Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Part III technology and Mends in amens E~l~me~
Women's Employment and Technological Change: A Historical Perspective CLAUDIA GOLDIN The unpact of technological change on female employment is an issue in two parts: that relating to all workers and that just to women. Technological change, in concert with the demand for output, ~ the driving force for all employment in the long run, and determines total employment, the share of employment in each sector of the economy, and the share of output received by labor. Female workers, however, have characteristics that distinguish them from their male counterparts and that cause them to be differentially affected by technological change. Among the distinctive characteristics of the female labor force are the lower participation rates of women as compared with those of men. Because of this difference the supply of women to the labor market is more responsive to changes in earnings and oc- cupational opportunities. Related to this point is that women, on average, have less labor market experience than men, and their lower degree of seniority might make them more vulnerable to rapid technological change. It is also the case, however, that women have less training specific to a particular technology and might be less at risk of losing income from changes in the produc- tion process. For various reasons women work in certain sectors 185 .
186 A HISTORICAL PERSPECTIVE and occupations more than in others. To the extent that certain sectors, industries, or occupations are "female intensive, techni- cal changes that increase employment in these areas relative to others will have positive employment effects for women. The precise employment impact of technological change, either on an individual sector or on the national economy, is a rather complicated affair. There are no a priori reasons to believe that technological change will reduce employment in any particular sector; its precise role is empirical in nature and is explored below. In addition, long-run and short-run effects may differ. In the long run, technological change alters the relative size of different sectors and the female intensity of the labor force in these sectors. We can assume, in the long run, that the economy adjusts to the various shocks imposed on it by technological change, returning eventually to the natural rate of unemployment. But over the short run there are displacement ejects, involving retraining and geographical mobility. Because the focus here is on historical changes over an extensive period of time, only the long-run effects are considered in detail. The conclusions of this research are many. Over the long run, technological advances, proxied by a measure called total fac- tor productivity, have been positively associated with the female intensity of a sector; that is, female-intensive sectors have had greater advances in technology. But sectors having the greatest advances in technology did not necessarily experience the largest expansion in employment. Technological changes associated with a greater division of labor, both in the early part of the nineteenth century in manufacturing and in clerical work a half century later, fostered the employment of relatively unskilled workers, especially women. Technological changes embodied in individuals, particu- larly advances in education, have provided the greatest impetus to the employment of women in the more recent period, most often in sectors undergoing general technological change. Without knowing the precise reasons for the expansion and evolution of the female labor force in general over the last two centuries it is difficult to ascribe a portion of it to technological change. But it is clear that advances in manufacturing in the nineteenth century increased the relative productivity of females to males (and children to adults) by substituting machinery and inanimate power for human strength. Such technological change frequently led to the substitution of capital and unskilled labor
CLAUDIA GOLDIN 187 for skilled (artisanal) labor, a factor substitution that is rather different from that observed later in the twentieth century. It is equally difficult to ascribe a portion of the changes in the sectoral distribution of labor in general over the last two cen- turies to technological change biased in favor of particular sectors. But it is clear that the relative decline in agriculture throughout the period and the rise of the tertiary (service) sector in the first decades of this century were instrumental in chin ging employment opportunities for women. Not all women's employment changed in the same direction in response to these movements. Black women, who have always been disproportionately employed in agriculture and domestic service, experienced decreases in employment with sectoral shifts that occurred early in this century, but these de- clines did not necessarily entail decreased well-being. In the most recent period, black women's employment, along with employ- ment of white women, has been greatly affected by educational advances. This paper explores various aspects of technological change and women's employment in the United States from 1800 to the present. ~ begin with a discussion of conceptual issues regarding technological change and then consider the evidence on female em- ployment over the last two centuries. Four aspects of the topic are then detailed: (~) the correlation over time and in cross-section between technological change and the proportion of a sector or an industry's labor force that is female; (2) the role of educational change in altering the employment of women; (3) changes in the or- ganization of work; and (4) the relative wages of females to males, a ratio measuring what is commonly termed the "gender gap." The focus is entirely on changes in technology outside the home, thus omitting changes in household production and contraceptive technology. One way of integrating the topics in this paper is to partition changes in the proportion of the total labor force that is composed of women, F/L, into that portion due to changes within sectors and that part due to the changing distribution of the labor force across sectors. Thus, d(F/L) = Ei~d(F`/Li)~(Li/L) + Ei(Fi/Li~d(Li/L), (1) where F ~ the female labor force (F = EFi), ~ = the total labor force (L _ EI,i), and i denotes sector or industry i. The change
188 A HIS T OR I CA L PERSPE C TI VE in the proportion of the labor force that is female is divided into two sources, and each can be studied in terms of the impact tech- nological change has on it. The first is the impact of technological change on changes in the female intensity of a sector or an industry (Fi/L`) times the relative importance of that sector or industry in tote] employment. The second is the impact of technological change on sectoral shares (Li/~) times the female intensity of that sector or industry. Each of these sources is discussed below.) TECHNOLOGICAL CHANGE AND EMPLOYMENT: A THEORETICAL FRAMEWORK To clarify the economist's conceptualization of technological change (see Mansfield, 1969; Rosenberg, 1972; Stoneman, 1983), let me offer this brief description. Technology is knowledge, and technological change is an increase in knowledge measured as an increase in output for a given quantity of inputs or, equivalently, a decrease in inputs required to produce a given amount of output. Technological change is identified with technological advance; all agents, in firms, sectors, and the economy as a whole, can reject any technological change that decreases outputs or increases the costs of production. One generally thinks of technological change as altering the demand for a particular factor, such as labor; it can also after the set of prices and wages in the economy. The employment ejects of technological change are complicated by the impacts such price changes have on the demands for input and output. Even a labor-saving technological change can result in an increase in the demand for labor if the quantity of output demanded increases sufficiently because of a lower price or a higher quality of output. Further, a technological change that is neutral in its effects on the various inputs can alter the economy's demand for labor if it tends to increase demand for outputs produced by relatively less labor-intensive sectors (or less intensive in the use of a particular type of labor women, blacks, skilled, unskilled). Production occurs when one combines particular inputs in some manner. Denoting the output as Q and the inputs as K ~ One can add another source, the increase in the female labor force participation rate, by scaling each source. One can scale, for example, the female intensity of each sector by expressing it as an index in relationship to the female intensity of the economy as a whole. This index is used in the empirical work below.
