¹ The notion that abundant land provided an incentive for labor-saving technical change can be traced back to many early nineteenth century observers. The best known recent exposition of this argument is Habakkuk, H. J., American and British Technology in the Nineteenth Century: The Search for Labour Saving inventions (London, 1967);Google Scholar however, see Clarke, Richard and Summers, Lawrence, “The Labour Scarcity Controversy Reconsidered,” Economic Journal, 90 (03 1980), 129–39.CrossRefGoogle Scholar for a more formal modeling of the role of land availability in the process of industrialization. While the major emphasis has been on the use of capital to conserve on labor, Habakkuk recognized that females and children were also used (p. 65). Lebergott, Stanley, Manpower in Economic Growth: The American Record Since 1800 (New York, 1964) contains an insightful discussion of the substitution of female and child for adult male labor and the workings of this early market in unskilled labor (pp. 125–29).Google Scholar The literature on the substitution of females and children for male labor in the British experience is extensive. See, for example, Pinchbeck, Ivy, Women Workers in the Industrial Revolution (London, 1930),Google Scholar and Mathias, Peter, The First Industrial Nation: An Economic History of Britain, 1700–1914 (London, 1969), Chapters 1 and 5.Google Scholar

² See Dublin, Thomas, Women at Work: The Transformation of Work and Community in Lowell Massachusetts, 1826–1860 (New York, 1979)Google Scholar and Nickless, Pamela J., “Changing Labor Productivity and the Utilization of Native Women Workers in the American Cotton Textile Industry: 1825–1860,” unpublished Ph.D. dissertation, Purdue University, 1976,Google Scholar among other recent additions to this literature. Abbott, Edith, Women in Industry: A Study in American Economic History (New York, 1910) has coverage of most of the important industries, but with little systematic analysis of the pre-1850 period.Google Scholar The employment of women in the paper industry is discussed in McGaw, Judith A., “A Good Place to Work.’ Industrial Workers and Occupational Choice: The Case of Berkshire Women,” Journal of Interdisciplinary History, 10 (Autumn 1979), 227–48.CrossRefGoogle Scholar The cotton textile industry has also dominated the literature on the substitution of child and female labor in Britain, e.g., Nardinelli, Clark, “Chikl Labor and the Factory Acts,” this JOURNAL, 40 (12 1980), 739–55.Google Scholar

³ This conception of early industrialization seems consistent with that of Cochran, Thomas, Frontiers of Change: Early Industrialism in America (New York, 1981) who writes:Google Scholar“Anintermediate stage between the mechanized factory using water power and handwork done at home was the shop that brought together a large number of handworkers. Here there could be minute division of labor, hence a decreasing need for general skill, and constant supervision could ensure a more reliable volume of production (p. 57).”Google Scholar

⁴ This issue is discussed below in section III, but receives a more extensive treatment in Goldin, Claudia and Sokoloff, Kenneth, “The Relative Productivity Hypothesis of Industrialization: The American Case, 1820–1850,” Quarterly Journal of Economics (forthcoming).Google Scholar

⁵ Goldin and Sokoloff, “The Relative Productivity Hypothesis.”Google Scholar

⁶ U.S. House of Representatives, Documents Relative to the Statistics of Manufactures in the U.S., 2 Vols., Serial Set Numbers 222 and 223 (Washington, D.C., 1833),Google Scholar is commonly, and will be henceforth, referred to as the McLane Report, after the then Secretary of the Treasury, Louis McLane. Each of the three data sets has problems that complicate the analysis. Among the most serious defects are the quality of the coverage differed substantially by geographic region and that small manufacturing establishments are underrepresented (in nearly all areas) in the 1820 and 1832 data. These sample selection biases prevent a straightforward calculation of aggregate totals, but allow the computation of averages for classes of manufacturing firms, and a re-weighting of them to arrive at manufacturing sector averages. Another of our concerns is that the categories of information on employees and their wages reported vary from census to census (or survey in the case of the McLane Report). In the 1820 census, adult males and females were enumerated apart from children, often separately listed as boys and girls. However only a total (annual) wage bill was given. The 1832 McLane Report generally listed adult males separately from boys less than 17 years old, but grouped females of all ages together. Although coverage and detail varied considerably by state, wages in the McLane Report were typically given as an average daily (or weekly) wage for each class of employees. The 1850 census distinguished only between males of all ages and females of all ages, and reported the average monthly wage for each group. For further information concerning the 1820 and 1832 samples, see Sokoloff, Kenneth L., “Industrialization and the Growth of the Manufacturing Sector in the Northeast, 1820–1850,” unpublished Ph.D. dissertation. Harvard University, 1982.Google Scholar

