that the associ-ated increase in the stock of equipmentcapital led to skill-biased technicalchange. Moreover, they argue that thisrelative decline accelerated in the early1970s, and the associated accelerationin the stock of equipment capitalincreased the demand for skills.

Krusell et al. (2000) formalize theirapproach by assuming the followingproduction function:

Y = Ksα[b1Lµ + (1 − b1)(b2Ke

λ

+ (1 − b2)Hλ)µ ⁄ λ](1 − α) ⁄ µ

where Ks is structures capital (such asbuildings), and Ke is equipment capital(such as machines). The parameterσ1 = 1 ⁄ (1 − λ) is the elasticity of substitu-tion between equipment and skilledworkers, and σ2 = 1 ⁄ (1 − µ) is the elastic-ity of substitution between unskilledworkers and the equipment-skilledworker aggregate. If σ1 > σ2 (i.e., µ > λ),equipment capital is more complemen-tary to skilled workers than unskilledworkers, and an increase in Ke will

19 The college premium fell between 1940 and1970 in part because it is estimated to be very highin the 1940 census. There may be reasons to besuspicious of data quality from this census, be-cause (i) the education variable was different, (ii)there may have been an overstatement of years ofschooling, possibly by as much as a factor of 1.5 or2 for some cohorts, and (iii) there was no self-employment income in this census. But it is notclear whether any of these measurement problemswill cause an upward bias in the college premium.In any case, the level of the college premium fromthis census is not out of line with other historicalevidence (see, e.g., Goldin and Katz 2000; andPiketty and Saez 2001). Moreover Autor, Katz, andKrueger show that even ignoring data from the1940 census, there is evidence for an accelerationin the skill bias of technical change. For example,for the range of the values for the elasticity of sub-stitution between σ = 1 and 2, skill-biased techni-cal change appears more rapid during the 1980sthan in the 1970s and the 1960s.


increase the wages of skilled workersmore than the wages of unskilled work-ers. More formally, the skill premium inthis model is

ω =wH

wL

=(1 − b2)(1 − b1)Hλ − 1(b2Ke

λ + (1 − b2)Hλ)(µ − λ) ⁄ λ

b1Lµ − 1.

(12)

Differentiation of (12) shows that aslong as µ > λ, which corresponds toequipment capital being more comple-mentary to skilled workers than un-skilled workers, we have ∂ω ⁄ ∂Ke > 0. Soprovided that equipment capital is morecomplementary to skilled workers thanunskilled workers, an increase in thequantity of equipment capital will in-crease the demand for skills. Since thepost-war period has been characterizedby a decline in the relative price of

equipment goods, there will be an asso-ciated increase in the quantity of equip-ment capital, Ke, increasing the demandfor skills steadily.

Figure 6, which plots the log of thisrelative price series, shows the fasterproportional decline after the 1970s.The behavior of the relative price seriesthen suggests that there may have beenan acceleration in the substitution ofequipment capital for labor, causingmore rapid skill-biased technical change.

Nevertheless, because there are seri-ous difficulties in adjusting capitalprices for quality, we may want to becautious in interpreting this evidence.Another problem comes from the factthat, as I discuss in more detail below, avariety of other evidence does not sup-port the notion of faster technologicalprogress since 1974, which is a basic

96

Alternative Time Trends for the Relative Demand for Skillsyear

Figure 5. Estimates of Time Trends from Regressions of ln ω on ln (H/L), year, year2, year3 and year4 between 1939 and 1996 (with observations in 1939, 1949, 1959 from the decennial censuses and observations for 1963–96 from the March CPSs)

897969594939

linear time trendcubic time trend

quadratic time trendquartic time trend


tenet of this approach. Finally, onewould presume that if, in fact, the de-cline in the relative price of equipmentcapital is related to the increase in thedemand for skills, then in a regressionof equation (11), it should proxy for thedemand for skills and perform betterthan a linear time trend. Table 2 re-ports a series of regressions which showthat, on the contrary, the level or thelog of the relative price of equipmentcapital is not significant in such regres-sions. Column 1 shows the equivalent ofthe regression by Katz and Murphy(1992) with only a time trend and therelative supply of skills. Columns 2 and3 show regressions that replace the timetrend with the level and log of the rela-tive price of equipment capital. Theseterms are significant, but the fit of theregression is worse than the one in

column 1. The remainder of the tableshows that once these terms are enteredsimultaneously with the time trend, thetime trend is significant, while there isno evidence that the relative price ofequipment capital matters for the de-mand for skills. While this evidencemay simply reflect the fact that the rela-tive price of equipment is measuredwith error, it casts some doubt on theview that the relative price of equip-ment capital is directly linked to thedemand for skills and that its fasterdecline since the 1970s indicates anacceleration in skill bias.

A final piece of evidence for accelera-tion comes from the behavior of overalland residual inequality over the pastseveral decades. The sharp rise in bothoverall and residual wage inequalitysince the early 1970s, documented in

–.1

1992

Log

rel

ativ

e pr

ice

of e

quip

men

t cap

ital

year

Figure 6. Behavior of the Log Relative Price of Equipment Capital, 1963–92

198819831978197319681963

Log rel. price of equip. capitalPost-1974 trend

–.4

–.7

–1

–1.3

Pre-1974 trend


section 2, weighs in favor of a markedchange in labor market prices and de-mand for skills during recent decades.This argument is based on the view thatchanges in residual inequality reflectchanges in labor market prices, a thesisput forth by Chinhui Juhn, Murphy,and Brooks Pierce (1993).20 This view isimportant for an interpretation for therecent changes in wage structure fortwo reasons. First, as the evidence insection 2 indicates, much of the recentincrease in overall inequality is due tothe rise in residual inequality. If resid-ual inequality were unrelated to the de-mand and supply of skills, the frame-work here could only account for arelatively small fraction of the increasein overall inequality. The Juhn-Murphy-

Pierce view, instead, suggests that theincrease in residual inequality and thebulk of the rise in overall inequalityare related to changes in labor marketprices, and therefore to the supply anddemand for skills. Second, according tothis view, the rise in residual inequalityis an important piece of evidence in thedebate on acceleration: residual in-equality, which was stable during the1960s, began to increase rapidly duringthe early 1970s (figures 2 and 3), indi-cating a discontinuity in labor marketprices and, most likely, in the rate ofincrease of the demand for skills.

A concrete example is useful for clari-fying why residual inequality is linkedto labor market prices. Suppose thattwo otherwise identical individuals dif-fer in terms of their unobserved skills(for example, in terms of interpersonalskills, motivation, specific skills for theirjob, or IQ).21 Denote the unobserved

TABLE 2THE EFFECT OF THE RELATIVE PRICE OF EQUIPMENT ON SKILLED PREMIA

Dependent variable is log college premium

(1) (2) (3) (4) (5)

Relative supply –0.742(0.053)

–0.388(0.037)

–0.610(0.068)

–0.691(0.100)

–0.740(0.054)

Time 0.026(0.002)

0.022(0.007)

0.024(0.005)

Log relative price –0.323(0.024)

–0.051(0.084)

Relative price –0.875(0.086)

–0.056(0.167)

Adjusted R2 0.900 0.864 0.795 0.898 0.897

Note: This table reports the regression of the log college premium on a linear time trend,the log relative supply of skilled workers and various measures of the relative price ofequipment capital. For comparability, all data taken from Krusell et al. (2000).

20 Of course, an alternative—and more cynical—view would be to interpret residual inequality as“a measure of our ignorance” (as Moses Abramo-vitz 1957 did for TFP). When a standard wageregression such as (1) provides a good fit, theresiduals will be less disperse. Nevertheless, giventhe variety of skills that we are unable to measurein standard data sets, much of the residual willplausibly reflect rewards to some unobservedskills.

21 By unobserved skills, I mean skills that arenot observed by the econometrician. These skillscould be—and are likely to be—observed by em-ployers. These types of skills are often referred to


skill of individual 1 by a1 and that ofindividual 2 by a2 > a1, and assume thatwages are given by

ln wit = 2θtai + γthi, (13)where γt is the price of h skills at time t,while θt is the price of a skills. Sincethese individuals are identical in all re-spects other than their unobserved skills,a, we have the variance of log wages(or of residual wages) among these twoindividuals as:

Var(ln w) = θt2(a2 − a1)2.

Now if at a later date, t′, this varianceincreases to Var(ln w)′, and we know thatthese two individuals are still identicalin all other respects and that a2 − a1 hasnot changed, we can interpret the in-crease in Var(ln w) as reflecting an in-crease in the price of unobserved skills,θt. The discussion in the appendixshows that the bulk of the increase inoverall and residual inequality cannotbe explained by composition effects,i.e., by changes in a2 − a1, so this in-crease is most likely due to a rise in theprice of and demand for unobservedskills during the 1970s.22 Therefore,with this interpretation, the rapid risein the residual inequality during the1970s and the 1980s indicates a morerapid increase in demand for skillsduring these decades than earlier.

Overall, there is a variety of evidencesuggesting an acceleration in skill biasover the past 25–30 years. Although notall evidence is equally convincing, therise in the returns to schooling over thepast thirty years, despite the very rapidincrease in the supply of skills, and the

behavior of overall and residual in-equality since the 1970s suggest amarked shift in the demand for skillsover the past several decades. I thereforetentatively conclude:23

Conclusion 2. The behavior of returnsto schooling and residual inequalityover the past three decades suggests anacceleration in the demand for skillsbeginning in the 1970s or 1980s.

5. Acceleration in Skill Bias

What explains the more rapid in-crease in the demand for skills overthe past several decades? The first pos-sibility is a change in labor market insti-tutions. The second is the role of in-creased international trade. The third ismore rapid skill-biased technical change.I argue in section 6 that changes inlabor market institutions and the in-creased importance of internationaltrade cannot explain the change in labormarket prices by themselves. Moreover,the evidence discussed in section 4 isconsistent with new technologies play-ing an important role in changing thewage structure. So here I begin withchanges in technologies, and in particu-lar, I discuss “pure technological” ap-proaches where technology is the onlyfactor determining the demand forskills.

5.1 “Technological Revolutions” andAcceleration in Skill Bias

The first group of technological theo-ries links the acceleration in skill bias toexogenous technological developments,and argues that a “technological revolu-tion” led to more rapid skill-biasedtechnical change beginning in the 1970s

as “unobserved ability.” This does not imply thatthese unobserved skills are necessarily synony-mous with IQ or other single dimensional skillindices.

22 Put differently, the absence of large composi-tion effects indicates that the rise in residual in-equality likely reflects changes in the demandfor unobserved skills rather than changes in the“supply of unobserved skills.”

23 The most important caveat is that the avail-able data suggest relatively slow skill-biased tech-nical change during the second half of the 1990s(see Mishel, Bernstein, and John Schmitt 1998;and Murphy and Welch 2000).


or the 1980s. In terms of the model de-veloped above, this corresponds to amore rapid increase in Ah ⁄ Al during thisperiod, translating into greater skill pre-mia. Many of the proponents of thisview argue that the acceleration in skillbias is, at least in part, related to infor-mation technology and computers (forexample, Krueger 1993; Berman, Bound,and Griliches 1994; Autor, Katz, andKrueger 1998; Berman, Bound, andMachin 1998).

An interesting version of this story isthe one developed by Krusell et al.(2000) discussed above. They argue thatthe demand for skills accelerated as aresult of the more rapid decline inthe relative price of capital equipmentbeginning in the early 1970s.24 TheKrusell et al. theory is attractive since itprovides a unified framework in whichwe can identify both the cause of thesteady increase in the demand for skills,and the source of the more rapid skill-biased technical change, though the evi-dence provided in the previous sectioncasts some doubt on the link betweenthe relative price of equipment and thedemand for skills.

The main idea of all these theories of

acceleration is that new technologiesare more complementary to skilledworkers than to unskilled workers—forexample, there are more rapid advancesin the technologies used by skilledworkers, as captured by Ah above. Rapidtechnological progress then corre-sponds to an acceleration in skill bias.An alternative perspective, building onan idea originally suggested by Nelsonand Phelps (1966), also focuses on theeffect of rapid technical change on in-equality, but puts the emphasis on theability of skilled workers to deal withthe introduction of new technologies.According to this view, demand forskills will automatically increase duringperiods of rapid technological change.Welch (1970) gave an early succinctsummary of these two views. The firstview, which we may call the accelera-tion hypothesis, sees human capital as afactor of production, and the more rapidincrease in the demand for skills resultsfrom an acceleration in technical changethat favors this factor of production. Inother words:

[T]echnical change may not be neutralbetween skill classes. It may be that incre-ments in technology result in increments inthe relative productivity of labor that arepositively related to skill level. (Welch 1970,p. 38)

In contrast, the second view—theNelson-Phelps hypothesis—argues that25

24 I classify this approach as one of exogenoustechnology, since the driving force, the decline inthe relative price of equipment capital, is assumedexogenous.

An alternative interpretation of the approach byKrusell et al. (2000) is that the main determinantof the demand for skills is not technology-skillcomplementarity, but capital-skill complementarity.I believe that the distinction between technology-skill versus capital-skill is not very useful. Capital-skill complementarity could play an important roleonly in a model as in Greenwood et al. (1997) orKrusell et al. (2000), where new capital embedssuperior technologies. In this sense, it is a combi-nation of new capital and new technologies that isincreasing the demand for skills. Moreover, Autor,Katz, and Krueger (1998) show that demand formore educated workers across industries is af-fected by high-tech capital (e.g., computers), butnot by equipment capital, suggesting further thatit is new technologies, not simply capital intensity,that matters for inequality.

25 A recent paper by Samuel Bowles, HerbertGintis, and Melissa Osborne (2001) has a similarclassification of different approaches. They callthe first view Walrasian, and the second Schumpe-terian. In the first, skills that enable workers toproduce more are valued, while in the second, it isskills that enable workers to deal with changes ineconomic and social environments. They also pro-pose an alternative, Coasian, view, where “skills”are workers’ capacity to work within organizationsand follow authority. From the point of view ofthe analysis here, this view is not fundamentallydifferent from the Walrasian view, since workersare paid more for skills that increase their marginalcontribution to the profits of their employers.


[T]he productivity of education would bepositively related to the rate of change in use-ful technology (the ability to change) and tothe size of the technological gap (room forinnovation). In this case, if the rate of utiliza-tion of technology is accelerating, or if thetechnology gap is growing, the return toeducation will rise relative to other inputs.(p. 38)

Studies building on the Nelson-Phelps hypothesis include Galor andTsiddon (1997), Greenwood and Yoru-koglu (1997), Caselli (1999), Aghionand Howitt (1998, ch. 9), Galor andMoav (2000), Violante (2000), Rubin-stein and Tsiddon (1999), Aghion,Howitt and Violante (2000), and Gould,Moav and Weinberg (2000).26 These pa-pers argue that there has been a tech-nological revolution in the U.S. econ-omy starting in the 1970s and relate therise in inequality to the increaseddemand for skills resulting from thetechnological revolution. For example,Greenwood and Yorukoglu (1997) drawa parallel between the First and theSecond Industrial Revolutions and whathas been happening in the U.S. econ-omy since 1974. Caselli (1999) developsa similar theory where a technologicalrevolution increases the demand forworkers who can switch to the sectorsthat benefit from the introduction ofnew technologies.27

To get a basic understanding of these

approaches, it is useful to consider asimple model based on Galor and Moav(2000). Suppose that

Al = φl(g)a and Ah = φha (14)

where a is a measure of aggregate tech-nology, and g is the growth rate of a, i.e.,g ≡ a⋅ ⁄ a. The presumption that skilledworkers are better equipped to deal withtechnological progress can be capturedby assuming that φ′l < 0. Galor and Moav(2000) refer to this assumption as the“erosion effect,” since it implies thattechnical change erodes some of the es-tablished expertise of unskilled workers,and causes them to benefit less fromtechnological advances than skilled work-ers do. Substituting from (14) into (4),the skill premium is

ω =wH

wL=

Ah

Al

(σ − 1) ⁄ σ

HL

−1 ⁄ σ

=

φh

φl(g)

(σ − 1) ⁄ σ

HL

−1 ⁄ σ

.

(15)

Therefore, as long as φ′l < 0, more rapidtechnological progress, as captured by ahigher level of g, will increase the skillpremium.

Theories that explain the increase ininequality as a result of rapid tech-nological progress have a number ofattractive features.28 First, many econo-mists and commentators view the ad-vances in computer and informationtechnology as a break with the tech-nologies of the past, and so are open to

26 See Aghion (2001) for an approach that com-bines the Nelson-Phelps insights with the Schum-peterian notion of creative-destruction to discussthe impact of the diffusion of computers oninequality.

27 The explanation offered by Rubinstein andTsiddon (1999), and Aghion, Howitt, and Violante(2000) is somewhat different. They argue thatthere is increased uncertainty at times of rapidtechnological change, and more skilled workersare better able to cope with uncertainty. This ideais also related to a thesis first put forth by MichaelPiore and Charles Sabel (1984) that the oil priceshocks increased the uncertainty faced by produc-ers, and induced them to change the organizationof production toward organizational forms thatrequire adaptable workers.

28 These theories also predict that inequalityshould increase when new technologies are beingintroduced, but should decline when these newtechnologies are standardized and being used rou-tinely by many firms (see, for example, Galor andTsiddon 1997, or Aghion, Howitt, and Violante2000). So far, there seems to be no evidence of adecline in inequality in the United States, but per-haps the years to come will see a return to thelevels of inequality experienced during the 1960s,vindicating this approach.


the idea that we might be in the midstof a technological revolution. Second, avariety of evidence supports the notionthat skilled workers have a comparativeadvantage in coping with rapid tech-nical change. Ann Bartel and FrankLichtenberg (1987) show that firms in-troducing new technologies hire moreskilled workers. Bartel and NachumSicherman (1998) document that re-turns to unobserved ability appear to behigher in industries with more rapidtechnical change. Andrew Foster andMark Rosenzweig (1996) provide evi-dence from developing countries thatmore educated workers are betterplaced to take advantage of advances inagricultural technology.

The main difficulty with both thetheories based on the acceleration andthe Nelson-Phelps hypotheses is thatthey rely explicitly on rapid technicalchange in recent decades.29 There islittle direct evidence that the decadesbetween 1970 and 1995 have been a pe-riod of rapid technical change, how-ever. First, this period has experiencedsluggish TFP and output growth rela-tive to earlier periods. Greenwood andYorukoglu (1997) and Hornstein andKrusell (1996) argue that the slow TFPgrowth itself may be an outcome of themore rapid technical change. Accordingto this argument, new revolutionarytechnologies first reduce productivitygrowth as firms and workers spend timelearning to use these technologies.