CLAUDIA GOLDIN 189 (capital) and ~ (labor), Q = f(K,~) is a production function. The theoretical framework economists use to study technological change can be condensed by considering a particular production function known as Cobb-Douglas. This production function is a geometrically weighted average of the inputs multiplied by a constant term, T: Q= TK"L(~-~. (2) In the case shown, the exponents of the two inputs, K and A, sum to one, which ensures the characteristic of constant returns to scale (a proportional increase in the inputs will increase the output pro- portionally). The constant term, T. indicates the degree to which the same input levels lead to greater output. Thus, the change in T is the measure of technical change, when the outputs and the inputs are measured properly. T is known as total factor produc- tivity, and changes in T are known as total factor productivity (technological) change. Various types of technological change are absorbed in the constant term T. Technological change can be disembodied or embodied in a particular factor. Technological change can be disembodied in the sense that it does not require the inputs to change in any noticeable manner. One can think.of a disembodied technologi- cal change as some increase in general knowledge that allows an increase in output for the same quantity of physical inputs. Em- bodied technological change is more easily conceptualized than is disembodied. When technological change is embodied in a piece of capital equipment we usually think of the new input as hav- ing a particular vintage, with later ones being the most efficient. When technological change is embodied in labor, we generally think of labor as being more skilled, more educated, healthier, and so on. We can measure inputs in physical terms, as in hours of labor or persons, or we can measure inputs in efficiency units, as in an education-weighted index of person-hours. When one is able to measure inputs in these efficiency units, one can more easily distinguish between technological change that is embodied and disembodied. This procedure is required for understanding precisely the sources of technological change and the impacts on women workers. In the Cob~Douglas production function, if either K or ~ is multiplied by some amount A, we would not be able to discern
190 . . A HISTORICAL PERSPECTIVE empirically its impact on output (Q) from a change in the con- stant term T. Thus the Cob~Douglas form does not distinguish between disembodied change that augments K from that which augments L. Furthermore, if the inputs are measured in physical terms, rather than in their efficiency units, and if their efficiency units expand (because of technological change that is embodied in labor or capital) while the physical units stay constant, we would not be able to distinguish the impact on Q from a disembodied technological change. In the analysis that follows, ~ explore the impact of technical change on the employment of women across various sectors of the economy and within the manufacturing sector from 1890 to the present. The measure of technical change used is the rate of change in total factor productivity. Recall that total factor productivity indicates the degree to which more output is produced for a given level of inputs. Assume that the production technology can be represented by a Cobb-Douglas production function, given in Equation 2. When Equation 2 is expressed in rate-of-change form, where an asterisk (*) over a variable indicates its time derivative, the rate of change in total factor productivity is given by * * ~ * T = Qcork(1a)L. (3) In other words, the rate of change in total factor productivity (read, in technology) is given by the rate of change in the output minus a weighted average of the rates of change of the inputs. The definition of technological change summarized in this manner- that it is measured by how much output increases above and beyond increases in the inputs is very intuitive. .. The measurement of the inputs and the outputs pose certain difficulties. The outputs and the inputs ought to be measured in physical units rather than in value terms. The procedure is ordi- narily not possible, and one generally weights outputs by prices. Thus cars become the value of cars or value added in the auto- mobile sector. The inputs themselves are normally measured in conventional or physical terms (for example, hours of labor) rather than in efficiency units. Thus part of total factor productivity will be attributable to the increased efficiency of the factors (for exam- ple, from increased education, skills, or health), and disembodied and embodied technological change are not distinguished. The
CLAUDIA GOLDIN 191 rates of change of the inputs are weighted by the exponents, and to get estimates of the exponents one must appeal to economic the- ory. Under conditions of equilibrium in the factor (input) markets and competition, the exponents will be the shares of the factors in national income. These data are easily located for even Tong periods of time. Standard measures of total factor productivity exist for vari- ous sectors and industries in the extensive work of Kendrick (1961, 1973, 19833. One aspect of technological change that is necessarily hidden in the use of this methodology is factor bias. To explore issues of biased technological change requires a more flexible func- tional form than Cobb-Douglas, and some estimates of factor bias using the transiog production function are discussed below. TRENDS IN FEMALE EMPLOYMENT: 1800 TO 1980 It is instructive to review the historical record regarding the labor market involvement of women in the United States before examining the impact of technological change. Data on the oc- cupations of women were first collected in 1860, but the printed tabulations were aggregated at the state level and provide little detail by age and other characteristics. Marital status was not requested by the U.S. Census of Population until 1880. Readily available labor force data dictate that the period under study be- gins with 1890. The period before 1890, particularly that from 1832 to 1880, is explored with materials from the manufacturing censuses, which necessarily include only a portion of the labor force. Although substantial change in the conventionally measured female labor force has only recently surfaced, it has been rooted in a longer history of economic transition. Increases in educa- tion during the first three decades of this century and the related increase in the tertiary sector, particularly clerical occupations, were instrumental in the post-World War II increase in partici- pation rates of women. The data on manufacturing labor force participation rates indicate the importance of early factory devel- opment in the employment of young women, particularly in the New England and Middle Atlantic regions. These points are more fully developed below. Labor force data for the aggregate female population together with data by race, marital status, and nativity are given in Ta-
192 A HISTORICAL PERSPECTIVE ble 1. The overall trend for the aggregate from 1890 to 1980 is upward, but most of the movement comes from increases in the participation rate of married women, particularly after 1950. The increase is most apparent for white married women. Before turning to a more detailed exploration of the data for white women, those for black women must be given more careful attention. The labor force participation of black women across all marital statuses and ages (with the exception of single young women) has always been higher than that for white women. Several factors account for these differences (see Goldin, 1977, for a review and an exploration of the role of slavery). The lower family incomes of black Americans throughout the period are responsible for much of the difference that appears in Table 1, as are the rural, southern location of the black labor force in general and the crops produced in the South, especially cotton. The geographical location of black women, their greater par- ticipation rate early on, and their occupational structure make the analysis of the effects of technological change on their employment different from that for white women. In 1910, for example, fully 95 percent of all employed black women were in just two occupa- tional groupings agriculture and domestic and personal service; the comparable figures for white women were 43 percent for na- tive born and 27 percent for native born of foreign parentage. In 1940 77 percent of black working women were found in these two occupational groupings; in 1980 the figure was under 30 percent. Despite dissimilarities between the black and white female labor forces, various changes in the economy, such as increases in edu- cation and the rise of the clerical sector, have had strong impacts on both. The participation rate of young single women across the entire United States rose until around 1920, when it reached a plateau at about the 0.40 level. Participation rates for this group varied sig- nificantly by geographical location. The daughters of native-born white parents in urban areas in 1890, for example, experienced a participation rate of 0.43, almost twice the aggregate level for that group. These data suggest that the labor force participation rate for the entire population of young single women converged by 1920 .
193 o :s ~ o a, CO JO A ~ - - U] - ._ L' : Is o ._ ._ ._ C) o o to 1 ~ o _ ~ ~ oo ._ .> ~ as m 0 CO 0 U. o 00 mp 0 0 C53 _ 0 en C5) so ~ _ (D O _ _ _ . · - u~ ~ en us ~ at en co 0 ~ ~ ~ ~ . · . . . · . · . . . · ~ ~ Us ~ ~ ~ Ch UP CO ut u~ ~ C9 ~ ~ CD u~ u _ _ _ · . . c~ ~ oo O a) ~ u~ ~ u~ CD C~ ~ ~ . . . · . . · · . · . . . ~ ~ ~o ~ _ ~ ~ ~ ~ ~ ~ ~ ~ . ~ · ~ a~ t- oo O} t- {~) oo ~ ~ · . · . . . . ~ ~ 0 ~ ~ ~ ~ _ 0 co a~ ~ a~ e~ ~P ~ ~ a · · . 0 ~= ~ - ` ~ _ - ~ ~ ~ ~ · . . . · . . . · . · . . ~ ~ ~ ~ ~ 2 ~ ~ ~ ~ ~ ~ ~ 00 ~S 00 U:~ U:) U:~ (S~ CD C~':~ O~ ~ C~ - . . . · . . · · . · . . . U' ~= - ~ - ^~_ e~ ~ ~ ~ ~ co e~ ~ oo ~e~ ~ ~ oo ~ co ~ ~ ~ u~ oo . .·· .. · ··· ··- ~ ~_o mOm mmm m-m e~ ~ ~ e~ ~ ~ co u: _ · · . . ~ rS O ~ c9 ~ 0 _ u, oo ~ ~ ~ · . . . . . . . . . · . . e~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ · · · ~ cs ~cs~o me~= ~om c. ~ ~ · . . . · . . . · . . · . 2 ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ C5~ 0 CD U~ CO u~ ~ ~ U) ~ 00 0 o0 . . · . . · . . · · . · . oO C53 ~P O {D c~ 00 ~ e~ ~ ~ o~ 0 _ ~ ~ eo co e~ u~ _ Le ,Q ·a ~ ·~ ~ ~e ~ ·. =I bO t:~, ~ L. bl' O O ~ ~ ~ ~ ~ O ~ ~ O ~ ~ ~ ~ 3 ~ ~ L' ·~ C~ ~ 7 o ~Q ~ . _ ~ _ {D L' O . , ~ U) C) ·" ~o o o c. - aS o :~ o ~ _ o ~ aS ~ ~ o .~ to ~ . . o t' oOo <U ~ ~ ~Q _ oo ~ ° ~ ~ . P4 ~ o ~ o L. ~ I o =m I., ~ ~ - 4 L. o . a' _ ., ost ~ ~ a1 P~ . CO ~Q U] o . - o L. V o Ll o oo J ~ a~ · I.4 i~ ~ ~ oo _ 1 ~ ·~ ~ o _ ·~ ·e V o U] s"
194 A HISTORICAL PERSPECTIVE on the rate experienced by the urban native-born group as early as 1890. The data for the 1832 to 1880 period demonstrate that wherever manufacturing activity spread, labor force participation rates of young single women rose, and the elasticity of supply of this labor force was rather high. Despite the apparent stability in the percentage of single ur- ban women in the labor force over the twentieth century, there was substantial variation in the activities of the approximately 60 percent who were not employed. In 1890 young women not in the labor force were overwhelmingly occupied "at home," ostensibly helping their mothers. Data for 1900 indicate that about two- thirds of the young women not in the labor force were also not in school, and were listed in the census as being at home. With the increase in high school education, the percentage at home dropped rapidly over the early twentieth century to less than one-tenth by 1930. The increase in formal education exactly offset the decline in time devoted to home production by young single women. Participation rates of married women did not expand to any great extent until the 1920s. Also in contrast to the experience of young single women, participation rates for married women did not vary significantly by region or by industrial development. Al- though the 1920s marked the first discernible increase, it was only after 1950 that the employment of married women rose rapidly, first for women over age 35 and later for those under 35 years (see also Easterlin, 1980~. As part of a larger study of women's changing economic roles, have prepared a matrix of cross-section and time-series labor force participation rates by marital status, age, race, and national origin from 1890 to the present. Figure 1 is part of this larger matrix and summarizes the expansion of the labor force participation of white (native-born) married women, born from 1866 to 1955. (Comparable data for black women are discussed below.) Each set of solid (or dashed) lines represents the participation rate of a particular birth cohort, tracing its market role as it matures. Cross-sectional data can also be read from this figure by connecting the relevant points on the cohort lines; for example, the data for 1970 are given by the top set of dotted lines.