⁷ For example, Lebergott, , Manpower in Economic Growth, used the wage data in the McLane Report, and Chandler, Alfred, The Visible Hand: The Managerial Revolution in American Business (Cambridge, Massachusetts, 1977),Google Scholar employed the data on large enterprises from that manufacturing survey. See Bateman, Fred and Weiss, Thomas, A Deplorable Scarcity: The Failure of Industrializa tion in the Slave Economy (Chapel Hill, North Carolina, 1981), for a discussion of the collection and analysis of their sample of firm data from the 1850 census.Google Scholar

⁸ The proportion of the labor force employed in agriculture can be computed for 1820, 1840, and 1850 from information contained in U.S. Department of State, Census for 1820 (Washington, D.C., 1821);Google Scholar U.S. Department of State, Sixth Census or Enumeration of the Inhabitants of the United States (Washington, D.C., 1841);Google Scholar and U.S. Census Office, A Compendium of the Seventh Census (Washington, D.C., 1854)Google Scholar. The figures are reported below and they indicate that the share of the labor force in agriculture was shrinking after 1820, if not before. They are flawed in some respects, however, and thus provide only rough estimates.

Although the 1820 and 1840 censuses were designed to include men, women, and children in their labor force estimates, the 1850 data were explicitly confined to males over the age of 15.

⁹ Most manufacturing industries, both mechanized and non-mechanized, appear to have realized significant gains in average firm size, the capital to labor ratio, and labor productivity between 1820 and 1850. See Sokoloff, “Industrialization and the Growth of the Manufacturing Sector,” Chapters 2, 3, and 4.Google Scholar

¹⁰ The 1850 figure is 28.8 percent if one includes the workers classified as “clothiers and tailors,” but drops to 24.1 percent if one does not. Clothiers and tailors seem to have been overcounted in 1850 relative to both the 1832 McLane Report and the 1860 census, and it seems likely that home sewers were included in the 1850 figures. See note (d) to Table I for further information.Google Scholar

¹¹ The 1820 and 1832 proportions of the manufacturing labor force composed of women and children were estimated by weighting the proportions of particular categories of firms, to adjust for potential biases. Our method uses the definitionGoogle Scholar

where Pf is the aggregate proportion of the manufacturing labor force that is female, S_j is the percentage of the total manufacturing labor force employed in firms of size class j, X_ji is the percentage of the labor force in size class j working in industry i, and is the percentage of the labor force in firms of industry i and size class j that is female. The under-enumeration of small firms in both 1820 and 1832 is reflected in biased S_js and X_jis. We have assumed that the observed L_ijs are unbiased, but have varied the assumptions about the other weights.

The sample of firms from the 1820 Census of Manufactures was drawn from randomly selected counties. Since the omission of firms from the census was apparently a function of their size rather than industry, it seems reasonable to assume that the sample X_jis are unbiased, but that the S_js are biased by the under-enumeration of small establishments. The Table I estimate uses a conservative assumption about the S_j distribution, that 40 percent of the labor force was in the small category (1 to 5 workers), 40 percent in the medium (6 to 15 workers), and 20 percent in the large category. Assuming an equal division of the labor force among size classes (33.3 percent in each) gives 10.6percent of the manufacturing labor force composed of adult females and 25.0 percent composed of children. A very extreme assumption of 25 percent, 25 percent, and 50 percent in the respective classes yields 12.8 percent adult females and 27.4 percent children.

The sample drawn from the McLane Report overrepresents firms from New England, particularly Massachusetts, and therefore it is likely that both the X_jis and the S_js are biased. To check the sensitivity of our estimates, we have varied the assumptions about both distributions. The 1832 estimate in Table I is based on the sample X_jis and an assumption that 25 percent, 25 percent, and 50 percent of the labor force were employed in the respective size classes. An assumption of 33.3 percent in each class yields an estimate of 26.7 percent females and 7.6 percent boys; an assumption of 18 percent, 15 percent, and 68 percent (the distribution observed in the 1850 sample), yields 38.9 percent females and 8.0 percent boys. Changing the X_jis to those observed in 1850 and keeping the S_ijs also at their 1850 levels, yields 26.8 percent females and 7.7 percent boys. Using 25 percent, 25 percent, and 50 percent, for the S_js, but taking the X_jis from 1850 yields 22.6 percent females and 7.8 percent boys. Since several of the highly female-intensive industries, such as textiles, were in relative decline by 1850, this latter estimate is substantially biased downward.

