We use cookies to distinguish you from other users and to provide you with a better experience on our websites. Close this message to accept cookies or find out how to manage your cookie settings.
Published online by Cambridge University Press: 03 March 2009
Multiple regression analysis reveals that work in New England cotton textile mills during 1905–1912 raised age-adjusted mortality rates over those of non-mill- workers, and that worker mortality increased with years of mill experience. Mortality varied among groups because of differential self selection. Central age group native males with broad occupational choices had lower mortality rates than control groups. Young males, women, and the foreign born had restricted occupational choices. Hence they were less self selected and experienced higher mortality. Death rates were highest among married women workers who bore children. The combination of homework and millwork worsened their health and raised their mortality rates.
1 Hoffman and Frederick S. Crum, both of whom were Prudential executives, pioneered in the study of occupational mortality. For two among many examples of their work see Hoffman, Frederick L., “Mortality from Respiratory Diseases in the Dusty Trades,” Bureau of Labor Statistics, Bulletin 231 (Washington, D.C., 1918)Google Scholar, and Crum, Frederick S., “The Mortality from Diseases of the Lungs in American Industry” Address to the Second Conference of Industrial Physicians (Harrisburg, Pennsylvania, Department of Labor and Industry, May 18, 1916).Google Scholar Both the American Association for Labor Legislation (AALL) and the National Civic Federation were concerned with industrial health. See Bates, [Mrs.] Lindon, Mercury Poisoning in the Industries of New York City and Vicinity (New York, n. d.),Google Scholar and Andrews, John B., Secretary of the AALL, “Protection Against Occupational Diseases,” in Business and the Public Welfare, Proceedings of the American Academy of Political Science 11, 1912, 18–23.Google Scholar For an overview of the work of Alice Hamilton, see Grant, Madelaine P., Alice Hamilton, Pioneer Doctor in Industrial Medicine (New York, 1967).Google Scholar A moden analysis of occupational mortality during this period is by Uselding, Paul, “In Dispraise of the Muckrakers: United States Occupational Mortality, 1890–1910” in Uselding, ed., Research in Economic History, 1 (1976), 334–71.Google Scholar
2 A special concern with the health of women workers and of future generations runs through much of this literature. Combined with other concerns, the result was a massive retreat from laissez-faire—as a number of writers have emphasized. For representative examples of this literature see Goldmark, Josephine, Fatigue and Efficiency: A Study in Industrial New York (New York, 1912),Google Scholar and Irving Fisher, “Industrial Hygiene as a Factor in Human Conservation,” in Business and the Public Welfare, pp. 1–8. Fisher asserted that “the employment under improper conditions of women, especially of married women, means an even more serious waste of life and strength for future generations” (p. 3). Goldmark also cites Fisher as follows: “The present working day is a striking example of the failure to conserve national vitality … A typical succession of events is first fatigue, then colds, then tuberculosis, then death”, See Fatigue and Efficiency, p. 115. For Fisher's work on health within the larger context of his thought see Aldrich, Mark, “Capital Theory and Racism: From Laissez Faire to the Eugenics Movement in the Career of Irving Fisher,” Review of Radical Political Economy, 7 (Fall 1975), 33–42.CrossRefGoogle Scholar The classic argument that factory work impaired women's health is Felix Frankfurter, “The Case for the Shorter Work Day … Bunting vs. Oregon,” (New York: National Consumers' League, 1916). For arguments that the minimum wage was necessary for women workers' health see Seager, Henry R., “The Minimum Wage as Part of a Program for Social Reform,” Annals of American Academy of Political and Social Science, 48 (07 1913), 3–11,CrossRefGoogle Scholarand in the same volume, Brown, H. LaRue, “Massachusetts and the Minimum Wage” pp. 3–21.Google Scholar
3 The argument in the text is not an effort to denigrate the important work that was done. See for example Hamilton's, Alice pioneering studies of lead and other industrial poisoning such as “Lead Poisoning in the Smelting and Refining of Lead,” B.L.S. Bulletin 141 (Washington, D.C., 1914),Google Scholar and Industrial Poisons in the United States (New York, 1925).Google ScholarHoffman, , too, did important studies of silicosis such as the one cited above and also his “The Problem of Dust Phthisis in the Granite- Stone Industry,” B.L.S. Bulletin 293 (Washington, D.C., 1922).Google Scholar
