THE ORIGINS OF TECHNOLOGY-SKILL
COMPLEMENTARITY*
CLAUDIA GOLDIN AND LAWRENCE F. KATZ
Current concern with the impact of new technologies on the wage structure
motivates this study. We offer evidence that technology-skill and capital-skill
(relative) complementarities existed in manufacturing early in this century and
were related to the adoption of electric motors and particular production methods.
Industries, from 1909 to 1929, with more capital per worker and a greater
proportion of motive energy coming from purchased electricity employed relatively
more educated blue-collar workers in 1940 and paid their production workers
substantially more. We also find a strong positive association between changes in
capital intensity and the nonproduction worker wage bill from 1909–1919 implying
capital-skill complementarity as large as in recent years.
The recent coincidence of computerization and a widening
wage structure has led many to conclude that new technologies
and human capital are relative complements and that large
injections of education may be necessary to reduce the impact of
advancing technology on inequality.1 In a related literature,
physical capital and skill have been shown to be relative complements
both today and in the recent past.2 These findings, taken
* Seminar participants at Harvard University, Johns Hopkins University, the
Massachusetts Institute of Technology, Cornell University, the University of
California at Los Angeles, the University of Chicago, Northwestern University, the
University of Michigan, the University of Texas, and the NBER Summer Institute
offered useful advice. The critical comments of Olivier Blanchard, John Dunlop,
Edward Glaeser, Andrei Shleifer, and three referees have made this a better paper.
Bridget Chen, Mireille Jacobson, and Serena Mayeri provided research assistance.
Arthur Woolf provided us with his sample of (four-digit) industry-level data from
the Census of Manufactures for 1909 to 1929. We thank all those mentioned.
Goldin and Katz [1996] summarizes some of the preliminary findings from an
earlier draft of this paper. This research has been supported by the National
Science Foundation. This paper was completed when Goldin and Katz were
resident scholars at the Russell Sage Foundation.
1. For models and (indirect) evidence on the relationship between new
technologies and the relative demand for skill, see, e.g., Bound and Johnson
[1992], Greenwood [1997], Greenwood and Yorukoglu [1997], Katz and Murphy
[1992], and Krusell et al. [1997]. Tinbergen [1975] has characterized the evolution
of the wage structure as a ‘‘race between technological development and access to
education.’’ Direct intervention in wage setting by governments and labor unions
can, of course, also alter the wage structure.
2. The empirical research covering the post-World War II period, beginning
with Griliches [1969] and summarized by Hamermesh [1993], supports the
capital-skill and technology-skill complementarity hypotheses. Bartel and Lichtenberg
[1987] find, from a panel of manufacturing industries from 1960 to 1980, that
the implementation of new technologies, proxied by the age of the capital stock,
increases the share of the highly educated in total labor cost. Doms, Dunne, and
Troske [1997] show that U. S. manufacturing plants utilizing more advanced
technologies employ more educated workers. Berman, Bound, and Griliches [1994]
௠ 1998 by the President and Fellows of Harvard College and the Massachusetts Institute of
Technology.
The Quarterly Journal of Economics, August 1998
Page 693
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
together, have prompted a widely noted conjecture that technological
progress and skill have always been relative complements.3
Even though physical capital and more advanced technologies are
now regarded as the relative complements of human capital, were
they so in the more distant past?
Some answers have already been provided.Aliterature on the
bias to technological change across history challenges the view
that physical and human capital were relative complements
throughout the industrial past. Many of the major technological
advances of the nineteenth century, according to this literature,
substituted physical capital, raw materials, and unskilled labor,
as a group, for highly skilled artisans.4 Rather than being the
relative complement to skill, physical capital was, for some time, a
relative complement of raw materials and, together with unskilled
labor, substituted for highly skilled individuals [Cain and
Paterson 1986; James and Skinner 1985].5 The prototypical
conclude that the shift in relative labor demand from unskilled (production) to
skilled (nonproduction) workers within U. S. manufacturing industries in recent
decades has been positively associated with the rate of growth of the capital-output
ratio and with the level of investment in R&D and computers. Berndt, Morrison,
and Rosenblum [1992] reach similar conclusions. Autor, Katz, and Krueger [1997]
find a substantial and growing wage premium associated with computer use
despite large increases in the supply of workers with computer skills (see also
Krueger, [1993]). Autor, Katz, and Krueger also find greater shifts toward
college-educated workers in industries with more rapid growth in computer usage
and computer capital per worker. Fallon and Layard [1975] present evidence
consistent with capital-skill complementarity using cross-country data at both the
aggregate and sectoral levels. Positive correlations have also been found between
the utilization of formal company training, and technological change and sectoral
capital intensity [Bartel and Sicherman 1995; Mincer 1989].
3. By skill we mean higher levels of education, ability, or job training. When
we use the term technology-skill or capital-skill complementarity, we mean that
skilled or more-educated labor is more complementary with new technology or
physical capital than is unskilled or less-educated labor.
4. The term ‘‘artisan’’ is used in this paper to mean a worker who produces
virtually the entire good in a production process containing almost no division of
labor.
5. James and Skinner [1985] divide industries in 1850 into two categories:
‘‘skilled’’ (e.g., woodworking) and ‘‘unskilled’’ (e.g., clothing). They find that in both
‘‘skilled’’ and ‘‘unskilled’’ industries raw materials were the relative complements
of physical capital, although the effect was greater in the ‘‘skilled’’ sector. More
importantly for the capital-skill complementarity hypothesis is that skilled labor
was the better substitute for capital in its sector than was unskilled labor in its.
Thus, an increase in capital (or raw materials) would have decreased the relative
demand for skilled labor. Cain and Paterson [1986] do not consider skill differences
but find, analogous to James and Skinner, that capital and raw materials were
relative complements and that both together substituted for labor. In the same
vein, Abramovitz and David [1973, 1996] characterize the bias to technological
change in general as shifting from Harrod labor-saving (capital-deepening) in the
nineteenth century to ‘‘unconventional-capital’’ deepening in the twentieth century.
Williamson and Lindert [1980], however, assume capital-skill complementarity
and generate, in their C.G.E. model, rising inequality with capital-deepening
during the nineteenth century.
QUARTERLY JOURNAL OF ECONOMICS694
Page 694
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
example is gun making. Cheap lumber inAmerica fostered the use
of wood lathes and displaced hand fitting in the production of gun
stocks by skilled woodworkers [Hounshell 1984]. The butcher,
baker, glassblower, shoemaker, and smith were also skilled artisans
whose occupations were profoundly altered by the factory
system, machinery, and mechanization.6
If technological advance and human skill were not relative
complements in the distant past but are today, when did they
become so? We argue that technology-skill complementarity
emerged in manufacturing early in the twentieth century as
particular technologies, known as batch and continuous-process
methods of production, spread. The switch to electricity from
steam and water-power energy sources was reinforcing because it
reduced the demand for unskilled manual workers in many
hauling, conveying, and assembly tasks.
We postulate that manufacturing production, for certain
products, began in artisanal shops, then shifted to factories (1830s
to 1880s), to assembly lines (early 1900s), and more recently to
robotized assembly lines.7 For other types of goods, however, the
shift may have been from artisanal shops or factories to continuous-
and batch-process methods (1890s and beyond).8 The production
process shifts did not affect all goods similarly, and some were
never manufactured by more than one method. But manufacturing
as a whole progressed in the fashion we posit: from artisanal
shops, to factories (also assembly lines), and then to continuousprocess
(also robotized assembly-line) or batch methods.
We have in mind rather distinct notions for each process,
following a rich literature in the histories of technology and
business. The distinction between the artisanal shop and factory
6. According to Sokoloff [1986], some initial deskilling, for example in
shoemaking, involved little capital, and no mechanization. It was, rather, of the
Smithian pin-factory variety. Landes [1972] takes an opposing view. Braverman
[1974], among others, argues that industrialization and mechanization served to
deskill a host of artisanal trades and to reduce the relative earnings of craftsmen.
7. See Atack [1987] and Sokoloff [1984] on the transition from the artisanal
shop to the factory in the nineteenth century. Both make the important point that
the transition was slow in some industries and depended not just on technological
change in manufacturing (often in the organization of work) but also on decreases
in transport costs. In some industries (e.g., boots and shoes, clothing, furniture,
leather, meatpacking, tobacco) a significant minority of value added was produced
in artisanal shops (Ͻ7 employees with no power source) even as late as 1870. And
in some (e.g., saddlery), the majority of value added came from artisanal shops in
1870.
8. The term ‘‘batch’’ refers here to production ‘‘in a batch,’’ generally used for
liquids (e.g., liquors), semisolid liquids (e.g., oleomargarine), or molten metals
(e.g., steel, aluminum). It is not to be confused with another usage, the production
of items in batches (e.g., clothing pieces) for later assembly.
TECHNOLOGY-SKILL COMPLEMENTARITY 695
Page 695
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
is mainly in the degree of division of labor. Factories are larger,
with more specialized workers and often more capital per worker.
Batch operations are used for processing liquid, semisolid, or gaseous
matters (e.g., chemicals, liquors, dairy products, molten metals, wood
pulp). Continuous-process methods, pioneered in the late nineteenth
century, are used for products requiring little assembly and having
few or no moving parts, such as oats, flour, canned foods (e.g.,
condensed milk, soup), soap, film, paper, matches, and cigarettes.
Continuous-process and batch methods (e.g., Bonsack’s cigarette
machine,Fourdrinierpapermakingmachine)are‘‘black-box’’technologies,
precursors of modern robotized assembly lines. Raw materials
are fed in, and finished products emerge, with few hands intervening
in production. A corps of machinists and mechanics, however, attend
the machinery. Assembly lines, including their robotized versions,
generally produce goods constructed from solid components. Note,
however, that assembly lines differ in important ways from
continuous-process (and batch) methods. Assembly lines, with
their vast quantities of human operatives, are the fully rationalized
versions of the factory, with its extreme division of labor.
Few products went through all the stages we describe, but
those that did are illuminating. Automobile production began in
large artisanal shops. Like the carriages that preceded them,
automobiles were first assembled by craftsmen who hand-fitted
the various pieces.9 Technological advances then led to standardized
and completely interchangeable parts that were assembled in
factories, later equipped with assembly lines as at Ford in 1913,
by scores of less-skilled workers. Much later, the robotized assembly
line appeared using relatively fewer less-skilled operatives
and more skilled machine-crewmen. In the history of automobile
production, the first technological advances reduced the relative
demand for skilled labor, whereas later advances increased it.
The question we address is how these technological shifts
affected the relative demand for skill. The transition from the
artisanal shop to factory production probably increased the
capital-output ratio, but most likely decreased the demand for
skilled relative to unskilled labor in manufacturing. The technological
advances that later shifted production from the factory (or
assembly line) to continuous-process methods further raised the
capital-output ratio but also served to increase the relative
9. Braverman [1974, p. 146] quotes Eli Chinoy: ‘‘Final assembly, for example,
had originally been a highly skilled job. Each car was put together in one spot by a
number of all-around mechanics.’’ See also Hounshell [1984].
QUARTERLY JOURNAL OF ECONOMICS696
Page 696
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
demand for skilled labor. Reinforcing these technological shifts
was electrification, the adoption of unit-drive systems, and the
automation of hauling and conveying operations which decreased
the demand for ‘‘common laborers.’’10
Our central point is that the technological shift from factories
to continuous-process and batch methods, and from steam and
water power to electricity, may have been at the root of an increase
in the relative demand for skilled labor in manufacturing in the
early twentieth century.11 We explore the origins of the transition
to technology-skill complementarity, believed to be in full blossom
today.12
We begin with a formal statement of our framework and the
conditions under which the technological changes we have in
mind will increase the relative demand for skill. We emphasize
that our model applies to manufacturing and that our evidence is
also primarily confined to that sector. Manufacturing employed
32.4 percent of the nonagricultural labor force in 1910 and 29.1
percent in 1940 [U. S. Department of Commerce 1975, series
D152–166]. The sector, moreover, affords the measurement of
inputs, outputs, and technological change over a long period.
Our framework predicts that industries adopting advanced
technologies (e.g., continuous-process and batch methods) in the
first part of this century should have employed production workers
with higher average skills and a larger share of nonproduction
(white-collar) workers. They should have been more capitalintensive
and relied on purchased electricity for a larger share of
their horsepower. We assess these predictions using the earliest
available national data on the educational levels of workers by
industry (viz., the 1940 U. S. census of population) and data on the
characteristics of detailed industries (from the 1909, 1919, and
1929 censuses of manufactures).
10. Jerome [1934] illustrates this point for many industries. By ‘‘unit-drive’’
we mean separate electric motors for each machine.
