---
title: "Deskilling: Firm Evidence from Europe and the United States"
authors: ["Thor Berger"]
date: "2026-11-17"
source_pdf: "https://thorberger.se/wp/Berger%20(2026)%20Deskilling.pdf"
format: "Markdown text extracted from the PDF"
---

[PDF version](https://thorberger.se/wp/Berger%20(2026)%20Deskilling.pdf)

This Markdown file contains text extracted from the PDF, including references and appendices. Consult the PDF for figures and mathematical notation. Tables are retained as fixed-width text to preserve column alignment and coefficient signs.

<!-- PDF page 1 -->

## **Deskilling: Firm Evidence from Europe and the United States**<sup>∗</sup> 

Thor Berger<sup>†</sup> 

November 17, 2026 

##### **Abstract** 

Technological change raised the demand for skilled workers over the past century, but is believed to have reduced the demand for skills during the industrial revolution. I use census data covering half a million industrial firms in Europe and the United States to show that deskilling was pervasive: the rise of the factory was associated with a relative shift toward unskilled workers—women and children—in all countries and within virtually all industries. Yet the main driver of deskilling was not mechanization: firms using inanimate power employed relatively fewer unskilled workers, suggesting that the driver of deskilling may have been organizational rather than technological changes and that the origins of skill-biased technological change may be traced to the shift from hand to inanimate power in the nineteenth century. 

**Keywords:** Deskilling; human capital; industrialization; technological change **JEL Codes:** J24; O33; N60 

> ∗I am very grateful to Amanda Gregg for helpful comments and for sharing the Russian manufacturing census data and to Richard Hornbeck, Anders Humlum, and Martin Rotemberg for providing early access to the digitized manuscripts from the Census of Manufactures. I also thank numerous other colleagues and seminar participants that provided valuable comments and feedback. Funding from Riksbankens Jubileumsfond and Jan Wallanders och Tom Hedelius stiftelse samt Tore Browaldhs stiftelse is also gratefully acknowledged. Any remaining errors are my own. 

> †Lund University, SCAS, CEPR, and Research Institute of Industrial Economics. Email: thor.berger@ekh.lu.se.

<!-- PDF page 2 -->

## **1 Introduction** 

Technological change has increased the relative demand for skilled workers over the past century (Goldin and Katz, 2008), a process led by large firms adopting new technologies and organizational practices (Atack et al., 2008; Engbom et al., 2025; Nicholas, 2026). In the nineteenth century, large firms likewise spearheaded technological and organizational change (Chandler, 1977). Yet contemporaries believed that the rise of the large factory led to deskilling, as skilled artisans were displaced by unskilled workers and machinery (Allen, 2018; Frey, 2019; Acemoglu and Johnson, 2024). After a century of skill-biased technological change, concerns over deskilling have recently reignited, as skilled workers appear to be the ones most exposed to technological advances in the twenty-first century (Eloundou et al., 2024; Liu et al., 2025; Althoff and Reichardt, 2026). 

Influential contemporaries across the political spectrum—from Smith to Marx—viewed the rise of the factory as fundamentally deskilling. Today, economists largely echo that interpretation. For example, the canonical account by Goldin and Katz (2008, p. 124) concludes that “the industrial revolution with the shift into various types of factories was, overall, deskilling”, while Acemoglu (2002, p. 12) characterizes the factory as the most skillreplacing technology of the past two centuries. At the same time, a growing body of work shows that technological advances during industrialization often increased the demand for skills (for example, Caselli, 1999; Autor et al., 2020; Franck and Galor, 2022; Becker et al., 2026), while reviews of the literature conclude that there is no empirical support for the deskilling thesis (Brugger and Gehrke, 2018).<sup>1</sup> 

A lack of consensus on the skill bias of technological change during industrialization partly reflects the scope of existing evidence, which often examines a single country, industry, or technology. Yet as Acemoglu (2002, p. 8) noted: “[i]n the absence of detailed econometric evidence, it is difficult to generalize from the historical examples and conclude that technical change was definitely skill-replacing during the nineteenth century.” More than two decades later, the central empirical challenge remains: we lack data that permit a systematic econometric analysis of whether deskilling was a general feature in countries undergoing industrialization. 

This paper provides a systematic examination of the deskilling thesis by leveraging recently digitized firm-level census data for more than half a million manufacturing firms in 

> 1The lack of consensus is underscored by the fact that even leading authorities have advanced opposing conclusions over time: while Goldin and Katz (2008, p. 124) concluded that “the switch to the factory most likely _decreased_ the demand for skilled relative to unskilled labor in manufacturing”, Katz (2025, p. 2) more recently argued that “[t]he shift from the artisanal shop to the factory and larger-scaled enterprises with water and steam power [...] increased the demand for more-educated workers” (also see Autor et al. (2020)).

<!-- PDF page 3 -->

nineteenth-century Europe and the United States. The data enable a systematic analysis of the relationship between firm size, technology use, and the composition of the workforce in Russia, Sweden, and the United States—three countries that differ considerably in terms of culture, industrialization, and labor-market institutions (see Online Appendix A)—allowing me to examine whether deskilling was a general feature of industrialization. 

To examine whether the rise of the factory involved a deskilling of the industrial workforce, I focus on the employment of women and children—the only consistently available proxy for skill in historical census data. On average, women and children had lower levels of skills than adult male workers, earned lower wages, and disproportionately occupied less-skilled roles on the factory floors (Goldin and Sokoloff, 1984; Goldin, 1990; Burnette, 2008; Burnette and Stanfors, 2020; Humphries and Weisdorf, 2015; Delabastita et al., 2025). Moreover, women and children constitute a relevant measure of unskilled workers because (i) they constituted a large share—about a quarter to a third among the firms in the sample—of the industrial workforce; and (ii) because contemporary concerns about deskilling focused precisely on the substitution of women and children for skilled adult male workers in the factory (Ure, 1835; Gaskell, 1833; Marx, 1976).<sup>2</sup> 

I show that the rise of the factory was associated with a substantial deskilling of the workforce by comparing firms active within the same detailed industry and location, which absorbs a wide range of factors that may shape firms’ employment decisions that vary either across locations or industries. In all countries and within virtually all industries, I find a greater share of unskilled workers in larger firms, which is corroborated in a battery of additional checks. Figure 1 contrasts these historical patterns with the modern context, by showing that the skill intensity across the firm size distribution has reversed over the past century. 

To go beyond a purely descriptive account, I use two complementary approaches. First, I construct an instrument for firm size using a leave-one-out logic based on average size of firms in the same granular industry located in other regions, which uses the fact that some goods are efficiently produced in larger establishments than others. Second, I use probabilistic linkage techniques to track Swedish and U.S. firms across censuses. Both instrumenting for firm size and exploiting changes in size within firms show that larger firms increased their employment of women and children and reduced the average wages paid to workers. 

Why was the rise of the large factory associated with deskilling? I examine two broad explanations. First, a traditional explanation going back to contemporaries such as Gaskell 

> 2However, it is an open question whether the employment of women and children correlates with other measures of worker skills. Using matched worker-firm data for Swedish firms in the early twentieth century, I show that the share of women and children in a firm is associated with a lower share of educated workers and a larger share of unskilled jobs (see Online Appendix F).

<!-- PDF page 4 -->

<!-- Start of picture text -->
45 15<br>170<br>% Without High-School Degree<br>(CPS, 2020-2025) 13<br>35 150<br>11 130 CPS, 2020-2025<br>25<br>9 110<br>15<br>7 90<br>Census of Manufactures, 1880<br>% Women and Children<br>(Census of Manufactures, 1880)<br>5 5 70<br>Under 10 10 to 24 25 to 99 100 to 499 500 to 999 1000+ Under 10 10 to 24 25 to 99 100 to 499 500 to 999 1000+<br>Firm size (workers) Firm size (workers)<br>(a) Unskilled Workers (b) Average Wages<br>Figure 1: Firm Size and Skill Intensity in the Nineteenth and Twenty-First Centuries<br>Notes: This figure shows that larger manufacturing firms in the nineteenth century employed relatively more<br>unskilled workers (Panel A) and paid lower average wages (Panel B) compared to smaller firms, whereas<br>larger manufacturing firms today employ relatively more skilled workers and pay substantially higher wages.<br>The historical data are drawn from the 1880 Census of Manufactures restricted to firms operating full-time,<br>year-round, with at least one worker and reported wages. The modern data come from the IPUMS 2020–2025<br>CPS ASEC samples (FloodFlood et al.,, 2025),), restricted to full-time wage and salary workers in manufacturing<br>aged 16–64 with one employer in the previous calendar year. For these respondents, I use reported total<br>pre-tax wage and salary income and an indicator for lacking a high-school degree, and calculate averages<br>across firm-size categories using the supplied person weights. Both samples are trimmed by excluding the<br>top and bottom wage percentiles.<br>(1833), Ure (1835), and Marx (1976) is that the main driver behind deskilling during the<br>rise of the factory was the adoption of steam power that facilitated mechanization (Landes,<br>1969). Second, an alternative explanation instead highlights organizational changes—that<br>occurred both in mechanized and non-mechanized firms—that enabled a greater division of<br>labor and task simplification, a more disciplined work regime, and managerial control, which<br>facilitated the employment of unskilled workers such as women and children (Pollard, 1963,<br>1965; Braverman, 1974; Marglin, 1974; Goldin and Sokoloff, 1982; Atack et al., 2024).<br>% women and children<br>% without high-school degree<br>Average wages (under 10 = 100)<br><!-- End of picture text -->

_Notes:_ This figure shows that larger manufacturing firms in the nineteenth century employed relatively more unskilled workers (Panel A) and paid lower average wages (Panel B) compared to smaller firms, whereas larger manufacturing firms today employ relatively more skilled workers and pay substantially higher wages. The historical data are drawn from the 1880 Census of Manufactures restricted to firms operating full-time, year-round, with at least one worker and reported wages. The modern data come from the IPUMS 2020–2025 CPS ASEC samples (FloodFlood et al.,, 2025),), restricted to full-time wage and salary workers in manufacturing aged 16–64 with one employer in the previous calendar year. For these respondents, I use reported total pre-tax wage and salary income and an indicator for lacking a high-school degree, and calculate averages across firm-size categories using the supplied person weights. Both samples are trimmed by excluding the top and bottom wage percentiles. 

I first provide suggestive evidence that larger firms had a greater division of labor, more intensive work regimes, and more developed formal supervision. Then, I systematically document that American, Swedish, and Russian firms that did _not_ rely on inanimate power exhibit a similarly strong link between firm size and unskilled labor, which I corroborate by examining the hand-powered brickmaking industry where larger brickworks employed greater shares of women and children, which historical sources mainly attribute to organizational differences. In contrast to traditional explanations, I show that firms that relied on steam power employed substantially _fewer_ women and children and paid higher average wages, both consistent with a greater skill intensity. Together, these findings suggest that organizational

<!-- PDF page 5 -->

changes may have been an important driver of deskilling. 

I contribute systematic evidence that deskilling accompanied the rise of the factory. As discussed in more detail in the next section, existing studies have often focused on individual countries, industries, or technologies (Goldin and Sokoloff, 1982; Katz and Margo, 2014; Cockriel, 2024; Vipond, 2025; Allen, 2018; Atack et al., 2024; Acemoglu and Johnson, 2024; Schneider, 2026), leaving open whether deskilling was a more general feature of industrialization. I show deskilling was prevalent in three very different industrializing countries and within a wide range of industries, which provides a stronger empirical foundation for the canonical view that industrialization involved deskilling (Braverman, 1974; Acemoglu, 2002; Goldin and Katz, 2008). 

However, my findings contrast with the traditional view that the main driver of deskilling was technological change by showing that the transition to steam was associated with a greater demand for skill, consistent with a growing body of work (Caselli, 1999; Atack et al., 2004; Autor et al., 2020; De Pleijt et al., 2020; Franck and Galor, 2022; Becker et al., 2026; Berger, 2026). Instead, deskilling appears to mainly have been driven by organizational changes consistent with evidence of deskilling before steam became widespread and in industries with limited mechanization (Marglin, 1974; Goldin and Sokoloff, 1982; Sokoloff, 1984; Katz and Margo, 2014), as well as Atack et al. (2024) showing that organizational changes were a key driver of deskilling. 

An implication is that when discussing the impacts of the factory on the demand for skills, one should make a distinction between the factory as an organizational form (reducing the demand for skill) and the technologies adopted within it (raising the demand for skill), which could potentially reconcile disagreements over how the rise of the factory shaped the demand for skills. A second implication is that past technological change may not have been so different from the technologies of the twentieth and twenty-first centuries—electricity, the internet, or industrial robots—which reduced the relative demand for unskilled workers (Goldin and Katz, 1998; Akerman et al., 2015; Acemoglu and Restrepo, 2020; Restrepo, 2024). 

## **2 Context: Deskilling and the Rise of the Factory** 

Industrial production was historically dominated by skilled artisans who carried out the entire production process relying on hand tools. However, by the early twentieth century, the factory—combining inanimately-powered machinery with a large workforce—had largely displaced artisanal production. 

Classical economists including Smith and Ricardo believed the factory entailed a per-

<!-- PDF page 6 -->

vasive deskilling of the industrial workforce, by decomposing production into simpler tasks and substituting unskilled workers and machinery for craft knowledge. In particular, the concerns and moral outrage were directed at the adoption of powered machinery and the rising employment of women and children that displaced adult male workers (Humphries, 2010; Tuttle, 2021), which was manifested in worker resistance to new technologies (Nuvolari, 2002; Frey, 2019; Caprettini and Voth, 2020). Notably, according to contemporary observers this was not an incidental consequence of the factory system but one of its defining features: 

It is, in fact, the constant aim and tendency of every improvement in machinery to supersede human labour altogether, or to diminish its cost, by substituting the industry of women and children for that of men; or that of ordinary labourers, for trained artisans. (Ure, 1835, p. 23) 

However, the subsequent empirical literature is inconclusive regarding whether the rise of the factory generally led to deskilling. An influential strand of the empirical literature supports the still canonical view of the classical economists (Acemoglu, 2002; Goldin and Katz, 2008). In early U.S. manufacturing, Goldin and Sokoloff (1982) shows that women and children became increasingly important and that larger establishments relied more heavily on unskilled labor (also see Katz and Margo (2014)). At the same time, other evidence points to a more direct substitution of capital for skill (James and Skinner, 1985) and that technological breakthroughs in specific industries reduced the earnings and employment of incumbent workers (Cockriel, 2024; Vipond, 2025; Allen, 2018; Acemoglu and Johnson, 2024; Schneider, 2026). 