4 Robert Higgs argues that cycles in nineteenth-century urban mortality were the result of immigration waves in part because new immigrants were likely to be especially susceptible. See his “Cycles and Trends of Mortality in 18 Large American Cities, 1871–1900”, Explorations in Economic History, 16 (1979), 381–408.CrossRefGoogle Scholar
5 There is no evidence that byssinosis (brown lung) resulted in excess mortality among these workers. See my “Mortality from Byssinosis Among New England Textile Workers During the Progressive Era,” Journal of Occupational Medicine (forthcoming).Google Scholar
6 Higgs, Robert and Booth, David, “Mortality Differentials Within Large American Cities in 1890,” Human Ecology, 7 (1979), 353–69, show that mortality rates are positively associated with population density.CrossRefGoogle ScholarPubMed
7 The term “kissing the weft” refers to the way weavers threaded the shuttle on a non-automatic loom. When a shuttle became empty a new thread would be placed next to the eye and then the weaver would use his/her lips to suck it through. Hence the phrase. Since more than one weaver typically tended a loom, this was an excellent method of transmitting disease, and it may not be an accident that, as we note below (see fn. 25), the weaving room experienced significantly higher mortality rates than other parts of the mill. It is an interesting footnote in the history of technology that the Northrup automatic loom which did away with hand threading and which was despised by weavers for increasing their workload, may well have improved their health. See Draper, George, Labor Saving Looms (Hopedale, Massachusetts, 1911), p. 191.Google Scholar
8 Barnum, Gertrude, “The Story of a Fall River Mill Girl,” in Taft, Philip and Stein, Leon, When Workers Speak (New York, 1971), p. 29.Google Scholar
9 Some modern information on the healthy worker effect is contained in McMitchael, A. J., “Standardized Mortality Ratios and the ‘Healthy Worker Effect’: Scratching Below the Surface” Journal of Occupational Medicine, 18 (03 1976), 165–68.CrossRefGoogle Scholar
10 Following Thaler and Rosen, let w(p, E, k) represent a family of earnings/risk indifference curves, where w is the acceptance wage, p the job risk and k the income loss due to job risk. Differences in susceptibility can be thought of as causing differing values for k, and “those individuals for whom k is small apply for the risky jobs.” Richard Thaler and Sherwin Rosen, “The Value of Saving a Life: Evidence from the Labor Market,” in Terleckyj, Nestor, ed., Household Production and Consumption, NBER Studies in Income and Wealth, 40 (New York, 1975), pp. 265–301. The quote is from footnote 6.Google Scholar
11 For evidence that a sickly worker effect characterized shoemaking in Britain, see Stewart, Alice, “Problems of Tuberculosis in Industry: A Study of the Shoemaking Trade in Northamptonshire,” British Journal of Tuberculosis and Diseases of the Chest (07 1953), 1–6.Google ScholarPubMed
12 Barnum, “Story of a Fall River Mill Girl,” pp. 29–30.Google Scholar
13 These studies are U. S. Department [then Bureau] of Labor, Report on the Condition of Women and Child Wage-Earners in the United States, 14 “Causes of Death Among Women and Child Cotton Mill Operatives,” (Washington, D. C., 1912),Google Scholar and U.S.B.L.S., “Preventable Death in Cotton Manufacturing Industry”, Bulletin 251 (Washington, D. C., 1919).Google Scholar A medical doctor, Arthur R. Perry, authored both studies. Apparently the investigators took care to see that the data gathered would be accurate. In the first work, every decedent was investigated and death certificates corrected where necessary. In the second work a census of all living mill workers was taken.
Still, errors and biases remain. The foreign-born who returned home to die were excluded, and since for Fall River at least a higher proportion of operatives than non-operatives aged 15–44 were foreign born (64 percent vs. 47 percent), this exclusion biases the study against finding an excess mortality for operatives. Also decedent ex-millworkers who were out of the mill more than two years were treated as non-operatives, and this too biases the study against finding an excess mortality for mill operatives.