11. Our emphasis here is on the manufacturing sector. Reinforcing the shift in
manufacturing was an increased demand for educated labor to sell, install, and
service technologically advanced products. In the past ten years individuals who
sell computer software and hardware have changed from nerds to slick sales
personnel. Similarly, in the early 1920s those who sold radios had to deal with the
‘‘radio nuts,’’ those who built their own radios. Not much later, radio sales
personnel were hired to increase sales, not fraternize with the wireless amateurs
[Electrical Merchandising 1922].
12. A walk through most any factory today—one that assembles autos or their
parts, makes high-grade steel, or fabricates just about anything except clothing—
will reveal few production operatives but many capital-maintenance workers. In
one auto assembly plant we visited, an engineer proudly reported that any human
welder we saw would soon be replaced by robots.
TECHNOLOGY-SKILL COMPLEMENTARITY 697
Page 697
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
We document substantial differences in the education levels
of blue-collar production workers by industry in 1940 that cannot
be attributed to the geographic distribution of production. Industries
in 1909, 1919, and 1929 with more-skilled blue-collar
workers (proxied by contemporaneous mean wages for production
workers and education levels for blue-collar workers by industry
in 1940) had more capital per worker and used purchased
electricity to power a greater fraction of their horsepower. Many of
the industries our data reveal to be capital-intensive and higheducation
have been classified elsewhere as using continuousprocess
and batch methods [Chandler 1977]. We also examine
within-industry changes from 1909 to 1919 and find a positive
relationship between changes in capital-intensity and purchased
electricity utilization, on the one hand, and the wage-bill share of
white-collar workers, on the other.
Overall, the evidence we present from both across- and
within-industry analyses of data for 1909 to 1940 is consistent
with the notion that the transition from the factory to continuousprocesses
increased the relative demand for skilled workers. The
previous transition, from the artisanal shop to the factory, appears
to have involved an opposite force. Many industries that
remained artisanal (e.g., engraving, jewelry, clocks and watches)
had far lower capital intensity but higher worker skill (education)
than the majority that shifted to factory production.
I. A FRAMEWORK TO UNDERSTAND TECHNOLOGY-SKILL
COMPLEMENTARITY
In considering our argument, that shifts between production
processes change the relative demand for skill, it is useful to
envision manufacturing as having two distinct stages: a machineinstallation
and maintenance segment (termed ‘‘machine-maintenance’’)
and a production or assembly portion (termed ‘‘production’’).
The two stages together comprise ‘‘manufacturing.’’ Capital
and skilled (educated) labor, we will argue, are always complements
in the machine-maintenance segment of manufacturing for
any technology. Machinists, for example, are needed to install
machinery and make it run.13 The ‘‘workable’’machines created by
13. Machine-maintenance may itself have a life-cycle: ‘‘In the first stages of
such mechanization when machines are somewhat crude, their operation requires
the watchful care of skilled machinists. Likewise, to make those improvements
which increase machine efficiency, competent machinists are employed to observe
QUARTERLY JOURNAL OF ECONOMICS698
Page 698
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
skilled labor and raw capital are then used by unskilled labor to
create the final product in the production or assembly segment of
manufacturing.
The adoption of a particular technology may increase or
decrease the capital to output ratio. Whether or not its adoption
increases or decreases the relative demand for skilled workers
will depend on the degree to which the demand for skilled labor in
machine-maintenance is offset by the demand for unskilled labor
in production. We outline a more formal framework showing that
the transition from the artisanal shop to the factory probably
increased the capital-output ratio but decreased the demand for
skilled relative to unskilled labor. The shift from the factory to
continuous-process (or batch) methods, however, raised the capitaloutput
ratio and increased the relative demand for skilled labor.
But within any technology (artisanal, factory, continuous-process)
an increase in the ratio of unskilled to skilled wages will always
induce an increase in the capital-output ratio, the capital-labor
ratio, and the relative employment of skilled labor.
Our framework posits three technologies (one with two
phases): the artisanal shop or hand production (H), the factory (F)
which has a technologically advanced stage called the assembly
line (A), and continuous-process or batch (C) methods.14 There are
three inputs: raw (or physical) capital (K), skilled or educated
labor (Ls), and unskilled labor (Lu), with corresponding prices r,
ws, and wu. The manufacturing process contains two distinct
segments: (1) raw capital must be installed and maintained,
(‘‘machine maintenance’’) and (2) goods must be assembled or
created (‘‘production’’). All workers in the ‘‘production’’ segment
are unskilled, whereas all workers in ‘‘machine-maintenance’’ are
skilled. The machine-maintenance portion is Leontief, and, thus,
physical capital and human capital are always strict complements
in the creation of usable, workable machines for all technologies
(H, F or A, and C). The usable, workable machines are K*.15 The
the machines in operation under practical working conditions. But, as machines
increase in importance, they must be further improved in efficiency so as to require
little attention and a minimum number of stoppages for repairs or overhauling.
This increasing efficiency of the machine itself tends, in the long run, to eliminate
much of the work of that large corps of machinists which was required when
machines were first installed to displace hand workers’’ [Anderson and Davidson
1940, p. 228].
14. As we noted before, one can also include the robotized assembly line in
this last group.
15. We also assume, realistically, that K* cannot be shifted from one
technology to another but is specific to the technology for which it was created.
TECHNOLOGY-SKILL COMPLEMENTARITY 699
Page 699
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
production portion of manufacturing uses K* and Lu to manufacture
Q and is Cobb-Douglas. Thus, the creation of K* is a
separable part of Q production.16
The various technologies differ in straightforward and relatively
intuitive ways. Consider first the differences in the creation
of K*. Factories and continuous-process methods do not necessarily
require different ratios of skilled labor to capital to create
workable machines, but the artisanal shop, in which each artisan
maintains his own tools, has a higher ratio. Thus, we assume that
the same factor proportions are needed to create K* in the F (or A)
and C technologies, but that the H technology requires a larger
amount of Ls relative to K (as in Figure I, Part A). In the
production of Q, however, each of the technologies requires a
different ratio of unskilled labor to workable capital. The greatest
16. Our framework produces results that are qualitatively similar to those of
a more general production function such as a two-level CES of the form,
Q ϭ Ai[␣i (K*)␳i ϩ (1 Ϫ ␣i)Lu
␳i
]1/␳i ␳i Յ 1
K* ϭ [␤i (K)␪i ϩ (1 Ϫ ␤i) Ls
␪i
]1/␪i ␪i Յ 1
as long as ␳i Ͼ ␪i for i ϭ H, F, A, C. We analyze the special case for which ␳i ϭ 0 and
␪i ϭ Ϫϱ.
FIGURE I
A Simple Framework for Understanding the Relationships among Capital,
Technology, and Skill
QUARTERLY JOURNAL OF ECONOMICS700
Page 700
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
is in the factory (and assembly line), the next is for continuousprocess
methods, and the least is in the artisanal shop, which used
little unskilled labor. Thus, production of Q using the H technology
is the most intensive in K*, next is C, and last is F (or A), as in
Figure I, Part B.17 More formally, we assume that K* is generated
by
(1) K* ϭ min (␭i
1
· Ls, ␭i
k
· K) i ϭ H, F, A, C
such that ␭F, A,C
1
Ͼ ␭H
1
, and ␭F, A,C
k
Ͻ ␭H
k
, where 1/␭i
1
, for example, is
the unit skilled-labor requirement for K* creation in production
process i. Q production is, likewise, given by
(2) Q ϭ ␥i · (Lu)␣i (K*)(1Ϫ␣i)
i ϭ H, F, A, C
such that ␣H Ͻ ␣C Ͻ ␣F, A.
Even though skilled labor and capital are strictly complements
in machine-maintenance for all technologies, as in equation
(1), a change in technology (meaning production process)
need not increase the relative demand for skilled labor.18 The
reason can be seen with reference to Figure I (see also Table I).
Part B of Figure I is drawn so that all isoquants represent equal
amounts of Q. It has also been drawn for a very special case of K*
production, one in which the average cost of K* for all technologies
is the same (and thus for which all isocost lines in part B are
parallel). If r*i is the average cost of K* for technology i, the
condition for which r* is equal for the H and F (or A or C)
technologies is
(3) r/ws
ϭ [(␭F
1
)Ϫ1
Ϫ (␭H
1
)Ϫ1
]/[(␭H
k
)Ϫ1
Ϫ (␭F
k
)Ϫ1
].
That condition, although not necessary for our results, makes the
17. Our schematic representation of manufacturing as consisting of two
related stages, machine maintenance and production, is easily applied to the
factory, assembly-line, and continuous-process methods. But it makes the least
sense for the artisanal shop because artisans are involved in both processes. For
our schematic representation to make sense for the artisanal shop, it must be that
the machine-maintenance segment includes some production and that the production
segment finishes the product. In the manufacture of window-glass, which
remained artisanal into the early twentieth century, an artisan would maintain
his capital and blow the ‘‘cylinder.’’ Less-skilled workmen would then cut the
cylinder and make window-glass from it. Even though all window-glass was made
by hand in 1900, only 44 percent was in 1914; the industry had been transformed
by a new ‘‘factory’’ technology of rolled glass [Jerome 1934, p. 101].
18. By a technological advance from artisanal production to the factory
system, we mean an increase in ␥F relative to ␥H. The appearance of continuous or
batch processes, likewise, comes from an increase in ␥C relative to ␥F.
TECHNOLOGY-SKILL COMPLEMENTARITY 701
Page 701
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
geometry far simpler.19 Without it we will need a less restrictive
condition for one of the results.
Consider the factory system to be in its infancy in period 0, at
which time continuous processes will not have been invented and
the artisanal shop is the dominant mode of production. We begin,
therefore, at point H0 with production taking place only in the
artisanal shop. Note that just because K* is relatively high at H0
does not mean that much raw capital per unit of output is used. As
can be seen in Part A of Figure I, and as per equation (1), the H
technology uses relatively more Ls than do the other technologies
in producing K*.
Technological change makes the factory system a rival mode
of Q production, as can be seen from a Hicks-neutral technical
change that shifts F0 to F1 (resulting from an increase in ␥F). As
this type of technological change proceeds, firms will eventually
switch from the H to the F technology. The shift from artisanal
shops to factories initially causes an increase in (K/Q), in (K/Ls),
probably also in K/(Ls ϩ Lu), and a decrease in Ls /(Ls ϩ Lu).20 In
the more general case of our framework, in which the isocost lines
19. If the r*s were different, the isocost lines would not be parallel. The condition
simply states that the degree to which the cost of a unit of K exceeds that for a unit
of Ls must equal the degree to which F conserves on Ls, relative to H, compared
with how much H conserves on K, relative to F.
20. See Table I for the various conditions. We are considering here a small
factor-neutral technological change around F1. Greater technological change, as in
the movement to A, can decrease K/Q.
TABLE I
PREDICTIONS OF THE FRAMEWORK
Technological change K/Q K/(Ls ϩ Lu ) Ls/(Ls ϩ Lu)
(a) Shift from artisanal or hand trades (H) to
factory production (F) >a ?b <c
(b) Shift from factory (F) to assembly-line (A)
production (Hicks-neutral technical change) < = =
(c) Shift from assembly-line (A) to continuousprocess
(or batch) methods (C) > > >
K ϭ capital stock.
Ls ϭ skilled or more-educated labor.
Lu ϭ unskilled or less-educated labor.
a. The prediction is obtained when (␭F
k
/␭H
k
) Ͻ [(1 Ϫ ␣F)/(1 Ϫ ␣H)] · (rH
* /rF
*). That is, considering the
restrictive case discussed in the text of equal r* for H and F, the prediction is correct only if the higher
K*-intensity for the H technology is outweighed by the greater use of K in the creation of K* in the F
technology.
b. The impact of (a) on [K/(Ls ϩ Lu )] is ambiguous in the case when [Ls/(Ls ϩ Lu)] declines.
c. The prediction holds in the restrictive case of equal r* for H and F. When the r*s differ, the condition is
(rH
* /rF
*) Ͻ [(␣F/␣H)] · [(1 Ϫ ␣H)/(1 Ϫ ␣F)] · (␭F
1
/␭H
1
).
QUARTERLY JOURNAL OF ECONOMICS702
Page 702
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
in Part B are not parallel for the H and F technologies, a weaker
condition than that given by equation (3) is required for the
conclusion that a shift to the factory system increases the relative
demand for unskilled labor. The condition requires that the share
of unskilled labor in total product be sufficiently larger for the
factory technology than it is for the artisanal shop technology. It is
likely that such a condition held.21
Additional Hicks-neutral technical change shifts the F isoquant
to A2 through an organizational innovation: the assembly
line. The shift does not change the capital-labor ratio nor the
relative demand for skilled labor, but it does lower the capitaloutput
ratio.