It is difficult to imagine how a newand radically more profitable technol-ogy will first lead to 25 years of substan-tially slower growth. Although, in aninfluential paper, Paul David (1990)argues that the spread of electricity toAmerican manufacturing was also slowand productivity gains from electrifica-tion were limited until the 1920s, theparallel with the recent productivityslowdown should not be overstated.First, though productivity growth fromelectrification was sluggish during theearly 1900s, the U.S. economy overallhad a much higher level of outputgrowth than the growth levels experi-enced over the past three decades. Datafrom table A-XVIII of John Kendrick(1961) imply that output growth be-tween 1899 and 1909 in the U.S. econ-omy was 4.2 percent a year, while be-tween 1909 and 1919, it was 3 percent,and between 1920 and 1929, outputgrew by 3.6 percent a year. Second, asnoted by Stephen Oliner and DanielSichel (1994), computers and other ad-vanced office equipment have only beena trivial part of the aggregate capitalstock of the U.S. economy until the mid-1990s. It is therefore unlikely that thewhole of the U.S. economy has beenadapting to the changes in this rela-tively small part of the capital stock. Fi-nally, as shown by Ernie Brendt, Cath-erine Morrison, and Larry Rosenblum(1994), more computerized sectors didnot perform any better in terms of laborproductivity growth over this period;this pattern is also difficult to recon-cile with a computer-led technologicalrevolution.

It is also useful to note that althoughcomputers have no doubt increased ourstandards of living and quality of lifeover the past thirty years, they may bemuch less radical innovations than cer-tain previous new technologies (seeGordon 1998). To gain perspective,

29 Another argument that applies only againstapproaches based on the Nelson-Phelps hypothe-sis is that historical evidence is not necessarily inline with a view that demand for skills always in-creases during times of rapid technical change. Asdiscussed in the introduction, the major techno-logical changes of the early nineteenth century ap-pear to have been largely skill-replacing (unskill-biased), even though they seem as radical ascomputer technology. This suggests that it isthe skill bias of technology, not merely its rapidarrival, that is important for the demand for skills.


consider the difference that the tele-graph made to a world in which thefastest medium of communication waspigeons. Mokyr (1990, p. 124) describesthis as follows:

The telegraph had an enormous impact on19th-century society—possibly as great asthat of the railroad. Its community and politi-cal value was vast, as was its effect in coordi-nating international financial and commoditymarkets. Unlike the railroad, it had no closesubstitutes, the closest being homing pigeonsand semaphore.

Or consider the difference that the auto-mobile and air conditioning made to thequality of life, and electricity and in-terchangeable parts made to the manu-facturing sector. As also pointed out byGordon (1998), compared to these improve-ments, the switch from mainframes toPCs, or from telephone to e-mail, or fromthe typewriter to the word processor maybe modest.30

A final problem for all of the ap-proaches based on exogenous techno-logical developments is the coincidencein the timing of this change, and therapid increase in the supply of skilledworkers. Recall that there was a verylarge increase in the supply of collegegraduate workers during the late 1960sand the early 1970s (figure 1 and table1 show the large increase in the em-ployment share of college workers be-tween 1970 and 1980). So the accelera-tion in skill bias is either concurrentwith, or immediately follows, this large

increase in the supply of skills. There isno a priori reason to expect the accel-eration in skill bias to coincide with therapid increase in the supply of skills.Those who want to subscribe to theexogenous technological progress viewhave to explain this as a chance event.31

5.2 Endogenous Skill-Biased TechnicalChange

The theories discussed so far pre-sume technical change to be skill-biased by nature (or, at the very least,recent technologies to have increasedthe demand for skills due to exogenousreasons). A different perspective is to linkthe types of technologies that are devel-oped and adopted to (profit) incentives,or to demand pull as emphasized bySchmookler (1966).32

Historical evidence is consistent withthe notion that profit incentives and op-portunities are important for the de-velopment and introduction of new

30 Another argument is that our ability to mea-sure TFP growth may have deteriorated followinga change in technological regime. However, asMartin Bailey and Gordon (1988) document, pro-ductivity slowdown has been concurrent in manysectors, some of them with little problems inmeasuring output or output quality. It is interest-ing to note, however, that evidence in favor of thishypothesis may yet emerge, especially since pro-ductivity growth has been quite rapid during thepast three years (but see Gordon 1998, and morerecently, Dale Jorgensen and Kevin Stiroh 2000,on this).

31 One could argue that the supply of skilledworkers increased because, during the 1960s,workers anticipated that there was going to be atechnological discontinuity in the decades tocome, and responded to this by increasing theireducation. This story appears quite unreasonable,however. There is no evidence that anyone, letalone teenagers, foresaw the technological devel-opments of the 1970s and the 1980s as early as the1960s. Moreover, the increase in the supply ofskills can be largely explained by two factors.First, the Vietnam era draft laws encouragedyoung males to stay in college longer (and indi-rectly also influenced female enrollments). Sec-ond, college enrollments were on an upward trendsince the early 1950s, and much of the increase inthe supply of college graduate workers is ac-counted for by the interaction of this upwardtrend with the very large relative size of the babyboom cohorts.

32 This is also the approach taken by the endog-enous growth theory, which determines the overallrate of technical change—but not the degree ofskill bias—from profit incentives (e.g., Romer1990; Gene Grossman and Elhanan Helpman1991; Aghion and Howitt 1992). An advantage ofthe endogenous-technology perspective in thiscontext is that it provides clear reasons for the de-gree of skill bias to increase even without a changein the overall pace of technological improvement—see below.


technologies. Fernand Braudel (1984,p. 566) took a strong position on this:

[T]he efficient application of technologylags, by definition, behind the general move-ment of the economy; it has to be called on,sometimes several times, to meet a preciseand persistent demand.

An interesting example of the timingof technological development respond-ing to profit incentives is given by theintroduction of the electric street car inU.S. cities during the late nineteenthcentury. In his history of electricity inthe United States, David Nye (1990)describes this as follows:

Cities grew larger, better transportationwas needed, so the [electric] trolley was in-vented, called into being by the crowded latenineteenth century cities . . . . By the 1870slarge cities had ceased to be accessible byfoot, or built to the scale of pedestrians, andtraffic congestion was terrible. (p. 85)

This created the profit opportunities todevelop and introduce the electric trol-ley. The technological requirements hadbeen met long before, and awaited theseprofit opportunities. Nye writes, “How-ever great the need for the electric trol-ley after 1870, it was hardly a new idea;it had been the object of experimentduring four decades” (p. 86).

Another example of the type of inno-vation responding to profit incentives isprovided by the cotton gin. In the lateeighteenth and early nineteenth centu-ries, only green seed cotton, which wasdifficult to clean, could be grown inmost of the American South, and Brit-ain imported most of its cotton from theWest Indies, Brazil, and India. A ma-chine to remove the seeds was essentialfor the success of American cotton. Incontrast to almost all other textile inno-vations that were taking place in Eng-land and Europe, such a machine, thecotton gin, was developed in the UnitedStates in 1793 by Eli Whitney in re-sponse to this need. The impact of the

cotton gin on the South was nothingshort of spectacular. In the court caseover the patent rights, Judge Johnsonwrote:

“[. . . as a result of the cotton gin] . . .individuals who were depressed with poverty,and sunk with idleness, have suddenly risen towealth and respectability. Our debts have beenpaid off, our capital increased; and our landsare treble in value.” (Quoted in ConstanceGreen 1956, p. 92)

Within a short time, Eli Whitney’s ginturned the United States from a cottonimporter into the largest cotton exporterin the world.

Schmookler (1966) provides a famousargument for the importance of demandpull in the development of many tech-nologies. He documents rapid innova-tions in railroads following increasedpurchases of railroad equipments, andmore generally argues that industrieswith greater investments experiencefaster technological progress becausethe returns to such progress are greater.A natural next step is then to argue thatthe degree of skill bias in technical changeis also determined by profit opportuni-ties and by the demand for differenttypes of technologies. Here, by endoge-nous (skill bias) technology approach Imean the view that the degree of skillbias in technical change is influencedby profit incentives.33

A key determinant of profitability ismarket size. As Schmookler (1966)stated in the title of two of his chapters:“The amount of invention is governed bythe extent of the market.” The most suc-cessful businessmen have always beenaware of this. For example, Matthew

33 An alternative explanation for the increase inwage inequality based on endogenous technicalchange is offered by Huw Lloyd-Ellis (1999). Inhis model, it is the interaction between endoge-nous technical change and the slowdown in thegrowth of labor quality that leads to increased in-equality. Skill bias is not endogenous in Lloyd-Ellis’ model.


Boulton wrote to his business partner,James Watt, “It is not worth my whileto manufacture your engine for threecountries only, but I find it very wellworth my while to make it for allthe world” (quoted in Michael Scherer1984, p. 13). Schmookler (1966) simi-larly provided many examples wheremarket size was crucial in determiningthe directions of technical change. Thehorseshoe is perhaps the most interest-ing one. Schmookler documented thatthere was a very high rate of innovationthroughout the late nineteenth and earlytwentieth centuries in this very ancienttechnology, invented in the second cen-tury B.C., and no tendency for inven-tors to run out of additional improve-ments. On the contrary, inventions andpatents increased because demand forhorseshoes was high. Innovations cameto an end only when “the steam tractionengine and, later, internal combustionengine began to displace the horse . . .”(p. 93).

According to this reasoning, the de-velopment of skill-biased technologieswill be more profitable when they havea larger market size—i.e., when thereare more skilled workers. Therefore,the equilibrium degree of skill biascould be an increasing function of therelative supply of skilled workers. Anincrease in the supply of skills will thenlead to skill-biased technical change.Furthermore, an acceleration in thesupply of skills can lead to an accelera-tion in the demand for skills. It is usefulto link this approach to technologicaldevelopment with the above framework.While the above supply-demand frame-work explains the prices of skills by sup-ply and technology, the perspective ofendogenous skill bias relates technologyto the supply of skills.34 Tinbergen, in

his pioneering study of the supply-demand framework, in fact, foresaw thispossibility, and wrote (1975, p. 61): “. . .an inequality-furthering phenomenon istechnological development. But need itbe? Increasingly we get the feeling thattechnological development is not simplysomething given, but that it may beguided, within limits.”

At some level, the idea that there willbe more technologies developed, cre-ated and adopted for skilled workers—“within limits”—when there are moreskilled workers to use them is quite ap-pealing. An extreme form of this view iscaptured by my model in Acemoglu(1998), where forward-looking profit-maximizing firms create new technolo-gies anticipating the profitability ofthese different investments. Accordingto this view, it would be the VietnamWar draft laws and the high college en-rollment rates of the baby boom cohortsthat induced the development of com-puters. Such an interpretation is not lit-eral. A more plausible interpretation maybe that new technological platforms—macroinventions to use Joel Mokyr’sterm, or General Purpose Technologiesto use Bresnahan and Trajtenberg’s term—stem from advances in basic science orfrom labs with little profit-maximizingincentive. The development of the mi-crochip would be such a macroinven-tion. But what matters for most workersin the labor force is how this new tech-nological platform is developed, i.e.,the microinventions that follow themacroinvention. At the expense of over-simplifying, we can say that the micro-chip could have been used to developadvanced scanners to increase the pro-ductivity of unskilled workers, or ad-vanced computer-assisted machines tobe used by skilled workers to replaceunskilled workers. The theory of endog-enous skill bias requires that the extent ofthe advances in these two technologies

34 Naturally, it is also possible to link the supplyof skills to skill premia. See below for a discussion.


is affected by profit opportunities. Whenthere are more college graduates, com-puters become relatively more profit-able to develop than scanners, and thisexplains the acceleration in skill bias.35

The endogenous response of firms tothe increase in supply will raise the de-mand for skills. In fact, supply may notsimply create its own demand, but theresponse of firms could be so pro-nounced that demand could overshootthe supply. In this theory, therefore,the increased supply may be the causeof the increase in the skill premium(see Acemoglu 1998, and also MichaelKiley 1999). Here I outline a simplifiedversion of this theory based on the aboveframework.

Suppose that consumers have a utilityfunction defined over Y = [Yl

ρ + Yhρ]1 ⁄ ρ,

and that Yh = NhH and Yl = NlL where Nh

and Nl can be interpreted as the num-ber of specialized machines used withskilled and unskilled workers, respec-tively. This is equivalent to the abovesetup with Ah = Nh and Al = Nl. An in-crease in Nh relative to Nl will corre-spond to skill-biased technical changeas long as σ = 1 ⁄ (1 − ρ) > 1. From con-sumer maximization, the relative priceof skill-intensive goods is

p ≡ph

pl=

NhHNlL

ρ − 1

, (16)

where once again ph denotes the price ofgood Yh and pl is the price of Yl .

Suppose now that these specializedmachines are created and sold by

profit-maximizing monopolists. Creat-ing a new machine costs B units of thefinal good Y, and the marginal cost ofproducing these machines, once cre-ated, is zero. The marginal willingnessto pay for an additional machine in thetwo sectors is given by the derivativesof phYh and plYl with respect to Nh andNl, i.e.,

phH and plL. (17)I assume that the creator of each newmachine obtains this “market” marginalwillingness to pay (see Acemoglu 1998,for a more rigorous and detailed discus-sion). Equation (17) therefore highlightstwo effects that encourage the creationof new technologies.

1. The price effect: technologies produc-ing more expensive goods will be im-proved faster. Since goods using thescarce factor will command a higherprice (see equation 16), this effectimplies that there will be more inno-vation directed at the scarce factor—i.e., directed at unskilled workersduring the 1970s and the 1980s.

2. The market size effect: a larger clien-tele for a technology leads to moreinnovation. Since the clientele for atechnology is the number of workerswho use it, the market size effect en-courages innovation for the moreabundant factor, and encouragesmore technologies for skilled andhighly educated workers during the1970s and the 1980s.

The creation of new machines willstop when the marginal increase inprofits is equal to the marginal cost ofinnovation in both sectors. This impliesthat in equilibrium

phHplL

= 1, (18)

i.e., the price and market size effectshave to be balanced in equilibrium. Howcan equation (18) be satisfied? Since H/L

35 There is some evidence that the compositionof R&D shifted toward more skill-biased tech-nologies during the period of the rapid increase inthe supply of college-educated workers. From theR&D expenditure data reported by the NSF, in1960 company funded R&D for office computingwas 3 percent of the total company funded R&Dexpenditure. This ratio had increased to 13 percentby 1987, suggesting that during this period ofrapid increase in the supply of skills, there wassignificantly more R&D directed to one of thetechnologies most complementary to skills.


is fixed, equation (18) can only hold ifthe relative price of skill-intensive goods,p = ph ⁄ pl, adjusts. From equation (16), thiscan only happen if Nh ⁄ Nl changes. There-fore, in this economy, the skill bias of tech-nology has to adjust in order to “clear thetechnology market.” Combining (16) and(18), we obtain equilibrium skill bias as

Nh

Nl=

Ah

Al=

HL

ρ ⁄ (1 − ρ)

. (19)

This equation shows that when ρ > 0, i.e.,when skilled and unskilled good are grosssubstitutes, the market size effect willdominate the price effect, and a greaterrelative supply of skilled workers will leadto more skill-biased technologies—higherNh ⁄ Nl.

Finally, by substituting from equation(19), the skill premium in this economyis

ω =phNh

plNl=

HL

(2ρ − 1) ⁄ (1 − ρ)

=

HL

σ − 2

where the final expression is obtained bycombining (16) and (19).

The most important result is that ifρ > 1 ⁄ 2, i.e., if the elasticity of substitu-tion σ is greater than 2, the skill pre-mium will be an increasing function ofthe relative supply of skills.36 This is be-cause an increase in H/L encourages somuch skill-biased technical change thatthe demand for skills increases more thanenough to offset the increase in thesupply of skills. As a result, the (long-run) relative demand for skills is an up-ward-sloping curve as drawn in figure 7,and an increase in the supply of skilledworkers increases the skill premium.

There are a number of implicationsthat follow from this approach. First, asthe relative supply of skilled workershas been growing throughout the pastsixty years, we expect technology to en-dogenously respond by becoming moreskill-biased over time. If the elasticityof substitution between skilled and un-skilled workers is greater than 2, i.e.,ρ > 1 ⁄ 2, the increase in the demand forskilled workers would be more thanenough to offset the increase in thesupply of skilled workers, and the econ-omy would be moving steadily along anupward-sloping relative demand curvefor skills. This would explain why re-turns to college have been increasingover the past half century.

A new theory for the acceleration inskill bias also emerges from this simplemodel. According to this theory, therapid increase in the supply of collegeeducated workers during the 1970s cre-ated a more pronounced shift towardskill-biased technologies, further in-creased the demand for skill, and raisedthe college premium. This story be-comes more interesting once we recog-nize that the equilibrium skill bias oftechnologies, Nh ⁄ Nl, is a sluggish vari-able determined by the slow buildupand development of new technologies.In this case, a rapid increase in the sup-ply of skills first reduces the skill pre-mium as the economy moves along aconstant technology (constant Nh ⁄ Nl)curve as drawn in figure 7. After a whilethe technology starts adjusting, and theeconomy moves back to the upward-sloping relative demand curve, with avery sharp increase in the college pre-mium. This theory therefore gives aninterpretation for both the decline inthe college premium during the 1970sand its subsequent surge, and relatesboth to the large increase in the supplyof skilled workers.

For the key insights of this theory,

36 The result that the elasticity of substitutionneeds to be greater than 2 for the long-run rela-tive demand to slope upwards is a feature of thesimple model here, and does not generalize toricher environments. In any case, there are a num-ber of estimates above 2, and a somewhat upwardsloping relative demand curve for skills, even withthis model’s parameterization, is an empiricalpossibility.


that increases in the relative supply ofskills induce skill-biased technicalchange, we do not need the long-runrelative demand curve to be upward-sloping. When ρ < 1 ⁄ 2, increases in thesupply of skills still induce skill-biasedchange, but this technical change is notenough to prevent the skill premiumfrom falling. Further “exogenous” skill-biased technical change is also neces-sary to explain why returns to schoolinghave risen over the past sixty years.With a downward-sloping long-run de-mand curve, the story for the 1970s andthe 1980s is also different. The large in-crease in the supply of skills againmoves the economy along a steeplydownward-sloping constant technologydemand curve. The response of technol-ogy then moves the economy to a lesssteep long-run demand curve as drawnin figure 8, raising the skill premium.

Additional exogenous skill-biased tech-nical change is then necessary for theskill premium to increase above itsinitial level.