By contrast, a recent body of work instead emphasizes that the rise of the mechanized factory often raised the demand for skill (Caselli, 1999; Bessen, 2011; Autor et al., 2020; Katz, 2025). Evidence from the diffusion of the steam-powered factory points to greater human capital accumulation and demand for skills in Britain, France, and Germany (De Pleijt et al., 2020; Franck and Galor, 2022; Becker et al., 2026), while other studies show that mechanization reduced reliance on child labor (Brown et al., 1992; Ridolfi et al., 2023; Berger, 2026). Atack et al. (2004) show that wages were higher in more capital-intensive and steam-powered U.S. firms, indicating a greater skill intensity, while Autor et al. (2020) interpret the evolution of relative wages as reflecting a rising relative demand for educated labor, echoing evidence that there was no clear deskilling in the aggregate labor market (Katz and Margo, 2014; Heikkuri, 2026). More broadly, a review of the empirical evidence by Brugger and Gehrke (2018, p. 686) concludes: 

[W]e found no evidence that technical change was on average predominantly deskilling in this period. Hence, the often-viewed opinion that nineteenth-century

<!-- PDF page 7 -->

technical change had a strong overall bias against skilled labour may be due to a generalisation of catchy stories, but lacks empirical support. 

Whether deskilling was a general feature of industrialization thus remains an open question, not least because it is hard to generalize from existing studies that focus on individual countries, industries, or technologies. 

## **3 Data** 

I construct a firm-level dataset that allows for a systematic analysis of a firm’s scale, technology use, and workforce composition by drawing on recently digitized manufacturing census data for the United States, Sweden, and Russia. For simplicity, although the censuses were collected at the establishment level, I refer to these units as “firms” throughout the paper, since multi-unit firms were rare in the nineteenth century. 

**United States** I use the complete surviving firm-level manuscripts of the 1870 and 1880 Census of Manufactures from Hornbeck et al. (2024). Enumerators were instructed to record all manufacturing firms with at least $500 in annual sales, which implies that only very small firms were excluded. The manuscripts report the average number of adult men (above 16), women (above 15), and children/youth employed (Atack and Bateman, 1999), which I use to construct the share of women and children in each firm. I also observe each firm’s power source and horsepower, capital and output, total wage payments, operating time, and county/state and industry.<sup>3</sup> To link U.S. firms between the 1870 and 1880 censuses, I use a probabilistic record linkage algorithm that links firms with similar firm/owner names that are active in the same county and industry, as described in more detail in Online Appendix H. 

**Sweden** I use the recently digitized Swedish manufacturing census that collected annual information on all firms that were active within manufacturing proper (Almås et al., 2026). Each firm reports employment by sex (men and women) and age (above and below 18), which I use to calculate the share of women and children in the workforce. Firms also reported their industry, ownership, output, and use of power technologies (steam, water, and animal) and total horsepower. In the panel regressions, I rely on a balanced sample of linked firms that are linked over time using probabilistic linkage algorithms and manual review based on 

> 3I define annual earnings per worker as the reported annual wage bill divided by reported average employment following the adjustments in Atack et al. (2002) and Atack et al. (2004), while excluding firms with missing or non-positive wage bills, employment, or required operating-time inputs.

<!-- PDF page 8 -->

firm/owner names, geographical location, and industry, as described in Almås et al. (2026). Throughout, I restrict the sample to 1864–1895, starting with the major institutional reforms of the Freedom to Trade Act of 1864 and ending in 1895 before the manufacturing census underwent substantial changes. 

**Russia** I use the 1894 manufacturing census compiled by the Imperial Ministry of Finance that provides data on factories in the Russian Empire (Gregg, 2020). The census aimed to cover factories, rather than small workshops: information was to be collected for firms with at least 15 employees, a steam boiler/engine, or other factory devices. While the 1894 census therefore has better coverage of larger firms, it also includes smaller firms that rely on steam and many smaller firms without reported machinery (Gregg and Matiashvili, 2022). The census reports data on firms’ employment of women and children (under 18), which I use to construct the share of women and children in each firm. The census also contains information on location (city or _uezd_ ), industry, output, the number of working days per year, as well as power technologies and horsepower. 

**Sample Definitions** After merging the census data, the dataset contains information on more than half a million industrial firms. I restrict the sample to firms with at least two workers (to exclude artisans working on their own) and no more than 1,000 workers (to exclude firms that likely jointly reported information for several different establishments) and harmonize the granular industry classifications into eight broad industry groups and twenty detailed industry groups. Online Appendix Table B1 provides summary statistics and industry classifications for the approximately 344,000 firms included in the main sample. 

## **4 Analysis** 

### **4.1 Empirical Strategy** 

I first provide descriptive OLS regressions that compare the employment of women and children in industrial firms across the firm size distribution: 



<!-- Start of picture text -->
K<br>Uit =  α  +  λt  + � θk 1 { ln(Size it )  ∈ Bk}  +  εit. (1)<br>k =1<br><!-- End of picture text -->

where _Uit_ is the percentage share of workers who are women or children, _λt_ is a year fixed effect, and Size _it_ denotes the number of workers. I create bins of log firm size in intervals of 0.25 among firms with at least five workers and group firms with 100 or more workers in the top bin; the estimates of _θk_ capture how the employment share of women and children

<!-- PDF page 9 -->

varies in larger firms relative to the omitted category of firms with fewer than five workers, but could reflect a number of omitted factors that may be correlated with firm size. 

I therefore use three complementary empirical approaches. First, I estimate pooled OLS regressions of the form: 



<!-- Start of picture text -->
Uit =  δtrj +  β  ln(Size it ) +  εit, (2)<br><!-- End of picture text -->

where _Uit_ is the percentage share of workers who are women or children, and ln(Size _it_ ) is the natural log of the number of workers in firm _i_ in year _t_ . A key concern is that the association between firm size and the employment share of women and children partly reflects underlying differences across regions (for example, child labor laws, poverty rates, or urbanity) or industries (for example, due to social norms shaping women’s employment opportunities). The regressions therefore include region-by-detailed industry-by-year fixed effects ( _δtrj_ ), where _r_ denotes region—city or _uezd_ in Russia, district ( _fögderi_ ) in Sweden, and county in the United States—and _j_ denotes the consistently defined 20 detailed industry groups. The identifying variation thus comes from differences across firms in the same detailed industry, region, and year, which absorbs a wide variety of potential omitted factors that vary across regions and industries. 

A remaining empirical challenge is that firm size is an endogenous choice. Even among firms operating in the same region, industry, and year, unobserved productivity or managerial capacity may jointly shape an establishment’s scale and workforce composition. Moreover, access to inexpensive female and child labor may itself allow firms to expand, while measurement error in reported employment would attenuate the OLS estimates. 

To address these concerns, the second empirical approach instruments firm size using the leave-one-out average size of firms in the same granular industry in other regions, shifting the identifying variation toward differences in an industry’s efficient scale of production rather than firm-specific labor demand.<sup>4</sup> The exclusion restriction may nevertheless fail if granular industries characterized by larger firms also differ in technologies or social norms that directly affect their reliance on women and children. I address this concern by controlling for power use and residual female and child employment in the same granular industry elsewhere, although these controls cannot rule out all unobserved industry-level differences. 

Third, I leverage the longitudinal nature of the Swedish and U.S. data to examine how the employment of women and children changes as firms grow and adopt new technologies. I estimate firm-level panel OLS regressions: 

4The granular industries are based on the most detailed industrial classifications in each respective census and include 469, 170, and 303 different industries in Russia, Sweden, and the United States, respectively.

<!-- PDF page 10 -->

<!-- Start of picture text -->
30 50<br>25<br>40<br>20<br>30<br>15<br>20<br>10<br>5 Russia 10 Russia<br>Sweden Sweden<br>0 U.S. 0 U.S.<br>Firm size (workers) Firm size (workers)<br>(a) All Firms (b) Non-Powered Firms<br>Figure 2: Firm Size and Unskilled Labor<br><5 5 10 20 50 100+ <5 5 10 20 50 100+<br>% women and children % women and children<br><!-- End of picture text -->

_Notes:_ This figure displays differences in the percentage of workers who are women or children across bins of log firm size, relative to firms with fewer than five workers. Panel A includes all firms; Panel B restricts the sample to firms that use neither steam nor water power. Among firms with at least five workers, bins are 0.25-wide intervals of log firm size; firms with 100 or more workers are grouped in the top bin. The x-axis reports worker counts corresponding to selected log-size bins. The Sweden and U.S. specifications include year fixed effects, while the Russia specification is a single cross section. Shaded areas denote 95% confidence intervals based on standard errors clustered at the city/ _uezd_ level in Russia, district-by-year in Sweden, and county-by-year in the United States. 



<!-- Start of picture text -->
Uit =  αi  +  λt  +  β  ln(Size it ) +  γ 1Steam it  +  γ 2Water it  +  εit, (3)<br><!-- End of picture text -->

where _αi_ and _λt_ denote firm and year fixed effects, _Uit_ is the percentage share of women and children in firm _i_ in year _t_ , and ln(Size _it_ ) is the natural log of the number of workers. In the Swedish panel, _Steam_ and _Water_ switch to one when adoption of the respective technology likely began, whereas in the U.S. panel they capture reported power use in each census year. Additional specifications also include a wider set of region and industry-by-year fixed effects and flexibly control for baseline firm characteristics—ln firm size, an indicator for corporate ownership, and the percentage share of women and children—that may have shaped a firm’s decision to expand and adopt new technologies (Gregg, 2020; Berger and Ostermeyer, 2025). 

### **4.2 Main Results** 

#### **4.2.1 OLS Estimates** 

Figure 2, Panel A, displays estimates of _θk_ in Equation (1) that capture differences in the percentage share of workers who were women or children across bins of log firm size. In all three countries, women and children constituted a greater share of employment in larger firms.<sup>5</sup> In the top bin—firms with 100 or more workers—the share of women and children was 

> 5Online Appendix Figure B1 shows that the pattern is similar when examining raw shares across the firm size distribution in all three countries.

<!-- PDF page 11 -->

about 25 to 26 percentage points higher than among firms with fewer than five workers.<sup>6</sup> However, these estimates are merely descriptive and I therefore next show that a similar association persists when comparing firms operating in the same detailed industry, region, and year. 

Table 1, Panel A, presents estimates of Equation (2). A strong positive link between firm size and the employment share of women and children persists when comparing firms active in the same detailed industry and region in columns 1, 3, and 6. The estimated OLS slopes are similar across countries and imply that a doubling in firm size (say from 10 to 20 workers) is associated with about a 2.1 to 2.3 percentage point increase in the employment share of women and children. 

I provide a battery of robustness checks in Online Appendix C.1 showing that the association between firm size and the share of women and children appears in most industries, when using a variety of measures of firm size, when considering the employment of women and children separately, and when examining firms that differ in terms of ownership, geography, or seasonality. Online Appendix F further uses matched worker-firm data to show that larger firms employed relatively more unskilled workers also when using other other measures of worker skills—including education and unskilled occupations—and that these firm-level skill measures are highly correlated with the share of women and children in the workforce. 

#### **4.2.2 IV Estimates** 

Table 1, Panel A, Columns 2, 4, and 7 report the 2SLS estimates for Russia, Sweden, and the United States, respectively, and confirm that the share of women and children sharply increased with firm size, which is evident across industries (Online Appendix Figure D2).<sup>7</sup> Online Appendix Figure D1 and Table D1 document a strong reduced-form association between the instrument (the average firm size among firms in the same granular industry located in other regions) and the share of women and children. Consistently, the 2SLS estimates in Table 1 are larger than the corresponding OLS estimates, consistent with measurement error or omitted variables biasing OLS downward as discussed above. 

A key concern with the instrument is that differences in firm size may be correlated 

> 6To gauge the aggregate relevance of the finding that large firms employed relatively more women and children, Online Appendix Figure B3 shows that large firms accounted for the largest shares of women’s and children’s employment, suggesting that the shift from small artisan shops to large factories was central to the rise in the employment of women and children. 

> 7A strong first stage association exists and the inference is robust to accounting for weak instruments: the first-stage F-statistics reported at the bottom of Table D1 are large and Online Appendix Table D2 shows that the estimates remain statistically significant at the 5% level using Anderson–Rubin inference in all three countries, though the Swedish estimate is less precise.