The shift of most interest to us is that to continuous-process
methods, given by isoquant C3.22 Although its machine-maintenance
technology is identical to that of the A or F processes,
continuous-process methods have a higher output elasticity with
respect to K* than do the A or F processes (␣C Ͻ ␣F, A). That is, C
conserves on (unskilled) labor in the production segment and thus
uses K* more intensively. The shift from the A or F technologies to
C increases (K/Q), K/(Ls ϩ Lu), and Ls /(Ls ϩ Lu). With the adoption
of C we observe capital-skill complementarity and technologyskill
complementarity, at given factor prices, in the sense that the
shift to a more advanced technology is associated with increases
in both capital intensity and the relative employment of moreskilled
workers.23
21. More formally, the condition is that (rH*/ rF*) Ͻ (␣F/␣H) · [(1 Ϫ ␣H)/
(1 Ϫ ␣F)] · (␭F
1
/␭H
1
). Intuitively, the condition for the shift from H to F to decrease the
relative demand for skilled labor, given Q, will not occur if (rH*/rF*) is too high. The
cost of installed capital will be relatively high in the H sector when the cost of
skilled labor is great in comparison with raw capital. When that occurs, the H
sector will have an incentive to conserve on expensive K* in the production of Q,
by substituting unskilled labor for K*. But the condition will hold, even when
(rH*/rF*) is high, if ␣H is sufficiently low in comparison with ␣F. The shift to the F
technology will, then, still increase the relative demand for unskilled labor because
the marginal product of Lu is lower in the H technology (given Q and Lu). The effect
is further helped by the assumption that ␭F
1
Ͼ ␭H
1
. It does seem reasonable that, for
given Q and Lu, the marginal product of unskilled labor was lower in the artisanal
shop mode of production compared with the factory. After all, the reason for having
the factory system, with its more intricate division of labor, was to increase the
marginal product of unskilled labor.
22. The shift from A to C could result from technological advance (growth in
␥C relative to ␥A, F) or to an increase in the relative price of unskilled labor (increase
in wu/r*), as occurred with the expansion of high schools after 1910 [Goldin 1998]
and the reduction of immigration after 1914 [Goldin and Katz 1995].
23. Although we do not have a fourth input—raw materials—it is easy to
compare our framework with that in James and Skinner [1984]. Their skilled
sector is our artisanal sector, and it shifts from the H to the F technology, thereby
increasing K/Q, probably also K/L, but decreasing the fraction of the workforce
TECHNOLOGY-SKILL COMPLEMENTARITY 703
Page 703
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
The increased reliance on electricity as a source of horsepower
and the introduction of unit-drive appears to have had
effects similar to the movement to continuous-process methods.
The mechanization of the hauling and conveying of materials
decreased the relative demand for unskilled labor and increased
capital intensity, and the faster and hotter running of machinery
powered by electric motors required relatively more machinemaintenance
personnel.24
In summary, the role played by skilled labor in machinemaintenance
means that capital and skilled labor are relative
complements within any given manufacturing production process.
But capital and skilled workers may be relative complements
or substitutes in shifts among different manufacturing
processes depending on the nature of the technological change.
The movement from artisanal production to factories in the
nineteenth century involved the substitution of capital and unskilled
labor for skilled (artisanal) labor. The adoption of continuous-process
and unit-drive methods in the twentieth century
involved the substitution of capital and skilled (educated) labor
for unskilled labor.
A complementary hypothesis, advanced by Greenwood and
Yorukoglu [1997], is that all technological change increases the
relative demand for skilled labor, who facilitate the adoption of
new technologies. We, instead, emphasize the skill-biased or
unskilled-biased nature of distinct technological innovations rather
than simply the rate of technological change. Thus, we argue that
the deskilling involved in the transition from artisanal to factory
production offset the need for more flexible workers who assisted
in the introduction of factory methods. The observed evolution of
relative wages, employment, and wage bill shares by education
and occupation throughout the twentieth century would appear to
require a rapid and persistent secular growth in the relative
demand for more-educated workers [Goldin and Katz 1995; Autor,
Katz, and Krueger 1997; Johnson 1997]. The skill-biased nature
of electrification and computerization, as emphasized by our
that is skilled. The cause of the technological shift in their case is the lower relative
price of raw materials in the United States versus Britain.
24. According to Jerome [1934, pp. 402–403]: ‘‘The construction, installation,
repair, maintenance, and adjustment of machines . . . is requiring an everincreasing
army of relatively skilled men. . . . On the whole the shift will continue
to be from the emphasis on the trade skill typical of the handicraftsman to, on the
one hand, the alertness and intelligence required in handling fast and intricate
machinery and, on the other, to the more formal training required in the
engineering and production planning departments.’’
QUARTERLY JOURNAL OF ECONOMICS704
Page 704
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
framework, is consistent with the continuing rise in the relative
demand for more-skilled workers within manufacturing.
II. TECHNOLOGY-SKILL AND CAPITAL-SKILL COMPLEMENTARITY:
1909 TO 1940
We turn now to the historical record to observe when technology,
capital, and skill became relative complements in the U. S.
manufacturing sector. Ideally we would use plant-level data on
the nature of technology and worker skills. But because such data
are not available for our period, we examine, instead, industrylevel
measures of technology indicators (capital intensity and
purchased electricity use) and utilize proxies for worker skills
(education levels, occupational mix, and wages).
Our framework predicts that industries farther along in the
transition from the factory to continuous-process methods should
have employed production workers with higher average skill
levels (e.g., machinists) and a larger share of white-collar workers.
They should have been more capital-intensive and used
purchased-electricity for more of their horsepower. We assess
these predictions using both cross-sectional comparisons between
industries and panel data on the determinants of differences in
the rate of within-industry changes in the relative utilization of
skilled labor.
Various measurement issues arise in our study (see also the
Data Appendix). Our primary sources for technology indicators
and workforce information by detailed (approximately four-digit)
industry are the U. S. censuses of manufactures for 1909, 1919,
and 1929. Information on purchased electricity usage is available
in 1909, 1919, and 1929, but data on the capital stock are only
available for 1909 and 1919.25 The censuses of manufactures
provide data on employment and total (wage and salary) expenses
for broad occupational categories (all wage earners and various
categories of nonproduction workers).26 The data, however, are
25. Capital was also inquired of in 1914, but the question on the capital stock
ended in 1919. In 1939 a question was asked on capital investment. For a
discussion of the accuracy and consistency of the capital stock variable, and in its
defense, see Creamer, Dobrovolosky, and Borenstein [1960, Appendix A]. Capital
was to include only the owned portions of land, buildings, machinery, tools,
inventories, and working capital (e.g., cash) and to be assessed at book value.
26. The groups enumerated for nonproduction workers are proprietors and
firm members; salaried officers; superintendents, managers, engineers, and other
technical experts; and clerks, stenographers, salesmen, and other salaried employ-
ees.
TECHNOLOGY-SKILL COMPLEMENTARITY 705
Page 705
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
not disaggregated into subgroups of blue-collar workers, except by
sex and broad age categories, and no direct evidence is given on
the average skill or education of workers by industry and occupation.
But beginning in 1940 the U. S. census of population asked
highest grade completed, industry, and occupation.27 The 1940
population census allows us to identify industries that used
disproportionate numbers of higher-educated (e.g., high school
graduate) workers in blue-collar occupations, and the census of
manufactures can reveal whether these high-education industries
were more capital- or electricity-intensive from 1909 to 1929.
A. Educated Labor in Blue-Collar Occupations
It is generally believed that production workers in manufacturing
for much of the twentieth century were less educated than
average and did not vary much in their formal education by
industry (e.g., Nelson and Wright [1992]).After all, for much of the
early part of this century most would have been foreign born or
the children of the foreign born. There is some truth to these
claims. But that with which we take issue is that production
workers in manufacturing were an undifferentiated group with
respect to their formal education. In fact, we will show, using the
1940 population census, that there were wide differences by
industry in the educational attainment of blue-collar manufacturing
workers and that these differences were directly related to
industry characteristics.
We limit the analysis of the 1940 data to 18 to 34-years old
men in blue-collar occupations (craft, operatives, laborers, service)
working in the manufacturing sector.28 The age limitation is
imposed because the educational attainment reported by older
Americans in the 1940 census appears overstated. Among all 18to
34-year old, male blue-collar manufacturing workers in the
1940 PUMS (Public Use Microdata Sample), 27.6 percent stated
they completed twelve years or more of schooling, most of whom
were graduates of high schools.29 As a fraction of the total
27. The 1940 U. S. population census was the first U. S. decennial census to
ask educational attainment. Only three state censuses did so prior to 1940 (Iowa
1915, 1925; South Dakota 1915).
28. Our substantive findings are similar if we look, instead, at the educational
attainment of all workers, all male workers or at all blue-collar workers, rather
than restrict attention, as we do, to 18 to 34-year old blue-collar males.
29. The sample is restricted to the currently employed. In 1940 more than 10
percent of those in the labor force were unemployed, and another 4 percent were on
work-relief. For some reasons why the 1940 census overstates the educational
attainment of older Americans, see Goldin [1998].
QUARTERLY JOURNAL OF ECONOMICS706
Page 706
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
blue-collar workforce, the more-educated were disproportionately
found in particular industries. The industries, moreover, fit the
categorization we offered previously, for many used continuousprocess
and batch technologies (petroleum refining, dairy products,
paints and varnishes, nonferrous metals) or were hightechnology
industries and those producing recently innovated
goods (aircraft, business machinery, scientific and photographic
equipment).30
We list, in Table II, industries by the percentage of their
blue-collar male workers (18 to 34-years old) who were high school
graduates, giving those in the top and bottom 20 percent by
employment.31 The industries divide into two main categories. At
the low end of the education spectrum are the products of the first
industrial revolution as well as those used in construction: cotton,
woolen, and silk textiles, boots and shoes, lumber, stone, clay, and
cement. At the high end are certain products of the second
industrial revolution (e.g., chemicals, petroleum), many in the
machine-producing group, and some crafted in settings similar to
the artisanal shop (i.e., clocks, watches, jewelry, and even aircraft).
Finally, there is that perennial among high-education
industries: printing and publishing.32
Before making further sense of the findings regarding highand
low-education industries, we must address whether the
results are generated by geographic differences in both education
and industry or by age differences in employment and educational
attainment. We have regressed an indicator variable for high
school graduation (twelve or more years of schooling) on a full set
of age, state, metropolitan status, and industry dummies using
30. The list would probably include many others in the batch and continuousprocess
group if the 1940 census tabulated finer categories of industries, e.g.,
distilled liquors, pharmaceuticals. Industries are defined as high-technology if a
large percentage of their total labor force were engineers, chemists, and other
scientific personnel similar to currently used definitions.
31. The 1940 PUMS separately identifies 61 manufacturing industries that,
for the most part, correspond to current three-digit SIC industries and some of the
larger four-digit industries.
32. The results differ trivially if years of education instead of the percentage
graduating from high school were used. In addition, sectors other than manufacturing
display similar patterns. More educated blue-collar workers in the communications,
transportation, and public utilities sectors were found in telephone, wire and
radio, air transportation, petroleum and gasoline pipe lines, electric light and power,
and radio broadcasting and television. In retail trade new products and those that
were time-sensitive or more valuable per unit were sold, delivered, and serviced by
high school graduates. For example, drivers for jewelry stores and drugstores were
more educated than were other drivers. Blue-collar workers in radio stores, and
even gas-station attendants, were far more educated than the average blue-collar
worker.
TECHNOLOGY-SKILL COMPLEMENTARITY 707
Page 707
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
TABLE II
PERCENTAGE HIGH SCHOOL GRADUATES BY INDUSTRY, 18 TO 34-YEAR OLD MALE
BLUE-COLLAR WORKERS: 1940
Three-digit SIC
manufacturing
industries
% H.S.
grad.
Number
of obs.
Three-digit SIC
manufacturing
industries
% H.S.
grad.
Number
of obs.
High-education industries (from high
to low)
Low-education industries (from low
to high)
Top 20% by employment Bottom 20% by employment
Aircraft and parts 52.7 541 Cotton manufac-
tures
10.8 1512
Printing and pub-
lishing
44.7 1289 Tobacco 11.6 144
Office machinery 43.7 166 Logging 11.7 706
Petroleum refining 43.3 415 Sawmills and
planing mills
14.1 1941
Dairy products 43.2 417 Not specified textile
mills
15.6 128
Scientific and photographic
equipment
40.8 227 Silk and rayon
manufactures
16.6 350
Electrical machinery 40.5 977 Carpets and rugs 16.9 107
Misc. nonmetallic
mineral products
36.2 135 Misc. fabricated
textiles
17.0 94
Paints and varnishes 35.9 107 Cut-stone and stone
products
17.1 101
Clocks, watches,
jewelry
34.7 197 Misc. textile goods 17.6 117
Shipbuilding 34.4 528 Structural clay
products
18.8 271
Miscellaneous
machinery
33.5 1669 Cement and concrete,
gypsum, and
plaster products
19.2 263
Nonferrous metals 33.1 342 Hats, except cloth
and millinery
20.5 60
Dyeing and finishing
textiles
20.6 191
Misc. wooden goods 21.4 475
Footwear industries
except rubber
22.9 680
Woolens and
worsteds
23.1 368
The sample is limited to 18 to 34-year old, currently employed males in blue-collar occupations (craft,
operative, laborer, service) in manufacturing. The mean for the entire sample of 31,531 is 27.6 percent. The
industry names are those given in ICPSR [1984]. High-education (low-education) industries are obtained by
ranking industries by their share of 18 to 34-year old, male, blue-collar workers with twelve or more years of
schooling and selecting off industries from the top (bottom) until 20 percent of manufacturing employment (for
all workers) is represented. The 1940 PUMS sampling weights are used in all calculations.