There are also other historical epi-sodes in which a large increase in thesupply of skills appears to have affectedthe direction of technical change. Highschool enrollment and graduation ratesdoubled in the 1910s. Goldin and Katz(1995) argue that increased enrollmentswere mostly driven by supply side fac-tors; changes in the location and curric-ula of schools and improvements intransportation technology. The skillpremium fell sharply in the 1910s. But,despite the even faster increase in thesupply of high school skills during the1920s, the skill premium levelled offand started a mild increase. Goldin andKatz (1995) conclude that the demandfor high school graduates must have

Figure 7. The Dynamics of the Relative Wage of Skilled Workers in Response to an Increase in the Supply of Skills with Endogenous Skill-Biased Technical Change

Exogenous Shift in Relative Supply

Long-run Rel. Wage

Relative Wage

Initial Rel. Wage

Long-run RelativeDemand for Skills

Short-run Response


expanded sharply starting in the 1920s,presumably due to changes in officetechnology and higher demand fromnew industries, such as electrical ma-chinery, transport and chemicals (seealso Goldin and Katz 1998).37

Another interesting case study comesfrom the response of the Israeli labormarket to the influx of large numbers ofhighly educated immigrants from theformer Soviet Union. The size of thisinflux was enormous: migration in-

creased the Israeli population by 12percent in the first half of the 1990s. Atheory with exogenous technologywould predict a large decline in therelative wages of educated workers,very much as in the case of Palestinianlabor discussed above.38 In practice, theeducation premium did not fall (e.g.,Rachel Friedberg 1997). This seems tobe mainly because most industries in-creased their employment of moreskilled workers during this large influx(Neil Gandal, Gordon Hanson, andMatthew Slaughter 2000; and Sarit

Figure 8. The Dynamics of the Relative Wage of Skilled Workers in Response to an Increase in the Supply of Skills with Limited Endogenous Skill-Biased Technical Change

Exogenous Shift in Relative Supply

Long-run Rel. Wage

Relative Wage

Initial Rel. Wage

Long-run RelativeDemand for Skills

Short-run Response

37 As Goldin and Katz (2000) show using datafrom Iowa Prairies, returns to education weremost likely higher in 1915 than in 1950. Althoughthis evidence suggests that the long-run relativedemand curve for skills was downward slopingover this period, it is consistent with the notion ofskill-biased technical change induced by an in-crease in the supply skilled workers. Specifically,during this period, demand for skills expandedvery rapidly to accommodate the very large in-crease in the supply of high school graduates (seeGoldin and Katz 1995, 2000).

38 The key difference between the two episodesis that Palestinian labor is a relatively small frac-tion of the Israeli work force, so we expect muchless of a technology response to changes in theeducational composition of Palestinian labor. Fur-thermore, Palestinian college graduates are not aclose substitute for Israeli college graduates, andonly a limited range of occupations are open tothem.


Cohen and Chang-Tai Hsieh 2000).This response suggests a change inthe production structure toward moreskilled workers, consistent with thetheories outlined in this section.39

Despite this evidence showing simul-taneous increases in the supply of, anddemand for, skills in a number of episodes,it is difficult to distinguish exogenousand endogenous technical change. Theexogenous technical change theorymaintains that technical change is oftenskill-biased. Endogenous technical changetheory instead suggests that new tech-nologies should be skill-biased whenthe supply of skills increases. Since thesupply of skills has increased most ofthe time over the past one hundred orso years, the implications of the twotheories are quite similar. The increasein the supply of unskilled labor in theEnglish cities during the early nine-teenth century provides an interestingcontrasting event for the two ap-proaches. Bairoch (1988, p. 245) de-scribes this rapid expansion as follows:“. . . between 1740 and 1840 the popu-lation of England . . . went up from 6million to 15.7 million. . . . while theagricultural labor force represented 60–70% of the total work force in 1740, by1840 it represented only 22%.” Habak-kuk (pp. 136–37) also emphasizes theincrease in the supply of unskilled laborin English cities, and attributes it tofive sources. First, enclosures releasedsubstantial labor from agriculture. Sec-ond, “population was increasing veryrapidly” (p. 136). Third, labor reserves

of rural industry came to the cities.Fourth, “there was a large influx oflabor from Ireland” (p. 137). Finally,“technical changes in agriculture in-creased the supply of labor available toindustry” (p. 137).

According to the endogenous tech-nology hypothesis, this increase in thesupply of unskilled labor should encour-age unskill-biased technical change. Andas predicted by this approach, therewere major skill-replacing (unskill-biased) technologies introduced duringthis period, most notably the factorysystem replacing tasks previously per-formed by skilled artisans. Moreover, inhis classic study, Habakkuk argues thatthe increase in the supply of labor wasan important inducement to the devel-opment of factory methods. He alsoquotes an American historian, Handlin,to explain why the adoption of factorymethods in the United States wassomewhat delayed. Handlin wrote:

[N]o matter what degree of standard-ization technical process of manufacturingreached, the absence of a cheap labor supplyprecluded conversion to factory methods.(p. 146)

Habakkuk placed much more impor-tance on the role of wages in determin-ing innovation decisions, a view thatlater became known as the “Habakkukhypothesis.” But he also emphasized thedifferent availabilities of skilled labor inBritain and the United States. He wrote:

[I]n both countries this provided manufac-turers with an incentive to adopt and devisemethods which replaced skilled by non-skilled . . . [but ] . . . the English had astronger incentive than the Americans toreplace skilled by unskilled labor. (p. 152)

With a similar reasoning to Habakkukand Schmookler, the endogenous tech-nology view suggests that nineteenth-century businessmen took advantage ofthe rapid increase in the supply of labor

39 Since Israel can be approximated by a smallopen economy, another possibility is a change inthe output mix and trade patterns. Gandal, Han-son, and Slaughter (2000) and Cohen and Hsieh(2000) find no evidence for this, and show thatdemand for skills increased in all Israeli sectors.Cohen and Hsieh (2000) also argue that becausemany Russian immigrants initially worked in low-skill occupations, the supply of skills to the Israelieconomy may not have increased by much.


in the cities by developing the factorysystem. According to this view, there isa close link between the skill-replacingtechnologies of the nineteenth centuryand the change in the factor suppliesfaced by employers—“the reserve armyof unskilled labor.” Although these his-torical examples are informative, theydo not reveal whether endogenous tech-nology choices are important in under-standing more recent skill-biased tech-nical change. A systematic study of howtechnologies respond to large changesin the relative supply of skills is clearlya worthwhile future research project.40

An important aspect of the endoge-nous technology theory is that it makesrelatively tight predictions regardingthe future path of technical change.While, with exogenous skill bias theo-ries, there is no clear reason to expect afurther acceleration or deceleration inthe skill bias of new technologies, ac-cording to this endogenous skill biastheory, the future path of technicalprogress should be closely tied to thepath of the supply of skills. If the relativesupply of skills continues to increase,we should expect further skill-biasedtechnical change.41

As discussed above, a major problemfor the “technological revolution” mod-els is the slowdown in TFP. Since theendogenous technology approach placesthe emphasis on which type of tech-nologies are developed, it is not incon-sistent with the slow growth in TFPduring the past thirty years: an accel-eration in the skill bias of new technolo-gies does not require faster technicalprogress. The evidence presented inRichard Newell, Adam Jaffee, andRobert Stavins (1999) is consistent withthe notion that changes in the directionof technical change can happen withoutan increase in the overall rate of pro-ductivity growth. They show that inno-vation in air conditioners responded tochanges in energy prices by becomingmore energy-efficient, but find no in-crease in the rate of productivitygrowth. In fact, Acemoglu (1998) andKiley (1999) show that the increasedeffort of firms to develop more skill-biased technologies could run into de-creasing returns, and hence may cause aslowdown in TFP growth. Intuitively,the overall productivity growth in theeconomy is maximized with a balanceddistribution of resources toward devel-oping skill-biased and unskill-biasedtechnologies (due to decreasing returnsto each activity). During periods ofrapid skill-biased technical change, allresources go into developing skill-biased machines, and cause a decline inadvances in unskill-biased technologies.Because of the decreasing returns to

40 It is also useful to note that the skill bias oftechnology most likely responds not only tochanges in the relative supply of skills, but to avariety of other factors. Recent work by MarcusMobius (2000) and David Thesmar and MathiasThoenig (2000) shows how the size of the productmarket, the degree of competitive pressure andinstability facing firms may affect the way firmschoose to organize, and therefore the demand forskills. Moreover, Mobius suggests that thesechanges reduced the demand for skills during thenineteenth century as there was greater stan-dardization of products, but increased this de-mand during the past several decades as the needfor flexibility increased. How the organization ofproduct markets and the extent of competition af-fect technology choices and the demand for skillsis a very promising area for future research.

41 Although evidence from the 1990s suggeststhat skill-biased technical change is now slower,there is not yet sufficient evidence to decidewhether the rapid skill-biased technical change of

the 1980s has come to an end. It also has to benoted that the increase in the relative supply ofcollege educated workers in the 1990s may havebeen less than the previous trend (see figure 1),and this may have affected the technology choicesof firms. In particular, with the increased labor forceparticipation of less skilled workers, there maynow be a sufficient number of unskilled workerssupplying labor at low wages to make the furtherdevelopment of unskilled-labor-complementarytechnologies quite profitable.


scale, improvements in the other sectorwill not fully offset this decline, andoverall TFP growth will fall.42

Finally, it is useful to discuss brieflythe response of skills to technology.The analysis so far treated the supply ofskills as exogenous, and investigated theimplications of the supply on the de-mand for skills. Naturally, the supplyof skills will also respond to economicincentives: more workers are likely toacquire skills when skill premia arehigher. Such supply choices can be eas-ily incorporated into this framework.Suppose, for example, that the relativesupply of skills, H/L, is an increasingfunction of the skill premium, ω. In thiscase, if the long-run demand curve forskills is upward-sloping, we can havean equilibrium path in which both therelative supply of skills and the skillpremium increase together over time(see Acemoglu 1998). This equilibriumconfiguration gives us an attractive in-terpretation for the joint behavior ofcollege skills and the college pre-mium in the U.S. economy over the pastsixty years (see figure 1): high returnsto schooling encourage education,which in turn induce more skill-biasedtechnical change, increasing returns toschooling again.

5.3 A Puzzle: The Decline in Wagesof Low-Skill Workers

A common shortcoming of all thepure technology approaches discussedin this section is that they do not natu-rally predict stagnant average wagesand/or falling wages for unskilled work-

ers.43 In the basic framework of section3, average wages always increase whenthe supply of educated workers in-creases, and all wages should rise in re-sponse to an increase in the productiv-ity of skilled workers, Ah. Yet, over thepast thirty years the wages of low-skillworkers have fallen in real value, whichcontrasts with their steady increase inthe previous thirty years.

Models of faster technological prog-ress would naturally predict that un-skilled workers should benefit from thisfaster progress. The endogenous tech-nology approach discussed in the pre-vious subsection, on the other hand,predicts that there may be no improve-ments in the technologies for unskilledworkers for an extended period of timebecause skill-biased innovations aremore profitable than unskill-biased in-novations. Yet in that case, their wagesshould be stagnant or increase slowly,but not fall.

Some of the studies mentioned abovehave suggested explanations for the fallin the wages of low-skill workers. Forexample, recall that Galor and Moav(2000) argue that faster technologicalchange creates an “erosion effect,” re-ducing the productivity of unskilled work-ers. Using equation (3) from above, inthe simplified version of their modeldiscussed in section 5.1, the unskilledwage is wL = φl(g)a[1 + φh

ρ(H ⁄ L)ρ](1 − ρ) ⁄ ρ,where the φl function captures the ero-sion effect. The rate of growth of un-skilled wages will be w⋅ L ⁄ wL = g(1 + εφ),where εφ is the elasticity of the φl func-tion which is negative by the assump-tion that φ′l < 0. If this elasticity is lessthan –1, an acceleration in economicgrowth can reduce the wages of42 The view that too much effort toward im-

proving the skill-biased technologies may be re-lated to the TFP slowdown is consistent with thepattern of sectoral TFP growth observed recently.As Gordon (1998) and Jorgensen and Stiroh (2000)document, there has been rapid TFP growth incomputer-producing sectors, but mediocre, oreven disappointing, TFP growth in other sectors.

43 However, recall that if the increase in non-wage benefits is taken into account, average wagesincreased over this period. So the more robust factmight be the fall in the real wages of low-skillworkers.


low-skill workers due to a powerfulerosion effect.

Acemoglu (1999a) and Caselli (1999)derive a fall in the wages of less-skilledworkers because the capital-labor ratiofor low-education/low-skill workers fallsas firms respond to technological devel-opments. In Caselli’s model this hap-pens because the equilibrium rate ofreturn to capital increases, and in mypaper, this happens because firms de-vote more of their resources to openingspecialized jobs for skilled workers.

Consider the following simple exam-ple to illustrate this point. There is ascarce supply of an input K, whichcould be capital, entrepreneurial talentor another factor of production. Skilledworkers work with the productionfunction

Yh = AhαKh

1 − αHα, (20)while unskilled workers work with theproduction function

Yl = AlαKl

1 − αLα, (21)where Kl and Kh sum to the total supplyof K, which is assumed fixed. For sim-plicity, assume that Yl and Yh are perfectsubstitutes. In equilibrium, the marginalproduct of capital in the two sectors hasto be equalized, hence

Kl

AlL=

K − Kl

AhH.

Therefore, an increase in Ah relative toAl will reduce Kl, as this scarce factorgets reallocated from unskilled to skilledworkers. The wages of unskilled workers,wL = (1 − α)Al

αKl1 − αLα − 1, will fall as a

result.An innovative version of this story is

developed by Paul Beaudry and DavidGreen (2000). Suppose that equation(21) above is replaced by Yl = Al

ηKl1 − ηLη,

with η < α, and K is interpreted asphysical capital. This implies that un-skilled workers are more “dependent”on capital than skilled workers. Beaudry

and Green show that an increase in H/Lcan raise inequality, and depress thewages of low-skill workers. Althoughthis is related to the effects of the in-crease in the relative supply of skills onthe path of technological progress dis-cussed in the last subsection, the mech-anism in Beaudry and Green’s paper isquite different. The increase in H/L in-creases the demand for capital, andpushes the interest rate up. This in-crease in the interest rate hurts un-skilled workers more than skilled work-ers because of the assumption thatη < α.

A potential problem with both theBeaudry and Green and Caselli storiesis that they explicitly rely on an increasein the price of capital. Although the in-terest rates were higher during the1980s in the U.S. economy, this seemsmostly due to contractionary monetarypolicy, and related only tangentially toinequality. Perhaps future research willshow a major role for the increase inthe interest rates in causing the declinein the wages of low-education workers,but as yet, there is no strong evidencein favor of this effect.44

Overall, a potential problem for mod-els based on technical change is to ac-count for the decline in the wages oflow-skill workers.45 I argue in the nextsection that the effect of technicalchange on the organization of the labor

44 Acemoglu (1999a), which is more in the spiritof the organizational theories discussed below, ob-tains the decline in the wages of unskilled workersthrough a change in the organization of produc-tion, which also entails a reallocation of capitalaway from them, but no increase in the rate ofreturn to capital.

45 Another possibility is that some of the techno-logical developments of the past two decades havebeen “truly labor-replacing,” for example, corre-sponding to an increase in Bl(t) or bt in terms ofthe production function in footnote 13. Autor,Levy, and Murnane (2000), for example, suggestthat computers have replaced unskilled routinetasks. This possibility has not been extensivelyresearched yet.


market both amplifies the effect oftechnology on wage inequality, andprovides a possible explanation for thisdecline.

6. Ramifications of Technical Change

This section discusses how technicalchange can affect labor market pricesby transforming the organization of thelabor market. The idea that technologyaffects the organization of production,and the institutions around it, is an oldone. Marx put it in a dramatic fashion:“The hand-mill gives you society withthe feudal lords; the steam-mill, societywith the industrial capitalist.” The argu-ment here is not as extreme, but re-lated: recent technological developmentsmay have led to important changes inthe organization of production.

My focus here is on three sets ofchanges that could account for the fallin the wages of low-skill workers: thetransformation of the organization offirms; change in labor market “insti-tutions,”46 particularly the decline inunionization; and the interaction betweeninternational trade and technical change.Organizational change often destroysthe types of jobs that pay high wages tolow-skill workers. Deunionization re-duces the bargaining power of low-skillworkers. And international trade withless developed countries (LDCs) in-creases the effective supply of unskilledlabor and depresses the marginal valueproduct of less skilled workers in theU.S. economy. Therefore, all threechanges could be responsible for thechanges in the U.S. wage structure andfor the decline in the wages of low-skillworkers. Nevertheless, I argue that thesefactors by themselves are not the major

cause of the increase in inequality. In-stead they have become powerful actorsonly by interacting with technicalchange, amplified the direct effect oftechnical change on inequality, andcontributed to the fall in the wages oflow-skill workers.

6.1 Organizational Change andInequality

A variety of evidence suggests thatimportant changes in the structure offirms have been taking place in the U.S.economy over the past 25 years. More-over, it seems clear that a major drivingforce for this transformation is changesin technologies (hence the view thattechnical change is essential for thechanges in the organization of firms).For example, team production andother high-performance productionmethods are now widespread in theU.S. economy (e.g., Casey Ichniowski,Giovanna Prennushi, and KathrynShaw 1997, or Eileen Applebaum andRosemary Batt 1994). Similarly, PeterCappelli and Steffanie Wilk (1997) showthat there has been an increase in thescreening of production workers, espe-cially from establishments that usecomputer technology and pay highwages.

Murnane and Levy (1996) report casestudy evidence consistent with thisview. From their interviews with humanresource personnel at a number of com-panies, they describe the change in thehiring practices of U.S. companies. Amanager at Ford Motor company in1967 describes their hiring strategy asfollows: “If we had a vacancy, we wouldlook outside in the plant waiting roomto see if there were any warm bodiesstanding there. If someone was thereand they looked physically OK andweren’t an obvious alcoholic, they werehired” (p. 19). In contrast, comparablecompanies in the late 1980s use a very

46 I am using the term institutions loosely here,to capture the rules of the game in the labor mar-ket, patterns of bargaining, as well as governmentlabor market policy.


different recruitment strategy. Murnaneand Levy discuss the cases of Honda ofAmerica, Diamond Star Motors, andNorthwestern Mutual Life. All threecompanies spend substantial resourceson recruitment and hire only a fractionof those who apply. Kremer and Maskin(1999) provide evidence of more segre-gation of workers across establishments.It seems that high-wage workers arenow much more concentrated in certainestablishments. Similarly, in Acemoglu(1999a) I documented a change in thecomposition of jobs over the pasttwenty years. Figure 9 here replicates apattern found in that paper, and plotsthe total percentage of workers em-ployed in the top 25-percent and bot-tom 25-percent industry-occupationcells (what I called weight-at-the-tailsof the job quality distribution). Theseare the cells (job types) that pay rela-

tively high or relatively low wages. In1983, 35 percent of employment was inthe top and bottom 25-percent jobcategories. By 1993, this number hadrisen to just under 38 percent. So, ap-proximately 2.5 percent more workersnow have either higher or lower qualityjobs rather than medium quality jobs.The actual changes in the distributionof jobs may be much larger than this,since substantial changes in the typesof jobs often take place within givenoccupations.