CLAUDIA GOLDIN 195 Although several qualifications2 should be kept in mind in interpreting the data in Figure 1, they make a more or less trans- parent statement: for every cohort born since 1866, participation in the labor force has increased within marriage, at least to about age 55. Despite the generally held notion that married women universally have experienced interruptions in their work careers, the majority who entered the labor force at mid-life had not ex- perienced labor force work since they were single, if even at that time. (Note that because the data are given only for married women, one cannot observe the participation rates of these women when they were single.) The notion of interrupted work careers as the norm has arisen, in part, from the pattern of double-peaked labor force participation characteristic of cros~sectional data for women's work experiences in many contemporary developed na- tions. This pattern emerged in the United States in 1950 and is illustrated by the top dotted line in Figure 1, giving participation rates for the cros~section of married women in 1970. Of most importance with regard to the msue of technology and women's employment is the variation in the increases for various cohorts. These variations indicate period effects, and technologi- cal change is likely to have been more important in some periods than in others. Cohorts born around 1900 achieved a considerably higher labor force participation rate than those born previously. Similarly, cohorts born after 1945 had substantially increased par- ticipation rates. All other cohorts, to be sure, contributed to the expansion of the female labor force participation rate, but the cohorts mentioned appear to have gone far beyond the trend line. 2 The qualifications are (1) while the data are for native-born white married (spouse present) women, there is heterogeneity in other respects: women enter the data when they marry, and thus the data are contaminated by selection bias if people who marry late or who are widowed early have different labor force participation rates; (2) intermittency of participation is disguised in these data and depends on the aggregate rate of labor turnover for married women; (3) there are numerous definitional issues, particularly the change in 1940 from the "gainful worked to the Labor forces construct and the omission in the early period of many women working on farms and in the home (Goldin, 1987, addresses these issues); and (4) the 10-year intervals mask certain changes, particularly those at the early ages. Related to this consideration is that the data concern only married women; participation rates for young single women are considerably higher than are those for young married women, particularly before the 1960s.
196 % in Labor Force 60 50 40 30 20 10 A HISTORICAL PERSPECTIVE 1936-1945 ~ 1946-1955 , / 1926-1935 / _ . ~ / / / / ~916-1925 \ 1906-1915 ~ \~\ 1 886-1 895 \\ 1876-1885 __ 60 Age FIGURE 1 Labor force participation rates of cohorts of white married women, born 1866 to 1955, for the entire United States. Notes: Dashed lines denote missing data. Data for 1890 to 1920 are for native-born women with nati~re-born parents. Dotted line is 1970 cross-section. Source: Derived from population census data. Data appendix on request from author. Similar data for black women have been compiled and rein- force the finding from the aggregated data in Table 1 that black married women's participation rates have always been substan- tially higher than those for white women. But beginning with the cohort born around 1910, black married women have also had large increases in participation rates. Unlike the data for this co- hort of white women, however, those for black women increase only for ages over 40. It is with the cohorts born around 1930 that increases in participation rates across the life cycle occur. These
CLAUDIA GOLDIN 197 findings complement those for white women. Even though the cohorts affected tend to be different, the underlying reasons for the expansions in these two labor forces are similar. Educational advances occurred at different times for each group, and the find- ings for each reinforce the importance of increases in education for women as a whole. The evidence that has been compiled for the period before 1890, for which the aggregate data are either entirely absent or lacking in detail, indicate that wherever significant industrial development occurred, labor force participation rates for young women rose. Furthermore, participation rates in industrial areas were extremely high even as early as the 1830s. Although the data are fragile at best for the period before 1860, estimates of a "manufacturing labor force participation rate" for young females in five northeastern states have been computed and are given in Table 2. The estimates must also be interpreted in light of the se- vere undercounting of manufacturing firms in the 1832 data for all states but Massachusetts. The 1832 manufacturing participation rate estimates, ranging across states from 12 to 27 percent, indi- cate that the manufacturing sector attracted a substantial portion of the population of young women in the Northeast. In the early industrializing state of Massachusetts, where the reporting was most complete, one-third of all young females were employed in the manufacturing sector by 1850, if not before. This level roughly equalled that prevailing in 1880. Given that some young women were employed in alternative pursuits such as domestic service and teaching, the crude manu- facturing labor force participation rates indicate that a high pro- portion of single women in New England had been drawn into the market economy by the 1830s. Comparable evidence on female la- bor force participation before 1832 is not available, but the levels implied by the 1830s data were most likely achieved quite rapidly, because opportunities for the employment of females were limited prior to industrial development.
198 A HISTORICAL PERSPECTIVE TABLE 2 The Manufacturing Employment of Females as a Percentage of 10- [or 15-] to 29-Year-Olds in Five States, 1832-1880 State 1832 1837 1850 1860 1870 1880a Connecticut [11.6Jb 22.6 23.1 18.4 28.5 [19.1] [33.7] Massachusetts 27.1b'C 40.2C 32.9 28.4 36.7 32.8 [18.73 [29.7] [44.0] [39.53 New Hampshire 11.6b'C 20.1 22.0 21.7 28.1 [10.5] t26.6] t36.6] New York 8.0 6.8 9.2 15.3 [10.8] [18.7] Rhode Island 26.6b 26.5 33.3 48.7 40.9 t24.6] [53~9] t45~0] Percentage of U.S. total female manufacturing in five states 70.3 62.9 60.7 56.5 t61.3] [57.7] aChildren were allocated between boys and girls as given by the 1880 population figures for children in manufacturing employment by states. The bracketed figures express the number of females employed in Manufacturing as a percentage of those 15 to 29 years old in the population. The returns for Rhode Island listed women and female children separately. The Massachusetts and New Hampshire returns did not, and the estimates assume that 45 percent of all working children were female. The bracketed figures give the employment of adult women as a percentage of those 15 to 29 years old. The Connecticut estimate is only for adult women as a percentage of those 15 to 29 years old. In all cases, the population figures for 1832 are for white females only. -The estimates include women in home workshop employment, mainly palm leaf hats and straw hats, bonnets, and braids; the bracketed figures exclude them. SOURCE: Goldin and Sokoloff (1982:Table 8~.
CLAUDIA GOLDIN TOTAL FACTOR PRODUCTIVITY AND FEMALE EMPLOYMENT TECHNICAL CHANGE AND FEMALE INTENSITY 199 Cross-section and time series data on female employment for 1890 to 1980 are presented in Table 3 in two ways: as a percentage of the sector or industry labor force that is female ant] as an index, which divides the industry percentage by the percentage for the entire economy. The index is the more appropriate indicator of change over time and measures the extent to which particular sectors and industries expand their female employment at a rate greater than the national average. The information contained in Table 3 is easily summarized. Most sectors and industries have been either male- or femaTe- intensive for the entire century under consideration, although sev- eral have undergone shifts in labor force composition. -Farming, mining, transportation, construction, and public utilities, among the sectors, have been male intensive.3 Lumber, furniture, chem- icals, petroleum, stone, metals, machinery, and transportation, among the manufacturing industries, have been male~intensive. Large shifts in the composition of the labor force have occurred in the communications, professional, and clerical sectors. How has the index of female intensity for the economy's sectors and industries been affected by changes in total factor productiv- ity, the measure of technological change used in this analysis?4 The answer can be found in Table 4, which presents the results of a pooled cross-section, time series regression in which the de- pendent variable is the relative proportion of the labor force in a sector, at time t, that is female (given by the index in Table 3~. The explanatory variables in the regressions are total factor produc- tivity change (TFP), lagged by 5 to 10 years; sector and industry 3 Male and female intensities are defined here in terms of whether the index (times 100) is less than or greater than 100. 4 The data on total factor productivity are discussed in several volumes by Kendrick (1961, 1973, 19833. Total factor productivity data have been computed for the private sector, and those for the service sector join to- gether three major but disparate occupational groups- professionals, clerical workers, and domestic workers. Thus the professional, clerical, and domestic occupations cannot be included, and only the private sector can be consid- ered. The omission of services from the analysis is unfortunate because of the preponderance of women in these three areas of employment.