Source. 1940 Public Use Micro-data Sample, 1/100; ICPSR [1984].
QUARTERLY JOURNAL OF ECONOMICS708
Page 708
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 swei
our sample of young male, blue-collar manufacturing workers
from the 1940 PUMS. The estimated coefficients on the industry
dummy variables can be viewed as industry differences in bluecollar
educational attainment adjusted for differences in urbanization,
the regional distribution of production, and age. A more
extreme adjustment uses the mean industry residuals from the
analogous regression, excluding the industry dummies. The adjusted
industry coefficients and the mean industry residuals
retain an almost identical set of rankings to that of the raw data
presented in Table II. Furthermore, the differences in educational
attainment of blue-collar workers across industries are substantial
even after adjusting for differences in urbanization, regional
location of production, and age structure. The standard deviation
of the actual share of high school graduates among young male
blue-collar workers across manufacturing industries is 0.086 as
compared with 0.080 for the adjusted industry coefficients and
0.071 for the mean industry residuals.
Another possibility is that the distribution of educated workers
by industry in 1940 was a product of the Depression. Firms
that employed high-educated workers, according to this logic,
would, in the face of decreased demand, have fired low-educated
workers and replaced them with their higher-skilled employees.
Such occupational changes are well documented for the Depression
years. But even in 1950 and 1960, most of the high-education
industries and sectors we identify in 1940 remained higheducation.
Our categorization is not simply a product of the
Depression.
We have established that, in 1940, firms in particular industries
hired disproportionate numbers of educated blue-collar
workers. The high- and low-education industries in Table II
appear to fit the prediction of the framework that high-technology,
continuous-process, batch-process, and artisanal modes of production
should use more skilled production workers than the factory
processes of the first industrial revolution.33 Our framework also
33. Data on employment by detailed (four-digit) industries from the 1909 to
1929 Censuses of Manufactures indicate that the majority of employees in five
1940 census industry categories (beverages, dairy products, grain-mill product,
paints and varnishes and petroleum refining) worked in industries classified by
Chandler [1977] as using continuous-process or batch-production methods. Consistent
with our framework, these industries employed a disproportionate share of
more-educated blue-collar workers in 1940: 36.0 percent of young, blue-collar
workers in these continuous-process and batch production industries had twelve
or more years of schooling as compared with 27.1 percent in the remainder of
manufacturing.
TECHNOLOGY-SKILL COMPLEMENTARITY 709
Page 709
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
suggests that industries with more educated blue-collar workers
should also have more skilled workforces overall as proxied by a
greater utilization of white-collar workers. In fact, there is a
strong positive correlation between the education level (share
with twelve or more years of schooling) of blue-collar workers and
the white-collar share of total employment across manufacturing
industries in 1940: the unweighted correlation is 0.61 and the
employment-weighted correlation is 0.73 for the 61 manufacturing
industries separately enumerated in the 1940 PUMS.
To investigate more directly capital-skill and technology-skill
complementarity in the 1909 to 1940 period, we have merged the
education data by industry from the 1940 population census with
that on industry attributes from the 1909, 1919, and 1929
censuses of manufactures. The industry categories in the 1940
population census are far broader than those in the censuses of
manufactures, and we have aggregated the earlier data to conform
to the 1940 categories for our analysis of the determinants of
industry variation in education levels (see the Data Appendix).
Table III presents regressions of the educational level of 18 to
34-year old male blue-collar workers in 1940 (adjusted for differences
in age composition, urbanization, and geographic distribution
of employment) on the capital to labor ratio (in 1909 and
1919), a measure of horsepower electrification (averaged over
1909, 1919, and 1929), and various controls for worker and
industry characteristics.34
The ratio of capital to wage-earners in 1909 and 1919 is
positively related to the education of workers by industry in 1940
(see Table III, columns (1) and (2)), and the effect is economically
significant. Increasing the capital to labor ratio by the equivalent
of its difference between, for example, the lumber and timber and
oleomargarine industries in 1909 increases the high school graduation
rate by seven percentage points or by 25 percent in 1940.35
Thus, more capital-intensive industries employed a more highly
educated labor force some twenty years later.
We find similar effects with respect to the use of purchased
electricity, entered either as the fraction of horsepower that was
34. The results are quite similar when we use the actual high school graduate
share of 18 to 34-year old, blue-collar males (unadjusted for differences in age
composition and geography) as the dependent variable.
35. These industries were chosen because oleomargarine production used a
continuous-process technology, whereas lumber and timber was mainly factory
produced. (Capital/wage-earners) in lumber and timber was $1693, and was $5871
in oleomargarine in 1909.
QUARTERLY JOURNAL OF ECONOMICS710
Page 710
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
TABLE III
EDUCATION, CAPITAL INTENSITY, AND ELECTRICITY USAGE, 1909 AND 1919
Adjusted fraction high school graduates among 18 to
34-year old males in blue-collar occupations, by industry
Capital intensity Capital and electrification
(1)
1909
(2)
1919
(3)
1919
(4)
1919
log (K/L) .0589 .0496 .0632 .0592
(.0169) (.0202) (.0194) (.0205)
% hp purchased electricity .199
(.0531)
log (hp purchased electricity/L) .0359
(.0151)
log (other horsepower/L) Ϫ.0405
(.0088)
log (total horsepower/L) Ϫ.0043
(.0149)
d log (employment)1909,1929 .0313 .0311
(.0126) (.0128)
% artisan .187 .189 .118 .122
(.0336) (.0355) (.0295) (.0295)
% female .142 .0932 Ϫ.0442 .0086
(.0537) (.0524) (.0636) (.0652)
% children Ϫ1.56 Ϫ1.56 Ϫ.660 Ϫ.804
(.490) (.515) (.463) (.487)
Constant .203 .193 .0921 .185
(.0238) (.0361) (.0366) (.0307)
Number of observations 57 57 57 57
R2 .482 .428 .711 .703
Standard errors are in parentheses. The unit of observation is a 1940-industry group. Industries from the
1909, 1919, and 1929 Censuses of Manufactures are aggregated up to the 1940 groupings, e.g., beverages in
1940 contains the 1909 categories of distilled, malt, and vinous liquors, and mineral and soda waters. Each
observation is weighted by the industry share of total blue-collar employment in manufacturing averaged over
1909, 1919, and 1929. The dependent variable is the adjusted percentage of 18 to 34-year old, male blue-collar
workers in the industry in 1940 who graduated high school. The ‘‘adjustment’’ is as follows. A regression of an
indicator variable for high school graduation (twelve or more years of schooling) on a full set of state dummies,
year-of-age dummies, indicator variables for central city and metropolitan area residence, and a full set of
three-digit 1940 census industry dummies was run on all employed 18 to 34-year old, male blue-collar workers
in the manufacturing, communications, transportation, and utilities sectors in the 1940 PUMS (sample
size ϭ 38,940, weighted by the 1940 PUMS sampling weights). The coefficient on each of the 57 industry
dummies is the adjusted high school graduate share of young, male blue-collar workers for each 1940 industry.
log (K/L) ϭ log of capital stock (000, in current dollars) per wage earner in each industry for 1909 and
1919, respectively, as indicated by the column headings.
% hp purchased electricity ϭ fraction total horsepower run by purchased electricity, averaged over 1909,
1919, and 1929.
log (hp purchased electricity/L) ϭ log of the horsepower of motors run by purchased electricity per wage
earner averaged over 1909, 1919, and 1929.
log (other horsepower/L) ϭ log of total horsepower of prime movers per wage earner averaged over 1909,
1919, and 1929; the total horsepower of prime movers includes the horsepower of steam engines, steam
turbines, water wheels, internal combustion engines, and so on.
log (total horsepower/L) ϭ log of total horsepower of prime movers ϩ the horsepower of motors run by
purchased electricity per wage earner averaged over 1909, 1919, and 1929.
d log (employment)1909,1929 ϭ change in log of total employment from 1909 to 1929.
% artisan ϭ fraction of wage earners in the 1940 industry categories who were in a disaggregated industry
(in 1909, 1919, 1929) classified as an artisanal trade, averaged over 1909, 1919, and 1929. Artisan is defined as
working in: gold and silver, leaf and foil; jewelry; photo-engraving; stereotyping and electrotyping; cardcut
design; wood and die engraving; glass; glass-cutting, staining and ornamenting; instruments; optical goods;
and statuary art. Printing and publishing is also included because of its special feature of demanding a literate
labor force.
% female ϭ average fraction of wage earners female in 1909, 1919, and 1929.
% child ϭ average fraction of wage earners child in 1909 and 1919.
Sources. U. S. Department of Commerce, Bureau of the Census [1913, 1923, 1933] supplemented with
data provided by Arthur Woolf. See Woolf [1980].
1940 Public-Use Microdata Sample, 1/100; ICPSR [1984].
TECHNOLOGY-SKILL COMPLEMENTARITY 711
Page 711
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
run by purchased electricity or as the log of the electricity variable
per wage earner (columns (3) and (4)). The effect, moreover, is
obtained even when the growth of industry employment from
1909 to 1929 is held constant. Production-worker skill was related
not just to industry growth and the resulting newness of plant and
equipment. Rather, the usage of purchased electricity affected the
relative demand for skill independent of industry growth. An 18
percentage point increase in the purchased electricity share of
horsepower (a one-standard-deviation change) is associated with
a 3.6 percentage point increase in the share of young blue-collar
workers who were high school graduates. Note that increases in
total horsepower did not have the same effect. The type of power,
not simply its amount, affected the skill level.
Electricity use in manufacturing grew rapidly during the
1909 to 1929 period. The percentage of all horsepower in manufacturing
driven by electric motors was 23 percent in 1909 but soared
to 77 percent by 1929 [Du Boff 1979]. Electric power was both
purchased from power plants and generated by firms using their
prime-movers (e.g., steam engines). Motors powered by purchased
electricity grew the fastest of the two, from 9 percent of all
horsepower in 1909 to 53 percent in 1929. Even though the growth
of purchased electricity was greater than that of all electricity,
generated electricity was still a sizable fraction of all horsepower
in 1929 (24 percent). For various reasons, the measurement of
horsepower driven by generated electricity is imprecise. Using the
best estimates provided for the two types, we find a much greater
impact on education levels from purchased electricity.36
There are many potential reasons for these findings, but no
hard evidence. One possibility is that purchased electricity was
associated with larger changes in the machinery of the factory
than was generated electricity. Electricity and separate motors
(unit-drive) enabled many industries to automate conveying and
hauling operations and thereby eliminate substantial numbers of
common laborers. Many industries were prompted to introduce
labor-saving methods with the onset of World War I, and chief
among them were iron and steel, brick manufacturing, pottery,
Portland cement, pulp and paper, rubber tires and tubes, slaugh-
36. The generated electricity variable is estimated using the procedure
described in Du Boff [1979, Appendix A], although see Jerome [1934] for another
method. The problem is that motors powered by generated electricity are often
rated above their actual use, which is limited by the horsepower of the prime
movers, whereas those powered by purchased electricity can be run at or above
their rating.
QUARTERLY JOURNAL OF ECONOMICS712
Page 712
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
tering and meatpacking, lumber manufacture and woodworking,
and mining [Jerome 1934; Nelson 1987; Nye 1990].37 Ample and
cheap electricity rendered feasible the production of various
materials, such as aluminum and other electrochemicals, that
disproportionately used skilled labor. Cheap electricity also encouraged
a more intensive use of machines thereby increasing demand
for the skilled personnel who maintained them.38 But purchased
electricity may also have been associated simply with newer
factories and technological advances built into a newer capital
stock.39
Note that we do not mean to imply that there was any direct
relationship between the individual workers in 1909 (or 1919) and
1940. In fact, education levels in Table III refer to men 18 to
34-years old in 1940 who could not have been in the 1909 labor
force. We are claiming, however, that there is something about
these industries that increased the value of secondary-school
education during the 1909 to 1940 period.