The view that changes in the organi-zation of firms have had a fundamentaleffect on the labor market is often ex-pressed in the popular press, and in theorganizational literature (e.g., ShohanaZuboff 1988). This organizational ap-proach is formalized by Kremer andMaskin (1999), Acemoglu (1999a), Mobius(2000), Thesmar and Thoenig (2000),

0.38

1993

Wei

ght-

at-t

he-t

ails

Year

Figure 9. The Evolution of the Percentage of Employment in the Top and Bottom 25 Percentile Industry-Occupation Cells (weight-at-the-tails of the job quality distribution)

19911989198719851983

0.375

0.37

0.365

0.36

0.355

0.35


and Gilles Duranton (2001). Kremerand Maskin consider a production func-tion which distinguishes between man-agers and workers. They show that achange in technology or an increase inthe dispersion of skills may encouragehigh-skill workers to match with otherhigh-skill workers, rather than work asmanagers in establishments employinglow-skill workers.

Here I outline a simple model, syn-thesizing Kremer and Maskin (1999)and Acemoglu (1999a), that capturesthe effect of the technical change onthe organization of production, and viathis channel, on wage inequality. Thebasic idea is simple. As the productivityof skilled workers increases, it becomesmore profitable for them to work bythemselves in separate organizationsrather than in the same workplace asthe unskilled workers. This is becausewhen the skilled and the unskilled worktogether, their productivities interact,and unskilled workers may put downwardpressure on the productivity of skilledworkers (for example, because unskilledworkers have to implement part of theproduction process designed by skilledworkers).

Specifically, suppose that firms haveaccess to the following productionfunctions

the old-style production function:Y = Bp[(AlL)ρ + (AhhO)ρ]1 ⁄ ρ,

the new-organization production function:

Y = BsAhβhN,

where hO is the number of skilled work-ers employed in the old-style firms, andhN is the number of skilled workers em-ployed in new organizations separatelyfrom unskilled workers. The fact thatwhen they are employed in the samefirm these two types of workers affecteach other’s productivity is captured bythe CES production function. This for-

mulation implies that if the productivity(ability) of unskilled workers, Al, is verylow relative to Ah, they pull down theproductivity of skilled workers. Incontrast, when they work in separatefirms, skilled workers are unaffected by theproductivity of unskilled workers. More-over, β > 1, which implies that improve-ments in the productivity of skilled work-ers has more effect on the productivityof new-style organizations. The parame-ters Bp and Bs capture the relative effi-ciency of old and new-style productionfunctions.

The labor market is competitive, sothe equilibrium organization of produc-tion will maximize total output, givenby Bp[(AlL)ρ + (AhhO)ρ]1 ⁄ ρ + BsAh

β(H − hO),where hO ∈ [0, H] is the number ofskilled workers employed in the old-styleorganizations. For all cases in whichhO > 0, the solution to this problem willinvolve

wH = BpAhρhO

ρ − 1[AlρLρ + Ah

ρhOρ ](1 − ρ) ⁄ ρ

= BsAhβ,

(22)

i.e., skilled workers need to be paid BsAhβ

to be convinced to work in the samefirms as the unskilled workers. Theunskilled wage is

wL = BpAlρLρ − 1[Al

ρLρ + AhρhO

ρ ](1 − ρ) ⁄ ρ

< wH.(23)

Now consider an increase in Ah. Dif-ferentiating (22) yields ∂hO ⁄ ∂Ah < 0, thatis, there will be fewer skilled workersworking with the unskilled. Moreover,differentiation also gives ∂(AhhO) ⁄ ∂Ah < 0,so the efficiency units of skilled workersthat the unskilled work with declines.From (23), this implies that dwL ⁄ dAh < 0.Therefore, skill-biased technical changeencourages skilled workers to work bythemselves, and as a result, unskilledwages fall. Intuitively, because, in theold-style organizations, the productiv-ity of skilled workers depends on the


ability of unskilled workers, when theskilled become even more productive,the downward pull exerted on their pro-ductivity by the unskilled workers be-comes more costly, and they prefer towork in separate organizations. This re-duces the ratio AhhO ⁄ L and depressesunskilled wages. As a result, improve-ments in technology, which normallybenefit unskilled workers as in section3, may hurt unskilled workers becausethey transform the organization ofproduction.

An increase in Bs ⁄ Bp, which raises therelative profitability of the new organi-zational form, also leads to further seg-regation of skilled and unskilled work-ers in different organizations. This lastcomparative static result is useful sinceBresnahan (1999) and Autor, Levy andMurnane (2000) argue that by replacingworkers in the performance of routinetasks, computers enabled a radical changein the organization of production.47

This is reminiscent of a technologicalchange that makes the new-organizationproduction function more profitable.

These organizational stories are at-tractive since they provide a unified ex-planation for the changes in the wagestructure and the apparent changes inthe organization of firms. An interestingrecent paper by Eve Caroli and VanReenan (1999) provides evidence sug-gesting that changes in wages have beenaccompanied by changes in organi-zational forms. Acemoglu (1999a) andKremer and Maskin (1999) also provideevidence suggesting a number of orga-nizational changes in the U.S. economyduring the past 25 years. Nevertheless,

this evidence does not yet enable an as-sessment of whether changes in organi-zational forms have been an importantcontributor to the changes in labormarket prices, and future research isrequired to determine the role oforganizational change in the rise ininequality.

6.2 Institutional Change

Labor market “institutions,” includ-ing minimum wage laws, the impor-tance of collective bargaining in wagedetermination, and perhaps socialnorms, undoubtedly have an importanteffect on the distribution of wages. TheGreat Compression in the wage distri-bution after the Second World War isdifficult to explain without invokingchanges institutions and social norms(see, e.g., Goldin and Robert Margo1992). Could changes in labor marketinstitutions also account for the in-crease in wage inequality and the dis-continuity in that demand for skillsduring the past decades?

Two major changes in labor marketinstitutions over the past 25 years arethe decline in the real value of state andfederal minimum wages and the reducedimportance of trade unions in wage de-termination. Although many economistssuspect that these institutional changesmay be responsible for the changes inthe structure of the U.S. labor market(see Freeman 1991; DiNardo, Fortin,and Lemieux 1995; Lee 1999), I willargue that these changes are unlikelyto be the major cause of the recentincrease in U.S. wage inequality.

The real value of the minimum wagewas eroded throughout the 1980s as thenominal minimum wage remained con-stant for much of this period. Since theminimum wage is likely to increase thewages of low-paid workers, this declinemay be responsible for increased wagedispersion. DiNardo et al. (1995) and

47 A related perspective is offered by Aghion(2001), who also argues that computers replaceunskilled tasks. He suggests that computers are a“general-purpose technology,” so their diffusionfollows an inverted S shaped pattern. He sug-gests that as more and more firms adopted com-puters over the past decades, demand for unskilledworkers fell rapidly.


Lee (1999) provide evidence in supportof this hypothesis. Although the contri-bution of the decline in the minimumwage to increased wage dispersion can-not be denied, most economists believethat this is unlikely to have been a ma-jor factor in the changes in the U.S.wage structure. First, only a small frac-tion of male workers are directly af-fected by the minimum wage (even in1992, after the minimum wage hike of1990–91, only 8 percent of all workersbetween the ages of 18 and 65 werepaid at or below the minimum wage).Although the minimum wage may in-crease the earnings of some workerswho are not directly affected, it is un-likely to affect the wages above the me-dian of the wage distribution. But asfigure 3 shows, the difference betweenthe 90th percentile and the medianbehaves very similarly to the differencebetween the median and the 10th per-centile.48 This implies that whateverfactors were causing increased wagedispersion at the top of the distributionare likely to have been the major causeof the increase in wage dispersionthroughout the distribution. Second, theerosion in the real value of the mini-mum wage started in the 1980s,whereas, as shown above, the explosionin overall wage inequality began in theearly 1970s (subject to the caveat raisedin footnote 11).

The declining importance of unionsmay be another important factor in the

increase in wage inequality. Unionsoften compress the structure of wagesand reduce skill premia (see for exampleLloyd Reynolds 1951; or Freeman andJames Medoff 1984). Throughout thepostwar period in the U.S. economy,unions negotiated the wages for manyoccupations, even indirectly influencedmanagerial salaries (see DiNardo,Kevin Hallock, and Pischke 2000).Unions also explicitly tried to compresswage differentials. This suggests that thedecline of unions may have been a majorcause of the changes in the structure ofwages.

Although deunionization could inprinciple be an important factor in thestructure of wages, the extent and tim-ing of deunionization suggests that it isnot the major driving force of the in-crease in inequality. First, wage in-equality increased in many occupationsin which prices were never affected byunions (such as lawyers and doctors).Moreover, in the United States, de-unionization started in the 1950s, a periodof stable wage inequality. During the1970s, though unionization fell in theprivate sector, overall unionization ratesdid not decline much because of in-creased unionization in the public sec-tor. Overall union density was approxi-mately constant, around 30 percent ofthe work force, between 1960 and 1975.It was the anti-union atmosphere of the1980s and perhaps the defeat of the air-traffic controllers’ strike that led to thelargest declines in unionization, datingmuch of deunionization after the rapidincrease in inequality during the early1970s.49 Evidence from other countriesalso paints a similar picture. For exam-ple, in the United Kingdom, wage in-equality started its sharp increase in the

48 Except during the early 1980s when there is amore rapid increase in inequality at the bottom ofthe wage distribution, most likely due to the fall-ing real value of the minimum wage. There aresubstantially larger increases in inequality at thebottom of the distribution among female workersor in a sample that combines male and femaleworkers. This reflects the larger effect of mini-mum wage laws on female earnings. Figure 3shows that there are factors, other than the de-cline in the real value of the minimum wage, hav-ing a major effect on the top and the bottom ofthe male wage distribution.

49 An interesting recent paper by Henry Farberand Bruce Western (2000) dates the major declinein union activity to the early 1980s, a few monthsbefore the air-traffic controllers’ strike.


mid-1970s, while union density in-creased until 1980 and started a rapiddecline only during the 1980s (AmandaGosling 1998). In Canada, while union-ization rates increased from around 30percent in the 1960s to over 36 percentin the late 1980s (Riddell 1993, table4.1), wage inequality also increased(see for example Freeman and KarenNeedels 1993, figure 2.4).

Although the timing of deunioniza-tion does not appear to be consistentwith this institutional change being themajor cause of the changes in wage in-equality, deunionization could clearlyaffect the wages of unskilled workers.One possibility is that deunionization isa contributing factor. But in this case,why did the unions decline while tech-nology was rapidly becoming more skill-biased? Acemoglu, Aghion, and Violante(2001) suggest that deunionization mayhave been caused by the technologicaldevelopments of the past decades. Ac-cording to this perspective, technicalchange is the driving force of thechanges in the wage structure, but italso leads to deunionization, and the re-sulting deunionization can have a sig-nificant indirect effect on wage inequal-ity, causing the wages of less-skilledworkers to fall.

To see the basic argument, suppose thatproduction can be carried out either inunionized or nonunionized firms. Innonunionized firms, workers receivetheir marginal products, which I denoteby Ah and Al for skilled and unskilledworkers. Assume that unions alwayscompress the structure of wages—i.e.,they reduce wage differentials betweenskilled and unskilled workers. Thiswage compression could be driven by avariety of factors. Acemoglu and Pischke(1999), for example, argue that unionsencourage productive training, and suchtraining is incentive compatible for firmsonly when the wage structure is com-

pressed. Alternatively, collective deci-sion making within a union may reflectthe preferences of its median voter, andif this median voter is an unskilledworker, he will try to increase unskilledwages at the expense of skilled wages. Itis also possible that union memberschoose to compress wages because ofideological reasons or for social cohe-sion purposes. The empirical literaturesupports the notion that unions com-press wages, though it does not distin-guish among the various reasons (seeFreeman and James Medoff 1984). Icapture wage compression in a reducedform way with the equation

ω =wH

wL≤ ψ

Ah

Al, (24)

where ψ < 1. Unions could never attractskilled workers unless they providedsome benefits to them to compensate forthe wage compression. Here I simply as-sume that they provide a benefit β to allworkers, for example, because unions in-crease productivity (e.g., Freeman andMedoff 1984; and Freeman and EdwardLazear 1995), or because they encouragetraining. Alternatively, β could be part ofthe rents captured by the union. Thezero-profit constraint for firms would be:(wH − β)H + (wL − β)L ≤ AhH + AlL (in thecase where β stands for rents, the zero-profit condition ensures that the firmdoes not wish to open a non-unionplant). Combining this equation with(24), and assuming that both hold asequality, we obtain

wH =(Ah + β)H + (Al + β)L

AhH + ψ−1AlLAh. (25)

Skilled workers will be happy to be partof a union as long as wH given by (25) isgreater than Ah. As Ah ⁄ Al increases—i.e.,as skill-biased technical change raisesthe productivity of skilled workers rela-tive to the unskilled—wH will fall relativeto Ah. Therefore, skill-biased technical


change makes wage compression morecostly for skilled workers, eventually de-stroying the coalition between skilledand unskilled workers that maintainsunions.

The important point is that deunion-ization causes a decline in the wages of un-skilled workers from wL = (Ah + β)H + (Al + β)L

AhH + ψ−1AlLψAl

to Al. Unskilled workers, who were pre-viously benefiting from wage compres-sion imposed by unions, now experi-ence a fall in real earnings. Therefore,technical change not only affects wageinequality directly, but also induces achange in labor market institutions. Theeffect of this change in institutions oninequality can be potentially larger thanthe direct effect of technical change,and explain the decline in the realwages of less-skilled workers.

Although Acemoglu, Aghion, and Vio-lante (2001) provide some evidenceconsistent with these patterns, whetherdeunionization was important in the de-cline of the wages of low-skill workers andwhether technical change is responsiblefor deunionization are open questions.It might also be interesting to investi-gate whether changes in technologymay have also affected the coalitionsupporting the minimum wage, andhence played a role in the decline inthe minimum wage during the 1980s.For example if unions, which have tra-ditionally supported minimum wages,were weakened by the technological de-velopments of that decade as argued inthis subsection, technical change mayhave indirectly contributed to the weak-ening of the coalition in support of theminimum wage.50

6.3 Trade, Technical Change,and Inequality

Finally, I discuss another majorchange affecting the U.S. economy: theincreased volume of international trade.I argue that increased internationaltrade by itself is not the cause of thechanges in the U.S. wage structure, buttrade may have been an important fac-tor in the rise of wage inequality byaffecting the degree of skill bias oftechnical change. This subsection issomewhat different from the previous two,since it is not about the effect of tech-nology on the organization of the labormarket, but on the effect of a majorchange in regulations on technologicaldevelopments.

Standard trade theory predicts thatincreased international trade with less-developed countries (LDCs), which aremore abundant in unskilled workers,should increase the demand for skills inthe U.S. labor market. So the increasein international trade may have been anunderlying cause of the changes in U.S.wage inequality.

To discuss these issues, consider thetwo-good interpretation of the model insection 3. Consumer utility is definedover [Yl

ρ + Yhρ]1 ⁄ ρ, with the production

functions for two goods being Yh = AhHand Yl = AlL. Both goods are now tradable.For simplicity, let me just compare theU.S. labor market equilibrium withoutany trade (as characterized in section 3)to the equilibrium with full internationaltrade without any trading costs.

Before trade, the U.S. relative priceof skill-intensive goods, ph ⁄ pl, is given by

pUS =ph

pl=

AhHAlL

ρ − 1

, (26)50 Finally, both the decline in the role thatunions play and in the value of the minimum wagemay have been caused by changes in certain socialnorms, for example, the norm of “equal workequal pay.” The decline of the same norms couldalso be responsible for the increase in inequality(e.g., the emergence of the winner-take-all societyas claimed by Robert Frank and Philip Cook

1996). Unfortunately, there is currently little re-search on the effect of social norms on inequalityand on why inequality norms may have changedover the past thirty years.


where H and L denote the supplies ofskilled and unskilled labor in the UnitedStates. The skill premium is then simplyequal to the ratio of the marginal valueproducts of the two types of workers,that is, ωUS = pUSAh ⁄ Al. Next, supposethat the United States starts trading witha set of LDCs that have access to thesame technology as given by Ah and Al,but are relatively scarce in skills. Denotethe total supplies of skilled and unskilledworkers in the LDCs by H and L whereH ⁄ L < H ⁄ L, which simply reiterates thatthe United States is more abundant inskilled workers than the LDCs.

After full trade opening, the productmarkets in the United States and theLDCs are joined, so there will be aunique world relative price. Since thesupplies of skill- and unskill-intensivegoods are Ah(H + H) and Al(L + L), therelative price of the skill-intensive goodwill be

pW =

Ah(H + H)Al(L + L)

ρ − 1

< pUS. (27)

The fact that pW > pUS follows immedi-ately from H ⁄ L < H ⁄ L. Intuitively, oncethe United States starts trading withskill-scarce LDCs, demand for skill-intensive goods increases and pushes theprices of these goods up.

Labor demand in this economy is de-rived from product demands. The skillpremium therefore follows the relativeprice of skill-intensive goods. Aftertrade opening, the U.S. skill premiumincreases to

ωW = pWAh

Al> ωUS (28)

where the fact that ωW > ωUS is an imme-diate consequence of pW > pUS. There-fore, trade with less developed countriesincreases wage inequality in the UnitedStates.

The skill premium in the LDCs willalso be equal to ωW after trade opening

since the producers face the same rela-tive price of skill-intensive goods, andhave access to the same technologies.Before trade, however, the skill pre-mium in the LDCs was ω = pAh ⁄ Al,where p = (AhH ⁄ AlL)ρ − 1 is the relativeprice of skill-intensive goods in theLDCs before trade. The same argumentas above implies that p > pW, i.e., tradewith the skill-abundant United States re-duces the relative price of skill-intensivegoods in the LDCs. This implies thatωW < ω; after trade wage inequalityshould fall in the LDCs.