¹² See, for example, Cole, Arthur H., The American Wool Manufacture, 2 vols. (Cambridge, Massachusetts, 1926);Google ScholarHazard, Blanche, The Organization of the Boot and Shoe Industry in Massachusetts Before 1875 (Cambridge, Massachusetts, 1921);Google Scholar and Ware, Caroline F., The Early New England Cotton Manufacture: A Study in Industrial Beginnings (Boston, 1931).Google Scholar

¹³ As shown in Table 2, even Massachusetts, the textile center of the United States, experienced a decline in the proportion of all employed females in manufacturing working n textiles, from 48 percent in 1837 to 36 percent in 1850.Google Scholar

¹⁴ This is not a comprehensive list of all manufacturing industries in which 30 percent of the workers in 1850 were female. The problem of underenumeration of small firms in the 1820 census and the 1832 McLane Report prevents us from calculating precise estimates, at the industry level, of the proportion of the labor force composed of women and children in those years. To illustrate that these classes of workers were prominent in many industries, we present the raw figures (unadjusted for sample selection bias) for selected industries:

¹⁵ Sokoloff, “Industrialization and the Growth of the Manufacturing Sector,” Chapter 2.Google Scholar

¹⁶ All three of these ways of saving on a scarce factor of production were discussed by Habakkuk in his American and British Technology. The ensuing debate on his work, however, focused solely on the substitution of capital for labor, in spite of Habakkuk's clear acknowledgment of there having been several classes of labor, some of which were associated with the use of capital (see p. 65 in particular). For examples of the debate, see Temin, Peter, “Labor Scarcity and the Problem of American Industrial Efficiency in the 1850s,” this JOURNAL, 26 (09. 1966), 277–98,Google Scholar and Fogel, Robert W., “The Specification Problem in Economic History,” this JOURNAL, 27 (09. 1967), 283–308.Google Scholar

¹⁷ Regressions of the log (fixed capital/value added) across our 1832 firms indicate that only in textiles (wool and cotton) and paper was capital intensity significantly and positively related to the percentage of the labor force composed of women and children and to the total size of the labor force (constructed as a weighted average of the three classes of labor). When the regression is estimated without industry dummy variables, the coefficients on both the labor force variable and the percentage of employees that were women and children are significantly negative.Google Scholar

¹⁸ Davis, Pearce, The Development of the American Glass Jndusry (Cambridge, Massachusetts, 1949), p. 48.Google Scholar

¹⁹ Hazard, The Organization of the Boot and Shoe Industry, pp. 85–86.Google Scholar

²⁰ Sex and age are not merely proxies for skill, they are also proxies for opportunity costs. For a more extensive discussion of this topic see Goldin and Sokoloff, “The Relative Productivity Hypothesis,” and for a caution about the use of sex as a proxy for skill see Nickless, Pamela J., “A New Look at Productivity in the New England Cotton Textile Industry, 1830–1860,” this JOURNAL, 39 (12. 1979), 889–910.Google Scholar

²¹ While much of the literature on the role of the factory has stressed the importance of machines in accounting for the increase in the scale of firm from cottage industry, another segment has pointed to the role of discipline and supervision. Marglin, Stephen, “What Do Bosses Do? The Origins and Functions of Hierarchy in Capitalist Production,” Review of Radical Political Economics, 6 (Summer 1974), 33–60 is the best source on this point.Google Scholar

²² For two recent, but different, views on the causes of industrial expansion and of regional variation in industrialization see Field, Alexander, “Sectoral Shift in Antebellum Massachusetts: A Reconsideration,” Explorations in Economic History, 15 (04. 1978), 146–71,CrossRefGoogle Scholar and Zevin, Robert Brooke, The Growth of Manufacturing in Early Nineteenth Century New England (New York, 1975).Google Scholar

²³ In a conventional two-sector, two-input model, when one sector grows more rapidly than the other, it necessitates bidding inputs away from the other sector. As a result, the return to the input in which the more rapidly growing sector is intensive, will rise relative to the return of the other factor. See Goldin and Sokoloff, “The Relative Productivity Hypothesis,” for a more formal treatment.Google Scholar

²⁴ Wage rate data for women and children prior to industrialization of the Northeast are scarce, and even when such data are found, they have often already been affected by the quick upward response of female and child wages that occurred when opportunities for work in manufacturing establishments became available. In his article on the Brandywine area of Delaware, which industrialized very early, Adams reports nearly identical estimates of (W_p/W_m) for agriculture (domestic work) and manufacturing. See Adams, Donald Jr, “Workers on the Brandywine: The Response to Early Industrialization,” Working Papers from the Regional Economic History Research Center (1980), Vol. 3, No. 4.Google ScholarWare, The Early New England Cotton Manufacture, p. 241, reviews the evidence on the increase in the wages of domestics during this period.Google Scholar