<!-- PDF page 12 -->

Table 1: Firm Size, Technology Use, and Unskilled Labor 



```text
                                                           Outcome: % women and children

                                       Russia (1894)          Sweden (1864–1895)               U.S. (1870–1880)

                                       OLS       2SLS        OLS       2SLS      Panel      OLS        2SLS      Panel
                                        (1)        (2)        (3)       (4)       (5)        (6)        (7)        (8)
Panel A. All Firms
Firm size (ln)                       3.218∗∗∗   7.619∗∗∗   2.995∗∗∗   5.124∗∗  5.972∗∗∗   3.287∗∗∗   7.046∗∗∗   4.288∗∗∗
                                     (0.397)    (1.262)    (0.837) (2.350) (0.694)         (0.181)    (0.600)   (0.366)
Region × industry × year FE             Yes       Yes        Yes        Yes       No         Yes        Yes        No
Firm FE                                 No         No         No        No        Yes        No         No         Yes
Year FE                                 No         No         No        No        Yes        No         No         Yes
Kleibergen–Paap F-statistic             —        271.19       —       103.54       —          —      2377.04       —
Mean outcome                           15.73     15.52      22.90      21.20     25.34      9.99       9.98      14.16
Observations                          14079      13883      58542     55746      17787     246960     246925     15676

Panel B. Non-powered firms
Firm size (ln)                      5.181∗∗∗   19.174∗∗∗   3.162∗∗∗   0.539    6.798∗∗∗   4.453∗∗∗  15.295∗∗∗   4.938∗∗∗
                                     (0.944)    (2.005)    (0.615) (2.297) (0.837)        (0.344)    (1.648)     (0.446)
Region × industry × year FE            Yes        Yes        Yes       Yes        No        Yes        Yes         No
Firm FE                                No         No          No        No        Yes       No          No         Yes
Year FE                                No         No          No        No        Yes       No          No         Yes
Kleibergen–Paap F-statistic            —        104.82        —       17.29       —          —       3851.88       —
Mean outcome                          19.70      19.46      21.77     20.83      25.63     11.92      11.92       13.98
Observations                          6754       6593       29213     28392     12267     145274     145238      13300

Panel C. All Firms: Technology Use
Firm size (ln)                     3.673∗∗∗   9.222∗∗∗   3.785∗∗∗   5.880∗∗   6.180∗∗∗   3.812∗∗∗   10.906∗∗∗   4.549∗∗∗
                                   (0.401)    (1.380)    (0.724)    (2.617)    (0.691)   (0.218)     (0.959)     (0.384)
Steam power (=1)                  -2.860∗∗∗  -7.814∗∗∗  -4.225∗∗∗  -6.973∗∗∗  -2.392∗∗   -3.465∗∗∗-11.063∗∗∗    -2.794∗∗∗
                                   (1.027)    (0.912)    (0.928)    (2.548)    (0.960)   (0.355)     (1.229)     (0.968)
Water power (=1)                  -3.528∗∗∗  -5.691∗∗∗    -1.404     -3.143     0.962    -2.466∗∗∗  -5.589∗∗∗   -4.417∗∗
                                   (0.920)    (0.913)    (1.303)    (2.071)    (1.190)   (0.245)     (0.510)     (2.105)
Region × industry × year FE          Yes        Yes        Yes        Yes        No        Yes         Yes         No
Firm FE                               No         No         No         No        Yes        No         No          Yes
Year FE                               No         No         No         No        Yes        No         No          Yes
Kleibergen–Paap F-statistic           —        170.31       —        113.40      —          —        1655.16       —
Observations                        14079      13883      58542      55746      17787    246960      246925      15676
```

_Notes:_ Firm-level OLS, 2SLS, and panel OLS regressions of the percentage of workers who are women or children. Panel A reports the baseline estimates. Panel B restricts the sample to firms that use neither steam nor water power; panel firms remain non-powered in every observed year. Panel C includes steam- and water-power indicators. In the Swedish panel specification (column 5), the indicators switch to one when adoption of the respective technology likely began, whereas in the U.S. panel specification (column 8) they capture reported power use in each census year. Both panel samples exclude firms reporting either power source at entry. Columns 2, 4, and 7 instrument firm size using average ln firm size in the same granular industry in other regions; Panel B constructs the instrument among non-powered firms. Region is city/ _uezd_ , district ( _fögderi_ ), and county in Russia, Sweden, and the United States, respectively. Industry is the harmonized set of 20 detailed groups. Standard errors are clustered at the region level in columns 1–4 and 6–7 and at the firm level in columns 5 and 8.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1. 

with, for example, the degree of mechanization or the stigma relating to women and children working in particular industries. Online Appendix Table D1 reports estimates when controlling for the use of steam and water power, defined analogously to the instrument, and

<!-- PDF page 13 -->

the residual employment shares of women and children that proxy for potential stigma in their employment across industries. There remains a large positive link between firm size and the employment of women and children, though the results for Sweden are imprecisely estimated. To examine robustness to potential violations of the exclusion restriction, I follow Caprettini and Voth (2023) and apply the method of Conley et al. (2012) to show that the direct effect of the instrument has to be substantial to turn the IV estimates insignificant.<sup>8</sup> 

#### **4.2.3 Panel Estimates** 

Table 1, Panel A, Columns 5 and 8 use the linked Swedish and U.S. samples and present estimates of Equation (3) showing that the employment of women and children increased as firms grew larger. A doubling in firm size is associated with an increase of about 4.1 and 3.0 percentage points in the employment share of women and children in Sweden and the U.S., respectively.<sup>9</sup> Similar to the IV estimates, the within-firm association between firm size and the share of women and children is larger compared to the baseline OLS estimates, which again is consistent with a downward bias in the OLS estimates discussed above. 

A natural concern is that the employment of women and children may shape future firm growth, yet Online Appendix Table E1 shows that estimates are similar when controlling for a wide range of baseline firm characteristics including the share of women and children. Another concern given that firms are probabilistically linked is that linkage quality may shape the estimated association. However, Online Appendix Figure E2 shows that the estimated association between firm size and the employment of women and children remains similar in magnitude when restricting the U.S. sample to higher-quality links. 

**Alternative Measures of Deskilling: Wages** As shown in Figure 1, Panel B, there exists a negative cross-sectional association between firm size and average wages across U.S. firms in 1880 (also see (Atack et al., 2004)), which contrasts modern patterns and indicates a lower skill intensity in large firms historically. Table 2 presents corresponding panel estimates: the elasticity between average wages and firm size in column 1 is -0.119, indicating that a 10 percent increase in firm size is associated with about 1.2 percent lower average wages. The elasticity remains similar when adding county-by-industry-by-year fixed effects 

> 8In the baseline specification, the direct effect of the instrument would have to account for about 70% of the reduced-form relationship in Russia and 85% in the United States before the 95% union confidence set includes zero; these thresholds remain similar when including the full IV controls (Online Appendix Figure D3). However, the corresponding threshold is about 10% in the baseline specification and 15% with the full controls in Sweden, indicating that the Swedish estimates are less robust to moderate violations of the exclusion restriction. 

> 9Again, the positive link is evident in most industries (Online Appendix Figure E1).

<!-- PDF page 14 -->

Table 2: Firm Size, Technology Use, and Wages in the United States Outcome: ln FTE annual wages per worker 



```text
                                           All frims         Non-powered frims             All firms

                                         (1)         (2)         (3)         (4)         (5)         (6)

Firm size (ln)                        -0.119∗∗∗  -0.111∗∗∗   -0.124∗∗∗   -0.095∗∗∗    -0.135∗∗∗  -0.128∗∗∗

                                       (0.008)    (0.009)     (0.012)     (0.013)     (0.009)     (0.010)
                                                                                            ∗∗∗         ∗∗∗
Steam power (=1)                                                                      0.138       0.158
                                                                                      (0.019)     (0.022)
Water power (=1)                                                                        0.032      0.026
                                                                                      (0.034)     (0.039)

Firm FE                                  Yes         Yes        Yes         Yes          Yes        Yes
Year FE                                  Yes         Yes        Yes         Yes          Yes        Yes
County × industry × year FE              No          Yes         No         Yes          No         Yes
Mean outcome                            5.95        5.98        5.95        5.98        5.95        5.98
Observations                           25954       21776       12568        9988       25954       21776
```

_Notes:_ Firm-level panel OLS regressions for linked U.S. firms observed in 1870 and 1880. The dependent variable is the log of full-time-equivalent annual wages per worker. Columns 1–2 use all linked firms, columns 3–4 restrict the sample to firms that used neither steam nor water power in either year, and columns 5–6 use all linked firms and control for steam and water power, which capture within-firm changes. The county-byindustry-by-year fixed effects compare firms in the same county and harmonized 20-group industry within each census year. Standard errors clustered at the firm level are reported in parentheses.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1. 

in column 2, when restricting the sample to firms operating full-time (accounting for seasonality) in both years (Online Appendix Table E2), or to higher-quality links (Online Appendix Table E3). 

Together, these results show that firms that grew larger also paid substantially lower average wages further suggesting that the rise of the large factory was associated with deskilling. However, these estimates do not shed light on the underlying mechanisms. 

### **4.3 Mechanisms: Organizational and Technological Changes** 

A positive association between firm size and the employment of women and children could reflect two broad mechanisms: organizational and technological changes. First, it is widely emphasized that a key driver of deskilling was the greater division of labor and factory discipline enabled by the rise of larger enterprises (Pollard, 1963, 1965; Braverman, 1974; Marglin, 1974; Goldin and Sokoloff, 1982; Clark, 1994; Humphries, 2003; Atack et al., 2024). Second, many influential contemporaries such as Gaskell (1833), Ure (1835), and Marx (1976) and later scholarship have argued that the adoption of steam power was the underlying key driver

<!-- PDF page 15 -->

of deskilling (Landes, 1969), which is underscored by the fact that the adoption of steam was spearheaded by larger firms (Atack et al., 2008; Berger and Ostermeyer, 2025). While technological and organizational changes in many cases were deeply intertwined (Gregg, 2020; Juhász et al., 2024; Berger and Ostermeyer, 2025), the two margins are conceptually distinct as evident by the fact that many firms underwent substantial organizational change without adopting new technologies.<sup>10</sup> 

#### **4.3.1 Organizational Changes** 

A key limitation in historical census data is the lack of direct measures of the internal organizational structure of firms. Using unique matched worker-firm data based on detailed surveys of early twentieth century Swedish printing firms, Online Appendix F documents suggestive evidence on organizational differences across the firm size distribution: the division of labor, work intensity, and development of managerial hierarchies were substantially greater among larger firms. However, these differences could be driven by the greater technology use within large enterprises. 

To provide more systematic but indirect evidence that organizational factors shaped the employment of unskilled workers even in the absence of the adoption of new power technologies, I examine the share of women and children among firms that used neither steam nor water power. Figure 2, Panel B, shows that the firm-size gradient is similar in all three countries, which is evident within most broad industries (Online Appendix Figure C2). Table 1, Panel B, further shows that the positive size gradient persists among non-powered firms. The OLS estimates imply that a doubling in firm size is associated with a 2.2 to 3.6 percentage point increase in the share of women and children, while the 2SLS and panel estimates imply larger increases, though the Swedish 2SLS estimate is small and imprecise. 

To make the role of organizational innovations more concrete, I zoom in on brickmaking in Online Appendix G—an industry producing a relatively homogeneous good with limited adoption of inanimate power. Yet there existed stark organizational differences between small and large non-mechanized brickyards: small brickyards and large urban brickworks used broadly similar methods of clay preparation, molding, and firing, relying on hand molds or simple presses, and comparable kiln technologies. However, large firms employed more unskilled labor—including women and children, often in family gangs—to perform tasks such 

> 10At least three major organizational innovations reshaped firms that did not adopt inanimate power. First, the decomposition of skilled artisanal work into simple repetitive tasks that unskilled workers could perform (Marglin, 1974; Braverman, 1974; Goldin and Sokoloff, 1982; Sokoloff, 1984; Atack et al., 2024). Second, the concentration of formerly dispersed workers under one roof lowering the cost of monitoring and coordination (Pollard, 1965; Marglin, 1974), and the related imposition of factory discipline—fixed hours, attendance rules, and continuous supervision (Pollard, 1963; Clark, 1994). Third, the rise of managerial structures that coordinated workers within larger firms (Chandler, 1977; Engbom et al., 2025).

<!-- PDF page 16 -->

as gathering clay, hauling bricks, and assisting male molders with edging and spatting. As described in the Online Appendix, the expansion in scale among these hand-powered firms rested on a greater division of labor and more developed supervision. 

Taken together, the fact that deskilling is similarly evident among firms that did not rely on new technologies is difficult to reconcile with the idea that technological change was the fundamental driver behind deskilling. Indeed, I next show that the use of steam power is _negatively_ associated with the reliance on unskilled workers. 

#### **4.3.2 Technological Changes** 

Table 1, Panel C, compares industrial firms in Russia, Sweden, and the United States showing that the association between a firm’s use of steam power and the share of women and children is large and negative. The estimated magnitudes are large in relative terms: U.S. firms relying on steam power employed about 3.5 percentage points fewer women and children in the OLS specification, which is sizable compared to the mean share of women and children of about 10 percent in the sample. Similarly, water power use is also negatively related to unskilled labor, but less precisely estimated in Sweden. Columns 5 and 8 of Table 1 report difference-in-differences estimates showing that the adoption of steam power was associated with a decline in the share of women and children of about 2.1 and 2.8 percentage points in Sweden and the U.S., respectively, which remain relatively large effects compared to the sample means. Water-power use is also associated with large declines in the U.S., though the reductions are smaller and imprecisely estimated in the Swedish sample.<sup>11</sup> Similarly, Table 2, Columns 5–6, show that the adoption of steam power is associated with a 14 to 16 log point increase in average wages per worker among U.S. firms, which again is consistent with a greater skill intensity after a firm transitions to steam technology. 

Together, these results suggest that while the rise of the large factory involved widespread deskilling, the main driver was not the adoption of new inanimate power technologies, which contrasts traditional accounts that ascribe deskilling primarily to the transition to inanimate power. Instead, it appears that deskilling can more plausibly be ascribed to broader organizational changes that took place both within mechanized and non-mechanized firms alike. 

> 11While the U.S. sample contains only two periods, the Swedish data allow me to examine trends in the employment of women and children before the transition to steam power. Online Appendix Figure E3 presents standard event studies and using the estimator of de Chaisemartin and D’Haultfœuille (2026) showing that there are no systematic pre-trends prior to the transition to steam, while there is a sharp decline after steam is installed.

<!-- PDF page 17 -->

<!-- Start of picture text -->
12<br>Sweden<br>U.S.<br>10<br>8<br>6<br>4<br>2<br>Year<br>1864 1868 1872 1876 1880 1884 1888 1892 1896<br>Coefficient on ln firm size<br><!-- End of picture text -->

Figure 3: Firm Size and Unskilled Labor: Gradients over Time 

_Notes:_ This figure displays year-specific coefficients from firm-level regressions of the percentage of workers who are women or children on ln firm size for Swedish manufacturing firms observed between 1864 and 1895, with comparison estimates for U.S. manufacturing firms in 1870 and 1880. The regressions control for steam and water power. All samples are restricted to firms with 2–1,000 workers. Vertical lines denote 95% confidence intervals based on standard errors clustered by registration district in Sweden and county in the United States. 

## **5 Concluding Questions: Why Did Deskilling End?** 

I use data on hundreds of thousands of industrial firms in Russia, Sweden, and the U.S. to show that larger firms employed substantially more unskilled workers across countries and industries, supporting the canonical deskilling view of Acemoglu (2002) and Goldin and Katz (2008). The findings raise the question: when did the positive association between firm size and unskilled labor in the nineteenth century reverse to the modern negative association shown in Figure 1, where large firms rely more intensively on skilled labor? 

Figure 3 presents estimates from regressions of unskilled labor on firm size separately by year, which show that the firm-size gradient in unskilled labor weakened over time, falling by about 53 percent in Sweden between 1864 and 1895 and by about 14 percent in the United States between 1870 and 1880. While suggestive, these patterns are consistent with large firms becoming relatively more skill-intensive over the latter half of the nineteenth century and a reversal taking place sometime in the early twentieth century. 