Although in theory increased tradewith the LDCs can be the cause of therapid increase in the demand for skills,most evidence suggests that the directeffect of increased international tradeon the U.S. labor market has been rela-tively minor. First, as equation (27)shows, the effect of international tradeworks through a unique interveningmechanism: more trade with the LDCsincreases the relative price of skill-intensive goods, p, and affects the skillpremium via this channel. In fact, inthis simple framework, the percentageincrease in the skill premium is propor-tional to the percentage increase in therelative price of skill-intensive goods.Perhaps the most damaging piece ofevidence for the trade hypothesis is thatmost studies suggest the relative priceof skill-intensive goods did not increaseover the period of increasing inequality.Robert Lawrence and Slaughter (1993)found that during the 1980s the relativeprice of skill-intensive goods actuallyfell. Jeffrey Sachs and Howard Shatz(1994) found no major change or aslight decline. A more recent paper byKrueger (1997) criticized the methodsand data used by these studies, andfound an increase in the relative priceof skill-intensive goods. Nevertheless,the increase in these prices is relativelysmall, so would not be able to account for


the large increase in the skill premiumexperienced in the U.S. economy.

Second, with trade as the drivingforce, increased production of skill-intensive goods should be drawingworkers away from other sectors. Incontrast, as documented by Murphyand Welch (1993), Berman, Bound, andGriliches (1994) and Autor, Katz, andKrueger (1998), all sectors, even thoseproducing less-skilled goods, increasedtheir demands for more-educated work-ers. This pattern is not consistent withtrade being the main driving force ofthe increase in the demand for skilledworkers (though one has to bear inmind the increase in outsourcing ininterpreting this fact; see RobertFeenstra and Hanson 1999).

Third, a direct implication of thetrade view is that, as shown above,while demand for skills and inequalityincreased in the United States, the con-verse should have happened in theLDCs that have started trading withthe more skill-abundant U.S. economy. Theevidence, however, suggests that moreof the LDCs experienced rising in-equality after opening to internationaltrade (see Hanson and Ann Harrison1994; or Donald Robbins 1995). Al-though the increase in inequality in anumber of cases may have been due toconcurrent political and economic re-forms, the preponderance of evidenceis not favorable to this basic implicationof the trade hypothesis.

Finally, a number of economists havepointed out that U.S. trade with theLDCs is not important enough to have amajor impact on the U.S. product mar-ket prices and consequently on wages.Paul Krugman (1995) illustrates thispoint by undertaking a calibration of asimple North-South model. Katz andMurphy (1992), Berman, Bound, andGriliches (1994) and George Borjas,Freeman, and Katz (1997) emphasize

the same point by showing that the con-tent of unskilled labor embedded in theU.S. imports is small relative to thechanges in the supply of skills takingplace during this period.51

Although the above arguments sug-gest that increased international tradewith the LDCs is not the major cause ofthe changes in the wage structure by it-self, they do not rule out a powerful ef-fect of international trade when it inter-acts with technical change. In a worldwith endogenous technical change, in-creased international trade could affectthe types of technologies developed andadopted by firms, and have a large ef-fect through this channel. This possibil-ity was first raised by Wood (1994), whoargued that trade with the LDCs willlead to defensive skill-biased innova-tions. Wood, however, did not developthe mechanism through which such de-fensive innovations could occur. I nowillustrate how trade causes skill-biasedtechnical change using the endogenoustechnology model developed in section5.2 (this analysis draws on Acemoglu1999b).

Suppose that the United States startstrading with the LDCs as discussedabove, and assume that the LDCs al-ways use U.S. technologies. Therefore,the supply of skilled and unskilledgoods in the LDCs is Yh = AhH andYl = AlL where as before H ⁄ L < H ⁄ L.The immediate effect will be anincrease in the relative price ofskill-intensive goods as illustrated by

51 This is probably the weakest criticism againstthe trade view, and many studies have pointed outhow international trade could have a larger effecton U.S. labor market prices in the presence of la-bor market rents. For example, George Borjas andValerie Ramey (1995), Dani Rodrik (1997), andDube and Reddy (1999) have argued that thethreat of international trade may reduce wages,especially in sectors with substantial rents, andthis change in bargaining power may affect theearnings of unskilled workers more, increasinginequality.


equation (27). Now, recall from theanalysis in section 5.2 that there is arelative price effect on the direction oftechnical change (because developingtechnologies to produce the more ex-pensive good is more profitable). There-fore, trade, by making the skill-intensivegoods more expensive, encourages moreskill-biased technical change.

To determine exactly how the direc-tion of technical change will be affectedby trade, we need to know the marketsizes for new technologies after tradeopening. It is plausible to assume thattrade opening with the LDCs will nothave a major effect on the enforcementof intellectual property rights in theSouth. In that case, trade opening willinduce skill-biased technical change inthe United States. Specifically, as longas after trade opening the United Statesdoes not start producing technologiesfor unskilled workers in the LDCs, therelative market sizes for the two typesof technologies remain at H/L. This im-plies that the technology market clear-ing condition, equation (18), no longerholds. In particular, since ph

WH > plWL

from equation (27), there will now onlybe incentives for skill-biased technicalchange. This process continues untilpW = ph

W ⁄ plW = (H ⁄ L)−1, so that incentives

to develop the two different types oftechnologies are balanced. This impliesthat the skill bias of technical change isstill determined by equation (18) fromsection 5.2, i.e., by U.S. domestic rela-tive supplies alone. Intuitively, the rela-tive price of skill-intensive goods playstwo roles in this model. The first is toclear the market for goods (i.e., equa-tion 27), and the second is to ensureequilibrium in the technology market(cfr. equation 18). Since the technologymarket clearing condition relates therelative price of skill-intensive goods tothe relative supplies in the U.S. market,which do not change, the long-run equi-

librium price of skill-intensive goodscannot change either. Combining equa-tions (18) and (27) gives the post-tradeskill bias of technology as

AhW

AlW =

(H + H)(L + L)

−1

HL

1 ⁄ (1 − ρ)

>Ah

US

AlUS =

HL

ρ ⁄ (1 − ρ)

where AhUS ⁄ Al

US is the pre-trade skill biasof technology in the United States.

The implication is that when the di-rection of technical change is endoge-nous, trade between the United Statesand the LDCs will induce skill-biasedtechnical progress. The result is notonly that trade leads to an increase inskill premia, but that this can happenwithout the counterfactual implicationsof the standard trade models discussedabove.

The first implication of this inducedskill bias is that the impact of trade onlabor markets may be much larger thanpredicted by the standard trade models,which helps against the criticism thatthe amount of trade the United Statesundertakes with the LDCs is not largeenough. Second, because trade causesskill-biased technical change, the factthat all sectors have increased the em-ployment of skilled workers is consis-tent with trade being the underlyingcause of the increase in inequality.Third, for the same reason, there is aforce counteracting the decline in in-equality in the LDCs implied by trade:these economies use U.S. technologies,which are becoming more skill-biased.Finally, and quite strikingly, tradeleaves the relative price of skill-intensivegoods in the United States unchangedin the long-run. Recall that changes inrelative prices are the usual interveningmechanism in trade models, so a num-ber of studies have concluded that trade


has not been an important factor in theincrease in inequality because the rela-tive price of skill-intensive goods hasnot increased much (e.g., Lawrence andSlaughter 1994; Sachs and Shatz 1995).In this model, the long-run relativeprice of skill-intensive goods in theUnited States is unaffected by trade.More generally, induced skill-biasedtechnical change in the United Statesimplies that trade will increase theprice of skill-intensive goods by onlya limited amount, while still havinga major effect on the U.S. labormarket.

Notice finally that the interaction be-tween international trade and technicalchange may help to explain the declinein the wages of low-skill workers. In-creased international trade acts as anincrease in the supply of unskilledworkers, and as shown in our basicframework in section 3, this would putdownward pressure on unskilled wages.Borjas, Freeman, and Katz (1987), forexample, provide evidence that in-creased international trade during the1980s reduced the wages of highschool graduate workers (though theyalso suggest that the effect of the immi-gration of less-skilled workers wasgreater).

Overall, international trade could stillbe a major driving force of the changesin the wage structure. However, forincreased trade to have such a largeeffect on the structure of wages—and toavoid the aforementioned counterfac-tual implications—it must cause achange in the path of technologicalprogress. So the combination of tradeopening and endogenous technicalchange gives an alternative theory foracceleration: skill-biased technical changeaccelerates, neither exogenously nor inresponse to the changes in the supplyof skilled workers, but due to tradeopening.

7. Changes in Residual Inequality

The previous sections highlightedthat there are major unanswered ques-tions regarding the causes of the in-crease in inequality, the reasons for thefaster skill-biased technical change inthe past few decades, and the determi-nants of the fall in the wages of low-skill workers. These questions awaitingfurther research notwithstanding, wehave a reasonably simple and usefulframework, and the beginnings of con-sistent answers. In contrast, in this andthe next section, I discuss areas whereanswers are much more tentative, andthere is a greater need for future re-search. I begin with residual inequality.A major issue that most models dis-cussed so far failed to address is the dif-ferential behavior of returns to school-ing and residual inequality during the1970s. I argue in this section that anexplanation for this pattern requiresmodels with multi-dimensional skills.

7.1 A Single Index Modelof Residual Inequality

The simplest model of residual in-equality is a single index model, inwhich there is only one type of skill,though this skill is only imperfectly ap-proximated by education (or experi-ence). Expressed alternatively, in a sin-gle index model observed andunobserved skills are perfect substi-tutes. Consider, for example, the modeldeveloped above, but suppose that in-stead of skills, we observe education,e.g., whether the individual is a collegegraduate, which is imperfectly corre-lated with true skills. A fraction φc ofcollege graduates are highly skilled,while a fraction φn < φc of noncollegegraduates are highly skilled. Denotethe skill premium by ω = wH ⁄ wL. Thecollege premium, i.e., the ratio ofaverage college to noncollege wages, is


ωc =wC

wN=

φcwH + (1 − φc)wL

φnwH + (1 − φn)wL

=φcω + (1 − φc)φnω + (1 − φn)

,

while within-group inequality, i.e., theratio of the wage of high-wage collegegraduates (or noncollege graduates) tothat of low-wage college graduates (ornoncollege graduates), is ωwithin = ω (sincehigh-wage workers in both groups earnwH, while low-wage workers earn wL).As long as φc and φn remain constant, ωc

and ωwithin will always move together.Therefore, an increase in the returns toobserved skills—such as education—willalso be associated with an increase in thereturns to unobserved skills.

This framework provides a naturalstarting point, linking between- andwithin-group inequality. It predicts thatwithin- and between-group inequalityshould move together. But during the1970s, returns to schooling fell whileresidual (within-group) inequality in-creased sharply. We can only accountfor this fact by positing a decline in φc

relative to φn of a large enough magni-tude to offset the increase in ω; thiswould ensure that during the 1970s thecollege premium could fall despite theincrease in within-group inequality. Alarge decline in φc relative to φn wouldpredict a very different behavior of thecollege premium within different co-horts. Yet the appendix shows little evi-dence in favor of this. I therefore con-clude that the single index model cannotexplain the changes in residual inequalityduring the 1970s and the 1980s.

7.2 Sorting and Residual Inequality

Another approach would combineeducational sorting with an increase inthe demand for skills. Suppose thatwages are given by ln wit = θtai + γthi + εit

where hi is a dummy for high education,ai is unobserved ability, and εit is a mean

zero disturbance term. Here γt is the priceof observed skills, while θt is the priceof unobserved skills. The educationpremium can be written as

ln ωt ≡ E(ln wit | hit = 1) − E(ln wit | hit = 0)= γt + θt(A1t − A0t)

where A1t ≡ E(ln wit | hit = 1) and A0t isdefined similarly. Residual inequalitycan be measured by Var(Ait | h = 0) andVar(Ait | h = 1) .

To simplify the discussion, assumethat there is perfect sorting into educa-tion, in the sense that there exists athreshold a− such that all individualswith unobserved ability above a− obtaineducation. Then within-group inequal-ity among high and low-education work-ers will move in opposite directions aslong as the price of observed skills, θ, isconstant: as a− declines (and averageeducation increases), Var(Ait | h = 1) willincrease, but Var(Ait | h = 0) will fall. In-tuitively, there are more and more“marginal” workers added to the high-education group, creating more unob-served heterogeneity in that group andincreasing within-group inequality. Butin contrast, the low-education group be-comes more homogeneous. Therefore,without a change in the prices for unob-served skills, this approach cannot ac-count for the simultaneous increasein inequality among both low- andhigh-education groups.52

7.3 Churning and Residual Inequality

Another approach emphasizes thatworkers of all levels of education mayface difficulty adapting to changes. This

52 A natural variation on this theme would be asituation in which γ and θ move together. How-ever, this will run into the same problems as thesingle index model: if γ and θ always move to-gether, then such a model would predict thatwithin-group inequality should have fallen duringthe 1970s. Therefore, models based on sorting alsorequire a mechanism for the prices of observedand unobserved skills to move differently duringthe 1970s.


has been argued by Piore and Sabel(1984), and more recently by Aghion,Howitt, and Violante (2000) and Gould,Moav, and Weinberg (2000). Violante(2000) offers an alternative, but relatedtheory, where increased churning isdriven by the greater productivity gapbetween different firms (because offaster embodied technical change) andincreased skill dynamics of ex anteidentical workers (which he contrastswith the argument that it is innate abil-ity differences that matter). Accordingto all these approaches, an increase ininequality results from more rapid tech-nical change, not because of skill biasbut because of increased “churning” inthe labor market. Aghion, Howitt, andViolante (2000), for example, suggestthat only some workers will be able toadapt to the introduction of new tech-nology, and this will increase wageinequality.

An advantage of this approach is thatit is in line with the increased earningsinstability pointed out by Gottschalkand Robert Moffit (1994). However,there is relatively little evidence, otherthan this increase in earnings instabil-ity, supporting the notion that there ismore churning in the labor market. Thedata on job creation and job destructionreported by Steven Davis, John Halti-wanger, and Scott Schuh (1996) showsno increase in job reallocation duringthe 1980s or the early 1990s, and mostevidence does not indicate much of adecline in job stability over this period(e.g. Francis Diebold, David Neumark,and Daniel Polsky 1997; or HenryFarber 1995).53 Also theories based onchurning do not naturally predict a di-vergence between returns to educationand residual inequality during the

1970s. Therefore, a mechanism thatcould lead to differential behavior inthe prices to observed and unobservedskills is still necessary.54

7.4 A Two-Index Modelof Residual Inequality

Since models based on a single indexof skill (or models where different typesof skills are perfect substitutes) are in-consistent with the differential behaviorof returns to schooling and within-groupinequality during the 1970s, an obviousnext step is to consider a two-indexmodel where observed and unobservedskills are imperfect substitutes. Thereare good reasons to expect that in thereal world skills are multidimensional.For example, the social psychologistHoward Gardner (1988) makes a strongcase that we should always think ofskills as multidimensional, and that thestandard IQ measures fail to capturethis multidimensionality. There is along tradition in economics building onthe Roy model which uses models withmultidimensional skills to analyze wagesand returns to schooling (e.g., RobertWillis and Sherwin Rosen 1975). In thecontext of the determinants of earnings,a recent paper by Bowles, Gintis, and

53 More recent evidence indicates that theremay have been a decrease in job tenure during thelater parts of the 1990s (see Neumark, Polsky, andDaniel Hansen 1999).

54 An interesting theory similar to the churningmodels that could lead to such a differential be-havior is advanced by Galor and Tsiddon (1997).They draw a distinction between ability and edu-cation, and argue that returns to ability increasefaster during periods of rapid technologicalchange. If we view the 1970s as a period of rapidtechnological change, as suggested above, this the-ory would imply an increase in the returns to abil-ity (unobserved skills) during this period. Never-theless, this explanation is still not consistent withthe facts because high-ability individuals are morelikely to be high education, so rapid technologicalprogress should also increase returns to schooling.Perhaps a combination of this mechanism with dif-ferential sorting into education, or with imperfectsubstitution between high- and low-educationworkers, might be able to account for the diver-gence between returns to schooling and residualinequality during the 1970s, but such a model hasnot been developed yet.


Osborne (2001) also forcefully arguesthat there are many dimensions to skillsincluding the ability to function in hier-archies and other social situations. Todiscuss these issues, I now consider asimple two-index model (see Acemoglu1998). Suppose that there are fourtypes of workers, differentiated by botheducation and unobserved skills. Theeconomy has an aggregate productionfunction

Y = [(AluLu)ρ + (AlsLs)ρ + (AhuHu)ρ

+ (AhsHs)ρ]1 ⁄ ρ,

where Lu is the supply of low-skill low-education workers, and other terms aredefined similarly. Within-group inequal-ity corresponds to the ratio of the wagesof skilled low-education workers to thoseof unskilled low-education workers,and/or to the ratio of the wages of skilledhigh-education workers to those of un-skilled high-education workers. A naturalstarting point is to presume that the frac-tion of high-skill workers in each educationgroup is constant, say at φl = Ls ⁄ Lu andφh = Hs ⁄ Hu > φl, which implies that thereare more high-ability workers among high-education workers. With this assumption,within-group inequality will be

wLs

wLu=

Als

Alu

ρ

φl−(1 − ρ) and

wHs

wHu=

Ahs

Ahu

ρ

φh−(1 − ρ).

(29)

The college premium, on the otherhand, is

ω =φh

ρAhs

ρ + Ahuρ

φlρAls

ρ + Aluρ

1 + φl

1 + φh

ρ

HL

−(1 − ρ)

.

Using this framework and the idea of en-dogenous technology, we can provide anexplanation for the differential behaviorof returns to schooling and within-groupinequality during the 1970s. Recall thataccording to the endogenous technologyapproach, it is the increase in the supply

of more-educated workers that triggersmore rapid skill-biased technical change.Because technology adjusts sluggishly,the first effect of an increase in the sup-ply of educated workers, as in the 1970s,will be to depress returns to schooling.This will continue until technology haschanged enough to offset the direct ef-fect of the higher supply of skills (seefigure 7). This change in returns toschooling has no obvious implication forwithin-group inequality in a multi-skillset-up, since it is the education skillsthat are becoming abundant, not unob-served skills—in fact, in equation (29),within-group inequality is invariant tochanges in the supply of educated work-ers unless there is a simultaneous changein φh and φl.