Secondary schooling spread rapidly after 1910. Less than 10
percent of youths had high school diplomas in 1910, but by the
mid-1920s to mid-1930s 30 percent to 50 percent did, depending
37. In iron and steel ‘‘the proportion of common laborers was cut approximately
in half from 1910 to 1931. The evidence is unmistakable,’’ notes Jerome,
‘‘that recent progress has eliminated unskilled labor to a much greater extent than
other grades’’ [1934, p. 63]. In all these industries, the use of conveyors, traveling
cranes, jitneys, carriers, industrial trucks, and other handling devices reduced the
relative demand for unskilled labor. The changes, moreover, were evident as early
as 1916 to 1920. We find, from the 1910 and 1940 PUMS of the census of
population, that laborers as a share of total manufacturing employment declined
from 23.6 percent in 1910 to 14.3 percent in 1940 [ICPSR 1984, 1989].
38. Electricity played a complex role in increasing the relative demand for
skilled workers. Although Nye [1990, pp. 234–235] concludes that electricity
increased the relative demand for skill, he also describes opposite effects. ‘‘As the
electrified factory evolved it required a different mix of labor and management . . .
more middle management; more engineers and technicians; fewer artisanal
workers; and a more complex grading of worker skills, with many more semiskilled
laborers . . . and far fewer unskilled workers. . . . Boy mule drivers in coal mines,
carriers in tire factories, or shovelers of raw materials in steel mills saw their work
taken over by electric locomotives, conveyors, and cranes . . . as a few skilled men
using expensive machines did work formerly performed by a mass of the
unskilled.’’ The increase in purchased electricity use, moreover, decreased the need
for prime movers and the skilled labor that serviced them. See also Du Boff [1979]
and Devine [1983] on the transition from mechanical to electric drive and the
introduction of group and unit drive motors.
39. The adjustment in Table III for the growth of the industry will account for
some of this factor. The newness of the capital stock concerns the replacement of
the shafts and pulleys of the older system of power with separate machines
(unit-drive) associated with electricity. Note that either generated or purchased
electricity could have accomplished the same transformation. Data on the average
horsepower of motors suggest that firms with purchased electricity switched more
completely to unit-drive (smaller motors), whereas those generating their own
appear to have used group-drive more.
TECHNOLOGY-SKILL COMPLEMENTARITY 713
Page 713
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
on region (see Goldin [1998] and Goldin and Katz [1997]). The
increase in formal education expanded the supply of skilled
manufacturing workers and altered their training. Before the
spread of high schools, most skilled machine-maintenance occupations
(e.g., machinist, electrician, technician) had cognitive skills
(e.g., algebra, geometry, trigonometry, mechanical drawing) learned
on-the-job by reasonably able individuals. These skills were
precisely those taught to young people in high schools. Formal
education substituted for a combination of ability and job training,
and the expansion of secondary schooling greatly increased
the supply of skilled manufacturing workers.
Were these more-educated blue-collar employees paid commensurately
more? The earliest nationally representative sample
with data on earnings and education is the 1940 PUMS, which
contains earnings data for 1939. We estimate, using the 1940
PUMS, that the ‘‘rate of return’’ to a year of schooling was 8.3
percent, in a standard human capital (log) earnings equation for
young, male blue-collar workers (white, 18 to 34-years old). The
return for a similar group of ordinary white-collar workers was
8.9 percent.40 The coefficient on years of schooling is .065 (s.e.
.0012) for our young, male blue-collar sample in an expanded (log)
earnings equation including a full set of industry dummies,
potential experience, and its square. Thus, the positive earnings
differential for more-educated blue-collar workers in 1939 arose
both because they ended up in higher-wage industries and
because they earned a substantial educational wage premium
within industries.
The average blue-collar wage in 1909, 1919, and 1929 is also
strongly and positively related to the education level of the
industry’s blue-collar workers in 1940. The coefficient on years of
schooling is 12 percent in 1909, 10 percent in 1919, and 17.5
percent in 1929 in regressions presented in Table IV of the (log)
average wage in each year on average years of schooling among
blue-collar workers in 1940. Although it is tempting to interpret
the coefficient on years of schooling as the rate of return to an
40. The coefficient on years of schooling for the blue-collar sample is .083 (s.e.
.0013) in a log (full-time equivalent) weekly earnings equation that includes
potential-experience and its square. The analogous regression for young, male,
ordinary white-collar workers (white, 18–34 years old, sales and clerical occupations)
in manufacturing is .091 (s.e. .0028). The blue-collar and ordinary whitecollar
samples contain 27,942 and 4,892 observations, respectively. The estimated
return in the blue-collar sample is 7 percent when a full set of state dummies are
also included.
QUARTERLY JOURNAL OF ECONOMICS714
Page 714
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
individual’s education, it is, instead, a combination of that and the
return to working in an industry having more educated workers.
Using data sets containing more complete information, such as
the censuses beginning with 1940, one can reconstruct the
‘‘poorer’’ data we are forced to use and measure both the individual
and industry effects. Almost identical magnitudes are
obtained to those we present for 1929 even though the return to an
individual’s years of education is far less.41 In light of these
comparisons we interpret the coefficients for 1909 to 1929 as
41. We construct a similar regression using micro-level data from the 1940
PUMS by aggregating across industries. The analogous regression of the log(average
wage, production workers) by industry on average years of schooling of male
blue-collar workers, 18 to 34-years old, in that industry (and the percentage female
of blue-collar workers) yields a coefficient of .185 (s.e. .013). In contrast, a
micro-level regression of individual log(earnings) on own schooling (plus a female
dummy) yields a coefficient of .040 (s.e. .0004). Because an experience variable is
not included, the coefficient on education is lower than in regressions that estimate
the return to education.
TABLE IV
RELATIONSHIP BETWEEN EARNINGS (1909, 1919, 1929) AND EDUCATION
(1940) FOR 18 TO 34-YEAR OLD BLUE-COLLAR MALES, BY INDUSTRY
Log (average annual, current $, wage)
in 1909, 1919, or 1929 industry
1909 1919 1929
Average years schooling among
blue-collar, 18 to 34-year old
males in 1940 industry grouping
.125
(.0111)
.0995
(.0128)
.176
(.0190)
Percentage (women ϩ children)
among wage earners in 1909 or
1919 industry; percentage female
in 1929 industry
Ϫ.494
(.0681)
Ϫ.605
(.111)
Ϫ.497
(.0917)
Constant 5.21 6.24 5.63
(.102) (.114) (.184)
Number of observations 191 191 191
Weighted mean of dependent
variable
6.24 7.00 7.13
R2 .699 .644 .657
␴ˆ .117 .133 .151
The number of 1940 industries is less than the number of 1909, 1919, or 1929 industries. Regressions are
weighted by the number of wage-earners in each 1909, 1919, 1929 industry. Numbers in parentheses are
Huber (White) standard errors allowing for grouped errors within 1940 industries. The education variable
(average number of years of school) is from the 1940 PUMS. The average wage is computed as the wage
bill/(average annual number of wage earners) for all years.
Source: U. S. Department of Commerce, Bureau of the Census [1913, 1923, 1933] supplemented with data
provided by Arthur Woolf. See Woolf [1980].
1940 Public Use Micro-data Sample, 1/100; ICPSR [1984].
TECHNOLOGY-SKILL COMPLEMENTARITY 715
Page 715
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
indicating that a return to education for blue-collar jobs existed,
but we are not certain of its magnitude.42
More direct evidence on the relationship between earnings
and education of blue-collar workers comes from a sample we have
collected from the Iowa state census of 1915, a unique set of
documents containing information on years of schooling, type of
schooling, annual earnings, and occupation (see Goldin and Katz
[1998]). Although industry information is not available, we can
identify nonfarm, blue-collar workers (laborers, operatives, and
craft workers). The coefficient on years of schooling in a log
(annual) earnings regression containing potential experience and
its square and an urban area dummy is .087 (s.e. .0045) for 18 to
34-year old, nonfarm, white male blue-collar workers in 1915.
When we separate years of education by type of schooling
(common school, grammar school, high school, and college), the
coefficient increases to .116 (s.e. .0093) for years of high school.43
Thus, the evidence from several sources indicates employers
apparently valued the education of blue-collar workers in the
1909 to 1939 period.
Educational attainment provides a useful proxy for the skill
levels of production workers, but it does not fully capture skill
differences related to on-the-job training, apprenticeship training,
and other sources. Earnings provide another (indirect) summary
measure that may capture all aspects of worker skill rewarded by
employers. The data on average earnings of blue-collar workers by
industry from the census of manufactures for 1909, 1919, and
1929 allow us to look at contemporaneous correlations of an
indicator of skills and industry characteristics. In Table V we
explore the relationship between average earnings per wage
earner and the capital and electricity variables in 1909, 1919, and
1929.44 We find, consistent with the previous results, a positive
relationship between the ratio of capital to labor and wages and,
similarly, a positive correlation between wages and the percent-
42. The wage regressions in Table IV (and those in Table V) use consistent
data on average annual earnings of blue-collar workers by industry for 1909, 1919,
and 1929. Because hours of work may have varied in a manner that could bias the
results, we have also estimated these regressions (for 1909 and 1919) using hourly
earnings, where hours worked in each industry is estimated from a distribution of
hours. The results are robust to the choice of hourly or annual data.
43. Our sample of 18 to 34-year old, white, male, nonfarm, blue-collar workers
for Iowa in 1915 contains 3021 observations. The mean of schooling in this sample
is 8.05 years, and 15.7 percent attended some high school.
44. We use the 1919 ratios of capital to labor for 1929.
QUARTERLY JOURNAL OF ECONOMICS716
Page 716
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
age of all horsepower from purchased electricity.45 Because World
War I caused a transitory compression in production-worker
wages, we prefer to concentrate on the 1909 and 1929 coefficients.
The magnitudes implied by the coefficients are substantial,
particularly with regard to purchased electricity use. The differ-
45. Within-industry variation over the 1909 to 1929 period leads to similar
conclusions to the between-industry regressions reported in Table V. There are
positive and significant relationships between the change in the (log) average
earnings of production workers from 1909 to 1929 and changes in capital intensity
and electricity use (conditional on controls for changes in the other covariates
included in the Table V regressions).
TABLE V
RELATIONSHIP BETWEEN PRODUCTION-WORKER EARNINGS AND INDUSTRY
CHARACTERISTICS, 1909, 1919, 1929
Log (average annual, current $, wage)
in 1909, 1919, or 1929 industry
1909
(s.e.)
1919
(s.e.)
1929
(s.e.)
Means, 1929
(s.d.)
log (K/L) .0910 .0417 .0480 1.44
(.0151) (.0169) (.0262) (.510)
% hp purchased electricity .439 .266 .546 .637
(.0556) (.0374) (.0548) (.211)
log (total horsepower/L) Ϫ.0213 Ϫ.00149 .0184 1.09
(.0115) (.0131) (.0189) (1.07)
log (employment/number of
establishments)
.0622
(.00633)
.0780
(.00577)
.0638
(.0103)
4.51
(1.08)
% female Ϫ.427 Ϫ.308 Ϫ.307 .210
(.0563) (.0613) (.0881) (.225)
% child Ϫ3.41 Ϫ6.91 Ϫ6.41 .014
(.377) (.697) (.927) (.016)
Artisan .136 .144 .273 .0563
(.0338) (.0336) (.0481) (.231)
Constant 5.99 6.65 6.56
(.0306) (.0386) (.0666)
Number of observations 228 225 228
Mean of weighted dependent
variable
6.23 7.03 7.15
R2 .791 .813 .667
␴ˆ .0991 .0973 .148
The average wage is computed as the (production-worker wage bill)/(average annual number of wage
earners) for all years. Independent variables are defined in Table III. Log (K/L) by industry for 1919 is used in
the 1929 regression; % child for 1919 is used in the 1929 regression. Regressions are weighted by the number
of wage earners in each year, as are the means for 1929.
Source. U. S. Department of Commerce, Bureau of the Census [1913, 1923, 1933] supplemented with data
provided by Arthur Woolf. See Woolf [1980].
TECHNOLOGY-SKILL COMPLEMENTARITY 717
Page 717
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
ence, for example, between the capital to labor ratios in the
oleomargarine and lumber and timber industries implies a 5
percent premium in wages for oleomargarine; the difference in
their purchased electricity use implies a 23 percent wage difference.46
The wage premiums we measure are largely due, we
believe, to compositional effects; a greater proportion of educated
workers was found in industries with more capital per worker and
more industry horsepower coming from purchased electricity.47
That high school graduate blue-collar workers were employed
in particular industries (more capital-intensive, using a greater
fraction of horsepower run by electricity, often producing newer
goods) and were paid substantially more, may come as a surprise.