²⁵ Firms in the McLane Report generally listed wage rates separately for men, females, and boys. We have checked the information provided by some firms with alternative sources of data, and have concluded that the reported wage rates were averages, across skill classes, of the wage rates of all workers in the particular category (i.e., females). For example, we checked the wage rates reported by the Hamilton Manufacturing Company in Lowell, Massachusetts, with the more extensive breakdown of female and male wages in that firm provided in Dublin, Women at Work, p. 66. A number of firms explicitly indicated that their labor force figures were annual averages. Since enumerators appear to have recognized the issues involved and preferred annual averages, we suspect that most firms sought to provide yearly averages. The firms included in the 1820 census also appear generally to have sought to estimate and report yearly averages.Google Scholar

There is no question that the wage rates from the 1850 census should be viewed as averages across skill classes for all workers in the particular category (males or females). Wage data were provided in the form of separate monthly wage bills for males and females. Firms in industries known to have employed many boys reported male wage rates (male wage bill/number of male employees) that were discernibly lower. It is unclear whether the labor force figures from 1850 are yearly averages or simply a count of workers during the most recent month. In any event, they do not involve the overcounting of workers with high annual turnover.

²⁶ Estimates of the wage of boys relative to that of adult males are not presented in Table 5, but are included in Goldin and Sokoloff, “The Relative Productivity Hypothesis,” Tables I and 2. These ratios increase from about 0.15 in agricultural New England in 1815 to between 0.41 and 0.45 in industrial New England in 1832. We cannot compute a comparable figure for 1850 because children were not listed as a separate category. See footnote 6.Google Scholar

²⁷ As indicated above, we cannot compute a wage rate for boys in 1850. Nevertheless, the increase in the relative wage of boys between 1815 and 1832, as well as various impressionistic evidence, suggest that the relative boy wage continued for a time to rise with the relative female wage.Google Scholar

²⁸ Our 1850 wage ratio (w_f/w_m) differs from that cited in Lebergott, Manpower in Economic Growth, and used by Paul David in his work on technical change in cotton textiles, David, Paul, Technical Choice, Innovation and Economic Growth: Essays on American and British Experience in the Nineteenth Century (London, 1975). The 1850 Census of Manufactures did not clearly state whether there was a lower age limit for the laborers included, and the 1880 Census of Manufactures, in a survey of trends, mistakenly claimed that the 1850 returns covered only adult laborers. In fact, the 1850 returns surveyed firms about all laborers, and thus the inclusion of boys in 1850 imparts a downward bias to the 1850 male wage when it is compared to the 1832 figure, for which boys are separated. Thus the Lebergott-David data indicate a marked increase in (w_p/w_m) from 1832 to 1850 and a sharp decrease in the male money wage over the same period. In constructing our 1850 figure we have used only industries in which boys were a small percentage ofthe labor force, thus eliminating the problem of noncomparability of wages over time. This confusion over the meaning of the 1850 wage rates may explain why Nickless, “A New Look at Productivity in the New England Cotton Textile Industry,” obtains different results from David by using skill classes rather than sex distinctions.Google Scholar

It should be noted as well that these wage ratios are lower than the ratio of the female wage to the adult male wage in the agricultural South. The southern ratio was approximately 0.58 in 1860. See Table I, Goldin and Sokoloff, “The Relative Productivity Hypothesis.”

²⁹ The evidence on the ratio (w_p/w_m) in manufacturing from 1885 to 1960 is from Goldin, Claudia, Economic Change and American Women (in progress).Google Scholar

³⁰ Some firms in 1832 did report wage rates separately for adult females and girls. In such cases, we used only the adult female wage in the regressions. Our method of computing the wage ratios in 1832 and 1850 has introduced another downward bias. The calculations were based on the assumption that the average size of firms employing females was the same as that of firms employing adult males. Since wage rates were positively related to firm size, and females generally worked in larger establishments than men, the estimated male wage is biased upward relative to the female wage. Hence, the estimated wage ratio will be biased downward slightly.Google Scholar

³¹ Gallatin, Albert, “Free Trade Memorial,” reprinted in Taussig, F. W. (ed.), State Papers and Speeches on the Tariff (Cambridge, Massachusetts, 1892), p. 129.Google Scholar

³² McLane Report, Vol. 2, p. 73 and p. 77.Google Scholar

³³ McLane Report, Vol. 1, p. 742 and Vol. 2, p. 22.Google Scholar

³⁴ Carey, Henry C., Essay on the Rate of Wages; With an Examination of the Causes of the Differences in the Condition of the Labouring Populations Throughout the World (Philadelphia, 1835), p. 26.Google Scholar