Under the plausible assumption thatmore skilled workers within each educa-tion group also benefit from skill-biasedtechnical progress, technical changespurred by the increase in the supplyof educated workers will immediatelybenefit workers with more unobservedskills, raising within-group inequality.Therefore, an increase in the supply ofeducated workers will depress returnsto schooling, while increasing within-groupinequality. After this initial phase, tech-nical change will increase both returns toschooling and within-group inequality.55

55 If the 1960s are also characterized by steadyskill-biased technical change, equation (29) sug-gests that there should have been an increase inresidual inequality during this decade as well. Thedata presented in section 2 do not support thisprediction. Therefore, it seems that to explain thebasic trends, one needs to posit that improvementsin technology take the form ln Aj(t) = γ ′j + γj ⋅ t withγhs = γhu > γls = γlu during regular times, but whenthere is an acceleration in skill bias, the patternchanges to favor workers with more unobservedskills, i.e., γhs > γhu and γls > γlu. Although this as-sumption can generate stable residual inequalitybefore the 1970s, and an increase in residual in-equality during the 1970s, it is simply reverse engi-neered to fit the facts. Future research to inves-tigate whether there are natural reasons for thispattern to arise would be useful.


Overall, single index models are notcapable of explaining the changes in re-sidual inequality over the past thirtyyears, and we do not yet know how im-portant various factors are. Analysis ofthe determinants of residual inequalityand the reasons why there was an explo-sion in overall inequality beginning inthe 1970s remains a major researcharea.

8. Cross-Country Patterns

So far, I have focused on U.S. wageinequality patterns and on the incen-tives to develop new technologies com-ing from the U.S. supply of skills. Thecross-country dimension presents anumber of challenges, in particular be-cause it is difficult to explain why in-equality increased much more in somecountries than others. In addition, forthe endogenous technology view, it isimportant to know whether it is therelative supply of skills in each countryor in the world as a whole that deter-mines the direction of technical change.I now briefly discuss these issues.

8.1 Differences in Inequality Patterns

Although the tendency towardgreater inequality has been a feature inmany developed and less developedcountries (see Freeman and Katz 1995;and Berman and Machin 2000), thereare also marked differences in the be-havior of within- and between-group in-equality across these countries. Katz,David Blanchflower, and Gary Loveman(1995), Murphy, Riddell, and Romer(1998), and Card and Lemieux (2000)show that the differential behavior ofthe supply of skills can go a long waytoward explaining the differences inthe returns to schooling, especially be-tween the United States, Canada, andthe United Kingdom. Nevertheless, it ispuzzling that wage inequality increased

substantially in the United States andthe United Kingdom, but remainedfairly stable in many continental Euro-pean economies (see for example Davis1995; Gottschalk and Timothy Smeeding1999).

The standard explanation for thisdivergent behavior, succinctly summa-rized by Krugman (1994) and OECD(1994), and sometimes referred to asthe Krugman hypothesis, maintains thatinequality did not increase as much (ornot at all) in Europe because labor mar-ket institutions there encourage wagecompression, limiting the extent of in-equality. This can be captured in thecompetitive framework of section 3,where firms are always along their rela-tive demand curve, by assuming thatlabor market institutions impose an ex-ogenous skill premium ω− = wH ⁄ wL. Thisimplies:

Hl

=

Ah

Al

ρ ⁄ (1 − ρ)

ω−−1 ⁄ (1 − ρ). (30)

where the level of employment of un-skilled workers, l, will generally be lessthan their labor supply L because ofwage compression. A more compressedwage structure—i.e., a lower ω−—will in-crease the unemployment of unskilledworkers, given by L − l.

The view that wages are more com-pressed in Europe clearly has somemerit. Francine Blau and LawrenceKahn (1995) show that the major differ-ence in overall inequality between theUnited States and many continentalEuropean economies is not in the 90–50 differential, but in the 50–10 differ-ential. This suggests that the minimumwage, strong unions, and generoustransfer programs in Europe are inpart responsible for the relative wagecompression in Europe.

Nevertheless, the Krugman hypothe-sis runs into two difficulties. First, unless


there are extremely rigid institutionsthat fix the skill premium exogenously,skill-biased technical change shouldincrease wage inequality irrespectiveof the degree of exogenously imposedwage compression. In contrast, in manycontinental European economies, mostnotably in Germany, wage inequalitywas very stable (see, e.g., Freeman andKatz 1995).

Second, the Krugman hypothesismakes an explicit prediction: profit-maximizing employment decisions offirms should lead to a decline in theemployment of unskilled workers rela-tive to that of skilled workers. In fact,skill-biased technical change mighteven reduce the unemployment rates ofskilled workers. Yet, in Europe, the un-employment of skilled and unskilledworkers increased together (e.g. StephenNickell and Brian Bell 1996; Kruegerand Pischke 1997), and unskilled em-ployment did not grow faster in theUnited States than in European econo-mies (Card, Kramartz, and Lemieux1996; Krueger and Pischke 1997).

It is possible that bargaining arrange-ments in Europe between firms andunions, imply not only wage compres-sion, but also deviations from the rela-tive demand curve for “skills” given by(30). This may be because European in-stitutions force firms to pay uniformwages to all educated workers irrespec-tive of their exact skills (marginal prod-uct), making the employment of skilledworkers less profitable as well. Alterna-tively, if unions represent both skilledand unskilled workers, and are commit-ted to wage compression, they may notwant to suffer a large decrease in theemployment of unskilled workers, andprefer to make certain concessions inwage levels in order to induce firms toemploy more unskilled workers at acompressed wage structure. Althoughsuch deviations from equation (30) are

a possibility, we have no direct evi-dence to assess how far off the relativedemand curve European economiesmay be, and how they would respond toskill-biased technical change when theyare away from the relative demandcurve for skills.56

My preferred approach to explainingcross-country differences is to considerthe effect of labor market institutionson technology choices. In particular,the European labor market institutions,which compress the structure of wages,will give greater incentives to adoptinglabor-complementary technologies, andwill reinforce wage compression. I givea simple example to illustrate the pointhere. Suppose the productivity of askilled worker is Ah = aη, whereas theproductivity of an unskilled worker isAl = a, where a is a measure of aggre-gate technology in use, and η > 1. Sup-pose that wages are determined by rentsharing, unless they fall below a legallymandated minimum wage, in whichcase the minimum wage binds. Hence,wj = min {βAj,w− }, where j = l or h, and βis worker’s share in rent sharing. Notethat the cost of technology upgradingdoes not feature in this wage equation,because rent sharing happens aftertechnology costs are sunk. To capturewage compression, suppose the mini-mum wage is binding for unskilledworkers in Europe. Now consider tech-nology adoption decisions. In particular,

56 An alternative view suggested by Nickell andBell (1996) explains the differences in the wagestructure across countries by differences in theskill distribution. According to this view, becauseof the relative weakness of the U.S. high schoolsystem, American noncollege workers are lessskilled than their European counterparts. How-ever, recent work by Dan Devroye and Freeman(2001) shows that differences in skill distributionhave little to do with cross-country differences inwage dispersion. They document that dispersionof internationally comparable test scores amongnative-born Americans are similar to those inEurope, but wage inequality among native-bornAmericans is much higher.


firms can either produce with someexisting technology, a, or upgrade to asuperior technology, a′ = a + α, at cost γ.The profit to upgrading the technologyused by a skilled worker is (1 − β)αη − γ,both in the United States and Europe.The new technology will therefore beadopted when

γ ≤ γS ≡ (1 − β)αη.

Note that there is a holdup problem, dis-couraging upgrading: a fraction β of theproductivity increase accrues to theworker due to rent sharing (Paul Grout1984; Acemoglu 1996).

The incentives to upgrade the tech-nology used by unskilled workers, onthe other hand, differ between theUnited States and Europe. In theUnited States, this profit is given by(1 − β)α − γ. So, the new technology willbe adopted with unskilled workers if

γ ≤ γU ≡ (1 − β)α.

Clearly, γU < γS, so adopting new tech-nologies with skilled workers is moreprofitable. The returns to introducingthe new technology are different inEurope because the minimum wage isbinding for unskilled workers. To sim-plify the discussion, suppose that evenafter the introduction of new technology,the minimum wage binds, i.e.,w− > β(A + α). Then, the return to intro-ducing the new technology in Europewith unskilled workers is α − γ, and firmswill do so as long as γ < α. Since α > γU,firms in Europe have greater incentivesto introduce advanced technologies withunskilled workers than in the UnitedStates. This is because the binding mini-mum wage in Europe makes the firm thefull residual claimant of the increase inthe productivity of unskilled workers.This highlights that in an economy with acompressed wage structure, firms mayhave a greater incentive to increase the

productivity of unskilled workers (seeAcemoglu and Pischke 1999 for thisargument in the context of training).

As long as the cost of upgrading tothe new technology, γ, is small, i.e., lessthan γU, firms will upgrade both withskilled and unskilled workers in theUnited States and Europe. In this case,cross-country inequality levels will bestable. This corresponds to the situationin the 1950s and the 1960s. In contrast,if γ is high, for example, because thetechnological improvements of the 1980sare more expensive to implement, theremay be a divergence in inequality be-tween the two economies. For instance,if γ ∈(γU,α), then new technology will notbe adopted with unskilled workers inthe United States, but it will be usedwith unskilled workers in Europe. As aresult, while wage inequality increasesin the United States, it will remainstable in Europe. Therefore, a simplestory for cross-country differences ininequality trends emerges from thismodel: wage compression encouragesthe use of more advanced technologieswith unskilled workers, and acts to rein-force itself in Europe. In contrast, tech-nological developments can harm theearnings of low-skill U.S. workers whoare not protected by this type of wagecompression. Whether the interactionbetween wage compression and tech-nology choice could be important in ex-plaining European inequality and un-employment patterns is an area forfuture study.

8.2 International Determinantsof Technology

The endogenous technology frame-work developed above links the skillbias of technology to the relative supplyof skills. There are a number of inter-esting and difficult issues that arisewhen we consider the international di-mension. Here I simply mention some


preliminary approaches, but clearlymuch theoretical and empirical workremains to be done.

A first extension of the endogenoustechnology idea to an international con-text might be to suppose that skill biasin each country is determined by thecountry’s relative supply of skills. Yet itmay be more plausible to imagine asituation where new technologies spreadacross countries. In this case, it may bethe incentives in the technologicallymost advanced country (the technologi-cal leader) that determine the skill biasof world technologies. This descriptionmay be adequate for understanding theskill bias of technologies used by lessdeveloped countries (see for exampleAcemoglu 1999b). But it is also possiblefor other technologically advanced econo-mies to pursue a different path of tech-nological development than the leader,in which case domestic incentives maybe important in shaping skill bias.

What determines the skill bias oftechnologies developed by the techno-logical leader? This depends on themarket sizes for different types of tech-nologies, hence on the internationalenforcement of intellectual propertyrights. For example, in the discussionon the effect of trade on technology, Isupposed that there were no intellec-tual property rights for United Statescompanies enforced in less developedeconomies. In this case, incentives todevelop new technologies are shaped bythe United States (or OECD) supplies.This may be a good starting point, sinceeven when property rights are enforced,there will be a number of difficultiesfacing U.S. companies marketing theirtechnologies in other countries, espe-cially in the case of technologies thatwill be used with relatively low-skillworkers. For example, technologies mayneed to be adapted to conditions in lo-cal markets, or producers in LDCs may

be unable to pay for these technologiesbecause of credit problems.

It is also worth noting that even whena country is using U.S. technologies, itseffective skill bias may be influenced byits domestic skill supply. This is be-cause U.S. technologies need to beadapted to local conditions, and firmswill have a greater incentive to do thiswhen there is a larger supply of workersto use these technologies. So it may benot only technological change that isendogenous to relative supplies, butalso technology adoption.

Finally, another interesting cross-country dimension comes from lookingat wage inequality trends in LDCs. Asdiscussed in section 6.3, the first-orderpredictions of the standard trade theoryare not borne out: instead of a declinein inequality, which would have beenexpected due to the greater integrationof these economies into world trade,inequality increased in most LDCs. Arecent paper by Berman and Machin(2000) shows an interesting pattern:while there has been rapid skill upgrad-ing in many middle income countries,there is much less evidence for rapidskill upgrading in the poorest econo-mies. A possible explanation for thesepatterns is that middle income countriesare adopting advanced technologiesmuch more rapidly than the poorestcountries, and since these technologiesare more skill-biased, these economiesare undergoing rapid skill upgradingand increases in inequality. Further-more, if, as claimed by Acemoglu andZillibotti (2001), new technologies de-veloped in the rich economies are typi-cally “too skill-biased” for LDCs, therecent acceleration in skill bias couldhave negative implications for theLDCs. More generally, the impact oftechnologies developed in the advancedeconomies on LDC labor markets is anarea that requires further research.


9. Conclusion

This essay discussed the link betweentechnical change and the labor market,with special emphasis on the recentchanges in the U.S. wage structure. Inthis process, I surveyed part of thelarge literature on the determinants ofthe rise in inequality, and put forth adifferent interpretation of the changesin technologies and their impact on thelabor market than the most widely ac-cepted view. It is difficult to summarizethis large literature, and therefore, evenmore difficult to summarize this es-say. Nevertheless, it may be useful toreiterate some of the main points:

1. The behavior of wages and returns toschooling in the United States indi-cates that technical change has beenskill-biased during the past sixtyyears, and probably for most of thetwentieth century.

2. Though more controversial, the evi-dence also points to an accelerationin skill bias during the past fewdecades.

3. In contrast, much of what we know sug-gests that technical change was not skill-biased during the nineteenth century,and most likely, it was skill-replacing.

4. We can understand the behavior oftechnical change by recognizing thatthe development and use of technol-ogy respond to profit incentives. Whendeveloping skill-biased techniques ismore profitable, new technology willtend to be skill-biased. According to thisperspective, the early nineteenth cen-tury was characterized by skill-replacingdevelopments because the increasedsupply of unskilled workers in theEnglish cities made the introduction ofthese technologies profitable.

5. In contrast, the twentieth century hasbeen characterized by skill-biasedtechnical change because the rapidincrease in the supply of skilled work-

ers has induced the development ofskill-complementary technologies.

6. The acceleration in skill-biased tech-nical change is then likely to havebeen a response to the rapid increasein the supply of skills during the pastseveral decades, though this perspec-tive does not suggest that we are nec-essarily in the midst of a “Technologi-cal Revolution”; what has changedis not necessarily the overall rate ofprogress, but the types of technologiesthat are being developed.

7. The recent technological develop-ments are also likely to have affectedthe organization of the labor mar-ket—including the way firms organizeproduction, labor market policies, andthe form of labor market “institu-tions”—and may have had a largeeffect on the structure of wagesthrough this channel.

Many of these conclusions are tenta-tive. There is much research to be doneto understand the process of technicalchange and how it impacts the labormarket. In particular, what determineswage differences among observationallysimilar workers, why the trajectories ofthe Anglo-Saxon and continental Euro-pean economies have diverged, and howtechnical change and institutionalchange interact are important areas forfuture research.

AppendixA1. Data Sources

The samples are constructed as in Katz andAutor (2000). I thank David Autor for providingme with data from this study. Data from 1939,1949, and 1959 come from 1940, 1950, and 1960censuses. The rest of the data come from 1964–97March CPSs. The college premium is the coeffi-cient on workers with a college degree or morerelative to high school graduates in a log weeklywage regression. The regression also includesdummies for other education categories, a quarticin experience, three region dummies, a nonwhitedummy, a female dummy, and interactions be-tween the female dummy and the nonwhite dummy


and the experience controls. The sample includesall full-time full-year workers between the ages of18 and 65, and except those with the lowest 1 per-cent earnings. Earnings for top coded observationsare calculated as the value of the top code times1.5. The relative supply of skills is calculated froma sample that includes all workers between theages of 18 and 65. It is defined as the ratio ofcollege equivalents to non-college equivalents, cal-culated as in Autor, Katz, and Krueger (1998)using weeks worked as weights. In particular, col-lege equivalents = college graduates + 0.5 × work-ers with some college, and noncollege equivalents =high school dropouts + high school graduates + 0.5 ×workers with some college.

Samples used for overall and residual wage in-equality include only white male full-time full yearworkers between the ages of 18 and 65, and ex-clude those earning less than half the real valueof the 1982 minimum wage converted from nomi-nal dollars using the personal consumption expen-diture deflator (see Katz and Autor 2000). Earn-ings for top coded observations are calculated asthe value of the top code times 1.5.

A2. The Behavior of Overall InequalityDuring the 1970s

In an important paper on the effect of labormarket institutions on inequality, DiNardo,Fortin, and Lemieux (1995) provide evidence sug-gesting that in the May CPSs, there is no increasein inequality during the 1970s. In table A1, I dis-play numbers from the survey by Katz and Autor(2000), who report changes in residual inequalityfor the past four decades from three different

sources: decennial censuses, and March CPSs andMay CPSs (and later Outgoing Rotation Groupfiles—ORGs). These numbers show no significantchange in residual or overall inequality during the1960s, and consistent increases in inequality from allsources during the 1970s and the 1980s. For example,the data from the Census and the March CPSs indi-cate that the 90–10 differential increased about 0.10 ayear between 1970 and 1979, while the 50–10 dif-ferential increased by about 0.11 a year during thesame period. The May CPS data show a smallerincrease in the 90–10 differentials during this pe-riod, but a comparable increase in the 50–10 dif-ferential. Overall, although there is less uniformityamong data sources regarding the behavior of re-sidual inequality than returns to schooling (seeKatz and Autor 2000), there is considerable evi-dence that residual and overall inequality startedto increase during the 1970s.

A3. Can Composition Effects ExplainInequality Changes?

A possible explanation for the patterns we ob-serve could be changes in the distribution of un-observed skills— or more concretely, compositioneffects. For example, the average ability of workerswith high education may have increased relative tothat of workers with low education over time.Here, I document that the increase in the returnsto education and residual inequality are not smplydue to composition effects. Note first that compo-sition effects cannot by themselves explain the re-cent changes in inequality: as noted in subsection7.2, composition effects suggest that inequalityamong educated and uneducated workers should

TABLE A1ANNUALIZED CHANGES IN OVERALL AND RESIDUAL WAGE INEQUALITY (FROM KATZ AND AUTOR)

Census March CPSs May CPSs (ORGs)

90–10 50–10 90–10 50–10 90–10 50–10

Changes in overall inequality1960s 0.10 0.03 –0.03 –0.11 — —1970s 0.10 0.11 0.10 0.11 0.01 0.101980s 0.17 0.06 0.20 0.09 0.26 0.101990s — — 0.11 –0.03 0.05 0.00

Changes in residual inequality1960s 0.03 0.01 –0.01 –0.01 — —1970s 0.09 0.05 0.11 0.08 0.11 0.081980s 0.07 0.02 0.12 0.06 0.15 0.081990s — — 0.07 0.03 0.06 0.02

Note: The numbers give 10 × annualized changes from Table 4 of Katz and Autor (2000). 90–10 is the differencebetween the 90th and 10th percent of the log wage or residual distribution, and 50–10 is the difference between themedian and 10th percent of the corresponding distribution. The residuals are estimated from log earnings regres-sions with nine education dummies, a quartic in experience and their interactions. See notes to Tables 3 and 4 inKatz and Autor (2000).


move in opposite directions. This suggests thatchanges in the true returns to skills have played atleast some role in the changes in inequality.