Rarely is the education of production workers mentioned in the
labor history literature. Yet there is qualitative evidence, complementing
our empirical findings, that cognitive skills were highly
valued in various trades. High school graduates were sought
because they could read manuals and blueprints, knew about
chemistry and electricity, could do algebra and solve formulas,
and, we surmise, could more effectively converse with nonproduction
workers in high-technology industries. Blue-collar positions
requiring some years of high school or a diploma were described as
needing cognitive skills such as ‘‘good judgment,’’ ‘‘skilled in
free-hand drawing,’’‘‘special ability to interpret drawings,’’‘‘[familiarity]
with the chemical formulas,’’ ‘‘general knowledge of chemicals
used,’’ ‘‘[ability] to mix the chemicals.’’ More technical skills
were also valued such as ‘‘knowledge of electricity’’ and ‘‘of electric
wire sizes and insulation,’’ ‘‘technical knowledge of the properties
of glass,’’ ‘‘general knowledge of photography.’’ Printing establishments
required that beginners be ‘‘well versed in grammar,
spelling, punctuation,’’ and noted that ‘‘an elementary knowledge
46. The wage premiums use the 1929 coefficients. The actual difference in
wages is 33 percent. Oleomargarine had a capital to labor ratio in 1919 of $8759
and 69.4 percent of its horsepower came from purchased electricity; the numbers
for lumber and timber are $3028 and 27.5 percent.
47. Another interpretation is that the wage differentials reflect premiums for
identically skilled individuals working in more capital- and electricity-intensive
industries in which there was greater worker bargaining power and managerial
discretion (e.g., Slichter [1950]). That may well be the case. But the strong
correlation between wages in the 1909 to 1929 period and education in 1940, by
industry, provides evidence that the compositional effect matters. There is, as well,
a relationship between the high-education industries given in Table II and the
percentage of the industry’s 1910 labor force that was ‘‘machine related’’ (e.g.,
machinist, electrician).
QUARTERLY JOURNAL OF ECONOMICS718
Page 718
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
of Latin and Greek is helpful.’’48 High school educated youths were
hired into skilled occupations, but they were also sought for
ordinary positions in many of the ‘‘high-education’’ industries.
B. Technology-Skill Complementarity and Nonproduction Workers
The relative size of the nonproduction (white-collar) group
provides an alternative proxy for skill levels. A larger nonproduction
worker share of employment is likely to be associated with an
increase in the average amount of skill required of all workers,
both because white-collar jobs tend to have higher education
requirements and because technical nonproduction workers (engineers
and chemists) tend to work with more-educated production
workers [Allen 1996]. Continuous-process and batch methods
required more managerial and professional employees, relative to
production workers [Chandler 1977], and our estimations above
provide evidence that such processes also required relatively more
skilled blue-collar workers.
The data from the census of manufactures for 1909 and 1919
allow a further assessment of the importance of technology-skill
complementarity early in this century by examining whether the
relative utilization of nonproduction workers increased with
capital intensity and reliance on purchased electricity.49 Crossindustry
comparisons for both 1909 and 1919 indicate robust and
strong positive partial correlations between the nonproduction
worker share of employment (or labor costs) and both the (log)
capital to labor ratio and the fraction of horsepower from purchased
electricity (in regressions analogous to those in Table V
including controls for log horsepower per worker, demographics,
and other industry characteristics).
We next examine whether within-industry changes in capital
intensity and purchased electricity usage generate similar patterns.
The approach allows a direct comparison with results using
more recent data. Various studies have demonstrated a positive
relationship between skill upgrading and both advanced technol-
48. See, for example, the descriptions of positions in electrical machinery,
glass, medicinal manufacturing, paint and varnish, and the printing trades in
U. S. Department of Labor [various years, 1918–1921].
49. Chiswick [1979] addresses a somewhat different question using crossstate
aggregate manufacturing data for both 1909 and 1919 and fails to find much
evidence of aggregate capital-skill complementarity (where skilled workers are
salaried officers and managers). Our data are for the same years, but are
disaggregated by industry. In addition, we analyze the composition of skill demand
within detailed industries in which all nonproduction workers (including clerical
workers) are in the skilled group.
TECHNOLOGY-SKILL COMPLEMENTARITY 719
Page 719
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
ogy indicators (computer usage, computer investment, R&D intensity)
and the growth of the capital to output ratio [Berman,
Bound, and Griliches 1994; Autor, Katz, and Krueger 1997]. The
precise estimating equation is motivated by a model in which
capital, over each ten-year horizon, is the quasi-fixed factor and
nonproduction (skilled or educated) and production (unskilled)
labor are the variable factors.50 If the variable cost function (total
labor cost function) for industry j is translog and production
exhibits constant returns to scale, then cost minimization produces
an equation for the nonproduction labor share of total labor
costs (S) in time t of the form,
(4) Sjt ϭ ␣j ϩ ␾j t ϩ ␥i ln (wn/wp)jt ϩ ␳j ln (K/Q)jt,
where wn (wp) is the wage of nonproduction (production) workers,
(K/Q) is the capital to output ratio, and ␾j measures the rate of
skill-biased technological change in industry j. Differencing equation
(4) to eliminate industry fixed-effects, yields the following
estimating equation for the change in the nonproduction share of
the wage bill in industry j (under the assumptions that ␥j and ␳j do
not vary across industries or, more realistically, that the average
coefficients across industries are being estimated):
(5) dSjt ϭ ␤0 ϩ ␤1 dln (wn/wp)jt ϩ ␤2 dln (K/Q)jt ϩ ⑀jt.
The coefficient ␤1 is greater than or less than 0 depending on
whether the elasticity of substitution between nonproduction and
production labor is less than or greater than 1. The condition ␤2 Ͼ
0 implies capital-skill complementarity; ␤0 captures the crossindustry
average of the skill-bias to technical change and (␤0 ϩ ⑀jt)
is the industry-specific bias to technical change plus measurement
error. The intuitive interpretation of ␤2 Ͼ 0 is that when
capital and skilled labor are relative complements, industries
experiencing a greater increase in capital intensity are those with
a larger increase in the nonproduction worker share of total labor
costs.
We estimate equation (5) as well as a modification that drops
the relative wage variable, because cross-sectional wage variation
could largely reflect skill differences, not exogenous wage varia-
50. See Berman, Bound, and Griliches [1994] and Brown and Christensen
[1981].
QUARTERLY JOURNAL OF ECONOMICS720
Page 720
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
tion.51 The data are matched at the (pseudo) four-digit SIC
industry level—there are 256—for 1909 and 1919, the last year
until 1957 that the census of manufactures inquired about the
capital stock. Nonproduction workers are listed in four groups:
proprietors, officers, managers (often including professionals),
and clericals (often including other salaried workers). Their share
of manufacturing employment increased from 13.9 percent in
1909 to 15.7 percent in 1919. Among the data issues we had to
confront are the treatment of proprietor earnings and the choice of
output and capital deflators.52 In some industries (e.g., bakeries),
proprietors were a large fraction of all workers and probably
performed managerial and clerical duties. We have chosen to
exclude their income from the wage bill.53 Our input (capital) and
output (shipments; value added, viz., shipments minus materials
cost) variables are both nominal measures. In the absence of
industry-specific output prices and materials and investment
deflators, we use the aggregate WPI for the 1909 to 1919 period.54
The results for 1909 to 1919, in Table VI, reveal that the
coefficient on the change in the (log) capital-output ratio (␤2) is
consistently positive and significant regardless of other controls
and independent of the output measure (shipments or value
added).55 Inclusion of the change in the relative wage reduces the
coefficient somewhat, but far less so when we purge the wage
differential of certain compositional effects.56 We also find that the
51. We also add an output term to account for cyclical differences in the extent
to which nonproduction and production labor are quasi-fixed factors. This also
allows the production function to be nonhomothetic.
52. The weighting of the regression is another data issue. Because classification
errors are more of a problem for small industries, we weight by the wage bill
share. Qualitatively similar results obtain without weights, and median regressions
without weights also yield similar results.
53. Two methods of accounting for proprietor earnings can be offered: one
imputes their earnings as if they earned the average nonproduction wage in the
census, and the second assumes that proprietor income entirely reflects returns to
capital and should not be included in the overall wage bill (it was not supposed to
be included). We present the results under the latter assumption; they are robust
to either.
54. We have tested the sensitivity of applying industry-specific versus
aggregate deflators using data from the NBER Manufacturing Productivity
Database for 1959 to 1989 and find that the capital-skill complementarity
coefficient (␤2) is robust to the choice of deflator.
55. Qualitatively similar results are obtained by using the employment share
of nonproduction workers as the dependent variable. We prefer the cost share
measure because it implicitly adjusts for changes in relative labor quality and
hours of work.
56. To mitigate the influence of compositional effects on changes in the
sectoral skill premiums, we use the clerical wage for the nonproduction workers
and adjust the change in the skill premium for changes in percent female and child
labor among production workers (see column (7) of Table VI).
TECHNOLOGY-SKILL COMPLEMENTARITY 721
Page 721
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
inclusion of a measure of technological advance—the change in
the percentage of horsepower run by purchased electricity—does
not greatly affect the coefficient on dln(K/Q) and is itself positive
and significant.
How large we interpret the effects in Table VI to be will
depend on their magnitude more recently, when many have
concluded on the basis of such regressions that there is capitalskill
complementarity. We present such a comparison in Table VII.
The first two columns repeat the results from columns (1) and (2)
in Table VI. The remaining six give as identical as possible
estimations using quite similar four-digit industry data from the
NBER Productivity Manufacturing Database for 1959–1969, 1969–
TABLE VI
CHANGE IN THE NONPRODUCTION-WORKER SHARE OF WAGE BILL, 1909 TO 1919
Variable (1) (2) (3) (4) (5) (6) (7) (8)
dln (K/Q) .059 .051 .044 .054 .048
(.011) (.012) (.010) (.010) (.012)
dln (Q) Ϫ.008 Ϫ.013
(.004) (.005)
dln (K/VA) .051 .042 .032
(.010) (.011) (.010)
dln (VA) Ϫ.0080
(.0044)
dln (wn/wp) .094 .092
(.013) (.014)
dln (wc /wp ) .099
(.014)
d(% electrified) .046
(.020)
Constant .017 .018 .016 .017 .034 .032 .016 .008
(.003) (.003) (.003) (.003) (.004) (.004) (.003) (.005)
R2 .100 .111 .093 .105 .248 .227 .247 .108
␴ˆ .0367 .0366 .0368 .0366 .0336 .0340 .0336 .0357
The dependent variable is the change in the nonproduction-worker share of the wage bill between 1909
and 1919. The number of observations is 256 for columns (1)–(7) and 253 for column (8). An observation is an
industry, generally at the (pseudo for 1909 and 1919) four-digit SIC level. Regressions are weighted by the
average wage bill share between 1909 and 1919. Standard errors are in parentheses. The weighted mean of
the dependent variable is 0.006.
K ϭ capital stock.
Q ϭ value of products (or shipments).
VA ϭ value added.
wn ϭ average wage of nonproduction workers.
wp ϭ average wage of production workers.
wc ϭ average wage of clerical workers.
dln (wc/wp) ϭ change in the (log) wage of clerical and productive workers adjusted for the change in the
percentage female and percentage child workers of production workers.
d(% electrified) ϭ change in the fraction of total horsepower run by purchased electricity.
Sources. U. S. Department of Commerce, Bureau of the Census [1913, 1923].
QUARTERLY JOURNAL OF ECONOMICS722
Page 722
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
1979, and 1979–1989.57 The striking finding is that the coefficient
on the change in the (log) capital-output ratio (␤2) is larger for the
1909–1919 period than more recently. Only one coefficient—that
for 1979–1989 in the estimation with the change in shipments
variable—is larger. The comparison of the results for 1909–1919
with those more recently suggests a strong movement during the
1910s toward capital-skill complementarity.
III. TECHNOLOGY-SKILL COMPLEMENTARITY: HISTORICAL
INSIGHTS AND CONCLUSION
We began with the notion, garnered from various sources,
that many technological advances of the nineteenth century
57. For consistency with the data from 1909 to 1919, the aggregate PPI for
finished goods is used to deflate all nominal quantities in the 1959 to 1989
comparison period.
TABLE VII
CHANGE IN NONPRODUCTION-WORKER SHARE OF WAGE BILL: 1909–1919,
1959–1969, 1969–1979, 1979–1989
Variable
1909–1919 1959–1969 1969–1979 1979–1989
(1) (2) (3) (4) (5) (6) (7) (8)
dln (K/Q) .059 .051 .005 .018 .023 .040 .024 .061
(.011) (.012) (.005) (.006) (.006) (.009) (.008) (.011)
dln (Q) Ϫ.008 .026 .021 .036
(.004) (.006) (.008) (.007)
Constant .017 .018 .010 Ϫ.003 .010 .005 .037 .030
(.003) (.003) (.002) (.004) (.002) (.003) (.002) (.002)
R2 .100 .111 .002 .040 .026 .040 .018 .071
␴ˆ .0367 .0366 .0354 .0348 .0415 .0412 .0438 .0427
Number of
observations 256 256 450 450 450 450 450 450
Mean of
weighted
dependent
variable .006 .006 .011 .011 .013 .013 .037 .037
The dependent variable is the change in the nonproduction-worker share of the wage bill between the
years given. An observation is an industry, generally at the (pseudo) four-digit SIC level. Regressions are
weighted by the average wage bill share between the years given. Standard errors are in parentheses.