200 s" o - x 1_ 0 ~5 ~ JO 1 o c°~ so .,Q all ·e ~ 0 TIC o ~ to `: A so Pa ~ ~ on o t~ ~ o a, Eel ·- m ~ ~ O ma O. cr. 00 ITS} ~ O L' of C53 o U] 003 o ~ O - 0 E ~ o `i ~ E~ no al - ·" o ~ o ~= in ~ so Go · . . ~ of ~ 0* en _ _ . . co ~ oo ~ ~ oo ~ 0 e~ 0 ~ e~ C~ ~ ~ ~ ~ ~ ~ ~ CO CO ~ 0 ~ ~ e~ 3 4:o 0 e~ a) a~ ~ e~ ~ ~ ~ ~ c~ ~ ~ * * °O ~ U. ~ ~ ~ e~ ~ ~ ~ t~ ~ _ _ _ _ ~ _I _ _I _ _ ~ oo. t- oo e~ t_ _~ co ~F CS~~ ~ O O oo 0 e~ oo * * _ ~, e~ u: O ~ ~ ~, _ _ oo o' ~ u, ~P ao 0 oo _ e~ - o. U, ~ ~ oo ~ 0 ~ oo cs 0 co 0 a, co ~ * * * * _ e~ _ _ _ _ e~ o. o, ~ oo ~ ~ ~ * _ e~ - i~ ~ ~ ~ eo ~P O O O 0O e~ ~ * - co. e~ 0. ~ =. ~ 0 ~ 0 co e~ ~ * - ~ oo _ t~ e~ ~ · · · . · ~ ~ ~ O O O O - - ~ oo · ~* ~0 oo _ oo - ·~ ~ E = ~- ~ ' ~ E ~ E =
201 _ '_ _ Hi ~ co ~ Cal co Ed * ~ oo oo ~ Hi ~ ~ oo CD CO O an co Cal ~ of co ~ ~ 0 co ~ co Cal ~ ~ ~ co en ~ Cal ~ co co en ~ en en _ _ {D ~ ~ O en A. ~ 0 ~ co ~ ~ en 0 ~ ~ ~ ~ co on 0 ~ 0 CO ~ ~ ~ ~ CD O * ~ ~ ~ O C9 en en on ~ en up on co ~ en ~ or ~ ~ ~ ~ en en co Cal ~ co ~ ~ ~ ~ on ~ _ _ _ - oo e~ e~ oo i~ oo co oo ~ ~ oo o. ~ ~ ~ e~ ~ ~ ~ ~ C5) =-o o, ~ us oO {9 * 00 ~ ~ u, ~ e~ cs ~ e~ oo ~ CD ~ ~ CO U: e~ u, oo O e~ ~ e~ ~ ~ ~ ~ e~ e~ ~ ~ c' ~ ~ ~ ~ co ~ _' _ _ _ c, ut ~ ~ ~ O ~ o ~ u' ~ ~ ~ co ~ O ~ u, CD O e~ u, ~ ~ c~ u, cs~ ~ u: ~ ~ * co c~ 0 ~ ut oo u~ C5) ~ cS~ ~ ~ co ~ a~ oo co e~ O ~ °O L, e~ ~ ~ e~ ~ e~ e~ ~ ~ e~ ~ ~ ~ ~ co _~ ,< _ O _ _ oi ~ ~ ooco~o~coooou~ooo~m~ oo~ cso . . _ . . . . . · . . . . . . . . . . . . * ~ ~ ~ ~ * ~ ~ ~ ~ ~ ~ o~ ~ ~ ~ 0 ~ e~ co r~ * ~ oo ~ e~ F ~ ~ ~ ~ ~ eo . . ~ . . . . . . . . . . . · . . . . . · . ~ ~eq~u~ooocDubc~ooo=~ a~ * * ·~t t_ ~ e~ co ~ ~ ~ co ~ e~ _ ~ o 0 - os `~=co` - a3U3o.~me~c~oo o~ C53c9 u) co * * _I ~ ~ e~ u:~ ~F c9 e~ ~ _1 a~ ~ c~ 00 ~ U, ~ ~ O CD CD eD ~ ~ e~ 0 ~ oo u'CO oO 00 ~ u, . · ~ . . . · · . . . . . · . . . . . . . . * ~ ~ ~ ~ ~ 49 ~ e~ 0 e~ ~ 0 co ~ co 0 0 ~ e~ 0 * ,~,~ 00 a~ e~ ~ ~ ~ co e~ ~ e~ - ~ e~ oo 0^ 0 ~ a~ ~ e~ eD C53 {D u: e~ 0. 00 e~ CD u~ ~ u: CO CO ~ ~i _ _i ~ ~ ~i u~ O CO ~ ~ ~ 0 {D CO e~ 0 u~ 0 u~ 0 * * ~ O ~ C~ CO ~ (D CO e~ ~ a~ e~ _~ ~ £ _ ~ ° - . £ o~ oo ~ oo c. ~ a, ~ U, ~ o oo o ~ o c: oo oo ~, * * mo ~ ~oo - - 'oDo - ~^u~o - ~ - o mo ~o - C: ~ ~ 3 .= .= ~ ~ V ~ ~ .0 ='E, ra C,n~x ° ~ ~ ~,.= ~ ~ ~ ~ ,E, E ~: ~ E '° > ~ ~m~E-¢~vP-~ ~ ~ ~ ~ ~ ~ ~ b~ ~ ~ ^, L ~4 ~ ~ _ ~ ~ ·= .
202 ._ o - CO m o h To 4: l o qJ :>, - o oo COD O U. O en a) g C5) O C~ oo ~0 V ~ U' h GO bO Ve ._ O U: CO CD . . . CO O * C~ CO t_ U:~ t- O . . . 03 ~ * 00 CD ~ 0 00 . . . O c~ e~ CD ~ * CO U. . . . CO* U: <9 e~ 00 . . . U: ~ * O . · . . co ~ ~ e~ ~ ~ ~ e~ 00 0O 00 ~ U. O U. o~ ~ e~ . . . . ~ CD CO O CD ~ CO . . . . U. 00 co ~ e~ o: oo ~ e~ . · . . CD~ e~ oo oo ._ a, 0= ~ . ~S o ~ ~ ~ ~ ~ ~ ~ V ~ ~ U) ~5 ,~ ~ h ~ · ~ ~ G, · ~ h O O O ~ o _I O C~ ~ O ~ ~ ._ 00 U, ~ ~ O ~ O >. ~ ~ h ~ ~5 ,= ° " C >. r~ ~~S =-t ~ ~ ~ ~ O h G~ ~ a] ~ ~ ~ . ~ O O ~, .0 ~ _.O ~ O S a.~., ,= E 0 ~ a D o 3 :, o, ~ S o ~ w =0-O m.° == -0 0 =3 ~ o ~ o, W ~ ~ ~ ~ ~ :^ · OO · h oo . ._ . 4) q.c b4 ~o ·s" ~ o ~ ~ e~ ~-E ~ o ~ ~ o · g O ·- 3 re o ._ o o & ~ oo s" . b,O V,~ ~ o , . oo 3° ¢, :~. O a, 0 E~ ~ o ~8 .~ ~ ~ ~a ~ ~ ~ O ~ v." _ O ~ ¢ ~ ~5 ~ . . ~ a, v ~o ~o 4,, 0 0
CLAUDIA GOLDIN TABLE 4 Explaining an Index of Female Employment Across Sectors and Industries, 1890 to 1980 Dependent Variable: Index of Female b b Employment (in year Pa Estimation 1 Estimation 2 Constant 0.789 (7.94) 0.805 (10.86) Total Factor Productivity (TFP) change at time (t-i), 5 <i < 10 years 0.0234 (1.74) -0.0317 (1.31) Time, years from 1900 -0.0020 (1.59) 0.0007 (0.91) Sector and industry dummies Farm -0.298 (1.38) -0.431 (5.33) Mine -0.496 (3.11) -0.709 (9.g8) Transportation (before 1940) -0.762 (1.92) -0.843 (6.95) Communications and public utilities (before 1940) 0.853 (1.62) 0.762 (5.03) Construction -0.463 (2.56) -0.718 (9.38) Rail -0.427 (1.58) -0.683 (7.34) Transportation (non-RR) -0.298 (1.36) -0.533 (6.50) Communications 0.032 (0.05) -0.282 (1.65) Public utilities -0.242 (0.92) -0.518 (5.56) Trade 0.403 (0.95) 0.141 (1.11) Finance 0.836 (4.38) 0.579 (7.40) Real estate 0.398 (1.40) 0.131 (1.35) Food 0.482 (2.25) Apparel 1.981 (7.50) Textiles 1.291 (5.3g) Interactions with TFP Farm * * Mine * * Transportation * 0.065 (1.60) Communications and public utilities * 0.174 (3.87) Construction * * Rail * 0.032 (1.01) Transportation (non-RR) * 0.068 (1.77) Communications 0.149 (1.22) 0.218 (5.38) Public utilities * 0.039 (1.36) Trade * * Finance * 0.071 (1.61) Real estate * 0.056 (1.01) Food -0.085 (1.15) Apparel * Textiles -0.131 (1.90) Nymber of observations 328 107 R .60 .97 NOTES: Absolute values of '"'-statistics are in parentheses. Ordinary least-squares estimation used. An asterisk (*) indicates coefficient is insignificant at even unconventional levels of significance testing. Manufacturing is the omitted sector in both estimations; two-digit industries other than food, apparel, and textiles are omitted from Estimation 1. The index is defined in the note to Table 3. bSee text for discussion. SOURCE: Calculated by the author based on data from Table 3 and Kendrick (1961, 1973, and 1983~. 203
204 A HISTORICAL PERSPECTIVE dummies; a time trend; and interactions between TFP and the dummy variables. The estimation indicates that female intensity has increased with increases in total factor productivitywhere technological advances have been greatest, women's employment share relative to the average has increased the most. Estimation 1, which includes all sectors and all industries within manufacturing, indicates that each percentage point in- crease in total factor productivity increases the index by 0.0234. (The mean of the index is 0.794, and the mean of the change in total factor productivity is 2.19.5) Estimation 2 excludes the 19 detailed] two-digit SIC manufacturing industries, reducing the number of observations from 328 to 107. While the coefficient on the change in total factor productivity is negative (-0.0317, but not very statistically significant), the interaction terms with the various sectors suggest strong positive net effects. The coefficients on the interactions with transportation (before 1940~; communications and public utilities (before 1940~; and transportation, communi- cations, trade, public utilities, and finance are large enough (and generally statistically significant) to outweigh the negative coeffi- cient on technical change. Only the farm, construction, mine, rail, trade, and real estate sectors appear to have either negative or zero net impacts of changes in total factor productivity on the index. The impact of total factor productivity change on female intensity is different in these sectors than for the economy as a whole, and the reasons in the cases of the farm, construction, mine, and rail sectors seem obvious. Several qualifications to this analysis must be noted. The re- lationship between the index and total factor productivity has not been explicitly modeled, and the analysis should be viewed as one that is exploratory rather than one of hypothesis testing.6 Total 5 The index would have a mean of I.0 if all sectors were included, because the index is merely the female intensity of a sector at time t divided by the female intensity of the economy's labor force at time t. The private service and the government sectors have been excluded because of the lack of total factor productivity data; taken together these sectors have an index greater than 1.0. 6 Kendrick (1983:44) regressed changes in total factor productivity on variables that included the proportion of the sector's labor force that was female, for the cross-section 1966 to 1979. The size of the coefficient on the percentage female is in the range of those in Table 4 (although note the reversal of the causality). Kend rick's motivation, it should be noted, was to