A reversal in the early twentieth century aligns with the view that attributes the shift to skill-biased technological change to electrification, which constituted a break with prior

<!-- PDF page 18 -->

deskilling technologies such as steam (Goldin and Katz, 1998). Yet evidence that the adoption of steam reduced the demand for unskilled labor already in the nineteenth century suggests that the shift from deskilling to skill-biased technological change may be traced to the earlier shift from hand to machine power. If technology always raised the demand for skills, it becomes less straightforward to interpret the reversal of skill intensity in Figure 1 simply as a story about new technologies. Instead, it could partly have reflected organizational changes associated with the rise of the managerial firm (Chandler, 1977). In that case, just like the onset of deskilling in the nineteenth century may partly have been driven by organizational changes, its undoing in the twentieth century may also have been driven by similar organizational forces.

<!-- PDF page 19 -->

## **References** 

- **Acemoglu, Daron** , “Technical Change, Inequality, and the Labor Market,” _Journal of Economic Literature_ , 2002, _40_ (1), 7–72. 

   - **and Pascual Restrepo** , “Robots and Jobs: Evidence from US Labor Markets,” _Journal_ 

   - _of Political Economy_ , 2020, _128_ (6), 2188–2244. 

   - **and Simon Johnson** , “Learning from Ricardo and Thompson: Machinery and Labor in 

   - the Early Industrial Revolution and in the Age of Artificial Intelligence,” _Annual Review of Economics_ , 2024, _16_ (1), 597–621. 

- **Akerman, Anders, Ingvil Gaarder, and Magne Mogstad** , “The Skill Complementarity of Broadband Internet,” _The Quarterly Journal of Economics_ , 2015, _130_ (4), 1781–1824. 

- **Allen, Robert C.** , “The Hand-Loom Weaver and the Power Loom: A Schumpeterian Perspective,” _European Review of Economic History_ , 2018, _22_ (4), 381–402. 

- **Almås, Ingvild, Thor Berger, Timo Boppart, Konrad B. Burchardi, Olof Ejermo, Björn Eriksson, Anders Larsson, Hannes Malmberg, Stefan Maukner, Mats Olsson, and Vinzent Ostermeyer** , “Historical Manufacturing Census of Sweden: Data Description and Quality Assessment,” _Historical Methods: A Journal of Quantitative and Interdisciplinary History_ , 2026, _59_ (1), 20–38. 

- **Althoff, Lukas and Hugo Reichardt** , “Task-Specific Technical Change and Comparative Advantage,” Working Paper 12403, CESifo 2026. 

- **Atack, Jeremy and Fred Bateman** , “Nineteenth-Century U.S. Industrial Development through the Eyes of the Census of Manufactures a New Resource for Historical Research,” _Historical Methods: A Journal of Quantitative and Interdisciplinary History_ , 1999, _32_ (4), 177–188. 

   - **, , and Robert A. Margo** , “Part-Year Operation in Nineteenth-Century American Manufacturing: Evidence from the 1870 and 1880 Censuses,” _The Journal of Economic History_ , 2002, _62_ (3), 792–809. 

   - **, , and** , “Skill Intensity and Rising Wage Dispersion in Nineteenth-Century American Manufacturing,” _The Journal of Economic History_ , 2004, _64_ (1), 172–192. 

   - **, , and** , “Steam Power, Establishment Size, and Labor Productivity Growth in Nineteenth-Century American Manufacturing,” _Explorations in Economic History_ , 2008, _45_ (2), 185–198.

<!-- PDF page 20 -->

- **, Robert A. Margo, and Paul W. Rhode** , “De-Skilling: Evidence from Late Nineteenth-Century American Manufacturing,” _Explorations in Economic History_ , 2024, _91_ , 101554. 

- **Autor, David, Claudia Goldin, and Lawrence F. Katz** , “Extending the Race between Education and Technology,” _AEA Papers and Proceedings_ , 2020, _110_ , 347–51. 

- **Becker, Sascha O., Christian Dustmann, and Hyejin Ku** , “The Virtuous Cycle Between Skills and Technology,” Discussion Paper Series 013/26, ROCKWOOL Foundation Berlin (RF Berlin), Berlin 2026. 

- **Berger, Thor** , “Engines of Liberation? Technology and the Decline of Child Labor,” 2026. Mimeo. 

   - **and Vinzent Ostermeyer** , “Institutional Innovation and the Adoption of New Tech- 

   - nologies: The Case of Steam,” _The Journal of Economic History_ , 2025, _85_ (3), 844–873. 

- **Bessen, James E.** , “Was Mechanization De-Skilling? The Origins of Task-Biased Technical Change,” Working Paper 11-13, Boston University School of Law 2011. 

- **Braverman, Harry** , _Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century_ , New York: Monthly Review Press, 1974. 

- **Brown, Martin, Jens Christiansen, and Peter Philips** , “The Decline of Child Labor in the U.S. Fruit and Vegetable Canning Industry: Law or Economics?,” _Business History Review_ , 1992, _66_ (4), 723–770. 

- **Brugger, Florian and Christian Gehrke** , “Skilling and Deskilling: Technological Change in Classical Economic Theory and Its Empirical Evidence,” _Theory and Society_ , 2018, _47_ (5), 663–689. 

- **Burnette, Joyce** , _Gender, Work and Wages in Industrial Revolution Britain_ Cambridge Studies in Economic History, Cambridge: Cambridge University Press, 2008. 

   - **and Maria Stanfors** , “Understanding the Gender Gap Further: The Case of Turn- 

   - of-the-Century Swedish Compositors,” _The Journal of Economic History_ , 2020, _80_ (1), 175–206. 

- **Caprettini, Bruno and Hans-Joachim Voth** , “Rage against the Machines: Labor-Saving Technology and Unrest in Industrializing England,” _American Economic Review: Insights_ , 2020, _2_ (3), 305–320.

<!-- PDF page 21 -->

- **and** , “New Deal, New Patriots: How 1930s Government Spending Boosted Patriotism 

   - During World War II,” _The Quarterly Journal of Economics_ , 2023, _138_ (1), 465–513. 

- **Caselli, Francesco** , “Technological Revolutions,” _American Economic Review_ , 1999, _89_ (1), 78–102. 

- **Chandler, Alfred D.** , _The Visible Hand: The Managerial Revolution in American Business_ , Cambridge, MA: Belknap Press of Harvard University Press, 1977. 

- **Clark, Gregory** , “Factory Discipline,” _The Journal of Economic History_ , 1994, _54_ (1), 128–163. 

- **Cockriel, William Mark** , “Technological Disruption in the 19th Century United States.” PhD dissertation, University of Chicago 2024. 

- **Conley, Timothy G., Christian B. Hansen, and Peter E. Rossi** , “Plausibly Exogenous,” _The Review of Economics and Statistics_ , 2012, _94_ (1), 260–272. 

- **de Chaisemartin, Clément and Xavier D’Haultfœuille** , “Difference-in-Differences Estimators of Intertemporal Treatment Effects,” _The Review of Economics and Statistics_ , 2026, _108_ (4), 863–880. 

- **De Pleijt, Alexandra, Alessandro Nuvolari, and Jacob Weisdorf** , “Human Capital Formation During the First Industrial Revolution: Evidence from the Use of Steam Engines,” _Journal of the European Economic Association_ , 2020, _18_ (2), 829–889. 

- **Delabastita, Vincent, Meredith M. Paker, and Benjamin Schneider** , “Women’s Wages and Job Quality in the 1920s United States,” Oxford Economic and Social History Working Paper 223, University of Oxford 2025. 

- **Eloundou, Tyna, Sam Manning, Pamela Mishkin, and Daniel Rock** , “GPTs Are GPTs: Labor Market Impact Potential of LLMs,” _Science_ , 2024, _384_ (6702), 1306–1308. 

- **Engbom, Niklas, Hannes Malmberg, Tommaso Porzio, Federico Rossi, and Todd Schoellman** , “Economic Development According to Chandler,” Working Paper 34483, National Bureau of Economic Research 2025. 

- **Flood, Sarah, Miriam King, Renae Rodgers, Steven Ruggles, J. Robert Warren, Daniel Backman, Etienne Breton, Grace Cooper, Julia A. Rivera Drew, Stephanie Richards, David Van Riper, and Kari C. W. Williams** , “IPUMS CPS: Version 13.0,” IPUMS, Minneapolis, MN 2025. Dataset.

<!-- PDF page 22 -->

- **Franck, Raphaël and Oded Galor** , “Technology-Skill Complementarity in Early Phases of Industrialisation,” _The Economic Journal_ , 2022, _132_ (642), 618–643. 

- **Frey, Carl Benedikt** , _The Technology Trap: Capital, Labor, and Power in the Age of Automation_ , Princeton University Press, 2019. 

- **Gaskell, Peter** , _The Manufacturing Population of England: Its Moral, Social and Physical Conditions, with an Examination of Infant Labour_ , Baldwin & Cradock, 1833. 

- **Goldin, Claudia** , _Understanding the Gender Gap: An Economic History of American Women_ , New York: Oxford University Press, 1990. 

   - **and Kenneth Sokoloff** , “Women, Children, and Industrialization in the Early Republic: 

   - Evidence from the Manufacturing Censuses,” _The Journal of Economic History_ , 1982, _42_ (4), 741–774. 

   - **and** , “The Relative Productivity Hypothesis of Industrialization: The American Case, 

   - 1820 to 1850,” _The Quarterly Journal of Economics_ , 1984, _99_ (3), 461–487. 

   - **and Lawrence F. Katz** , “The Origins of Technology-Skill Complementarity,” _The Quar-_ 

   - _terly Journal of Economics_ , 1998, _113_ (3), 693–732. 

   - **and** , _The Race between Education and Technology_ , Cambridge, MA: Belknap Press, 

   - 2008. 

- **Gregg, Amanda and Tamar Matiashvili** , “Modernization in Progress: Part-Year Operation, Mechanization, and Labor Force Composition in Late Imperial Russia,” _The Journal of Economic History_ , 2022, _82_ (4), 1143–1182. 

- **Gregg, Amanda G.** , “Factory Productivity and the Concession System of Incorporation in Late Imperial Russia, 1894–1908,” _American Economic Review_ , 2020, _110_ (2), 401–427. 

- **Heikkuri, Suvi** , “Technological Change and Structural Shifts in Sweden’s Labor Market, 1870–1930: Skill Upgrading or Deskilling?,” _Scandinavian Economic History Review_ , 2026, pp. 1–25. 

- **Hornbeck, Richard, Shanon Hsuan-Ming Hsu, Anders Humlum, and Martin Rotemberg** , “Gaining Steam: Incumbent Lock-in and Entrant Leapfrogging,” Working Paper 32384, National Bureau of Economic Research 2024. 

- **Humphries, Jane** , “Child Labor: Lessons from the Historical Experience of Today’s Industrial Economies,” _The World Bank Economic Review_ , 2003, _17_ (2), 175–196.

<!-- PDF page 23 -->

- , _Childhood and Child Labour in the British Industrial Revolution_ , Cambridge University Press, 2010. 

**and Jacob Louis Weisdorf** , “The Wages of Women in England, 1260–1850,” _The Journal of Economic History_ , 2015, _75_ (2), 405–447. 

- **James, John A. and Jonathan S. Skinner** , “The Resolution of the Labor-Scarcity Paradox,” _The Journal of Economic History_ , 1985, _45_ (3), 513–540. 

- **Juhász, Réka, Mara P. Squicciarini, and Nico Voigtländer** , “Technology Adoption and Productivity Growth: Evidence from Industrialization in France,” _Journal of Political Economy_ , 2024, _132_ (10), 3215–3259. 

- **Katz, Lawrence F.** , “Beyond the Race between Education and Technology,” 2025. Mimeo. 

   - **and Robert A. Margo** , “Technical Change and the Relative Demand for Skilled Labor: 

   - The United States in Historical Perspective,” in Leah Platt Boustan, Carola Frydman, and Robert A. Margo, eds., _Human Capital in History: The American Record_ , University of Chicago Press, 2014, pp. 15–57. 

- **Landes, David S.** , _The Unbound Prometheus: Technological Change and Industrial Development in Western Europe from 1750 to the Present_ , Cambridge: Cambridge University Press, 1969. 

- **Liu, Huben, Dimitris Papanikolaou, Lawrence D. W. Schmidt, and Bryan Seegmiller** , “Technology and Labor Markets: Past, Present, and Future; Evidence from Two Centuries of Innovation,” _Brookings Papers on Economic Activity_ , 2025, _2025_ (2), 215–272. 

- **Marglin, Stephen A.** , “What Do Bosses Do?: The Origins and Functions of Hierarchy in Capitalist Production,” _Review of Radical Political Economics_ , 1974, _6_ (2), 60–112. 

- **Marx, Karl** , _Capital: A Critique of Political Economy_ , Vol. 1, Harmondsworth, Middlesex: Penguin Books in association with New Left Review, 1976. Originally published 1867. Introduced by Ernest Mandel and translated by Ben Fowkes. 

- **Nicholas, Tom** , “Human Capital and the Managerial Revolution in the United States: Evidence from General Electric,” _The Review of Economics and Statistics_ , 2026, _108_ (2), 291–310. 

- **Nuvolari, Alessandro** , “The “Machine Breakers” and the Industrial Revolution,” _Journal of European Economic History_ , 2002, _31_ (2), 393–426.

<!-- PDF page 24 -->

- **Pollard, Sidney** , “Factory Discipline in the Industrial Revolution,” _The Economic History Review_ , 1963, _16_ (2), 254–271. 

   - , _The Genesis of Modern Management: A Study of the Industrial Revolution in Great Britain_ , Cambridge, MA: Harvard University Press, 1965. 

- **Restrepo, Pascual** , “Automation: Theory, Evidence, and Outlook,” _Annual Review of Economics_ , 2024, _16_ (1), 1–25. 

- **Ridolfi, Leonardo, Carla Salvo, and Jacob Weisdorf** , “The Effect of Mechanisation on Labour: Evidence from the Diffusion of Steam,” Working Paper 689, CAGE Research Centre 2023. 

- **Schneider, Benjamin** , “Technological Unemployment in the British Industrial Revolution: The Destruction of Hand-Spinning,” _Past & Present_ , 2026, _270_ (1), 105–149. 

- **Sokoloff, Kenneth L.** , “Was the Transition from the Artisanal Shop to the Nonmechanized Factory Associated with Gains in Efficiency? Evidence from the U.S. Manufacturing Censuses of 1820 and 1850,” _Explorations in Economic History_ , 1984, _21_ (4), 351–382. 