More generally, to get a sense of how importantcomposition effects may be, consider a variant ofequation (13) above with two education levels, highh = 1 and low h = 0, and suppose wages are given by

ln wit = ai + γthi + εit (31)

where hi is a dummy for high education, ai is un-observed ability, and εit is a mean zero disturbanceterm. Define the (log) education premium—thedifference between the average wages of high andlow education workers—can be written as

ln ωt ≡ E(ln wit | hi = 1) − E(ln wit | hi = 0)= γt + A1t − A0t

where A1t ≡ E(ai | hi = 1) and A0t is defined simi-larly. The increase in the education premium canbe caused by an increase in γt (a true increase inthe returns to skills) or an increase in A1t − A0t.There are basically two reasons for an increase inA1t − A0t: (1) changes in cohort quality, or (2)changes in the pattern of selection into education.

Consider changes in cohort quality first. If, asmany claim, the U.S. high school system has be-come worse, we might expect a decline in A0t with-out a corresponding decline in A1t. As a result,A1t − A0t may increase. Alternatively, as a largerfraction of the U.S. population obtains higher edu-cation, it is natural that selection into education(i.e., the abilities of those obtaining education)will change. It is in fact possible that those whoare left without education could have very low un-observed ability, which would translate into a low levelof A0t, and therefore into an increase in A1t − A0t.

Although these scenarios are plausible, theoret-ically the opposite can happen as well. For exam-ple, many academics who have been involved inthe U.S. education system for a long time com-plain about the decline in the quality of universi-ties, while the view that American high schoolshave become much worse is not shared universally(e.g., Krueger 1998). The selection argument isalso more complicated than it first appears. It istrue that, as long as those with high unobservedabilities are more likely to obtain higher educa-tion, an increase in education will depress A0t. Butit will also depress A1t. To see why, assume that

TABLE A2COMPOSITION EFFECTS

Born in 19– 06–10 11–15 16–20 21–25 26–30 31–35 36–40 41–45 46–50 51–55Year ↓ →

Panel A

1950 1.448 1.370 1.175 1.0931960 1.551 1.564 1.525 1.421 1.303 1.1321970 1.680 1.656 1.613 1.539 1.392 1.1531980 1.567 1.560 1.538 1.402 1.222 1.0631990 1.798 1.761 1.723 1.674

Panel B

∆ ln ω50−60 0.103 0.194 0.350 0.328∆ ln ω60−70 0.155 0.234 0.311 0.407∆ ln ω70−80 –0.047 0.021 0.146 0.249∆ ln ω80−90 0.260 0.359 0.500 0.611

Panel C

∆2 ln ω50−70 0.051 0.040 –0.040 0.079∆2 ln ω60−80 –0.201 –0.213 –0.165 –0.158∆2 ln ω70−90 0.307 0.338 0.354 0.362

Note: The top panel gives the college premium from the census indicated at the beginning of the row for cohortsborn in the five-year intervals indicated at the head of the column. For example, the first number is for individualsborn between 1906–10 from the census of 1950. The college premium is defined as the wages of workers from thatcohort with a college degree or more divided by the wages of workers from that cohort with twelve years ofschooling. The bottom panel gives the change in the college premium for a given cohort between the two indicateddates and the difference between the wage growth of two neighboring cohorts as indicated by equations (34) and(35). All data are from the decennial censuses for white males born in the United States.


there is perfect sorting—i.e., if an individual withability a obtains education, all individuals withability a′ > a will do so as well. In this case, therewill exist a threshold level of ability, a−, such thatonly those with a > a− obtain education. Next con-sider a uniform distribution of ai between b0 andb0 + b1. Then,

A0 =1

a− − b0∫b0

a−

ada =a− + b0

2

and

A1 =1

b1 − b0 − a− ∫a−

b0 + b1ada =

b0 + b1 + a−

2.

So both A0 and A1 will decline when a− decreases toa− ′. Moreover, A1 − A0 = b1 ⁄ 2, so it is unaffected bythe decline in a−. Intuitively, with a uniform distri-bution of ai, when a− increases, both A0 and A1 fallby exactly the same amount, so the compositioneffects have no influence on the education pre-mium. Clearly, with other distributions of ability,this extreme result will no longer hold, but it re-mains true that both A0 and A1 will fall, andwhether this effect will increase or decrease theeducation premium is unclear. Overall, therefore,the effects of changes in composition on educationpremia is an empirical question.

Empirically, the importance of composition ef-fects can be uncovered by looking at inequalitychanges by cohort (see McKinley Blackburn,David Bloom, and Freeman 1992; Juhn, Murphy,and Pierce 1993). To see this, rewrite equation(31) as

ln wict = aic + γthic + εcit (32)

where c denotes a cohort—i.e., a group of indi-viduals who are born in the same year. I have im-posed an important assumption in writing equa-tion (32): returns to skills are assumed to be thesame for all cohorts and ages; γt—though clearlythey vary over time. We can now define cohortspecific education premia as

ln ωct ≡ E(ln wict | hi = 1) − E(ln wict | hi = 0)= γt + A1ct − A0ct

where A1ct ≡ E(aic | hi = 1) and A0ct is defined simi-larly. Under the additional assumption that thereis no further schooling for any of the cohortsover the periods under study, we haveln ωct = γt + A1c − A0c, which implies

∆ ln ωc,t′ − t ≡ ln ωct′ − ln ωct = γt′ − γt, (33)

i.e., changes in the returns to education within acohort will reveal the true change in the returns.Yet, the assumption that returns to skills are con-stant over the lifetime of an individual may be toorestrictive. Murphy and Welch (1992), for exam-ple, show quite different age-earning profiles byeducation. Nevertheless, a similar argument canbe applied in this case too. For example, sup-pose ln ωcst = γst + A1c − A0c for cohort c of age s in

year t, and that γst = γs + γt (this assumption isalso not necessary, but simplifies the discussion).Then ∆ ln ωc,t′ − t = γs′ − γs + γt′ − γt, where obviouslys′ − s = t′ − t. Now consider a different cohort, c′′that is age s′ in the year t and age s in the year t′′.Then ∆ ln ωc′′,t − t′′ = γs′ − γs + γt − γt′′. So, the truechange in the returns to skills between the dates t′′and t′ is

∆2 ln ω ≡ ∆ ln ωc,t′ − t − ∆ ln ωc′′,t − t′′ = γt′ − γt′′. (34)

Using data from the 1950–90 censuses, table A2gives some of the single and double differences ofcohort inequality for white men aged 26–55. Thesingle differences show increases in the returns to

TABLE A3CHANGES IN INEQUALITY BY COHORT

(FROM JUHN ET AL. 1993)

Panel A: 90–10 Differentials for Log Weekly Wages

Year ofmarket entry 1964 1970 1976 1982 1988

1983–88 1.381977–82 1.27 1.381971–76 1.13 1.24 1.381965–70 1.08 1.12 1.29 1.421959–64 1.13 1.01 1.13 1.30 1.401953–58 1.02 1.07 1.16 1.32 1.431947–52 1.02 1.11 1.15 1.301941–46 1.02 1.07 1.161935–40 1.06 1.091929–34 1.09

Panel B: 90–10 Differentials for Log Wage Residuals

Year ofmarket entry 1964 1970 1976 1982 1988

1983–88 1.091977–82 1.06 1.161971–76 .96 1.09 1.181965–70 .86 .96 1.12 1.231959–64 .92 .86 .98 1.12 1.211953–58 .88 .91 .99 1.15 1.261947–52 .89 .94 .99 1.141941–46 .94 .94 1.051935–40 .95 .981929–34 .99

Note: This table replicates Table 3 of Juhn, Murphy,and Pierce (1993). The top panel reports the 90–10differential for log weekly wages of the cohorts thathave entered the labor market in the correspondingsix-year interval. Panel B gives the 90–10 differentialfor the residuals from a regression of log weekly wageson education controls.


college within most cohorts, with the exception ofthe years between 1970 and 1980. Therefore,these increases are likely to reflect differential ageeffects by education. In contrast, the numbers inpanel C for the 1950–70 period show no increases,suggesting that the double difference does a goodjob of controlling for composition effects. Thenumbers for the 1960–80 period are negative,which likely reflect the decline in the college pre-mium between 1960 and 1980. The final row givesthe most important results of this table. The 1970–90 double differences are large and positive, sug-gesting that the true returns to education in-creased over this time period. Interestingly,despite the well-known evidence that the collegepremium increased faster for younger workersover the 1980s, the results in table A2 show thatthe true increase in returns to skills between 1970and 1990 are comparable for cohorts born be-tween 1936 and 1955. These results therefore in-dicate that the major component of the increase inthat college premium during the 1980s and 1990swas changes in true skill prices, not compositioneffects.

Table A3, which replicates table 3 from Juhn,Murphy, and Pierce (1993), shows that the in-crease in overall and residual inequality cannot beexplained by composition effects either. Panel Ashows that the 90–10 differential for cohorts en-tering the market between 1935 and 1964 is ap-proximately constant between 1964 and 1970, butincreases sharply for each cohort between 1970and 1976, and then increases further between1982 and 1988. Panel B shows a similar picture forlog wage residuals. These results suggest that thechanges in the structure of wages observed overthe past thirty years cannot be explained by purecomposition effects, and reflect mainly changesin the true returns to observed and unobservedskills.

REFERENCES

Abramovitz, Moses. 1956. “Resource and OutputTrends in the United States since 1870,” Amer.Econ. Rev. 46, pp. 5–23.

Acemoglu, Daron. 1996. “A Microfoundation forSocial Increasing Returns in Human Capital Ac-cumulation,” Quart. J. Econ. 111:3, pp. 779–804.

———. 1998. “Why Do New Technologies Com-plement Skills? Directed Technical Change andWage Inequality,” Quart. J. Econ. 113:4, pp.1055–89.

———. 1999a. “Changes in Unemployment andWage Inequality: An Alternative Theory andSome Evidence,” Amer. Econ. Rev. 89:5, pp.1259–78.

———. 1999b. “Patterns of Skill Premia,” NBERwork. paper 7018.

———. 2000 “Labor- and Capital-AugmentingTechnical Change,” NBER work. paper 7544.

Acemoglu, Daron; Phillipe Aghion and GianlucaViolante. 2001. “Technical Change, Deunioniza-

tion and Inequality,” MIT mimeo, forthcomingCarnegie-Rochester Conf. Ser. Public Pol.

Acemoglu, Daron and Jorn-Steffen Pischke. 1999.“The Structure of Wages and Investment inGeneral Training,” J. Polit. Econ. 107:3, pp.539–72.

Acemoglu, Daron and Fabrizio Zilibotti. 2001.“Productivity Differences,” Quart. J. Econ.116:2, pp. 563–606.

Aghion, Philippe. 2000. “Technical Change, Insti-tutions, and the Dynamics of Inequality,” Har-vard mimeo; forthcoming Econometrica.

Aghion, Philippe and Peter Howitt. 1998. Endoge-nous Growth Theory. MA: MIT Press.

Aghion, Philippe; Peter Howitt and Gianluca Vio-lante. 2000. “Technology, Knowledge, and In-equality,” mimeo, Harvard U.

Ahmad, Syed. 1966. “On the Theory of InducedInvention,” Econ. J. 56, pp. 344–57.

Angrist, Joshua. 1995. “The Economic Returns toSchooling in the West Bank and Gaza Strip,”Amer. Econ. Rev. 85:5, pp. 1065–87.

Appelbaum, Eileen and Rosemary Batt. 1994. TheNew American Workplace. NY: ILR Press.

Autor, David; Alan Krueger and Lawrence Katz.1998. “Computing Inequality: Have ComputersChanged the Labor Market?” Quart. J. Econ.113:4, pp. 1169–213.

Autor, David; Frank Levy and Richard Murnane.2001. “The Skill Content of Recent Technologi-cal Change: An Empirical Exploration,” MITmimeo.

Bailey, Martin and Robert Gordon. 1988. “TheProductivity Slowdown: Measurement Issuesand Explosion of Computer Power,” BrookingsPap. Econ. Act., Macroecon. 2, pp. 379–431.

Bairoch, Paul. 1988. Cities and Economic Develop-ment: From the Dawn of History to the Present.Christopher Braider, trans. Chicago: U. Chi-cago Press.

Bartel, Ann P. and Frank Lichtenberg. 1987. “TheComparative Advantage of Educated Workersin Implementing New Technology,” Rev. Econ.Statist. 69:1, pp. 1–11.

Bartel, Ann and Nachum Sicherman. 1998. “Tech-nological Change and the Skill Acquisition ofYoung Workers,” J. Labor Econ. 16:4, pp. 718–55.

Beaudry, Paul and David Green. 1999. “What IsDriving U.S. and Canadian Wages: ExogenousTechnical Change or Endogenous Choice ofTechnique?” NBER work. paper 6853.

Berman, Eli; John Bound and Zvi Griliches. 1994.“Changes in the Demand for Skilled Laborwithin U.S. Manufacturing Industries: Evidencefrom the Annual Survey of Manufactures,”Quart. J. Econ. 109, pp. 365–67.

Berman, Eli; John Bound and Stephen Machin.1998. “Implications of Skill-Biased Technologi-cal Change: International Evidence,” Quart. J.Econ. 113, pp. 1245–80.

Berman, Eli and Stephen Machin. 2000. “SBTCHappens! Evidence on the Factor Bias of Tech-nological Change in Developing and DevelopedCountries,” Boston U. mimeo.


Blackburn, McKinley; David Bloom and RichardFreeman. 1992. “Changes in Earnings Differen-tials in the 1980s: Concordance, Convergence,Causes, and Consequences,” NBER work. pa-per 3901.

Blau, Francine D. and Lawrence M. Kahn. 1996.“International Differences in Male Wage In-equality: Institutions versus Market Forces,” J.Polit. Econ. 104:4, pp. 791–836.

Borghans, Lex and Bas Ter Weel. 2000. “Do WeNeed Computer Skills to Use a Computer: Evi-dence from the United Kingdom,” Maastrichtmimeo.

Borjas, George J.; Richard B. Freeman andLawrence F. Katz. 1997. “How Much Do Immi-gration and Trade Affect Labor Market Out-comes?” Brookings Pap. Econ. Act. 0:1, pp. 1–67.

Borjas, George J. and Valerie Ramey. 1995. “For-eign Competition, Market Power and Wage In-equality,” Quart. J. Econ. 110:4, pp. 1075–110.

Boskin Michael; Ellen Dulberger, Robert Gordon,Zvi Griliches and Dale Jorgenson. 1998. “Con-sumer Prices, the Consumer Price Index, andthe Cost of Living,” J. Econ. Perspect. 12:1, pp.3–26.

Bound, John and George Johnson. 1992. “Changesin the Structure of Wages in the 1980s: AnEvaluation of Alternative Explanations,” Amer.Econ. Rev. 82:3, pp. 371–92.

Bowles, Samuel; Herbert Gintis and Melissa Os-borne. 2001. “The Determinants of Earnings: ABehavioral Approach,” J. Econ. Lit. 39:4, pp.1137–76.

Braverman, Harry. 1974. Labor and MonopolyCapital: The Degradation of Work in the Twen-tieth Century. NY: Monthly Review Press.

Brendt, Ernie; Catherine Morrison and LarryRosenblum. 1994. “High-tech Capital Forma-tion and Labor Composition in U.S. Manufac-turing Industries: An Exploratory Analysis,” J.Econometrics 65, pp. 9–43.

Bresnahan, Timothy F. 1999. “Computerizationand Wage Dispersion: An Analytical Reinter-pretation,” Stanford U. mimeo.

Bresnahan, Timothy F.; Erik Brynjolfsson andLorin M. Hitt. 1999. “Information Technology,Workplace Organization and the Demand forSkilled Labor: Firm-level Evidence,” NBERwork. paper 7136.

Cappelli, Peter and Steffanie Wilk. 1997. “Under-standing Selection Processes: Organization De-terminants and Performance Outcomes,”mimeo, Wharton School, U. Pennsylvania.

Card, David E. and Alan Krueger. 2000. “SchoolQuality and Black–White Relative Earnings: ADirect Assessment,” Quart. J. Econ. 107, pp.151–200.

Card, David and Thomas Lemieux. 2001. “CanFalling Supply Explain the Rising Return toCollege for Younger Men?” Quart. J. Econ.116:2, pp. 705–46.

Card, David; Francis Kramartz and ThomasLemieux. 1996. “Changes in the Relative Struc-

ture of Wages and Employment: A Comparisonof the United States, Canada and France,” Can.J. Econ.

Caroli, Eve and John van Reenen. 1999. “WageInequality and Organizational Change,” UCLmimeo.

Caselli, Francesco. 1999. “Technological Revolu-tions,” Amer. Econ. Rev. 89:1, pp. 78–102.

Cohen, Sarit and Chang-Tai Hsieh. 2000. “Macro-economic and Labor Market Impact of RussianImmigration in Israel,” Princeton U. mimeo.

David, Paul. 1975. Technical Choice, Innovationand Economic Growth: Essays on American andBritish Experience in the Nineteenth Century.London: Cambridge U. Press.

David, Paul. 1990. “The Dynamo and the Com-puter: An Historical Perspective on the ModernProductivity Paradox,” Amer. Econ. Rev. 80:2,pp. 355–61.

Davis, Steven J. 1992. “Cross-Country Patterns ofChanges in Relative Wages,” in NBER Macro-economics Annual 1992. Olivier Jean Blanchardand Stanley Fischer, eds. Cambridge and Lon-don: MIT Press, pp. 239–92.

Davis, Steven J.; John Haltiwanger and ScottSchuh. 1996. Job Creation and Destruction.MA: MIT Press.

Devroye, Dan and Richard Freeman. 2000. “DoesInequality in Skills Explain Inequality of Earn-ings across Countries?” Harvard U. mimeo.

Diebold, Francis; David Neumark and DanielPolsky. 1997. “Job Stability in the UnitedStates,” J. Labor Econ. 15:2, pp. 206–33.