CLAUDIA GOLDIN 205 factor productivity (read technological) change is implicitly as- sumed to be independent of the female.intensity of a sector's labor force, but it may not have been. More female-intensive sectors may have had higher levels of total factor productivity change because of lower unionization and less-specific human capital, among other influences that might reduce resistance to technological advance. Several variables that rrught be included in the analysis, such as the education and the skills of the labor force and the level of unionization, have not been. In addition, the total factor produc- tivity measure of technological change is constructed as a residual, and as such it can measure more than the change in output for constant inputs. It may unintentionally measure changes in the inputs that are not observed.7 Leaving this source of potential bias aside, the findings from the estimation indicate that sectors that have increased total factor productivity more than average study the determinants of technological change rather than to understand its impact. 7 The labor force is measured by a weighted average of hours of work, where the weights are based on labor compensation (see Kendrick, 1961, Appendix A for further details). As the labor force increases because more women enter it, the measure will increase less than a labor force increase because men entered. The lower average compensation of women accounts for the difference. But if the lower compensation of women does not entirely reflect lower productivity, the labor force measure will be biased downward. Sectors and industries increasing their labor force by adding women will appear more productive in terms of the total factor productivity index, because their labor force will be increasing by less than their output even if there had been no change in technology. A related bias concerns the efficiency units of labor. Increases in education for society as a whole have had impacts on the labor force that differ for males and females. Certain occupations, it appears, enable workers to substitute training acquired off the job for that traditionally acquired on the job, and women, whose life-cycle labor force participation is more discontinuous and shorter than that of men, have tended to enter such occupations. Thus, as the average educational attainment of the population has increased, women have become a higher proportion of the labor force in certain sectors, such as those employing clerical workers. More highly educated workers will increase the measure of total factor productivity, because only the number of laborers or the hours of labor input, and not the quality-adjusted labor input, is used. Thus a positive correlation between the index and the measure of technological change can occur. This possible bias can be accounted for by including the mean level of education of the work force in each rector. Such data exist only for relatively recent periods of time.
206 A HISTORICAL PERSPECTIVE tend to have more female-intensive methods of production.8 The result stems primarily from the correlation found in the cross- sections, rather than from a positive relationship between total factor productivity change and female intensity in time series for given sectors. Indeed, the female-intensity indices of most of the sectors and industries are relatively constant. Why might there be a positive correlation between the index of female intensity and the change in total factor productivity? It is possible that the most innovative sectors require flexible labor forces, and that the female labor force, being less unionized and less experienced, has been more amenable and receptive to change in work organization. It may also be the case that total factor productivity change is correlated with an increase in the proportion of the labor force that is less skilled.9 The degree to which technological change increases or decreases a portion of the labor force is probably dependent on the degree to which the capital employed is a relative substitute or complement to that type of labor. Technological change in the clerical sector, in manufacturing earlier in this century, and in communications may well have involved the substitution of relatively unskilled for skilled labor, but other evidence suggests that skilled labor has more recently become a relative complement to capital in the economy as a whole (see, for example, the evidence cited in Williamson and Lindert, 1980~. What does the positive correlation between changes in total factor productivity and the relative female intensity of a sector - ~ Manufacturing data for the early nineteenth century confirm the positive correlation between the proportion of females in an industry and the rate of technological change found here. From 1820 to 1850, total factor productivity change was greatest in cotton textiles, hats, tanning, and wool and mixed textiles. All but tanning was a female-intensive industry. Total factor productivity change was the lowest in liquors, flour milling, and tobacco. Only the tobacco industry, among the three, hired female employees. (Indexes of total factor productivity for 1820 to 1860 were produced by Kenneth SokoloR as part of Sokoloff [1984] and were made available to me by the author.) The evidence on factor bias is more mixed (Cain and Paterson, 1986~. The industries with the greatest labor-saving technical change from 1850 to 1919 were apparel, rubber, leather, and metals (both primary and fabricated). The female intensities of the labor forces in these industries were diverse and did not change much over the 70-year period. 9 Kendrick does find a positive correlation between the proportion of the labor force consisting of production workers and the change in total factor productivity.