- **Tuttle, Carolyn** , _Hard at Work in Factories and Mines: The Economics of Child Labor During the British Industrial Revolution_ , Routledge, 2021. First published 1999 by Westview Press. 

- **Ure, Andrew** , _The Philosophy of Manufactures; or, an exposition of the scientific, moral, and commercial economy of the factory system of Great Britain. With illustrations_ , Charles Knight, 1835. 

- **Vipond, Hillary** , “Technological Unemployment in Victorian Britain: Young Workers and the Collapse of Entry,” 2025. Mimeo.

<!-- PDF page 25 -->

# **Online Appendix** 

# **Deskilling: Firm Evidence from** 

# **Europe and the United States** 



```text
A   Additional Context: Russia, Sweden, and the U.S.                                       A2
B   Additional Descriptive Results                                                         A7
C   Additional OLS Results                                                                A12
     C.1   Additional Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A14
D   Additional IV Results                                                                 A25
E   Additional Panel Results                                                              A30
     E.1   Event-Study Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A32
F   Alternative Measures: Skills and Organizational Change                                A37
     F.1   Firm Size and Skills in the Printing Industry . . . . . . . . . . . . . . . . . A37
     F.2   Firm Size and Organizational Change in the Printing Industry . . . . . . . A40
G   Firm Size and Deskilling in Brickmaking                                               A43
H   Linking U.S. Firms                                                                    A48
References                                                                                A50
```

<!-- PDF page 26 -->

## **A Additional Context: Russia, Sweden, and the U.S.** 

I focus on Russia, Sweden, and the United States partly because of the availability of comparable manufacturing census data, but also because they provide substantial cultural, economic, and institutional variation that is useful to examine whether deskilling was a general feature of industrialization across different historical contexts. 

Online Appendix Figure A1 displays GDP per capita and industrial development in the three countries over the nineteenth century until the outbreak of World War I. By the late nineteenth century, the United States was emerging as the world’s leading industrial economy underpinned by rapid integration of a vast internal market, the expansion of resourceintensive manufacturing, and urbanization (Wright, 1990; Donaldson and Hornbeck, 2016; Hornbeck and Rotemberg, 2024). In 1900, GDP per capita was more than twice as high in the United States as in Sweden and almost eight times as high as in Russia. U.S. industrial output per capita in 1913 was about twice the Swedish level and more than six times the Russian level, while U.S. urbanization rates were more than twice as high (see Table A1). 

While initially a poor agricultural economy, Sweden experienced a rapid industrial takeoff in the 1870s initially driven by an industrial expansion based around natural resources but increasingly revolving around advanced high-tech manufacturing toward the end of the century (Heckscher, 1954; Ljungberg and Schön, 2013; Schön, 2012). While it trailed far behind the U.S. in terms of GDP per capita, Swedish living standards as captured by heights and life expectancy surpassed the U.S. by the late nineteenth century (see Table A1). This take-off has partly been linked to Sweden’s position as Europe’s “impoverished sophisticate” (Sandberg, 1979), which is reflected in comparatively high levels of human capital both in terms of average years of education and numeracy considering its relatively low levels of GDP per capita (see Table A1). By contrast, Russia remained predominantly agrarian even though factory production expanded, which recent work attributes to the legacy of serfdom and limited human capital investments as key factors in explaining the limited industrial expansion and catch-up (Gregg, 2023; Markevich and Zhuravskaya, 2018; Zhuravskaya et al., 2024). 

The different economic and industrial trajectories of Russia, Sweden, and the U.S. are also reflected in more fundamental drivers evident in Table A1, which displays a set of measures of political institutions and proxies for women’s empowerment. Russia measures poorly in terms of institutional quality both compared to Sweden and the United States. Similarly, women’s civil liberties and political empowerment lagged far behind in Russia, while Sweden exhibited greater levels of women’s empowerment compared to the United States. The three countries also differed in their legislation of child labor. U.S. states began

<!-- PDF page 27 -->

to introduce restrictions on child labor in the latter half of the nineteenth century (Moehling, 1999), while Sweden and Russia introduced restrictions in the early 1880s (Gorshkov, 2009; Berger, 2026a). However, I account for state and national legislation through fixed effects in the analysis, and existing evidence suggests that child labor laws were largely ineffective in shaping the employment of children (Moehling, 1999; Berger, 2026b). 

Together, the descriptive account above suggests that the firms I study are active in three countries that differ considerably in economic development, human capital, industrialization, institutions, and women’s empowerment. The fact that the main result—the positive link between firm size and the employment of women and children—is so similar in these three countries strongly bolsters the interpretation that deskilling was a general feature of industrialization.

<!-- PDF page 28 -->

<!-- Start of picture text -->
100<br>Russia<br>Sweden<br>United States<br>80<br>World<br>60<br>40<br>20<br>0<br>1800 1825 1850 1875 1900<br>Year<br>(a) GDP per Capita<br>100<br>Russia<br>Sweden<br>United States<br>80<br>World<br>60<br>40<br>20<br>0<br>1800 1825 1850 1875 1900<br>Year<br>GDP per capita (U.S. 1913 = 100)<br>Industrial output per capita (U.S. 1913 = 100<br><!-- End of picture text -->



<!-- Start of picture text -->
(b) Industrialization<br><!-- End of picture text -->

Figure A1: Growth and Industrialization in Russia, Sweden, and the United States 

_Notes:_ This figure displays GDP per capita (Panel A) and industrial output per capita (Panel B) for Russia, Sweden, and the United States, with a world reference series in each panel. GDP per capita data are from Bolt and van Zanden (2025); industrial output per capita data are from Bairoch (1982).

<!-- PDF page 29 -->

Table A1: Development, Political Institutions, and Civil Liberties Russia Sweden United States 



```text
Economic and human development
GDP per capita, 1900                                    1,049     3,320         8,038
Industrial output per capita, 1913                      15.9      53.2          100.0
Urbanization rate, 1880                                 10.6      12.5           25.0
Height, 1880                                            166.6     171.2         169.5
Life expectancy, 1880                                   31.7      47.7           39.4

Human capital
Average years of education, 1880                        1.00      4.61           6.02
Numeracy, 1880                                          94.7      100.0          95.4

Political institutions and civil liberties
Electoral democracy, 1880                               0.032     0.313         0.333
Rule of law, 1880                                       0.411     0.974         0.886
Political corruption, 1880                              0.439     0.017         0.141
Freedom of expression, 1880                             0.167     0.797         0.762
Freedom of association, 1880                            0.077     0.840         0.862

Women’s empowerment
Women’s political empowerment, 1880/1900 0.219                    0.526         0.303
Women’s civil liberties, 1880                           0.163     0.747         0.435
Women’s civil society participation, 1880               0.304     0.491         0.362
```

_Notes:_ This table reports development, human-capital, institutional, and civil-liberties measures for Russia, Sweden, and the United States. GDP per capita is from the Maddison Project Database 2023 (Bolt and van Zanden, 2025) and is measured in 2011 dollars. Industrial output per capita is from Bairoch (1982) and is indexed so that the United States in 1913 equals 100. Urbanization is from Bairoch and Goertz (1986), Table 3, and is measured as the percentage of the total population living in cities of 5,000 or more inhabitants. Height, average years of education, and numeracy are from Clio Infra (Baten and Blum, 2015; van Leeuwen and van Leeuwen-Li, 2015; Baten, 2015). Clio Infra rows use 1880 when available, then 1900, and otherwise the nearest observation in 1870–1910. Height is male-height equivalent in centimeters, average years of education is for the population age 15 and above, and numeracy is the ABCC estimate in percent. Life expectancy is from the V-Dem Country-Year Dataset v16 (Coppedge et al., 2026) and is measured in years at birth. The political-institution and civil-liberties rows are also from V-Dem; women’s political empowerment uses 1880 for Russia and 1900 for Sweden and the United States because the 1880 observations are missing. The democracy, rule-of-law, corruption, expression, association, and gender rows are 0–1 V-Dem indices. Electoral democracy summarizes suffrage, clean elections, elected officials, association, and expression. Rule of law captures transparent and predictable law, impartial administration, judicial constraints, and access to justice. The gender rows measure women’s political empowerment, civil liberties, and civil-society participation.

<!-- PDF page 30 -->



<!-- PDF page 31 -->

## **B Additional Descriptive Results** 



<!-- Start of picture text -->
35<br>29<br>23<br>17<br>11<br>5<br>Firm size (workers)<br>(a) Russia<br>44<br>39<br>34<br>29<br>24<br>19<br>Firm size (workers)<br>(b) Sweden<br>37<br>31<br>25<br>19<br>13<br>7<br>Firm size (workers)<br>(c) United States<br>2--9 10--19 20--29 30--39 40--49 50--59 60--69 70--79 80--89 90--99 >=100<br>2--9 10--19 20--29 30--39 40--49 50--59 60--69 70--79 80--89 90--99 >=100<br>2--9 10--19 20--29 30--39 40--49 50--59 60--69 70--79 80--89 90--99 >=100<br>% women and children<br>% women and children<br>% women and children<br><!-- End of picture text -->

Figure B1: Women and Children by Firm Size 

_Notes:_ This figure displays the percentage of workers who were women or children across the firm size distribution in Russia, Sweden, and the United States. Percentages are calculated by dividing total employment of women and children by total employment within each country and size category. Note that the y-axis scales differ across countries, as do the age cutoffs used to define children (see Section 3).

<!-- PDF page 32 -->

<!-- Start of picture text -->
40 37 38<br>36<br>32<br>31<br>30 28<br>25<br>23 24 24<br>20 20<br>20 16 16 18 16<br>12<br>10 7 9 8 10 9 8 7<br>4 4 4<br>0<br>Russia (1894) Sweden (1864-1895) U.S. (1870-1880)<br>(a) Unweighted<br>70 66<br>60<br>51<br>50<br>43 41<br>40 37 3535 34 33<br>29<br>30 27 23 24 26 22 23<br>20<br>20 17 1817 16<br>12 12 12<br>10 7 9 8<br>0<br>Russia (1894) Sweden (1864-1895) U.S. (1870-1880)<br>(b) Employment-Weighted<br>Food & Beverages All TextilesPaper & PrintingLeatherGlass, Stone, etc.WoodMetals & MachineryChemicalsFood & Beverages All TextilesPaper & PrintingLeatherGlass, Stone, etc.WoodMetals & MachineryChemicalsFood & Beverages All TextilesPaper & PrintingLeatherGlass, Stone, etc.WoodMetals & MachineryChemicals<br>Food & Beverages All TextilesPaper & PrintingLeatherGlass, Stone, etc.WoodMetals & MachineryChemicalsFood & Beverages All TextilesPaper & PrintingLeatherGlass, Stone, etc.WoodMetals & MachineryChemicalsFood & Beverages All TextilesPaper & PrintingLeatherGlass, Stone, etc.WoodMetals & MachineryChemicals<br>% women and children<br>% women and children<br><!-- End of picture text -->

Figure B2: Employment of Women and Children Across Industries 

_Notes:_ This figure displays the average percentage of women and children in total employment for all manufacturing firms and broad industry groups in Russia (1894), Sweden (1864–1895), and the United States (1870 and 1880). Panel A reports unweighted firm-level means; Panel B reports employment-weighted means.

<!-- PDF page 33 -->

<!-- Start of picture text -->
75<br>60<br>45<br>30<br>15<br>0<br>Firm size<br>(a) Russia<br>75<br>Women and children, 1864<br>Women and children, 1895<br>All workers, 1864<br>60 All workers, 1895<br>45<br>30<br>15<br>0<br>Firm size<br>(b) Sweden<br>75<br>Women and children, 1870<br>Women and children, 1880<br>All workers, 1870<br>60 All workers, 1880<br>45<br>30<br>15<br>0<br>Firm size<br>(c) United States<br>2-4 5-9 10-19 20-49 50-99 100+<br>2-4 5-9 10-19 20-49 50-99 100+<br>2-4 5-9 10-19 20-49 50-99 100+<br>Share of workers (%)<br>Share of workers (%)<br>Share of workers (%)<br><!-- End of picture text -->

Figure B3: Employment of Women and Children Across Firm Sizes 

_Notes:_ This figure displays the distribution of women and children across firm-size categories. Bars report each size category’s share of total employment of women and children, while lines report each size category’s share of total employment. Panel A reports Russia in 1894; Panel B reports Sweden in 1864 and 1895; Panel C reports the United States in 1870 and 1880. Firm-size categories are defined using total employment.