Nominal quantities are deflated by the WPI for 1909–1919 and by the PPI for finished goods in the other three
periods.
K ϭ capital stock.
Q ϭ shipments.
Sources. U. S. Department of Commerce, Bureau of the Census [1913, 1923] for 1909–1919 data. NBER
Manufacturing Productivity Database, derived from the Annual Survey of Manufactures, as described in
Bartelsman and Gray [1996], for 1959 to 1989 data. Deflators. 1909, 1919: U. S. Department of Commerce,
Bureau of the Census [1975], series E-40. All other years: Economic Report of the President [1995], Table B-64,
p. 347.
TECHNOLOGY-SKILL COMPLEMENTARITY 723
Page 723
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
increased the relative demand for the lesser-skilled in manufacturing
but that studies using recent data generally find the reverse.
Capital and skills, and technology and skills, are deemed relative
complements today. Our goal was to locate the origins, and thus
the sources, of technology-skill and capital-skill complementarity
in manufacturing. We have demonstrated that such relative
complementarities existed from 1909 to 1929 and were associated
with technologies, such as continuous-process and batch methods,
and with the adoption of electric motors.58 To motivate the
empirical work, we devised a framework in which capital is
always a relative complement of skilled labor in the creation of
workable machines but in which unskilled labor is used with
machines to create output. Manufacturing-wide capital-skill
complementarity arises from a shift between particular technologies.
Our evidence has pertained mainly to the period since 1909
and to the transition from the factory (or assembly-line) system to
continuous-process and batch methods and electric motors in
manufacturing.
One often hears that we are living in a time of extraordinary
technological change. Yet the technologies that appeared or
diffused in the two decades around 1915 may have been more
consequential. Manufacturing horsepower in the form of purchased
electricity rose from 9 percent in 1909 to 53 percent in
1929; similar changes swept residential use. New goods proliferated:
automobiles, airplanes, commercial radio, aluminum, synthetic
dyes, and rayon; household electric appliances (e.g., refrigerators
and washing machines); office machinery (e.g., calculators,
dictating machines, and copying equipment). New techniques
improved the production of rubber, plate glass, gasoline, canned
condensed milk, and factory-made butter. Of the goods mentioned
all but the auto are among the higher-education industries we
identify for 1940 and which appear also to have been the
higher-skill industries in the 1910s and 1920s. Looking back at
Table II, the high-education industries include aircraft, office and
store machines, petroleum refining, dairy products, electrical
58. Using similar data from the 1890 census of manufactures, we have also
found a positive relationship between the ratio of capital to labor and production
worker wages, and a positive correlation between the growth of the nonproduction
worker wage share and capital’s share from 1890 to 1909. Thus, the origins of
capital-skill and technology-skill complementarity may have been earlier than we
discuss here, consistent with Chandler’s [1977] dating of the beginnings of
continuous-process and batch methods.
QUARTERLY JOURNAL OF ECONOMICS724
Page 724
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
machinery, paints and varnishes, miscellaneous machinery, and
nonferrous metals.
Input price changes may have been reinforcing in enticing
firms to adopt particular technologies. The decline in the price of
purchased electricity was the greatest change and perhaps most
significant.59 Cheaper power, purchased electricity, and unit-drive
meant that firms could adopt different types of capital and
production processes.60 Although the advent of electric utilities,
and thus purchased electricity, meant that less capital was needed
to generate power within firms, production-capital (as opposed to
power-generating capital) could run faster and hotter, thereby
requiring greater numbers of skilled workers to maintain it.
The technologically forward industries of the day also grew
the fastest. If we define a high-education industry (see Table II) as
one in which more than 33 percent of the young male blue-collar
workers had a high school degree in 1940, the share of manufacturing
employment in these industries rose from 20.7 percent in 1909
to 26.7 percent in 1929. The share in low-education industries
(defined as less than 23 percent with high school degrees in 1940)
declined from 29.0 percent in 1909 to 20.5 percent in 1929. The
share of manufacturing employment in the top five (two-digit)
industries by education (petroleum, chemicals, electrical machinery,
printing and publishing, and scientific instruments) continued
to expand from 10 percent to 16 percent over the 1910 to 1940
period [Goldin and Katz 1995, Table A4].61 Similar shifts have
occurred more recently, reinforcing the effects of capital- and
technology-skill complementarity.
If, according to our estimates, there was as much skill-biased
technical change in the manufacturing sector from 1909 to 1929
59. According to Woolf [1980, Table 2.1], nominal price declines for electricity
were 3.4 percent, and real declines exceeded 5 percent average annually from 1910
to 1929.
60. See Du Boff [1979] and Devine [1983] on why unit-drive (separate motors
for each tool) diffused slowly until the 1920s. Before electricity, mechanized firms
were powered by prime movers, and the power was often distributed throughout
the factory by a complex series of shafts and pulleys. Electricity and small motors
enabled firms to replace this system with one that was cleaner and safer, and in
which tools were more efficiently arranged. But large capital investments were
tied up in the previous arrangement.
61. Within-industry skill upgrading associated with the adoption of new
technologies was probably a more important source of increases in the relative
demand for skill than were shifts of employment between industries. For example,
the nonproduction worker wage bill share grew from 21.5 percent in 1909 to 23.6
percent in 1929. Of that change, we find, using our matched sample of industries
from the 1909 and 1929 Censuses of Manufactures, that approximately two-thirds
occurred within four-digit industries. This finding is similar to that observed in
manufacturing over the past few decades [Berman, Bound, and Griliches 1994].
TECHNOLOGY-SKILL COMPLEMENTARITY 725
Page 725
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
as more recently, was there also as much widening in the gap
between the skilled and the unskilled?62 We have reported elsewhere
that the wages of ordinary white-collar workers fell relative
to those for all production workers from the mid-1910s to the
early-1920s and then remained at this lower level through 1940
[Goldin and Katz 1995]. Because ordinary white-collar workers in
1910 were mainly high school graduates, the wage differences by
occupation are similar to those by education. That is, the wage gap
between the high school educated and the less educated first fell
and then remained at that level for some time. Rather than
widening during a period of skill-biased technical change, the
wage gap actually narrowed and then remained stable.
What accounts for the contrast between the labor market
response to skill-biased technical change from 1909 to 1940 and
today? One potential difference is that the supply of educated
workers greatly and rapidly expanded at an increasing rate
throughout the earlier period. In thousands of school districts
across the United States high schools mushroomed, curricula
modernized, and enrollments soared from 1910 to 1940 [Goldin
1998; Goldin and Katz 1997]. The vast majority of those who
graduated high school in the 1910s but did not continue to college,
entered a host of ordinary white-collar occupations, such as
clerks, bookkeepers, secretaries, and various sales positions. The
large premium to high school educated workers that existed at the
turn of the century, and which may have prompted the high school
expansion, was markedly reduced by the 1920s, although it
remained substantial.Aonce-expensive commodity—a high school
graduate—suddenly became a more reasonably priced input and
this change in relative input prices may have further fueled
skill-biased technical change. Most importantly, as the high
62. Note that the impact of skill-biased technical change is potentially even
more pervasive. Chandler [1977] notes that the managerial revolution was
brought about by many of the technologies we believe increased the relative
demand for educated labor on the shop-floor. The demand for more-educated farm
workers must also have increased as agricultural machinery became more
complex. Goldin and Katz [1995] infer from data on movements in the relative
wages of white-collar workers and relative skill supplies (under the assumption of
a stable elasticity of substitution between more- and less-skilled workers) that the
relative demand for workers with at least a high school degree grew rapidly from
1890 to 1940 in the manufacturing sector and the aggregate economy. In fact, the
rate of growth of the (log) relative wage bill share of nonproduction workers in
manufacturing (a measure of the growth in the relative demand for nonproduction
workers with an elasticity of substitution between nonproduction and production
workers of close to one) was almost as large from 1890 to 1929 (.0103 log points per
year) as it was in the more recent period of generally acknowledged rapid growth
in the relative demand for skills from 1959 to 1989 (.0128 log points per year).
QUARTERLY JOURNAL OF ECONOMICS726
Page 726
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
school movement continued to spread across America in the
1920s, the wage gap between the high school educated employee
and the less-educated worker may have been kept in check in an
era of skill-biased technical change.
DATA APPENDIX
Manufacturing Data 1909, 1919, 1929
The primary source for our data on the manufacturing sector
is the U. S. Census of Manufactures for the years 1909, 1919, and
1929. We use U. S. aggregates by industry for the capital stock
(1909, 1919 only), wage earners (average number employed over
the year), female (Ͼ15 years) and child (Ͻ16 years) workers,
expenses on all wage earners, employment of various groups of
nonproduction workers (proprietors, salaried officials, managers,
and clerks), expenses on nonproduction workers (salaries), total
employment, number of establishments, value of output, materials
expenses, value added, total primary horsepower, horsepower
of motors run by purchased electricity, and horsepower from
generated electricity, from U. S. Department of Commerce, Bureau
of the Census [1913, 1923, 1933]. For 1909 and 1929 we
employ a data set generously provided by Arthur Woolf (see Woolf
[1980]) for the power variables and some others, and matched the
information by industry to our data. We have also constructed a
similar data set used in some of our supplemental analyses from
the 1890 Census of Manufactures [U. S. Census Office 1895].
The census surveyed establishments (defined as ‘‘separate
plants operated under the same ownership . . . in different states
or counties and for those in different cities having a population of
10,000 or more’’) and, beginning in 1909, limited its purview to
‘‘factories’’ as opposed to ‘‘the neighborhood, hand, and building
industries.’’ Establishments with annual gross output of less than
$500 were not surveyed in 1909 and 1919, and the cutoff was
raised to $5000 in 1929 (see Fabricant [1940] on the absence of
biases resulting from the change). The data were averaged by the
census at the industry level across the entire United States. A
comprehensive standard industrial classification (SIC) system
had not yet been established by 1929 and the industries listed in
the three censuses are at various levels of aggregation depending
on the value of gross output produced. There is, however, considerable
overlap from one decade to the next. Because by 1929 the SIC
TECHNOLOGY-SKILL COMPLEMENTARITY 727
Page 727
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
system was being developed (see Beales [1930]), we have useful
sources for making concordances across the three census years
and also with the 1940 census of population.
The 1909 Census of Manufactures contains information on
260 separate industries. That for 1919 has more detail and lists
354 separate industries, many of which are also tabulated as
aggregated industries (e.g., tobacco, cigars, and cigarettes). The
Census of Manufactures for 1929 contains data on 326 separate
industries. During the period we study some industries emerged
(e.g., airplanes in 1919), whereas various smaller ones were later
grouped into a miscellaneous category (e.g., wood carpet). Still
other industries were separated over time (e.g., iron and steel,
processed; iron and steel, steel works and rolling mills), yet others
were merged (e.g., beverages). There are 61 manufacturing industries
in the 1940 Census of Population PUMS. Most are similar to
current three-digit SIC classifications (e.g., beverage industries).
We use various concordance procedures to produce consistent
samples that are as large as possible in all the empirical work
reported. That for Table III takes the 1909, 1919, and 1929
industries and maps them into 57 of the 61 industrial groups
given in the 1940 PUMS. To do so, we relied on U. S. Department
of Commerce, Bureau of the Census [1940] to link the 1909, 1919,
1929, and 1940 industries, and the insights of Fabricant [1940]
concerning products that underwent change from 1909 to 1940.
The procedure followed for Tables IV and V was to use
industries at the same level of aggregation for each of the three
years, 1909, 1919, and 1929, although for 1919 three of the
industries were dropped due to missing information on purchased
electricity. Across the three years (1909, 1919, and 1929) we can
match 228 (pseudo) four-digit industries covering 99 percent of
manufacturing employment in those years. The census documentation
provides detailed information on the relationship among
the industries in the three Census of Manufactures; we also
consulted Beales [1930]. For Table IV we were able to match 191 of
these industries to the industrial groups in the 1940 PUMS.
The concordance used for Tables VI and VII matches 256 of
the 260 industries in 1909 to the industries in the 1919 Census of
Manufactures, some of which had to be aggregated up to the 1909
categories. Because the 1919 data are given at a more disaggregated
level than those for 1909, the concordance was done by
mapping the 1919 industries into those for 1909. Three of these
QUARTERLY JOURNAL OF ECONOMICS728
Page 728
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
matched industries are missing purchased electricity data in
1919.
Manufacturing Data 1959, 1969, 1989
The data on the nonproduction worker share of the wage bill,
the capital stock, and shipments for 1959, 1969, and 1989 for 450
four-digit manufacturing industries are from the NBER Manufacturing
Productivity Database. Documentation on this database is
available in Bartelsman and Gray [1996]. For comparability with
the 1909 and 1919 data, we first convert the real capital stock
variable available in the NBER Productivity Database for 1959–
1989 into a nominal capital stock variable using the industryspecific
price deflators for new investment. We then deflate each
industry’s nominal capital stock by the aggregate PPI for finished
goods for 1959 to 1989.