DiNardo, John; Nicole Fortin and ThomasLemieux. 1995. “Labor Market Institutions andthe Distribution of Wages, 1973–1992: A Semi-parametric Approach,” Econometrica 64:5, pp.1001–44.

DiNardo, John; Kevin Hallock and Jorn-SteffenPischke. 2000. “Unions and Managerial Pay,”London School Econ. mimeo.

DiNardo, John and Jorn-Steffen Pischke. 1997.“The Returns to Computer Use Revisited: HavePencils Changed the Wage Structure Too?”Quart. J. Econ. 112:1, pp. 291–303.

Doms, Mark; Timothy Dunne and KennethTorske. 1997. “Workers, Wages and Technol-ogy,” Quart. J. Econ. 112, pp. 217–52.

Duranton, Gilles. 2001. “The Economics of Pro-duction Systems: Segmentation and Skill-BiasedChange,” mimeo, London School of Economics.

Entorf, Horst and Francis Kramartz. 1998. “DoesUnmeasured Ability Explain the Higher Wagesof New Technology Workers?” Europ. Econ.Rev. 41, pp. 1489–509.

Farber, Henry. 1995. “Are Lifetime Jobs Disap-pearing? Job Duration in the United States,”NBER work. paper 5014.

Farber, Henry and Bruce Western. 2000. “RoundUp the Usual Suspects: The Decline of Unionsin the Private Sector,” Princeton U. Ind. Rel.work. paper 437.

Feenstra, Robert and Gordon Hanson. 1999. “TheImpact of Outsourcing and High-Technology


Capital on Wages: Estimates for the United States,1979–1990,” Quart. J. Econ. 114:3, pp. 907–40.

Foster, Andrew and Mark Rosenzweig. 1996.“Technical Change in Human Capital Returnand Investments: Evidence from the GreenRevolution,” Amer. Econ. Rev. 86, pp. 931–53.

Frank, Robert H. and Phillip J. Cook. 1995. TheWinner-Take-All Society. NY: Free Press.

Freeman, Richard. 1976. The OvereducatedAmerican. NY: Academic Press.

———. 1986. “Demand for Education,” in Hand-book of Labor Economics, Vol. I. Orley Ashen-felter and Richard Layard, eds. North-Holland:Elsevier.

———. 1991. “How Much Has De-unionizationContributed to the Rise of Male Earnings In-equality?” NBER working paper 3826.

Freeman, Richard and Lawrence F. Katz. 1995.“Introduction and Summary,” in Differencesand Changes in Wage Structures. R. Freemanand L. Katz, eds. Chicago: U. Chicago Press.

Freeman, Richard and Edward Lazear. 1995. “AnEconomic Analysis of Works Councils,” inWorks Councils: Consultation, Representation,and Cooperation in Industrial Relations. JoelRogers and Wolfgang Streeck, eds. NBERCompar. Labor Markets Series, Chicago: U.Chicago Press.

Freeman, Richard and James Medoff. 1984. WhatUnions Do. NY: Basic Books.

Freeman, Richard and Karen Needles. 1993. “SkillDifferences in Canada in an Era of Rising La-bor Market Inequality,” in Small DifferencesThat Matter. D. Card and R. Freeman, eds.Chicago: U. Chicago Press and NBER.

Friedberg, Rachel. 1997. “The Impact of Mass Mi-gration on the Israeli Labor Market,” RI: BrownU. Population Studies and Training Centerwork. paper 97–11.

Galor, Oded and Daniel Tsiddon. 1998. “Techno-logical Progress, Mobility and EconomicGrowth,” Amer. Econ. Rev. 87, pp. 363–82.

Galor, Oded and Omer Maov. 2000. “Ability Bi-ased Technological Transition, Wage Inequalityand Economic Growth,” Quart. J. Econ. 115,pp. 469–98.

Gandal, Neil; Gordon H. Hanson and Matthew J.Slaughter. 2000. “Technology, Trade, and Ad-justment to Immigration in Israel,” mimeo,Dartmouth College.

Goldin, Claudia and Lawrence F. Katz. 1995. “TheDecline of Non-Competing Groups: Changes inthe Premium to Education, 1890 to 1940,”NBER work. paper 5202.

———. 1998. “The Origins of Technology–SkillComplementarity,” Quart. J. Econ. 113:3, pp.693–732.

———. 2000. “Education and Income in the Early20th-century: Evidence from the Prairies,”forthcoming, J. Econ. Hist.

Goldin, Claudia and Robert A. Margo. 1992. “TheGreat Compression: The Wage Structure in theUnited States at Mid-Century,” Quart. J. Econ.107:1, pp. 1–34.

Gordon, Robert. 1990. Measurement of DurableGoods Prices. IL: U. Chicago Press.

———. 1998. “Current Productivity Puzzles froma Long-term Perspective,” Northwestern U.mimeo.

Gosling, Amanda. 1998. “The Determination ofUnion Status and Wages in British Estab-lishments,” U. Essex, discuss. paper 491.

Gottschalk, Peter. 1997. “Inequality in Income,Growth and Mobility: The Basic Facts,” J. Econ.Perspect. 11, pp. 21–40.

Gottschalk, Peter and Robert Moffitt. 1994. “TheGrowth of Earnings Instability in the U.S.Labor Market,” Brookings Pap. Econ. Act. 2,pp. 217–72.

Gottschalk, Peter and Timothy M. Smeeding.1997. “Cross-National Comparisons of Earningsand Income Inequality,” J. Econ. Lit. 35:2, pp.633–87.

Gould, Eric D.; Omer Moav and Bruce A. Wein-berg. 2000. “Precautionary Demand for Educa-tion, Inequality, and Technological Progress,”mimeo, MIT.

Green, Constance M. 1956. Eli Whitney and theBirth of American Technology. MA: Little,Brown.

Greenwood, Jeremy; Zvi Hercowtiz and PerKrusell. 1997. “Long-Run Implications ofInvestment-Specific Techological Change,”Amer. Econ. Rev. 87:3, pp. 342–62.

Greenwood, Jeremy and Mehmet Yorukoglu.1997. “1974,” Carnegie-Rochester Conf. Ser.Public Pol. 46, pp. 49–95.

Griliches, Zvi. 1957. “Hybrid Corn: An Explora-tion in the Economics of TechnologicalChange,” Econometrica 25, pp. 501–22.

———. 1969. “Capital-Skill Complementarity,”Rev. Econ. Statist. 51, pp. 465–68.

Grossman, Gene and Elhanan Helpman. 1991. In-novation and Growth in the Global Economy.MA: MIT Press.

Grout, Paul. 1984. “Investment and Wages in theAbsence of Binding Contracts: A Nash Bargain-ing Approach,” Econometrica 52, pp. 449–60.

Habakkuk, H. J. 1962. American and British Tech-nology in the Nineteenth Century. London:Cambridge U. Press.

Hanson, Gordon H. and Ann Harrison. 1995.“Trade Technology, and Wage Inequality: Evi-dence from Mexico,” NBER work. paper 5110.

Hayami, Yujiro and Vernon Ruttan. 1970. “FactorPrices and Technical Change in AgriculturalDevelopment: the U.S. and Japan, 1880–1960,”J. Polit. Econ. 77, pp. 1115–41.

Hornstein, Andres and Per Krusel. 1996. “CanTechnology Improvements Cause ProductivitySlowdowns?” NBER Macroeconomics Annual,pp. 209–59.

Ichniowski, Casey; Kathryn Shaw and GiovannaPrennushi. 1997. “The Effects of Human Re-source Management Practices on Productivity:A Study of Steel Finishing Lines,” Amer. Econ.Rev. 87:3, pp. 291–313.

James, John A. and Jonathan S. Skinner. 1985.


“The Resolution of the Labor-Scarcity Para-dox,” J. Econ. Hist. 45, pp. 513–50.

Jerome, Harry. 1934. Mechanization in History.NY: NBER.

Johnson, George. 1997. “Changes in Earnings In-equality: The Role of Demand Shifts,” J. Econ.Perspect. 11, pp. 41–54.

Jorgensen, Dale W. and Kevin J. Stiroh. 2000.“Raising the Speed Limit: U.S. EconomicGrowth in the Information Age,” BrookingsPap. Econ. Act. 1, pp. 125–236.

Juhn, Chinhui; Kevin M. Murphy and BrooksPierce. 1993. “Wage Inequality and the Rise inReturns to Skill,” J. Polit. Econ. 101, pp. 410–42.

Katz, Lawrence and David Autor. 2000. “Changesin the Wage Structure and Earnings Inequal-ity,” in Handbook of Labor Economics vol. III.Orley Ashenfelter and David Card, eds. Amster-dam: Elsevier.

Katz, Lawrence and Kevin Murphy. 1992.“Changes in Relative Wages: Supply and De-mand Factors,” Quart. J. Econ. 107, pp. 35–78.

Katz, Lawrence; Gary W. Loveman and David G.Blanchflower. “A Comparison of Changes in theStructure of Wages in Four OECD Countries”in Differences and Changes in Wage Structures.R. Freeman and L. Katz, eds. IL: U. ChicagoPress.

Kendrick, John W. 1961. Productivity Trends inthe United States. NJ: Princeton U. Press.

Kennedy, Charles. 1964. “Induced Bias in Innova-tion and the Theory of Distribution,” Econ. J.74, pp. 541–47.

Kiley, Michael. 1999. “The Supply of Skilled La-bor and Skill-Biased Technological Progress,”Econ. J. 109, pp. 708–24.

Kremer, Michael and Eric Maskin. 1999. “Segre-gation by Skill and the Rise in Inequality,” Har-vard mimeo.

Krueger, Alan. 1993. “How Computers HaveChanged the Wage Structure: Evidence fromMicrodata, 1984–1989,” Quart. J. Econ. 110,pp. 33–60.

———. 1997. “Labor Market Shifts and the PricePuzzle Revisited,” Princeton U. Ind. Rel. work.paper 375.

———. 1998. “Reassessing the View That Amer.Schools Are Broken,” Princeton U. Ind. Rel.work. paper 395.

———. 1999. “Measuring Labor’s Share,” Amer.Econ. Rev. 89, pp. 45–51.

Krueger, Alan B. and Jorn-Steffen Pischke. 1997.“Observations and Conjectures on the U.S. Em-ployment Miracle,” NBER work. paper 6146.

Krugman, Paul. 1994. “Part and ProspectiveCauses of High Unemployment,” in ReducingUnemployment: Current Issues and Policy Op-tions. MO: Fed. Reserve Bank Kansas City.

———. 1995. “Growing World Trade: Causes andConsequences,” Brookings Pap. Econ. Act. 1,pp. 327–77.

Krusell, Per; Lee Ohanian, Victor Rios-Rull andGiovanni Violante. 2000. “Capital Skill Comple-

mentary and Inequality,” U. Rochester mimeo;forthcoming Econometrica.

Lawrence, Robert and Matthew Slaughter. 1993.“Trade and U.S. Wages: Giant Sucking Soundor Small Hiccup?” Brookings Pap. Econ. Act. 2,pp. 161–226.

Leamer, Edward E. 1995. “Trade, Wages andRevolving-Door Ideas,” NBER work. paper4716.

Lee, David S. 1999. “Wage Inequality in the U.S.During the 1980s: Rising Dispersion or FallingMinimum Wage?” Quart. J. Econ. 114, pp. 941–1024.

Lloyd-Ellis, Huw. 1999. “Endogenous Technologi-cal Change and Wage Inequality,” Amer. Econ.Rev. 89:1, pp. 47–77.

Machin, Stephen and John Van Reenen. 1998.“Technology and Changes in Skill Structure:Evidence from Seven OECD Countries,”Quart. J. Econ. 113:4, pp. 1215–44.

Marglin, Stephen A. 1974. “What Do Bosses Do?The Origins and Functions of Hierarchy inCapitalist Production,” Rev. Radical Polit. Econ.3, pp. 33–60.

Mishel, Lawrence and Jared Bernstein. 1994. “Isthe Technology Black Box Empty? An Empiri-cal Examination of the Impact of Technology onWage Inequality and the Employment Struc-ture,” mimeo, Econ. Policy Institute.

———. 1998. “Technology and the Wage Struc-ture: Has Technology’s Impact Acceleratedsince the 1970s?” Res. Lab. Econ. 17, pp. 305–55.

Mishel, Lawrence; Jared Bernstein and JohnSchmitt. 1998. “Wage Inequality in the 1990s:Measurement and Trends,” mimeo, Econ. Pol-icy Institute.

Mobius, Marcus. 2000. “The Evolution of Work,”mimeo, MIT.

Mokyr, Joel. 1990. The Levers of Riches: Techno-logical Creativity and Economic Progress. NY:Oxford U. Press.

———, ed. 1993. The British Industrial Revolu-tion: An Economic Perspective. CA: WestviewPress.

Murnane, Richard J. and Frank Levy. 1996.Teaching the Basic New Skills. NY: Free Press.

Murphy, Kevin M.; W. Craig Riddell and Paul M.Romer. 1998. “Wages, Skills and Technology inthe United States and Canada,” in General Pur-pose Technologies. E. Helpman, ed. MA: MITPress.

Murphy, Kevin M. and Finis Welch. 1992. “TheStructure of Wages,” Quart. J. Econ. 107, pp.255–85.

———. 1993. “Industrial Change and the RisingImportance of Skill,” in Uneven Tides. S.Danziger and P. Gottschalk, eds. NY: RussellSage.

———. 2000. “Wage Differentials in the 1990s: Isthe Glass Half Full or Half Empty?” unpub. pa-per.

Nelson, Richard and Edmund Phelps. 1966. “In-vestment in Humans, Technological Diffusion


and Economic Growth,” AEA Papers Proceed-ings 56, pp. 69–75.

Newell, Richard; Adam Jaffee and Robert Stavins.1999. “The Induce Innovation Hypothesis andEnergy-Saving Technological Change,” Quart.J. Econ. 114, pp. 907–40.

Nickell, Stephen and Brian Bell. 1996. “The Col-lapse in Demand for the Unskilled and Unem-ployment across the OECD,” Oxford Rev. Econ.Pol. 11, pp. 40–62.

Neumark, David; Daniel Polsky and Daniel Han-sen. 1999. “Has Job Stability Declined Yet? Evi-dence for the 1990s,” J. Lab. Econ. 17:4, S29–64.

Nye, David E. 1990. Electrifying America: SocialMeanings of a New Technology, 1880–1940.MA: MIT Press.

OECD. 1994. Jobs Study. Vols. 1 and 2. Paris andWashington, DC: OECD.

Oliner, Stephen and Daniel Sichel. 1994. “Com-puters and Output Growth Revisited: How BigIs the Puzzle?” Brookings Pap. Econ. Act. 0:2,pp. 273–317.

Pierce, Brooks. 2001. “Compensation Inequality,”forthcoming, Quart. J. Econ.

Piketty, Thomas and Emannuel Saez. 2001. “In-come Inequality in the United States, 1913–1998,” MIT and Harvard mimeo.

Piore, Michael and Charles Sabel. 1984. The Sec-ond Industrial Divide. NY: Basic Books.

Reynolds, Lloyd G. 1951. The Structure of LaborMarkets; Wages and Labor Mobility in Theoryand Practice. NY: Harper.

Riddell, Craig. 1993. “Unionization in Canada andthe United States: A Tale of Two Countries,” inSmall Differences That Matter. David Cardand Richard Freeman, eds. IL: U. ChicagoPress.

Robbins, Donald J. 1995. “Trade and Inequality inSeven Developing Countries: Summary andSynthesis of Country Studies,” Harvard, HIIDmimeo.

Rodrik, Dani. 1997. Has Globalization Gone TooFar? Washington DC: Institute Int. Econ.

Romer, Paul M. 1990. “Endogenous TechnologicalChange,” J. Polit. Econ. 98:5, S71–S102.

Rubinstein, Yonna and Dani Tsiddon. 1999. “Cop-ing with Technological Progress: The Role ofAbility in Making Inequality So Persistent,” TelAviv U. work. paper.

Sachs, Jeffrey and Howard Shatz. 1994. “Tradeand Jobs in U.S. Manufacturing,” BrookingsPap. Econ. Act. 0:1, pp. 1–69.

Samuelson, Paul. 1965. “A Theory of Induced In-novations Along Kennedy-Weisacker Lines,”Rev. Econ. Statist. 47, pp. 444–64.

Scherer, Michael. 1984. Innovation and Growth.MA: MIT Press.

Schiff, Ze’ev and Ehud Ya’ari. 1989. Intifada, ThePalestinian Uprising—Israel’s Third Front. NY:Simon and Schuster.

Schmookler, Jacob. 1966. Invention and EconomicGrowth. MA: Harvard U. Press.

Schultz, Theodore. 1975. “The Value of the Abilityto Deal with Disequalibria,” J. Econ. Lit. 13,pp. 827–46.

Thesmar, David and Mathias Thoenig. 2000. “Crea-tive Destruction and Firm Organization Choice:A New Look into the Growth–Inequality Rela-tionship,” Quart. J. Econ. 115:4, pp. 1209–38.

Tinbergen, Jan. 1975. Income Difference: RecentResearch. Amsterdam: North-Holland.

Violante, Giovanni L. 2000. “Technological Accel-eration, Skill Transferability and the Rise in Re-sidual Inequality,” University College Londonmimeo.

Welch, Finis. 1970. “Education in Production,” J.Polit. Econ. 78, pp. 312–27.

———. 1973. “Black–White Differences in Re-turns to Schooling,” Amer. Econ. Rev. 63:5, pp.893–907.

Williamson, Jeffrey. 1990. Coping with CityGrowth During the British Industrial Revolu-tion. MA: Cambridge U. Press.

Williamson, Jeffrey G. and Peter H. Lindert. 1980.American Inequality: A Macroeconomic His-tory. NY: Academic Press.

Willis, Robert and Sherwin Rosen. 1975. “Educa-tion and Self-Selection,” J. Polit. Econ. 87, pp.817–36.

Wood, Adrian. 1994. North–South Trade, Employ-ment and Inequality: Changing Fortunes in aSkill Driven World. Oxford: Clarendon Press.

Woolf, Arthur G. 1984. “Electricity, Productivity,and Labor Saving: American Manufacturing1900–1929,” Explor. Econ. Hist. 21:2, pp. 176–91.

Zuboff, Shohana. 1988. In the Age of the SmartMachine: The Future of Work and Power. NY:Basic Books.


Journal of Economic Literature, Vol. XL March 2002 ... · have technological advances been...

Documents

Transcript of Journal of Economic Literature, Vol. XL March 2002 ... · have technological advances been...