CLAUDIA GOLDIN 207 or an Industry imply about the impact of technological change on female employment? That female-intensive sectors have been those experiencing the greatest technological advances does not necessarily mean that these sectors have had the greatest increases in employment. Indeed, the opposite could have occurred. The regression analysis examined the first portion of Equa- tion 1, the impact on an index of female intensity of a sector or industry. The second part of that equation concerns the impact on total employment. Equations similar to those in Table 4 (but not presented here) have been estimated for the entire labor force, where the dependent variable was some function of the propor- tion of total employment in a particular sector. The dependent variable was expressed as the percentage, the change in the per- centage, or the percent change in the percentage. In almost all estimations the dependent variable was negatively related to total factor productivity in the preceding decade. That is, sectors with higher productivity growth had a smaller increase in their employ- ment share or a smaller share or a smaller proportional increase in the share. Simply put, employment growth was negatively as- sociated with technological advance. The agricultural sector had the largest impact on the regression and, in at least one variant of the estimation, was responsible for the entire effect. Agriculture experienced relatively small increases in total factor productivity until the mid-twentieth century when its technological advance was among the greatest in the economy. It has also been a sector of rapidly reduced employment. The example of agriculture is a reminder that the income and price elasticities of demand, which are only implicit in this simple analysis, are driving many of the results. It does appear to be the case that more female-intensive sec- tors have generally had expanded employment relative to all others because of less, and not greater, technological advance. That is, among the femaTe-intensive sectors, those with the least techno- logical change grew the most in employment. Services are a well- known example, although it has also been claimed that total factor productivity measures do an imperfect job of capturing technolog- ical advance in this diverse sector (National Research Council, 1979~. On balance, the role of technological change in the total employment picture is less one of encouraging the expansion of female-intensive sectors as it is one of allowing the reduction in
208 A HISTORICAL PERSPECTIVE the employment of the single largest sector early in this century, that of agricultural production. EDUCATION AND THE CHANGING EMPLOYMENT OF WOMEN The expansion of knowledge, termed technological advance here, is frequently embodied in a factor and expands it when the factor is expressed in efficiency units.~° Measures of total factor productivity do not ordinarily use efficiency units, and embodied technological change is then necessarily included in the residual measure of technological change. Denison (1962, 1974), among others, has computed the portion of the residual due to increases in education. from 1929 to 1957 real national income rose at 2.93 percent average annually. Labor, land, and capital, conventionally measured, accounted for 1.43 percentage points, with the residual measure of technological change equal to 1.50. Of this 1.50, almost one-half, or 0.67 percentage points, was due to the augmentation of labor in the form of increased education. In the American case, educational advances for males and females have paralleled each other, but the impact on the oc- cupational structure and labor force participation of women has probably been greater. ~ turn first to the data on educational change and then to the impact on employment and occupations. The two lines in Figure 2 give, from left to right, median years of schooling for female cohorts born from 1876 to 1952 and the percentage with 4 or more years of college. The data on median years of schooling clearly show a sharp rise in the educational attainment of young American women beginning approximately with the cohorts born between 1900 and 1910. During a very brief period, the median female increased her years of education by one-third, from about 9 to 12 years. The rapid rise in years of schooling was a product of the well-known increase in high school education, with these individuals leaving school from 1915 to 1928. I Inputs are generally measured in physical units, such as labor hours. To express a factor in efficiency units means that one measures, in some manner, the degree to which the factor is skilled or educated, for example. One method of doing this is to use a weighting scheme for, say, education. Years of education are valued using earnings at some date to form an index, and as the labor force becomes more educated its value rises.
CLAUDIA GOLDIN 13 12 ._ 0 1 1 En In t~ 10 Cal ._ ~ 9 R 209 Median Yeats Schooling 7 ~ 30 ~ 25 _ / _~ Percentage With > 4 Years College 1 20 _ 15 10 i 5 _ r T I T r I I ~ ~ I T T I T I I I I ~ 1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 Year of Birth a, o c' ct ~1 - cat - a) FIGURE 2 Educational attainment for cohorts of white women born from 1876 to 1952. Note: Horizontal lines indicate the width of the birth cohorts for which data on educational attainment are given. Source: Bureau of the Census, Current Population Reports, Series P-20 for years 1940, 1947, 1962, 1966, 1968, 1970, 1972, 1974, 1977. After the initial schooling increase with the cohorts born from 1900 to 1910, median years of education have increased only gradu- ally, with the exception of the most recent cohorts. Change in their educational attainment can be seen more clearly with reference to the graph in the far right of Figure 2, which gives the percentage of women with 4 or more years of college. That indicator has increased most rapidly with cohorts born after 1940. The employment ejects of increased education can be seen by comparing the educational data with those on cohort labor force participation in Figure 1. It is not at all coincidental that the cohorts experiencing the greatest increases in participation rates over their lifetimes were those with the greatest increase in education when young. The cohort born from 1900 to 1910 achieved an educational transformation in high school completion rates; it was precisely this cohort that experienced substantial increases in its labor force participation both during its early years and even more so during the 1950s when it was 40 to 50 years old. Although Figure 1 does not show the labor force participation
210 A HISTORICAL PERSPECTIVE rate of single women, the cohort born around 1910 also achieved a high rate of labor force participation when single, even though it had an increased educational attainment. It was able to achieve this both by spending considerably less time "at home," as the census termed those individuals not at school and not having an occupation. The data on education in Figure 2 are for the white female population. Increases for the black female population have been greater than those for the female population as a whole, beginning with cohorts born around 1900. Cohorts of black women born in 1900 had, on average, three years less schooling than their white counterparts; cohorts born around 1950 had one-half year less schooling (Smith, 1984~. The large increase in high school completion evident from Figure 2 for white women was achieved about two decades later for black women. Thus, differential rates of educational attainment might explain some of the differences pointed out earlier in labor force participation rates between black and white married women. How did educational change in the first three decades of this century affect the job opportunities of young women, and how did it later affect their labor force participation when married? Occu- pational data arranged by cohort suggest that young women with increased education moved into the clerical sector and, relatively, out of manufacturing employment in 1930 and 1950; and other data indicate that women, who in the 1920s and 1930s were em- ployed in such occupations, had a higher probability of remaining in or reentering the labor force when older (Goldin, 1986b). Table 5 gives various occupational distributions by age for 1930 to 1950, where the group described varies somewhat over time because of definitional changes in the census. Despite this complication, it is clear that the percentage of women in the clerical sector is lowest for those birth cohorts having the least amount of education, with the latter variable given by the data in Figure 2. Cohorts born after 1900 increased their percentages in the clerical sector from about 13 to 24 percent for married women in 1930 and from 17 to 23 percent for all marital statuses in 1950. Micro-level data from 1940 demonstrate that educational levels above high school substantially increased earnings in clerical-sector jobs (Goldin, 1984), and in general, occupations in the clerical sector have had employees with relatively high levels of education.
CLAUDIA GOLDIN TABLE 5 Percentage of the Female Labor Force Employed in the Clerical (and sales) Sector, 1930-1950 by Birth Cohort 211 Birth Cohorts Demographic Group 1916- 1911- 1906- 1896- 1886 1885- 1922 1915 1910 1905 1895 and earlier 1930 Married, native white 26.9 24.0 13.3 6.7 1930 Single, native white 26.1 35.6 1940 Married, whitea 35.5 36.4 34.2 28.8 19.5 1940 Married, white, urban 42.4 43.0 39.6 31.9 21.4 1926- 1921- 1916- 1906- 1896- 1891- 1886- 1930 1925 1920 1915 1905 1895 1890 1950, All marital statuses, white 48.9 39.7 30.3 26.4 23.1 17.3 14.0 ~Married, husband present. Employment away from home and not on public emergency works. Married, husband present, in cities with more than 100,000 persons. SOURCE: U.S. Censuses of Population, 1930-1950. Thus, increased educational standards for all Americans dur- ing the first few deca(les of this century expanded the employment of women at that time and later, as cohorts with increased edu- cation aged. Furthermore, the increase in educational attainment with the cohorts born after 1940 to 1945 is easily correlated with substantially higher labor force participation in the most recent decades. The influence of increased education on labor force par- ticipation can also be observed in cross-sectional analysis. Almost all labor supply studies find positive coefficients on education in participation or hours equations (see, for example, the studies in Smith, 1980~. The reasons for this positive cross-sectional rela- tionship are a bit different from those offered here. Education augments wages, and labor supply increases with the wage, hold- ing income constant or giving a relatively small income effect. The relationship for the time series analysis concerns the shift in occu- pations from those requiring on-thejob training to those for which ofl~-thejob training is a good substitute. CHANGES IN WORK ORGANIZATION Technological change also encompasses alterations in work or- ganization that enhance productivity. During the first half of the