14 The data in this paragraph are drawn from the Report on Women and Child Wage Earners, 14, ch. 1 and B.L.S., Bulletin251, Introduction and ch. 1.Google Scholar
15 A minor methodological problem stems from the fact that these are cross-section rather than cohort data. Comparing death rates of individuals aged 30 and 50 who were all born in 1880 is not the same thing as comparing death rates of people aged 30 and born in 1900 with individuals aged 50 and born in 1880. Advances in medicine, between 1880 and 1900 for example, could make the studies yield very different results. Since our comparisons in this work are largely of similar aged individuals but across sexes and occupations, however, this problem is minimized.Google Scholar
16 Life expectancies were calculated from data in Table I and using mortality rates by sex for ages 65 and up from Preston, Samuel H. and others, Causes of Death: Life Tables for National Populations (New York, 1972). Life expectancies for the U.S. population in 1910 are from the same source.Google Scholar
17 Data for these and equations in subsequent tables are cohort mortality rates by age, occupation, ethnicity, and sex, weighted by the ratio of the cohort's population to the average of all cohorts. All equations are estimated by OLS techniques. Because there is no strong justification for any particular functional form or set of variables, one must be wary of tailoring the hypothesis to the data. To avoid this, we partitioned the data and experimented on a subset (three cities for 1906). The equations yielding the best results were then applied to the full sample with the results contained in Table 2.Google Scholar
18 The inability to include nativity as a control for recency of immigration may not be especially serious. Other equations were fitted to Fall River data for 1908–1912 and to subgroups of those data partitioned by sex, occupation, and sex and occupation combined. Foreign birth was never statistically significant, and almost always had a negative sign.Google Scholar
19 If income truly was not a significant determinant of mortality this is a significant finding. Since our income data are for decedents only, however, they may not accurately reflect the incomes of the populations from which the decedents came.
We tried to evaluate our income data by comparing our figures for operative decedents with similar data for living millworkers drawn from the U.S. Immigration Commission, Report on Immigrants in Industries, Part 3, “Cotton Goods Manufacturing in the North-Atlantic States” (Washington, D.C., 1911),Google Scholar General Tables 1 and 18, and U.S. Department of Labor, Report on the Condition of Women and Child Wage-Earners in the United States, I, “Cotton Textile Industry” (Washington, D.C., 1910), p.52. Data for the three groups are as folows: 
Our sample shows significantly higher incomes than either of the other two! This “finding” may well be due to bad data, however. The incomes of decedents were the results of interviews and probably not verified. And both the Immigration Commission and the Report are probably biased downward because they oversampled the foreign-born. In addition, the Report sampled families of women and child wage-earners and it might be expected that they would be relatively poor. These results still leave us ignorant as to how accurate our own income data may be, however.Google Scholar
20 These numbers indicating the percentage impact of the dummy variable were computed as follows. If c is the coefficient of the dummy, its relative impact, g, is: g = exp (C − .5(V)) − 1, where V is the variance of c. The percentage effect of the dummy on deaths—the numbers in the text—is simply 100g. See Halvorsen, Robert and Palmquist, Raymond, “The Interpretation of Dummy Variables in Semilogarithmic Equations” American Economic Review, 70 (06 1981), 474–75,Google Scholar and Kennedy, Peter, “Estimation with Correctly Interpreted Dummy variables in Semilogarithmic Equations,” American Economic Review, 70 (09. 1981), 801.Google Scholar
21 Some multicollinearity is present in the equations in Table 3, which may account for the suspiciously low coefficient of Ln(Age). The simple correlation of Ln(Age) and xp26up is .59. But for age and Xp16to25 it is only .38, and for all the other experience variables it is negative. It is also possible that the low age coefficients partly reflect the healthy worker effect, as they are similar in size to the one for male operatives in equation (5).Google Scholar