<!-- PDF page 34 -->

<!-- Start of picture text -->
3.5 r = 0.67 3.5 r = 0.66 Clothing & Apparel 3.5 r = 0.73 Clothing & Apparel<br>2.5 Clothing & Apparel 2.5 2.5<br>1.5-.5.5 Lumber & Sawn WoodGrain, Flour & StarchBeverages & VinegarMeat, Fish, Dairy & CanningChemicals, Oils, Matches & ExplosivesLeather ProcessingStone, Clay, Ceramics & GlassTextile Finishing & DyeingMachinery & ElectricalWood Products & FurnitureSugar, Chocolate & CoffeeNon-Ferrous MetalsIron & SteelVehicles & ShipsInstruments & ClocksPaper & PrintingYarn, Spinning & WeavingShoes & Leather GoodsTobacco 1.5-.5.5 Lumber & Sawn WoodGrain, Flour & StarchBeverages & VinegarMeat, Fish, Dairy & CanningChemicals, Oils, Matches & ExplosivesLeather ProcessingStone, Clay, Ceramics & GlassTextile Finishing & DyeingMachinery & ElectricalWood Products & FurnitureSugar, Chocolate & CoffeeNon-Ferrous MetalsIron & SteelVehicles & ShipsInstruments & ClocksPaper & PrintingYarn, Spinning & WeavingShoes & Leather GoodsTobacco 1.5-.5.5 Leather ProcessingVehicles & ShipsLumber & Sawn WoodTextile Finishing & DyeingMachinery & ElectricalStone, Clay, Ceramics & GlassWood Products & FurnitureInstruments & ClocksGrain, Flour & StarchBeverages & VinegarIron & SteelChemicals, Oils, Matches & ExplosivesSugar, Chocolate & CoffeeNon-Ferrous MetalsPaper & PrintingShoes & Leather GoodsYarn, Spinning & WeavingMeat, Fish, Dairy & CanningTobacco<br>-1.5 -1.5 -1.5<br>-2.5 -2.5 -2.5<br>-3.5 -3.5 -3.5<br>-3.5 -2.5 -1.5 -.5 .5 1.5 2.5 3.5 -3.5 -2.5 -1.5 -.5 .5 1.5 2.5 3.5 -3.5 -2.5 -1.5 -.5 .5 1.5 2.5 3.5<br>Russia: standardized share Russia: standardized share Sweden: standardized share<br>(a) Russia and Sweden (b) Russia and the U.S. (c) Sweden and the U.S.<br>3 r = 0.72 3 r = 0.74 Clothing & Apparel 3 r = 0.85 Clothing & Apparel<br>2 Clothing & Apparel 2 Yarn, Spinning & Weaving 2 Yarn, Spinning & Weaving<br>Yarn, Spinning & Weaving<br>1 Meat, Fish, Dairy & CanningChemicals, Oils, Matches & ExplosivesShoes & Leather GoodsPaper & Printing Tobacco 1 Meat, Fish, Dairy & Canning Paper & Printing Tobacco 1 Paper & PrintingMeat, Fish, Dairy & CanningTobacco<br>-10 Lumber & Sawn WoodGrain, Flour BeveragesMachinery & ElectricalLeather ProcessingSugar, Chocolate & CoffeeStone, Clay, Ceramics & GlassVehicles & Ships &  Textile Finishing & DyeingNon-Ferrous MetalsVinegarStarchIron & SteelInstruments & ClocksWood Products & Furniture -10 Lumber & SawGrain, Flour & StarchBeverages & VinegarMachiLeather ProcessingSugar, Chocolate & Coffee n Stone, Clay, Ceramics & GlassVehicles & Ships WoodTextile Finishing & Dyeingery & ElectricalNon-Ferrous MetalsChemicals, Oils, Matches & ExplosivesIron & SteelInstruments & ClocksWood Products & FurnitureShoes & Leather Goods -10 Machinery & ElectricalVehicles & ShipsLeather ProcessingInstruments & ClocksWood Products & FurnitureSugar, Chocolate & CoffeeStone, Clay, Ceramics & GlassLumber & Sawn WoodIron & SteelBeverages & VinegarGrain, Flour & StarchTextile Finishing & DyeingNon-Ferrous MetalsChemicals, Oils, Matches & ExplosivesShoes & Leather Goods<br>-2 -2 -2<br>-3 -3 -3<br>-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3<br>Russia: standardized share Russia: standardized share Sweden: standardized share<br>(d) Russia and Sweden (e) Russia and the U.S. (f) Sweden and the U.S.<br>U.S.: standardized share U.S.: standardized share<br>Sweden: standardized share<br>U.S.: standardized share U.S.: standardized share<br>Sweden: standardized share<br><!-- End of picture text -->

Figure B4: Employment of Women and Children Across Industries and Countries 

_Notes:_ This figure compares standardized industry percentages of women and children across detailed industry groups for each pair of countries. In all panels, values are standardized within country across the detailed industry groups. The top row reports the unweighted mean firm-level percentage within each country-bydetailed-industry cell. The bottom row reports the aggregate detailed-industry percentage, computed as total employment of women and children divided by total employment within each country-by-detailed-industry cell. Labels report detailed industry names; dashed lines denote the 45-degree line.

<!-- PDF page 35 -->

Table B1: Summary Statistics 



```text
                                            Russia (1894) Sweden (1864–1895) U.S. (1870–1880)

Employment
% women and children                            15.81             22.99               9.92
Number of workers                               48.12             30.84              12.13

Power use
Steam power (0/1)                               0.42              0.32                0.27
Water power (0/1)                               0.11              0.21                0.15

Geography
Urban (0/1)                                     0.45              0.48                0.47

Broad industry
Food & beverages (0/1)                          0.42              0.16                0.18
Textiles (0/1)                                  0.16              0.17                0.15
Leather (0/1)                                   0.10              0.16                0.05
Wood (0/1)                                      0.06              0.09                0.26
Paper & printing (0/1)                          0.05              0.05                0.03
Glass, stone, etc. (0/1)                        0.08              0.16                0.06
Chemicals (0/1)                                 0.04              0.05                0.03
Machinery & metals (0/1)                        0.09              0.16                0.24

Detailed industry
Grain, flour & starch (0/1)                     0.17              0.04                0.11
Sugar, chocolate & coffee (0/1)                 0.03              0.01                0.01
Meat, fish, dairy & canning (0/1)               0.04              0.02                0.02
Beverages & vinegar (0/1)                       0.17              0.06                0.02
Tobacco (0/1)                                   0.01              0.04                0.03
Yarn, spinning & weaving (0/1)                  0.12              0.07                0.02
Textile finishing & dyeing (0/1)                0.02              0.08                0.00
Clothing & apparel (0/1)                        0.02              0.02                0.06
Shoes & leather goods (0/1)                     0.01              0.01                0.09
Leather processing (0/1)                        0.07              0.15                0.02
Lumber & sawn wood (0/1)                        0.04              0.03                0.14
Wood products & furniture (0/1)                 0.02              0.04                0.12
Paper & printing (0/1)                          0.05              0.05                0.03
Chemicals, oils, matches & explosives (0/1)     0.06              0.06                0.02
Iron & steel (0/1)                              0.02              0.06                0.09
Non-ferrous metals (0/1)                        0.02              0.02                0.04
Machinery & electrical (0/1)                    0.05              0.05                0.04
Vehicles & ships (0/1)                          0.01              0.01                0.06
Stone, clay, ceramics & glass (0/1)             0.08              0.16                0.06
Instruments & clocks (0/1)                      0.00              0.02                0.01

Observations                                   16019             72973              255044
```

_Notes:_ This table reports average outcomes for manufacturing firms in Russia (1894), Sweden (1864–1895), and the United States (1870 and 1880). The sample restricts to firms with at least two and no more than 1,000 workers. The urban row denotes establishments located in cities in Russia and Sweden; for the United States, where establishment-place urban status is not consistently observed, it is a county-level indicator equal to one if more than one-third of the county population lived in cities with at least 2,500 residents.

<!-- PDF page 36 -->

## **C Additional OLS Results** 



<!-- Start of picture text -->
Food & Beverages<br>Textiles<br>Leather<br>Wood<br>Paper & Printing<br>Glass, Stone, etc.<br>Chemicals Russia<br>Sweden<br>Metals & Machinery U.S.<br>-5 0 5 10 15<br>Coefficient on ln firm size<br><!-- End of picture text -->

Figure C1: Firm Size and Unskilled Labor Across Industries 

_Notes:_ This figure displays broad-industry-specific OLS estimates of the coefficient on ln firm size from regressions of the percentage of women and children in total employment on ln firm size. The Russia regressions include region fixed effects, while the Sweden and U.S. regressions include year _×_ region fixed effects. Region corresponds to city/ _uezd_ , district ( _fögderi_ ), and county for Russia, Sweden, and the United States, respectively. Industry corresponds to the harmonized set of eight broad industry groups. Horizontal bars denote 95% confidence intervals based on standard errors clustered at the region level.

<!-- PDF page 37 -->

<!-- Start of picture text -->
Food & Beverages<br>Textiles<br>Leather<br>Wood<br>Paper & Printing<br>Glass, Stone, etc.<br>Chemicals Russia<br>Sweden<br>Metals & Machinery U.S.<br>-5 0 5 10 15<br>Coefficient on ln firm size<br><!-- End of picture text -->

Figure C2: Firm Size and Unskilled Labor Across Industries: Non-Powered Firms 

_Notes:_ This figure displays broad-industry-specific OLS estimates of the coefficient on ln firm size from regressions of the percentage of women and children in total employment on ln firm size. The sample is restricted to firms that use neither steam nor water power. The Russia regressions include region fixed effects; the Sweden and U.S. regressions include year _×_ region fixed effects. Region corresponds to city/ _uezd_ , district ( _fögderi_ ), and county for Russia, Sweden, and the United States, respectively. Bars denote 95% confidence intervals based on standard errors clustered at the region level.

<!-- PDF page 38 -->

### **C.1 Additional Robustness** 

This section provides a set of robustness checks to document that the association between firm size and the employment of women and children is evident in most industries, that it is not sensitive to how I define firm size or a “factory”, or pooling women and children, and that it is broadly similar when accounting for differences in firm ownership, geographical location, and firm operation time. 

While most traditional accounts focus on deskilling within textiles, Online Appendix Figures B2 and B4 show that women and children were extensively employed in many other industries. Online Appendix Figure C1 shows that the estimated association emerges within virtually all broad industries in all three countries, with the exception of metals and machinery. A lack of a relationship in metals and machinery is consistent with the fact that the rise of machine tools and interchangeable parts within large firms was associated with a growing demand for skilled machinists, blacksmiths, and engineers trained in fitting, measuring, and tool-making (Hounshell, 1984). Indeed, the average firm in the metals and machinery sector employed relatively few women and children in all three countries (Online Appendix Figure B2). To address the more stringent concern that firms employing women and children sorted into particular tasks within broad industry groups, Online Appendix Table C1 estimates the baseline specification separately in five narrowly defined industries that can be matched across all three countries, restricting attention to industries with at least 100 firms in each country, documenting a positive size gradient. 

I next show that the results are similar when using alternative ways to measure firm size, or to define a factory. First, adding firm owners to the worker counts accounts for the possibility that the labor input of owners may not have been included in reported worker counts (Sokoloff, 1984; Margo, 2015), which leaves the association largely unchanged as shown in Online Appendix Table C2. 

The share of women and children was also more pronounced in large firms and (manu)factories when using typological measures to delineate different firm types as in Sokoloff (1984) and Berger and Ostermeyer (2025a): for example, Online Appendix Table C2 shows that a large factory (defined as a firm with at least 16 workers) had a 5.1 to 7.3 percentage point higher employment share of women and children compared to a smaller firm. Similarly, the share of women and children was substantially greater within manufactories (large firms not relying on inanimate power) and factories (firms relying on inanimate power), compared to artisan shops (small firms not relying on inanimate power). The highest shares of women and children are observed in manufactories, which did not rely on inanimate power but rather relied on achieving productivity gains through division of labor, factory discipline, and improved coordination (Sokoloff, 1984).

<!-- PDF page 39 -->

Online Appendix Table C2 further shows that results are broadly similar when measuring firm size using nominal output, though the association in Sweden is imprecisely estimated. Relatedly, Online Appendix Table C3 shows that results are similar when restricting the sample to firms with at least 15 workers, which makes the inclusion criteria more similar and accounts for the fact that smaller firms may not have been consistently included in the Russian census (see Section 3). 

Online Appendix Table C4 shows that the results are not driven by combining adult women and children since the same positive size gradient is evident when examining women and children separately. Moreover, Online Appendix F uses linked employer-employee data for Swedish printing firms to show that (i) the share of women within a firm is correlated with other measures of worker skills including educational attainment and occupation/task-based measures; and (ii) that large firms employed relatively more workers without a secondary education and in unskilled jobs. Together, these results suggest that the association between unskilled workers and firm size is not driven by a particular definition of unskilled workers. 

Online Appendix Table C5 replaces the indicators for the use of steam and water power with reported horsepower per worker from inanimate sources. The negative estimates are consistent with the argument that the transition to inanimate power, particularly steam power, was associated with a reduction in the share of unskilled workers. 

I then show that the pattern is robust across geographies and different types of firms. First, Online Appendix Table C6 shows that the positive association between firm size and the share of women and children is evident in both urban and rural samples.<sup>12</sup> Second, the type of firm ownership may be correlated both with firm size and the share of unskilled workers. For example, female owners were more likely to employ female workers than maleowned firms, though not children (Gozen et al., 2025). However, female-owned firms were substantially smaller on average, which would serve to bias upward the share of women in the left tail of the firm size distribution, rather than the right tail. Similarly, larger firms were more likely to be incorporated (Gregg, 2020; Berger and Ostermeyer, 2025b), which could shape the employment of women and children if corporations had more “modern” employment practices. Online Appendix Table C7 uses the fact that the Swedish census recorded information on owners and shows that the association between firm size and the share of women and children is robust to controlling for reported ownership characteristics. Third, Online Appendix Table C8 shows that results are similar both among Russian and U.S. firms that operate part-time and year-round, which may have been correlated with firm 

> 12Urbanity is defined based on a firm being located in an urban district in the Swedish and Russian census data and based on information on county urbanization levels from Haines (2010) to define an urban indicator for U.S. counties; I classify firms as urban if they are located in counties where more than a third of the population lives in a city with at least 2,500 inhabitants as urban.

<!-- PDF page 40 -->

size and the share of unskilled labor (Atack et al., 2002; Gregg and Matiashvili, 2022). 

Table C1: Selected Narrow Industries 

Outcome: % women and children 



```text
                           Russia (1894) Sweden (1864–1895) U.S. (1870–1880)

                                 (1)                  (2)                   (3)
Flour/grist milling           1.333∗∗∗             2.350∗∗∗              2.578∗∗∗

                               (0.259)             (0.790)                (0.311)
                               [2,086]              [1,265]              [20,576]
Leather tanning               3.594∗∗∗              1.155                 0.612∗∗

                               (0.734)             (0.795)                (0.297)
                                [786]              [10,122]               [4,410]
Sugar refniing                 3.816∗              3.662∗∗∗              3.250∗∗∗

                               (1.997)             (0.344)                (0.913)
                                [186]                [144]                 [856]
Glass manufacture              4.574∗               1.182                4.662∗∗∗

                               (2.598)             (3.748)                (0.942)
                                [144]                [618]                 [463]
Tobacco manufacturing         7.230∗∗∗             5.658∗∗∗              7.226∗∗∗

                               (1.955)             (1.512)                (0.526)
                                [141]               [2,196]               [7,616]

Region × year FE                 Yes                 Yes                    Yes
```

_Notes:_ Firm-level OLS regressions estimated separately for each country-by-industry sample. The dependent variable is the percentage of women and children in total employment. Each cell reports the coefficient on ln firm size; standard errors clustered at the region level are reported in parentheses, and firms are reported in square brackets. Region corresponds to city/ _uezd_ , district ( _fögderi_ ), and county for Russia, Sweden, and the United States, respectively. All specifications include region-by-year fixed effects; for Russia, which is observed in a single cross section, these fixed effects correspond to region fixed effects. The table restricts to five narrowly defined industries that can be matched across all three countries and have at least 100 firms and positive estimated coefficients in all three countries.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1.