212 A HISTORICAL PERSPECTIVE nineteenth century the employment of young women in the manu- facturing regions of America greatly expanded both in industries that were mechanized, such as textiles, and in those that were not, such as boots and shoes, paper, and clothing (Goldin and Sokoloff, 1982; Sokolo~, 1984~. In the mechanized industries, part of the increased scale of firms and the increased employment of women was due to technological change involving new and improved types of capital equipment. But even in these industries, and certainly in the nonmechanized ones, much of the increased employment of women wan attributable to an increase in the division of labor and thus to changes in the organization of work in nascent factories. Changes external to the firm, such as improved transportation of goods, better-integrated labor markets, and more efficient capi- tal markets, enabled firms to increase their scale of operations, and thus benefit from the increased division of labor and achieve greater efficiency. Within manufacturing, there have been wide differences in the organization of work for female and male employees. In 1890, for example, 47 percent of all female manufacturing operatives were paid by the piece, but only 13 percent of the males were. Females were therefore 3.5 times as likely to be employed on piece rates than were males. Furthermore, piece-rate payment almost always prevailed when males and females occupied the same position in the same firm. Examples from the textile industry are instructive. In only one out of the six predominantly male occupations in cotton textiles was payment generally made by the piece, but among four, in which both men and women were found, only one was paid by time. Also, mates, but not females, were frequently employed in teams and by the method of inside contracting, by which independent contractors organized labor within a firm. Goldin (1986a) explores the role of monitoring and supervisory costs in the adoption of different methods of work organization for male and female manufacturing employees. Firm-leve} data on supervisory costs and the numbers of male and female workers in piece and time rate positions suggest that differences in the costs of supervision influenced the form of work organization. In factories, male workers were more often paid time-based wages rather than piece rates and employed on teams because their time on the job was generally longer than that of females and thus they required less supervision. The ability of manufacturing enterprises both to use a technology with an intricate division of labor and to
CLAUDIA GOLDIN 213 monitor the output of workers therefore fostered the employment of females. The existence of monitoring and supervisory costs can also explain the transformation of occupations in the clerical sector that were "feminized" rapidly from 1900 to 1920. It has frequently been cIaimec} that this feminization was the result of technological changes, such as the mechanization of the office. But was the feminization of the office a function of the reduced level of skill required with the division of office work into tasks or was it a function of a reduced level of supervision needed to elicit some level of output? In the early history of the modern office various tasks were paid by the piece. Typewriters in the Graton and Knight Manu- facturing Company, for example, were equipped with cyclometers; "240 depressions of the typewriter keys or space bar twereJ equiva- lent to one point . . . 600 points twereJ considered base production and each point produced in excess twas] allowed for at the rate of one and one-half cents a point" (Coyle, 1928:23-24~. But piece rates did not prevail in this sector, and their decline was a tribute to the ability of employers to pretest employees whose training in commercial and high school courses was completed before job entry. Monitoring in the office became even simpler and cheaper than in the factory. Employers divided workers into homogeneous groups and paid each a set day rate. Standardization enabled em- ployers to screen workers prior to employment. Thus, it appears that women began to be employed in the clerical sector when its jobs could be more finely divided, as had occurred a full century before in manufacturing, and its output more cheaply monitored. (See RotelIa, 1981, for an analysis of clerical employment that stresses human capital aspects of the mechanization and feminiza- tion of the office.) RELATIVE EARNINGS OF FEMALES TO MALES, 1 8 1 5 TO 1 98 2 The degree to which technological change is biased within sectors or industries and the degree to which even neutral techno- logical change occurs in particular sectors or industries alters the relative earnings of inputs, such as capital and labor or mate and female labor. Such impacts can be ignored when only one sector or one industry is at issue, but not when the entire economy is being
214 A HISTORICAL PERSPECTIVE considered. Changes in female earnings alter female employment because their labor supply function is elastic. How have the earn- ings of females changed relative to those of males over the course of American history, and what role has been played by technological change? Table 6 and Figure 3 give the ratio of female to male earnings for the manufacturing sector from 1820 to 1970 and for all sectors from 1890 to 1982. The relative wage of females to males was fairly low in the northeastern states prior to industrialization but rose quickly wherever manufacturing activity spread (Goldin and Sokoloff, 1982, 1984~. Around 1815 the ratio of female to male wages in agriculture and domestic activities was 0.288 and rose to about 0.303 to 0.371 among manufacturing establishments at the inception of industrialization in the United States in 1820. By 1832 the average ratio in manufacturing was about 0.44, and it contin- ued to rise to 0.50 in the northeastern states by 1850. Nationwide the ratio rose slowly to about the year 1885, when it reached its 1970 value of approximately 0.56. Early industrialization, there- fore, increased the relative wage of females to males by almost 50 percent, and in the briefest of periods, a mere two decades, the gender gap in manufacturing and domestic employment narrowed by 15 percentage points. The ratio of female to male full-time earnings for the entire population increased from 0.463 to 0.603 over the period 1890 to 1970, using the series constructed from earnings in six sectors. The increase over time is somewhat greater when earnings are adjusted for hours of work among full-time employees, from 0.498 to 0.657. The gender gap has remained relatively constant over the period since 1950, with the exception of an initial decline in the ratio and a rise beginning in about 1980. The recent rise appears to be substantial in magnitude, and if the explanations in Goldin (1986b) and Smith and Ward (1984) are correct it will continue for some time. Goldin (1986b) and Smith and Ward (1984) present data on the life cycle labor force experience of the working population of women, which indicates that working women's labor market expe- rience has not increased substantially, if at all, during the period from 1950 to the present. The increase in the labor force partici- pation rate of women was substantial enough that new labor force entrants pulled down the average labor market experience accu- mulated by those already employed. Because wages are computed
CLAUDIA GOLDIN .70 a, .60 a' O .50 Is IL ID .40 - o .30 - a tar 215 / ~0 ~ _- - ~ '~ ~ :-l // · Agriculture Manulactunng, New England ~ Manufacturing, Middle Adanbc D Manufactunng, United States ~ Weighted average of six sectors, see Table 6 O Current Population Survey, Cl Current Population Survey, weekly ~ ~ 1 1 1 1 1 1 1 1 1800 25 50 75 1900 Year 25 50 75 FIGURE 3 The gender gap in historical perspective: the manufacturing sector and the entire United States, 1815 to 1982. Sources: see Table 6 and O'Neill (1985~. Only for those in the labor force, an increase in the relatively inex- perienced will prevent the average earnings of women from rising relative to those of men, all other things being equal. This scenario is coming to a close, however. As more and more women have en- tered the labor force, new entrants are a smaller percentage of the working population of women, and the lesser experience of these new entrants has a far smaller effect. Thus the experience of the working population of women has begun to increase, and with it the relative earnings of women. The same explanation holds for the educational attainment of the working population of women. Working women in each cohort have always been among the more educated. Therefore, as more and more women have entered the ranks of the employed, the average education of working women in each cohort has declined. This process, too, has come to an end, and the educational attainment of the working population of women is now increasing faster than that of men. What accounted for the decrease in the gender gap over the past century? Various studies have shown that earnings within oc- cupational groupings have increased for females relative to those
216 A HISTORICAL PERSPECTI VE TABLE 6 Wage Ratios for Males and Females in Manufacturing Employment, 1815 to 1970, and Across All Occupations, 1890 to 1982 Agriculture 1815 Manufaacturing 1820- 1832a 1850a 1885 1890 0.29 0.37-0.30 0.44-0.43 0.46-0.50 0.559 0.539 1899 1904 1909 1914 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 Full Time 0.535 0.536 0.536 0.535 0.536 0.535 0.536 Full Time Actual Weekly Hourly 0.536 0.535 0.537 0.534 0.536 0.536 0.536 0.568 0.592 0.559 0.645 0.617 0.653 0.612 0.677 0.607 0.672 0.593 0.664 0.592 0.657 0.585 0.662 0.597 0.652 0.573 0.645 0.575 0.637 0.578 0.635 0.612 0.621 0.653 0.618 0.661 0.656 0.688 0.704 0.653 0.700 Manufacturing All Occupations Constructed (median earnings) from Six Sectors Full Time Total Adjusted Actual for Hours 1890 0.463 1930 0.556 1939 0.539 0.513 0.581 1950 0.537 1951 0.532 1952 0.558 1953 0.512 1954 0.497 1955 0.580 0.526 0.639 0.689 1956 0.583 0.515 0.639 0.690 1957 0.554 0.496 0.633 0.680 1958 0.570 0.477 0.630 0.677 1959 0.580 0.613 0.664
CLAUDIA GOLDIN TABLE 6 (continued) 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 217 Manufacturing Constructed from Six Sectors All Occupations (median earnings) Full Time Total 0.559 0.534 0.557 0.544 0.547 0.532 0.524 0.563 0.549 0.544 0.540 Adjusted Actual for Hours 0.608 0.663 0.594 0.647 0.595 0.652 0.596 0.654 0.596 0.659 0.600 0.666 0.580 0.646 0.578 0.639 0.582 0.644 0.605 0.669 0.594 0.655 0.595 0.653 0.579 0.636 0.566 0.627 0.572 0.627 0.588 0.633 0.602 0.666 0.589 0.648 0.600 0.658 0.596 0.656 0.602 0.646 0.592 0.646 0.617 0.672 0.603 NOTE: Except where noted these ratios are based on mean earnings of full-time year-round employees. aThe range is for New England and the Middle Atlantic. SOURCES: This series has been compiled by the author from a variety of sources. See Goldin (1986c) for details. Data for All Occupations category from O'Neill (1985:Table 1).