22 Data on death rates by experience category for young males, and the percentage of the male population in the mills by age group are from B.L.S., Bulletin 251, General Tables 18 and 46. The decline with increasing age in the proportion of males in the mills is not the result of falling labor force participation, because overall participation rates were higher for males 25–65 than for the younger group. The quotation is from Stafford, Victor, “Influence of Occupation on Health During Adolescence”, U.S. Public Health Service, Bulletin 78 (Washington, D.C., 1916), p.17.Google Scholar
23 Data on the percentage of Fall River residents who were foreign or native born are from B.L.S. Bulletin 251, Table 18. Whether the overrepresentation of immigrant males in textiles was the result of discrimination or their relative dearth of human capital is not the issue here. Discussions of the role of human capital vs. discrimination in shaping the economic position of immigrants can be found in Higgs, Robert, “Race, Skills, and Earnings: American Immigrants in 1909,” this JOURNAL, 31 (06 1971), 420–28,Google Scholar and McGouldrick, Paul and Tannen, Michael, “Did American Manufacturers Discriminate Against Immigrants Before 1914”, this JOURNAL, 37 (09. 1977), 723–46.Google Scholar Neither, unfortunately, deals specifically with the issue of occupational segregation. Aldrich, Mark and Albelda, Randy, “Determinants of Working Women's Wages During the Progressive Era” Explorations in Economic History, 17 (1980), 323–41, argue that some new immigrant women did face job discrimination due to ethnicity, and also discuss female occupational segregation.CrossRefGoogle Scholar
24 Data on mill experience by sex and age from B.L.S. Bulletin 251, General Table 45.Google Scholar
25 If reduced occupational choices outside the mills raised death rates for female workers, the same cannot be said for occupational segregation within the mills. The Labor Department gathered data on mortality by both detailed occupation (e.g. ring spinner) and by workroom on the supposition that mortality was more likely to depend on the general conditions of labor than on the specific task being done. The workrooms were spinning, carding, spooling, and weaving. Regression analyses on pooled data of both male and female workers (not shown) do indicate a significant excess mortality in the weave room. But since women were not disproportionately bunched in the weave room, this cannot explain the relative excess of female mortality.Google Scholar
26 B.L.S. Bulletin 251, p. 86.Google Scholar
27 While the coefficient of marriage is not significant in equation (12), this is not a very firm result as the sample size is quite small. Hence, it is much harder to reject the null hypothesis than to fail to do so, and so we have more confidence in significant than in non-significant results.Google Scholar
28 The effect of marriage and childbirth on female mortality probably cannot simply be added to the earlier-discussed impact of long experience in the mills. While data limitations prevent inclusion of marriage and mill experience in the same equation, we do know that age and experience are correlated, and so are age and marriage. Hence, experience and marriage must be correlated and so to treat them as entirely separate would be to “overexplain” death rates. For women aged 15–35, few of whom had long mill experience, or for women over age 44, few of whom were currently bearing children, overexplanation is not likely to be serious. But for women aged 35–44, who were married, childbearing, and veterans of many years in the mills, experience and marriage could not have been entirely additive. They were not entirely duplicative either, however, for as the data in Table I show, it was in just this age bracket that women workers' death rates exceeded those of male workers by the largest margin.Google Scholar
29 Smuts, Robert, Women and Work in America (New York, 1959), p. 55.Google Scholar
30 See McGaw, Judith, “Technological Change and Women's Work,” in Trescott, Martha, ed., Dynamos and Virgins: Technological Change in History (Metuchen, New Jersey, 1979), 78–99.Google ScholarSmuts, Women Who Work, pp. 56–57, notes that the only two groups for whom it was usual for married women to work were Negroes and women in New England textile mill towns. He incorrectly asserts, however, that it was only immigrant women who worked in this latter group. In fact, over a fifth of native born Fall River married women worked in the cotton mills. Evidence on the incomes of Fall River decedents comes from B.L.S. Bulletin 251, General Table 55.Google Scholar
31 The percent of women employed in the mills and elsewhere is from B.L.S.,Bulletin 251, p. 17.Google Scholar
32 The Labor Department is silent on the source of these data on homework. Almost certainly they were asked retrospectively of family members and are surely subject to substantial error. Measurement error should bias the results toward zero, however.Google Scholar