³⁵ The extent of urbanization in the local county was calculated as the fraction of the county population residing in cities with a population of 2500 or more. Since the poor environmental conditions thought to have affected many industrial workers were often linked to urban areas, our variable should be a reasonable proxy. The size of the manufacturing firm (as measured by the number of workers) might also be a useful proxy for undesirable working conditions for which employees would require compensation. Early factories may have had higher levels of noise anddirt than did the small shops (or the farms) they replaced. These factories also appear to have been distinguished by a more regimented organization of work.Google Scholar

³⁶ For example, the urban-rural wage differential might be due to costs of migrating, or to higher average skill levels in urban areas. The positive relationship between male wages and firm size could be attributable to the disproportionate number of higher quality or more experienced workers in large firms.Google Scholar

³⁷ As has been emphasized above, the average female worked (within an industry) in a larger establishment than did the average male. The assumption that there was no difference between the two groups tends to bias our estimates of (w_f/w_m) downward slightly in each year. As for changes over time, it has the effect of exposing females and adult males to the same movement in the proxy for working conditions. Hence, compensatory payments due to increases in firm size over time could only account for the advance in the relative wages of females if a unit change in firm size had a greater effect (in percentage terms) on female wages than on male wages. As is clear from the regressions, this was not the case.Google Scholar

³⁸ The role of population density in the industrial development of New England is discussed in Ware, The Early New England Cotton Manufacture, p. 14.Google Scholar

³⁹ Goldin and Sokoloff, “The Relative Productivity Hypothesis.” This paper asks how exogenous differences between the agricultural sectors of two economies affect the pace and pattern of industrial development, with examples drawn from the histories of the U. S. North and South.Google Scholar

⁴⁰ In 1850, nearly 60 percent of the white male (over age 15) labor force in the South was still principally employed in the agricultural sector. The proportion in the Northeast was below 35 percent.Google Scholar

⁴¹ These figures were computed from information contained in U. S. Census Office, Abstract of Statistics of Manufactures, According to the Returns of the Seventh Census. The discrepancy is even greater if one makes the comparison between the most female-intensive industries and the least female-intensive industries or employs the number of workers (rather than firms) as the measure of development.Google Scholar

⁴² For example, within the female-intensive boots/shoes, cotton textile, hats/caps, paper, and wool textile industries, northeastern firms had, on average, 2.6 times as many employees as southern firms did in 1850. Within the male-intensive flour mill, glass, iron furnace, nail, and tanning industries, northeastern firms had on average, only 1.2 times as many employees.Google Scholar

⁴³ In most of the pre-industrial Northeast, women, and to a lesser extent children, seem to have worked only occasionally in the agricultural sector. See Bidwell, Percy W. and Falconer, John F., History of Agriculture in the Northern United States (Washington, D. C., 1925), especially p. 116 and p. 275,Google Scholarand our discussion in Goldin and Sokoloff, “The Relative Productivity Hypothesis.”Google Scholar

⁴⁴ Hamilton, Alexander, “Report on Manufactures,” reprinted in Taussig, State Papers and Speeches, p. 19.Google Scholar

⁴⁵ McLane Report, vol. 2, p. 141.Google Scholar

⁴⁶ There is, however, evidence that during the early period, 88 percent of women working in the large textile mills at Lowell were under 30 years old. See Dublin, Women at Work, p. 258, footnote 9. Even in 1888, after manufacturing had become far more concentrated in urban areas, about 86 percent of all female industrial workers were under 30 years old.Google Scholar See Wright, Carroll, Working Women in Large Cities: Fourth Annual Report of the Commissioner of Labor, 1888 (Washington, D. C., 1889).Google Scholar

⁴⁷ The manufacturing participation rates are computed from firm reports of the average number of female workers employed over the year. They are, accordingly, implicitly adjusted for the job turnover that Dublin has found among female workers in Lowell. See his Women at Work, pp. 59–60. To the extent that young women frequently went in and out of the labor force, our estimates will understate the percentage of them engaged in the market economy for some portion of the year.Google Scholar

⁴⁸ The decline in the female proportion of the manufacturing labor force has played a prominent role in the history of the cotton textile industry. See Dublin, Women at Work, and Ware, The Early New England Cotton Manufacture, both of whom stress immigration and technical change as causal factors.Google Scholar

⁴⁹ The data underlying the calculaton are from the 1832 sample and U. S. Census Office, Abstract of Statistics of Manufactures, According to the Returns of the Seventh Census. Using a slightly different formulation from that discussed in footnote II, the female share

and the sources of change have been averaged over the two ways of factoring p_f1850 – p_f1832.