218 A HISTORICAL PERSPECTI VE for males, particularly in the clerical and professional sectors. Furthermore, the skill differential across all occupations narrowed considerably after 1940 (Keat, 1960~. Thus, relative earnings for females to males have increased within occupations, and this fac- tor has been of overwhelming importance in accounting for the increase in the earnings ratio over time. It appears that the oc- cupational distribution mattered far less than earnings within oc- cupations in determining the overall earnings ratio of females to mates (Goldin, 1986c; Polachek, 1984~. This finding is particularly noteworthy, since it is generally presumed that the occupational distribution is the primary determinant of the gap between male and female earnings. The degree to which the occupational distribution matters in determining the aggregate ratio of female to mate earnings involves constructing two hypothetical cases. In one, all women have the occupational distribution of men, but their earnings for each occupation remain the same. In the other, counterfactual world, all men have the occupational distribution of the women but have the earnings for each occupation that they ordinarily have. There are then three (or more) measures of the percentage of the earnings gap that is explained by occupations; one minus this measure gives the difference in pay within occupations. The three measures include the two from the hypothetical cases and a third which averages the first two. This technique has been used by Polachek (1984) and Neiman and Hartmann (1981~. One problem with it is that the answer de- pends on how many occupations are used. Polachek uses 195 occupations and finds that somewhat more than 10 percent of the earnings gap is explained by occupational differences; Treiman and Hartmann use 222 occupations and find that somewhere be- tween 11 and 19 percent of the gap is explainecl. Neither of these measures seems particularly large. But when the number of occu- pations rises to 479, more is explained. If the first of the counter- factual worlds is used (and this seems to be the one that makes the most sense in this context), then 19 percent of the pay gap is explained; the second counterfactual world leads to 41 percent being explained.~i Even 19 percent seems rather low. Further- more, it is not clear whether it would be useful to increase the ii Note that these numbers diner from those cited in early printings of Treiman and Hartmann (1981:Table 9), which were in error.
CLAUDIA GOLDIN 219 number of occupations in the analysis beyond 222. The issue of the optimal or correct number of occupations to use is still very much unexplored.~2 Technological change narrowed the gender gap during the pe- riod of early industrialization, when the use of new forms of capital equipment and work organization raised the relative productivity of females and the young. The increase in the manufacturing sec- tor, which was relatively female intensive, also served to raise the overall ratio of female to male earnings. But changes in the earn- ings ratio after the first half of the nineteenth century are more complicated. Employment effects of technical change combined with differing demand, income elasticities expanded certain sec- tors and contracted others, and the increase in education raised the relative earnings of females within occupational groups. The large increase in female labor force participation over the last 30 years has meant that recent labor market entrants have had substantially less job training than have previous participants. Entrants have, until recently, also had less education than the average female population. Therefore, the average wages of the working population of women as a whole have been depressed by the wages of the new entrants, even though the wages of prior participants have been rising. In a period of rising labor force participation, a stable ratio of female to male earnings does not necessarily indicate an absence of economic and social progress for women. CONCLUDING REMARKS Technological advances have altered the employment of wom- en in the following ways throughout American history: 1. Increases in total factor productivity across sectors from 1890 to 1980 have been positively associated with an index of the female intensity of the labor force, but total employment of both males and females has not necessarily been positively relater] to technological advance. i2 One procedure would be to assess the cross-elasticity of substitution among occupations for individuals. Such a procedure would be similar to that used in antitrust cases in deciding what the market is for a particular good, and thus what the definition of a good is. As a general rule, one does not want to disaggregate occupations so finely that occupation itself is an exact proxy for earnings.
220 A HISTORICAL PERSPECTIVE 2. Increases in education have been positively associated with increases in female labor force participation and may be the single most important factor in altering the shape of participation rates for married women during their life cycles. 3. Changes in work organization, in particular an enhanced division of labor in manufacturing from 1820 and in the clerical sector from abound 1900, have increased the demand for female employees. 4. Relative earnings for females to males rose after 1820 and continued to rise to about 1930 or 1940. The aggregate ratio was virtually constant from 1950 to 1980, but has risen during the past half decade. The ratio for manufacturing employment has remained virtually constant since about 1885. Changes in tech- nology seem to be the most likely reason for the initial advance, ant} changes in the experience and education of the working pop- ulation of women appear to have been responsible for much of the recent rise. The secular increase in female employment in America has owed much to the relative growth of particular sectors in the econ- omy, such as those employing clerical and professional workers, and to the decline of others, such as agriculture. Further work will have to perform the major task of separating the role of tech- nological change from those of demand and income elasticities in altering the sectoral distribution of labor. REFERENCES Cain, Louis P., and Donald G. Paterson 1986 Biased technical change, scale, and factor substitution in American industry, 18501919. Journal of Ecorzemic History 56(March):153- 164. Coyle, Grace 1928 Present Fords ire the Clerical Occupations. New York: The Woman's Press. Denison, Edward F. 1962 The Sources of Economic Growth in the United States and the Alternatives Before Us. Supplementary Paper 13. New York: Com- mittee for Economic Development. 1974 Accounting for United States Economic Growth. 1929 to 1969. Washing- ton, D.C.: The Brookings Institution. Easterlin, Richard 1980 Birth and Fortur~c: The Impact of Numbers ore Pcraor~al Welfare. New York: Basic Books.
CLAUDIA GOLDIN 221 Goldin, Claudia 1987 The female labor force and American economic growth, 1890 to 1980. Ch. 10 in S.L. Engerman and R. Gallman, eds., Lor~g-Term Rend in the Americar' Economy. Chicago: University of Chicago Press. 1986a Monitoring and occupational segregation by sex: a historical anal- ysis. Journal of Labor Economics 4(January):1-27. 1986b Life-Cycle Labor Force Participation of Married Women: Historical Evidence and Implications, University of Pennsylvania. Revised version of National Bureau of Economic Research Working Paper #1251. Cambridge, Mass. December. 1986c The Earnings Gap Between Male and Female Workers: An Histor- ical Perspective. National Bureau of Economic Research Working Paper #1888. Cambridge, Mass. April. 1984 The historical evolution of female earnings functions and occupa- tions. Explorations in Economic History 21(January):1-27. 1983 The changing economic role of women: a quantitative approach. Journal of Intcrdi~ciplinary History 13(Spring) :707-733. 1977 Female labor force participation: the origin of black and white differences, 1870 to 1880. Journal of Economic History 37(March):87- 108. Goldin, Claudia, and Kenneth Sokoloff 1982 Women, children, and industrialization in the early republic: evi- dence from the manufacturing censuses. Journal of Economic History 42 (Decembe r) :741-774. 1984 The relative productivity hypothesis of industrialization: the American case, 1820 to 1850. 17`c Quarterly Journal of Ecor~orn~cs 44(August):461-487. Keat, Paul G. 1960 Long run changes in occupational wage structure, 1900-1956. Jour- nal of Political Economy 68(December):584~00. Kendrick, John W. 1961 Productivity fiends in the United Starts. National Bureau of Economic Research. Princeton: Princeton University Press. 1973 Postwar Productivity lFend~inthc United Stacl;ce, 1948-1969. National Bureau of Economic Research. New York: Columbia University Press. 1983 Intcrindwtry Diffcrcnece in Productivity Growth Washington, D.C.: American Enterprise Institute. Mansfield, Edwin 1969 The Economics of Technological Change. New York: Norton. National Research Council 1979 Mcasuremerd and Intcrpretation of Productivity. Report of the Panel to Review Productivity Statistics, Committee on National Statistics. Washington, D.C.: National Academy of Sciences. O'Neill, June 1985 The trend in the male-female wage gap in the United States, Jourr~al of Labor Ecorwmic~ 2(January, Suppl.~:49-70. Polachek, Solomon William 1984 Women in the economy: perspectives on gender inequality. Pp. 34-53 in U.S. Commission on Civil Rights, Comparable Worth Induce
222 A HISTORICAL PERSPECTIVE for the 80'e. Vol. 1. A consultation held June 6-7. U.S. Government Printing Office 1984-524-379. Washington, D.C.: U.S. Commission on Civil Rights. Rosenberg, Nathan 1972 Technology and American Economic Growth. New York: Harper & Row. Rotella, Elyce 1981 From Rome to Office: U.S. Women at Work, 187~1950. Ann Arbor: UMI Press. Smith, James, ed. 1980 Female Labor Supply: Theory and Estimation Princeton: Princeton University Press. Smith, James 1984 Race and human capital. American Economic Review 74 (Septem- ber):685-698. Smith, James, and Michael Ward 1984 Women's Wages and Work in the Twentieth Century. Santa Monica, Calif.: Rand. Sokoloff, Kenneth 1984 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. E~ploratior" in Econorn~c History 21 (October) :351-382. Stoneman, Paul 1983 The Economic Analysis of Technological Change. Oxford, England: Oxford University Press. Treiman, Donald J., and Heidi I. Hartmann, eds. 1981 Women, Work, and Wage`: Equal Payfor Jobs of Equal Value. Wash- ington, D.C.: National Academy Press. Williamson, Jeffrey, and Peter Lindert 1980 American Inequality: A Macroeconomic History. New York: Academic Press.