<!-- PDF page 41 -->

Table C2: Alternative Measures of Firm Size 



```text
                                                                                          Outcome: % women and children

                                                   Russia (1894)                                 Sweden (1864–1895)                                  U.S. (1870–1880)

                                     (1)       (2)       (3)        (4)      (5)       (6)       (7)       (8)       (9)      (10)      (11)      (12)      (13)      (14)       (15)
Firm size (ln)                    3.218∗∗∗                                          2.995∗∗∗                                          3.287∗∗∗

                                   (0.397)                                           (0.837)                                          (0.181)
Firm size + 1 (ln)                           3.341∗∗∗                                         3.232∗∗∗                                          3.713∗∗∗

                                             (0.437)                                           (0.910)                                          (0.196)
Large factory (=1)                                     6.371∗∗∗                                         5.115∗∗∗                                          7.282∗∗∗

                                                       (0.553)                                           (1.528)                                           (0.406)
Manufactory (=1)                                                 8.137∗∗∗                                          3.727∗∗∗                                         6.118∗∗∗

                                                                 (0.977)                                           (0.890)                                           (0.622)
Factory (=1)                                                     5.979∗∗∗                                           1.645                                           1.780∗∗∗

                                                                 (0.934)                                           (1.358)                                           (0.304)
Output (ln)                                                                0.411∗                                             0.102                                           0.330∗∗∗

                                                                           (0.245)                                           (0.463)                                           (0.094)

Region × industry × year FE          Yes       Yes       Yes       Yes       Yes       Yes       Yes       Yes       Yes       Yes      Yes       Yes        Yes       Yes       Yes
Mean outcome                        15.73     15.73     15.73     15.73     16.37     22.90     22.90     22.90     22.90     23.28     9.99      9.99      9.99      9.99       9.99
Firms                               14079     14079     14079     14079     12227    58542      58542     58542     58542    55384    246960     246960    246960    246958    246960
```

_Notes:_ Firm-level OLS regressions using alternative measures of firm size. The dependent variable is the percentage of women and children in total employment. Large factory equals one for firms with at least 16 workers, following Sokoloff (1984). The industrial-organization type indicators follow Berger and Ostermeyer (2025a): artisan shops are establishments with no steam or water power and fewer than seven workers; manufactories are establishments with no steam or water power and at least seven workers; factories are establishments using steam or water power. Artisan shops are the omitted category in the type-indicator columns. Region corresponds to city/ _uezd_ , district ( _fögderi_ ), and county for Russia, Sweden, and the United States, respectively. Industry corresponds to the harmonized set of 20 detailed industry groups. Standard errors clustered at the region level are reported in parentheses.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1.

<!-- PDF page 42 -->

Table C3: Restricted Sample 

Outcome: % women and children 



```text
                                    Russia (1894)     Sweden (1864–1895) U.S. (1870–1880)

                                     (1)      (2)       (3)         (4)          (5)       (6)
Firm size (ln)                     1.084∗   1.543∗∗∗  2.474∗∗    2.561∗∗∗     2.812∗∗∗   3.260∗∗∗

                                  (0.570) (0.588) (0.973)         (0.784)     (0.213)    (0.255)
Steam power (=1)                             -2.073               -0.451                -3.253∗∗∗

                                            (1.734)               (1.704)                (0.724)
Water power (=1)                             -2.183               -0.414                 -1.096∗

                                            (1.631)               (2.330)                (0.639)
Region × industry × year FE         Yes       Yes       Yes         Yes         Yes        Yes
Mean outcome                       23.63     23.63     32.07       32.07       21.58      21.58
Firms                               6939      6939     16316       16316       35478      35478
```

_Notes:_ Firm-level OLS regressions. The sample is restricted to firms with at least 15 workers in all countries. The dependent variable is the percentage of women and children in total employment. Region corresponds to city/ _uezd_ , district ( _fögderi_ ), and county for Russia, Sweden, and the United States, respectively. Industry corresponds to the harmonized set of 20 detailed industry groups. Standard errors clustered at the region level are reported in parentheses.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1.

<!-- PDF page 43 -->

Table C4: Female and Child Workers 

Russia (1894) Sweden (1864–1895) U.S. (1870–1880) 

Panel A. Outcome: % adult female workers 



```text
                                     (1)        (2)       (3)         (4)         (5)        (6)
Firm size (ln)                    2.496∗∗∗   2.551∗∗∗   2.239∗∗∗    2.567∗∗∗    2.419∗∗∗  2.933∗∗∗

                                   (0.182)   (0.196)    (0.534)     (0.467)     (0.277)    (0.334)
Steam power (=1)                              -0.330                -1.827∗∗              -3.385∗∗∗

                                             (0.456)                (0.760)                (0.456)
Water power (=1)                              -1.177∗                -0.137               -2.289∗∗∗

                                             (0.702)                (1.006)                (0.233)
Region × industry × year FE          Yes       Yes        Yes         Yes         Yes        Yes
Mean outcome                         8.10      8.10      11.83       11.83        6.46      6.46
Firms                               14079     14079      58542       58542      246960     246960
```

Panel B. Outcome: % children 



```text
                                     (1)        (2)       (3)         (4)         (5)        (6)
Firm size (ln)                     0.722∗∗   1.121∗∗∗    0.756      1.217∗∗     0.867∗∗∗  0.879∗∗∗

                                   (0.352)   (0.324)    (0.534)     (0.580)     (0.133)    (0.153)
Steam power (=1)                             -2.530∗∗∗             -2.398∗∗∗               -0.080

                                             (0.855)                (0.585)                (0.171)
Water power (=1)                             -2.351∗∗∗              -1.267∗∗               -0.177

                                             (0.684)                (0.635)                (0.137)
Region × industry × year FE          Yes       Yes        Yes         Yes         Yes        Yes

Mean outcome                         7.63      7.63      11.06       11.06        3.52      3.52
Firms                               14079     14079      58542       58542      246960     246960
```

_Notes:_ Firm-level OLS regressions. The dependent variable is the percentage of adult female workers in total employment in Panel A and the percentage of children in total employment in Panel B. Region corresponds to city/ _uezd_ , district ( _fögderi_ ), and county for Russia, Sweden, and the United States, respectively. Industry corresponds to the harmonized set of 20 detailed industry groups. Standard errors clustered at the region level are reported in parentheses.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1.

<!-- PDF page 44 -->

Table C5: Horsepower per Worker 

Outcome: % women and children 



```text
                                     Russia (1894)      Sweden (1864–1895) U.S. (1870–1880)

                                     (1)        (2)       (3)         (4)         (5)        (6)
Firm size (ln)                    3.293∗∗∗   3.288∗∗∗   2.810∗∗∗    2.935∗∗∗    3.284∗∗∗  3.278∗∗∗

                                   (0.419)   (0.414)    (0.839)     (0.818)     (0.183)    (0.191)
ln HP per worker (std)                       -1.558∗∗∗             -3.518∗∗∗              -2.109∗∗∗

                                             (0.421)                (0.615)                (0.135)
Region × industry × year FE          Yes       Yes        Yes         Yes         Yes        Yes

Mean outcome                        16.16     16.16      23.61       23.61       10.00      10.00
Firms                               13217     13217      53442       53442      243650     243650
```

_Notes:_ Firm-level OLS regressions. The dependent variable is the percentage of women and children in total employment. All specifications restrict the sample to firms with observed horsepower per worker. The ln(horsepower per worker + 1) measure is standardized within country. Region corresponds to city/ _uezd_ , district ( _fögderi_ ), and county for Russia, Sweden, and the United States, respectively. Industry corresponds to the harmonized set of 20 detailed industry groups. Standard errors clustered at the region level are reported in parentheses.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1.

<!-- PDF page 45 -->

Table C6: Urban-Rural Split 

Outcome: % women and children 



```text
                         Russia (1894)      Sweden (1864–1895) U.S. (1870–1880)

                        Urban      Rural      Urban       Rural       Urban      Rural
                          (1)       (2)        (3)          (4)         (5)        (6)
Firm size (ln)         4.401∗∗∗   3.719∗∗∗   4.504∗∗∗    4.363∗∗∗    4.483∗∗∗   4.224∗∗∗

                        (0.652)   (0.292)    (0.844)     (0.414)     (0.276)    (0.169)
Steam power (=1)       -3.835∗∗  -2.040∗∗∗  -5.201∗∗∗     -1.116    -4.419∗∗∗  -2.427∗∗∗

                        (1.692)   (0.733)    (1.258)     (1.036)     (0.598)    (0.267)
Water power (=1)       -6.865∗∗  -4.270∗∗∗  -4.967∗∗∗     -0.456    -3.514∗∗∗  -1.043∗∗∗

                        (2.687)   (0.844)    (1.836)     (0.923)     (0.678)    (0.289)
Industry × year FE        Yes       Yes        Yes         Yes         Yes        Yes
Mean outcome             20.56     11.94      26.80       19.50       14.06       6.19
Firms                    7192       8827      34835       38123      120731     134313
```

_Notes:_ Firm-level OLS regressions estimated separately for urban and rural firms. The dependent variable is the percentage of women and children in total employment. The Sweden and U.S. specifications include year _×_ detailed industry fixed effects; the Russia specification includes detailed industry fixed effects because the data are observed in a single cross section. Industry corresponds to the harmonized set of 20 detailed industry groups. Standard errors clustered at the region level are reported in parentheses.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1.

<!-- PDF page 46 -->

Table C7: Controlling for Firm Ownership 

Outcome: % women and children 



```text
                                        (1)        (2)        (3)        (4)

                                           ∗∗∗        ∗∗∗        ∗∗∗         ∗∗∗
Firm size (ln)                       2.995      3.001      3.309      4.038
                                      (0.837)    (0.826)   (0.740)     (0.658)
Female owner (=1)                                 0.468      0.235      0.128
                                                 (1.626)   (1.592)     (1.611)
Partnership (=1)                                            -0.775     -0.765
                                                           (0.617)     (0.623)
Corporation (=1)                                           -2.441∗     -2.112

                                                           (1.476)     (1.384)
Steam power (=1)                                                      -4.135∗∗∗

                                                                       (0.887)
Water power (=1)                                                       -1.274
                                                                       (1.260)
Region × industry × year FE             Yes        Yes        Yes        Yes

Mean outcome                           22.90      22.90      22.90      22.90
Firms                                 58542      58542      58542       58542
```

_Notes:_ Firm-level OLS regressions for Swedish manufacturing firms observed in 1864–1895. The dependent variable is the percentage of women and children in total employment. Region corresponds to district ( _fögderi_ ). Industry corresponds to the harmonized set of 20 detailed industry groups. Standard errors are clustered at the district level.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1.

<!-- PDF page 47 -->

Table C8: Year-Round and Part-Time Firms 

Outcome: % women and children 



```text
                            Russia (1894)                 U.S. (1880)

                       Year-round Part-time Year-round Part-time
                            (1)           (2)           (3)           (4)

Firm size (ln)           3.251∗∗∗      3.863∗∗∗      4.185∗∗∗      4.615∗∗∗

                          (0.447)       (0.371)       (0.224)       (0.359)
                                 ∗            ∗∗∗           ∗∗∗           ∗∗∗
Steam power (=1)          -2.406       -3.130        -4.349        -3.439
                          (1.373)       (0.712)       (0.572)       (0.595)
                                ∗∗∗           ∗∗∗           ∗∗∗           ∗∗∗
Water power (=1)         -4.890        -3.794        -4.308        -2.512
                          (0.990)       (0.978)       (0.533)       (0.486)

Industry FE                 Yes           Yes           Yes           Yes
Mean outcome               20.89         13.58         10.08          9.51
Firms                      4957          10838         76843         56644
```

_Notes:_ Firm-level OLS regressions estimated separately for year-round and part-time firms. In Russia, yearround firms are defined as those operating at least 280 working days; part-time firms operate fewer than 280 days. In the 1880 U.S. census, year-round firms are defined as those reporting 12 full-time months of operation; part-time firms report fewer than 12 full-time months. The dependent variable is the percentage of women and children in total employment. Industry corresponds to the harmonized set of 20 detailed industry groups. Standard errors are clustered at the city/ _uezd_ level in Russia and at the county level in the United States.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1.

<!-- PDF page 48 -->



<!-- PDF page 49 -->

## **D Additional IV Results** 



<!-- Start of picture text -->
1<br>6<br>.5 4<br>2<br>0<br>0<br>-.5 -2<br>-4<br>-1<br>-1 -.5 0 .5 1 -1 -.5 0 .5 1<br>Firm size elsewhere (residualized) Firm size elsewhere (residualized)<br>Coefficient = 0.665; SE = 0.040; KP F = 271.19 Coefficient = 5.068; SE = 0.778<br>(a) Russia: First stage (b) Russia: Reduced form<br>1<br>6<br>.5 4<br>2<br>0<br>0<br>-.5 -2<br>-4<br>-1<br>-1 -.5 0 .5 1 -1 -.5 0 .5 1<br>Firm size elsewhere (residualized) Firm size elsewhere (residualized)<br>Coefficient = 0.725; SE = 0.071; KP F = 103.54 Coefficient = 3.715; SE = 1.833<br>(c) Sweden: First stage (d) Sweden: Reduced form<br>1<br>6<br>.5 4<br>2<br>0<br>0<br>-.5 -2<br>-4<br>-1<br>-1 -.5 0 .5 1 -1 -.5 0 .5 1<br>Firm size elsewhere (residualized) Firm size elsewhere (residualized)<br>Coefficient = 0.849; SE = 0.017; KP F = 2377.04 Coefficient = 5.983; SE = 0.440<br>(e) United States: First stage (f) United States: Reduced form<br>Firm size (residualized)<br>Women and children (p.p., residualized)<br>Firm size (residualized)<br>Women and children (p.p., residualized)<br>Firm size (residualized)<br>Women and children (p.p., residualized)<br><!-- End of picture text -->

Figure D1: Instrument Relevance and Reduced Form 

_Notes:_ This figure displays binned scatterplots for the baseline IV specifications without IV controls. Panels A, C, and E show the first-stage relationship between own-firm ln size and average ln firm size in the same granular industry in other regions. Panels B, D, and F show the corresponding reduced-form relationship for the percentage of workers who are women or children. All variables are residualized using the fixed effects in the baseline specification. Each panel contains 20 equal-frequency bins; the solid line is estimated using the underlying firm-level observations. Panel annotations report the regression coefficient and standard error clustered at the baseline region level; the first-stage panels also report the Kleibergen–Paap F-statistic. Region corresponds to city/ _uezd_ , district ( _fögderi_ ), and county for Russia, Sweden, and the United States, respectively.