Education and Earnings Data 1940
Data on the education of manufacturing workers in 1940 and
the earnings of manufacturing workers in 1939 come from the 1%
Public Use Micro-Sample of the 1940 Census of Population
[ICPSR 1984]. High school graduates are defined as those with
twelve or more years of schooling; years of schooling are given by
the highest grade completed. Weekly earnings in 1939 are computed
as the ratio of annual wage and salary income to the
number of (equivalent full-time) weeks worked. The samples for
1939 earnings regressions are limited to wage and salary
employees.
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Abramovitz, Moses, and Paul A. David, ‘‘Reinterpreting Economic Growth: Parable
and Realities,’’ American Economic Review, LXIII (1973), 428–439.
Abramovitz, Moses, and Paul A. David, ‘‘Technological Change and the Rise of
Intangible Investments: The U. S. Economy’s Growth-Path in the Twentieth
Century,’’ in Employment in the Knowledge-Based Economy, D. Foray and
B.-A. Lundvall, eds. (Paris: OECD, 1996).
Allen, Steven G., ‘‘Technology and the Wage Structure,’’ NBER Working Paper No.
5534, 1996.
Anderson, H. Dewey, and Percy E. Davidson, Occupational Trends in the United
States (Stanford, CA: Stanford University Press, 1940).
Atack, Jeremy, ‘‘Economies of Scale and Efficiency Gains in the Rise of the Factory
in America, 1820–1900,’’ in Quantity and Quiddity: Essays in U. S. Economic
History, P. Kilby, ed. (Middletown CT: Wesleyan University Press, 1987), pp.
286–335.
TECHNOLOGY-SKILL COMPLEMENTARITY 729
Page 729
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
Autor, David, Lawrence F. Katz, and Alan Krueger, ‘‘Computing Inequality: Have
Computers Changed the Labor Market?’’ NBER Working Paper No. 5956,
1997.
Bartel, Ann P., and Frank Lichtenberg, ‘‘The Comparative Advantage of Educated
Workers in Implementing New Technology,’’ Review of Economics and Statistics,
LXIX (1987), 1–11.
Bartel, Ann P., and Nachum Sicherman, ‘‘Technological Change and the Skill
Acquisition of Young Workers,’’ NBER Working Paper No. 5107, 1995.
Bartelsman, Eric, and Wayne Gray, ‘‘The NBER Manufacturing Productivity
Database,’’ NBER Technical Working Paper No. 205, 1996.
Beales, LeVerne, Industry Classifications. Census of Manufactures: 1929 (Washington,
DC: G.P.O., 1930).
Berman, Eli, John Bound, and Zvi Griliches, ‘‘Changes in the Demand for Skilled
Labor within U. S. Manufacturing Industries: Evidence from the Annual
Survey of Manufacturing,’’ Quarterly Journal of Economics, CIX (1994),
367–397.
Berndt, Ernst R., Catherine J. Morrison, and Larry S. Rosenblum, ‘‘High-Tech
Capital Formation and Labor Composition in U. S. Manufacturing Industries:
An Exploratory Analysis,’’ NBER Working Paper No. 4010, 1992.
Bound, John, and George Johnson, ‘‘Changes in the Structure of Wages in the
1980s: An Evaluation of Alternative Explanations,’’ American Economic
Review, LXXXII (1992), 371–392.
Braverman, Harry, Labor and Monopoly Capital: The Degradation of Work in the
Twentieth Century (New York: Monthly Review Press, 1974).
Brown, Randall S., and Laurits R. Christensen, ‘‘Estimating Elasticities of
Substitution in a Model of Partial Static Equilibrium: An Application to U. S.
Agriculture, 1947 to 1974,’’ in Modeling and Measuring Natural Resource
Substitution, E. Berndt and B. Field, eds. (Cambridge, MA: MIT Press, 1981),
pp. 209–229.
Cain, Louis P., and Donald G. Paterson, ‘‘Biased Technical Change, Scale, and
Factor Substitution in American Industry, 1850–1919,’’ Journal of Economic
History, XLVI (1986), 153–64.
Chandler, Alfred D., Jr., The Visible Hand: The Managerial Revolution in
American Business (Cambridge, MA: Belknap Press of Harvard University
Press, 1977).
Chiswick, Carmel Ullman, ‘‘The Growth of Professional Occupations in U. S.
Manufacturing: 1900–1973,’’ Research in Human Capital and Development, I
(1979), 191–217.
Creamer, Daniel, Sergei P. Dobrovolosky, and Israel Borenstein, Capital in
Manufacturing and Mining: Its Formation and Financing (Princeton, NJ:
Princeton University Press for the NBER, 1960).
Devine, Warren, ‘‘From Shafts to Wires: Historical Perspective on Electrification,’’
Journal of Economic History, XLIII (1983), 347–372.
Doms, Mark, Timothy Dunne, and Ken Troske, ‘‘Workers, Wages, and Technology,’’
Quarterly Journal of Economics, CXII (1997), 253–290.
Du Boff, Richard B., Electric Power in American Manufacturing 1889–1958 (New
York: Arno Press, 1979).
Economic Report of the President (Washington, DC: G.P.O., 1995).
Electrical Merchandising, How to Retail Radio (New York: McGraw Hill, 1922).
Fabricant, Solomon, The Output of Manufacturing Industries, 1899–1937, Publication
No. 39 (New York: NBER, 1940).
Fallon, P. R., and P. R. G. Layard, ‘‘Capital-Skill Complementarity, Income
Distribution, and Output Accounting,’’ Journal of Political Economy, LXXXIII
(1975), 279–301.
Goldin, Claudia, ‘‘America’s Graduation from High School: The Evolution and
Spread of Secondary Schooling in the United States,’’ Journal of Economic
History LVIII (1998), 345–374.
Goldin, Claudia, and Lawrence F. Katz, ‘‘The Decline of Noncompeting Groups:
Changes in the Premium to Education, 1890 to 1940,’’ NBER Working Paper
No. 5202, 1995.
Goldin, Claudia, and Lawrence F. Katz, ‘‘Technology, Skill, and the Wage Structure:
Insights from the Past,’’ American Economic Review, LXXXVI (1996),
252–257.
QUARTERLY JOURNAL OF ECONOMICS730
Page 730
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
Goldin, Claudia, and Lawrence F. Katz, ‘‘Why the United States Led in Education:
Lessons from Secondary School Expansion, 1910 to 1940,’’ NBER Working
Paper No. 6144, 1997.
Goldin, Claudia, and Lawrence F. Katz, ‘‘Human Capital and Social Capital: The
Rise of Secondary Schooling in America, 1910 to 1940,’’ Journal of Interdisciplinary
History (1998, forthcoming).
Greenwood, Jeremy, The Third Industrial Revolution: Technology, Productivity,
and Income Inequality (Washington, DC: The AEI Press, 1997).
Greenwood, Jeremy, and Mehmet Yorukoglu, ‘‘1974,’’ Carnegie-Rochester Conference
Series on Public Policy, XLVI (1997), 49–95.
Griliches, Zvi, ‘‘Capital-Skill Complementarity,’’ Review of Economics and Statistics,
LI (1969), 465–468.
Hamermesh, Daniel S., Labor Demand (Princeton: Princeton University Press,
1993).
Hounshell, David, From the American System to Mass Production (Baltimore, MD:
Johns Hopkins Press, 1984).
Inter-university Consortium for Political and Social Research (ICPSR), Census of
Population, 1940 [United States]: Public Use Microdata Sample (ICPSR 8236)
(Ann Arbor, MI: ICPSR, 1984).
Inter-university Consortium for Political and Social Research (ICPSR), Census of
Population, 1910 [United States]: Public Use Microdata Sample (ICPSR 9166)
(Ann Arbor, MI: ICPSR, 1989).
James, John A., and Jonathan S. Skinner, ‘‘The Resolution of the Labor-Scarcity
Paradox,’’ Journal of Economic History, XLV (1985), 513–540.
Jerome, Harry, Mechanization in Industry (New York: National Bureau of Economic
Research, 1934).
Johnson, George E., ‘‘Changes in Earnings Inequality: The Role of Demand Shifts,’’
Journal of Economic Perspectives, XI (1997), 41–54.
Katz, Lawrence F., and Kevin M. Murphy, ‘‘Changes in Relative Wages, 1963–
1987: Supply and Demand Factors,’’ Quarterly Journal of Economics, CVII
(1992), 36–78.
Krueger, Alan, ‘‘How Computers Have Changed the Wage Structure: Evidence
from Micro-Data, 1984–1989,’’ Quarterly Journal of Economics, CVIII (1993),
33–60.
Krusell, Per, Lee E. Ohanian, Jose´-Vı´ctor Rı´os-Rull, and Giovanni L. Violante,
‘‘Capital-Skill Complementarity and Inequality: A Macroeconomic Analysis,’’
Federal Reserve Bank of Minnesota Staff Report No. 239, 1997.
Landes, David, The Unbound Prometheus: Technological Change and Industrial
Development in Western Europe from 1750 to the Present (Cambridge: Cambridge
University Press, 1992).
Mincer, Jacob, ‘‘Human Capital Responses to Technological Change in the Labor
Market,’’ NBER Working Paper No. 3207, 1989.
Nelson, Daniel, ‘‘Mass Production in the U. S. Tire Industry,’’ Journal of Economic
History, XLVII (1987): 329–339.
Nelson, Richard R., and Gavin Wright, ‘‘The Rise and Fall of American Technological
Leadership: The Postwar Era in Historical Perspective,’’ Journal of
Economic Literature, XXX (1992), 1931–1964.
Nye, David E., Electrifying America: Social Meanings of a New Technology
(Cambridge, MA: MIT Press, 1990).
Sokoloff, Kenneth, ‘‘Was the Transition from the Artisanal Shop to the NonMechanized
Factory Associated with Gains in Efficiency? Evidence from the
U. S. Manufacturing Censuses of 1820 and 1850,’’ Explorations in Economic
History, XXI (1984), 351–382.
, ‘‘Productivity Growth in Manufacturing during Early Industrialization,’’ in
Long-Term Factors in American Economic Growth, Studies in Income and
Wealth, NBER, Vol. 51, S. Engerman and R. Gallman, eds. (Chicago:
University of Chicago Press, 1986).
Slichter, Sumner H., ‘‘Notes on the Structure of Wages,’’ Review of Economics and
Statistics, XXXII (1950), 80–91.
Tinbergen, Jan, Income Differences: Recent Research (Amsterdam: North-Holland,
1975).
TECHNOLOGY-SKILL COMPLEMENTARITY 731
Page 731
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary
U. S. Census Office, Report on Manufacturing Industries in the United States at the
Eleventh Census, 1890. Part II. Statistics of Cities (Washington, DC: G.P.O.,
1895).
U. S. Department of Commerce, Bureau of the Census, Thirteenth Census of the
United States, 1910. Vol. VIII, Manufactures, 1909, General Report and
Analysis (Washington, DC: G.P.O., 1913).
U. S. Department of Commerce, Bureau of the Census, Fourteenth Census of the
United States, 1920, Vol. VIII, Manufactures, 1919, General Report and
Analytical Tables (Washington, DC: G.P.O., 1923).
U. S. Department of Commerce, Bureau of the Census, Fifteenth Census of the
United States, 1930. Vol. 1, Manufactures, 1929, General Report (Washington,
DC: G.P.O., 1933).
U. S. Department of Commerce, Bureau of the Census, Alphabetic Index of
Occupations and Industries: Occupation and Industry Classifications Based
on the Respective Standard Classifications, prepared by Alba M. Edwards
(Washington, DC: G.P.O., 1940).
U. S. Department of Commerce, Bureau of the Census, Historical Statistics of the
United States, from Colonial Times to the Present (Washington, DC: G.P.O.,
1975).
U. S. Department of Labor, Bureau of Labor Statistics, Descriptions of Occupations:
Coal and Water Gas, Paint and Varnish, Paper, Printing Trades, Rubber
Goods; Electrical Manufacturing Distribution and Maintenance; Glass; Medicinal
Manufacturing, prepared for the U. S. Employment Service (Washington,
DC: G.P.O., various years 1918–1921).
Williamson, Jeffrey G., and Peter H. Lindert, American Inequality: A Macroeconomic
History (New York: Academic Press, 1980).
Woolf, Arthur George, ‘‘Energy and Technology in American Manufacturing:
1900–1929,’’ Ph.D. thesis, Department of Economics, University of Wisconsin,
1980.
QUARTERLY JOURNAL OF ECONOMICS732
Page 732
@xyserv2/disk4/CLS_jrnlkz/GRP_qjec/JOB_qjec113-3/DIV_017a06 mary