<!-- PDF page 50 -->

<!-- Start of picture text -->
Food & Beverages Food & Beverages<br>Textiles Textiles<br>Leather Leather<br>Wood Wood<br>Paper & Printing Paper & Printing<br>Glass, Stone, etc. Glass, Stone, etc.<br>Chemicals Chemicals<br>Metals & Machinery Metals & Machinery<br>-20 0 20 40 60 -40 -20 0 20 40<br>Coefficient on ln firm size Coefficient on ln firm size<br>(a) Russia (b) Sweden<br>Food & Beverages<br>Textiles<br>Leather<br>Wood<br>Paper & Printing<br>Glass, Stone, etc.<br>Chemicals<br>Metals & Machinery<br>0 10 20 30<br>Coefficient on ln firm size<br>(c) United States<br><!-- End of picture text -->

Figure D2: Firm Size and Unskilled Labor: IV Estimates 

_Notes:_ This figure displays broad-industry-specific IV estimates of the coefficient on ln firm size from regressions of the percentage of women and children in total employment on ln firm size. Panels A–C report estimates for Russia, Sweden, and the United States, respectively. The instrument is average ln firm size in the same granular industry in other regions. The Russia regressions include region fixed effects, while the Sweden and U.S. regressions include year _×_ region fixed effects. Region corresponds to city/ _uezd_ , district ( _fögderi_ ), and county for Russia, Sweden, and the United States, respectively. Industry corresponds to the harmonized set of eight broad industry groups. Horizontal bars denote 95% confidence intervals based on standard errors clustered at the region level.

<!-- PDF page 51 -->

Table D1: Firm Size, Technology Use, and Unskilled Labor: 2SLS Estimates 



```text
                                                                                            Outcome: % women and children

                                                    Russia (1894)                                  Sweden (1864–1895)                                   U.S. (1870–1880)

                                     RF       2SLS      2SLS       2SLS       2SLS       RF      2SLS     2SLS      2SLS       2SLS       RF       2SLS      2SLS       2SLS        2SLS
                                     (1)       (2)       (3)        (4)        (5)       (6)      (7)       (8)       (9)       (10)      (11)      (12)      (13)       (14)        (15)
Firm size elsewhere (ln)          5.068∗∗∗                                             3.715∗∗                                         5.983∗∗∗

                                   (0.778)                                             (1.833)                                          (0.440)
Firm size (ln)                               7.619∗∗∗  7.493∗∗∗  9.222∗∗∗   7.426∗∗∗            5.124∗∗  3.608∗∗   5.880∗∗    3.631∗∗             7.046∗∗∗  5.266∗∗∗  10.906∗∗∗    5.257∗∗∗

                                             (1.262)   (1.122)    (1.380)    (1.102)            (2.350) (1.601)    (2.617)    (1.556)             (0.600)   (0.450)    (0.959)     (0.442)
Steam power (=1)                                                 -7.814∗∗∗  -5.517∗∗∗                             -6.973∗∗∗  -3.837∗∗                                 -11.063∗∗∗  -4.318∗∗∗

                                                                  (0.912)    (0.788)                               (2.548)    (1.784)                                  (1.229)     (0.505)
Water power (=1)                                                 -5.691∗∗∗  -5.262∗∗∗                               -3.143    -0.855                                   -5.589∗∗∗  -2.310∗∗∗

                                                                  (0.913)    (0.942)                               (2.071)    (1.929)                                  (0.510)     (0.289)
Region × industry × year FE          Yes       Yes       Yes        Yes        Yes       Yes      Yes      Yes       Yes        Yes       Yes       Yes       Yes        Yes         Yes
IV controls                          No        No        Yes        No         Yes       No       No       Yes        No        Yes       No        No        Yes         No         Yes
Kleibergen-Paap F-statistic           —      271.19     246.78    170.31     272.04      —      103.54    66.56     113.40    112.12       —      2377.04 1099.57      1655.16     1966.36
Firms                               13883     13883     13845     13883      13845      55746    55746    55670     55746      55670    246925    246925    246923      246925     246923
```

_Notes:_ Firm-level reduced-form OLS and 2SLS regressions. The dependent variable is the percentage of workers who are women or children. Firm size is instrumented using average ln firm size among firms in the same granular industry located in other regions. IV controls are residual child/female employment among firms in the same granular industry in other regions, where residuals net out ln firm size, own-firm steam and water power, and region-by-detailed-industry fixed effects, as well as steam and water shares among firms in the same granular industry in other regions. For Sweden and the United States, leave-region-out measures pool all sample years within country. Region corresponds to city/ _uezd_ , district ( _fögderi_ ), and county for Russia, Sweden, and the United States, respectively. Country-specific granular industries are used to construct the leave-region-out measures; absorbed industry fixed effects use the harmonized set of 20 detailed industry groups. Standard errors clustered at the region level are reported in parentheses.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1.

<!-- PDF page 52 -->

Table D2: Weak-Instrument-Robust Inference 

Russia (1894) Sweden (1864–1895) U.S. (1870–1880) 



```text
                                                (1)                  (2)                   (3)
Firm size (ln)                               7.619∗∗∗             5.124∗∗               7.046∗∗∗

                                              (1.262)             (2.350)                (0.600)
Region × industry × year FE                     Yes                 Yes                    Yes

Anderson–Rubin 95% confidence set [5.234, 10.229]              [0.155, 9.610]        [5.899, 8.254]
Anderson–Rubin p-value (β = 0)                <0.001               0.044                 <0.001
Kleibergen–Paap F-statistic                   271.19               103.54                2377.04
Mean outcome                                   15.52               21.20                  9.98
Firms                                         13883                55746                 246925
```

_Notes:_ Firm-level baseline 2SLS regressions of the percentage of workers who are women or children. Firm size is instrumented using average ln firm size among firms in the same granular industry located in other regions. For Sweden and the United States, the instrument pools all sample years within country. The table reports conventional 2SLS point estimates and standard errors in parentheses, 95% Anderson–Rubin confidence sets, and Anderson–Rubin p-values for the null hypothesis that the coefficient on firm size is zero. Both conventional and Anderson–Rubin inference cluster standard errors at the baseline region level. Region corresponds to city/ _uezd_ , district ( _fögderi_ ), and county for Russia, Sweden, and the United States, respectively. Country-specific granular industries are used to construct the instrument; absorbed industry fixed effects use the harmonized set of 20 detailed industry groups.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1.

<!-- PDF page 53 -->

<!-- Start of picture text -->
15 95% union confidence set 15 95% union confidence set<br>10 10<br>5 5<br>0 0<br>-5 -5<br>-10 -10<br>0 25 50 75 100 125 0 25 50 75 100 125<br>Direct-effect bound (% of reduced form) Direct-effect bound (% of reduced form)<br>(a) Russia: Baseline (b) Russia: Full IV controls<br>15 95% union confidence set 15 95% union confidence set<br>10 10<br>5 5<br>0 0<br>-5 -5<br>-10 -10<br>0 25 50 75 100 125 0 25 50 75 100 125<br>Direct-effect bound (% of reduced form) Direct-effect bound (% of reduced form)<br>(c) Sweden: Baseline (d) Sweden: Full IV controls<br>15 95% union confidence set 15 95% union confidence set<br>10 10<br>5 5<br>0 0<br>-5 -5<br>-10 -10<br>0 25 50 75 100 125 0 25 50 75 100 125<br>Direct-effect bound (% of reduced form) Direct-effect bound (% of reduced form)<br>(e) United States: Baseline (f) United States: Full IV controls<br>Figure D3: Sensitivity to Violations of Exclusion Restriction<br>Effect of ln firm size Effect of ln firm size<br>Effect of ln firm size Effect of ln firm size<br>Effect of ln firm size Effect of ln firm size<br><!-- End of picture text -->

_Notes:_ This figure displays 95% union confidence sets from plausibly-exogenous 2SLS specifications. Panels A, C, and E report the baseline specifications for Russia, Sweden, and the United States, respectively; Panels B, D, and F report the corresponding specifications with full IV controls. The horizontal axis gives the upper bound on a same-sign direct effect of the instrument as a percentage of its reduced-form coefficient. For each value of the bound, the shaded region is the union of conventional 95% confidence intervals obtained by allowing the direct effect to range between zero and that bound. The vertical dashed line denotes a bound equal to the full reduced-form coefficient. Full IV controls include own-firm steam- and water-power indicators and residual child/female employment and steam- and water-power shares among firms in the same granular industry in other regions. All specifications absorb the fixed effects used in the baseline IV regressions. Standard errors are clustered at the baseline region level: city/ _uezd_ in Russia, district ( _fögderi_ ) in Sweden, and county in the United States.

<!-- PDF page 54 -->

## **E Additional Panel Results** 



<!-- Start of picture text -->
Food & Beverages<br>Textiles<br>Leather<br>Wood<br>Paper & Printing<br>Glass, Stone, etc.<br>Chemicals<br>Metals & Machinery<br>-5 0 5 10 15 20<br>Coefficient on ln firm size<br>(a) Sweden<br><!-- End of picture text -->



<!-- Start of picture text -->
Food & Beverages<br>Textiles<br>Leather<br>Wood<br>Paper & Printing<br>Glass, Stone, etc.<br>Chemicals<br>Metals & Machinery<br>-5 0 5 10 15 20<br>Coefficient on ln firm size<br>(b) United States<br>Figure E1: Panel OLS Regressions: By Industry<br><!-- End of picture text -->

_Notes:_ This figure displays broad-industry-specific panel estimates of the coefficient on ln firm size from regressions of the percentage of women and children in total employment on ln firm size. Panel A reports estimates for linked Swedish firms observed in 1864–1895; Panel B reports estimates for linked U.S. firms observed in 1870 and 1880. Both samples exclude firms using steam or water power when first observed, and both specifications include firm and year fixed effects. Industry corresponds to the harmonized set of eight broad industry groups. Horizontal bars denote 95% confidence intervals based on standard errors clustered at the firm level.

<!-- PDF page 55 -->

<!-- Start of picture text -->
8<br>6<br>4<br>2<br>0<br>0.95 0.96 0.97 0.98 0.99<br>Minimum match score<br>Coefficient on firm size (ln)<br><!-- End of picture text -->

Figure E2: Panel OLS Regressions: Match Quality in the U.S. Linked Sample 

_Notes:_ This figure plots the coefficient on ln firm size and 95% confidence intervals from panel regressions of the percentage of women and children on ln firm size, restricting the U.S. linked sample to matches with a bigram similarity score above the indicated threshold. The baseline threshold used in the main analysis is 0.95. All specifications include firm fixed effects and year _×_ county _×_ broad industry fixed effects. Industry corresponds to the harmonized set of eight broad industry groups. Standard errors are clustered at the firm level.

<!-- PDF page 56 -->

### **E.1 Event-Study Estimates** 

Figure E3 presents conventional two-way fixed-effects event-study estimates alongside estimates obtained using the dynamic difference-in-differences estimator of de Chaisemartin and D’Haultfœuille (2026). Online Appendix Figure E3 includes a barebones specification and a specification that flexibly controls for factors that may have shaped the adoption of steam. Across all four specifications, joint tests fail to reject the null that the pre-treatment coefficients are zero. After the transition to steam, the share of women and children declines sharply: the average post-treatment estimates range from _−_ 2 _._ 73 to _−_ 2 _._ 18 percentage points. These estimates are broadly similar in magnitude to the standard difference-in-differences estimate reported in Table 1, Panel C, Column 5.

<!-- PDF page 57 -->

<!-- Start of picture text -->
5 5<br>4 Pre-trends: p = 0.600Avg. post effect: -2.73 (0.99) 4 Pre-trends: p = 0.257Avg. total effect: -2.61 (1.04)<br>3 3<br>2 2<br>1 1<br>0 0<br>-1 -1<br>-2 -2<br>-3 -3<br>-4 -4<br>-5 -5<br>-6 -6<br>-4 -3 -2 -1 0 1 2 -4 -3 -2 -1 0 1 2<br>Years since steam transition began Years since steam transition began<br>(a) TWFE: Baseline (b) DCDH: Baseline<br>5 5<br>4 Pre-trends: p = 0.578Avg. post effect: -2.61 (0.99) 4 Pre-trends: p = 0.340Avg. total effect: -2.18 (0.96)<br>3 3<br>2 2<br>1 1<br>0 0<br>-1 -1<br>-2 -2<br>-3 -3<br>-4 -4<br>-5 -5<br>-6 -6<br>-4 -3 -2 -1 0 1 2 -4 -3 -2 -1 0 1 2<br>Years since steam transition began Years since steam transition began<br>(c) TWFE: Controls (d) DCDH: Controls<br>Figure E3: The Transition to Steam Power and Unskilled Labor<br>Point estimate and 95% CI Point estimate and 95% CI<br>Point estimate and 95% CI Point estimate and 95% CI<br><!-- End of picture text -->

_Notes:_ This figure displays establishment-level event-study estimates of changes in the percentage of women and children around the transition to steam power in Swedish manufacturing firms. Event time zero is defined as one year before steam use is first reported, allowing for installation and reporting lags; the treatment indicator remains equal to one thereafter. Panels A and C use the eventdd estimator, with endpoints binned at _t ≤−_ 4 and _t ≥_ 2 and _t_ = _−_ 1 omitted. Panels B and D use the dynamic difference-in-differences estimator of de Chaisemartin and D’Haultfœuille (2026), implemented by did_multiplegt_dyn, and plot the corresponding placebo and effect horizons after residualizing the outcome on the controls and fixed effects used in the respective panel. All panels control for contemporaneous ln total employment and include firm and year fixed effects. Panels C and D additionally include county-by-year and detailed-industry-by-year fixed effects. Firm controls in Panels C and D are entry ln total employment, an indicator for corporate ownership at entry, and the entry percentage of women and children, each interacted with year fixed effects. Industry corresponds to the harmonized set of 20 detailed industry groups. Each panel reports the _p_ -value from a joint test of the displayed pre-treatment coefficients and a post-treatment summary, with its standard error in parentheses. Panels A and C report the equal-weight average of the eventdd estimates for _t_ = 0 _,_ 1 _,_ 2. Panels B and D report the average total effect produced by did_multiplegt_dyn across the displayed effect horizons. Connected markers denote point estimates. Vertical bars denote 95% confidence intervals based on standard errors clustered at the firm level.

<!-- PDF page 58 -->

Table E1: Panel Estimates: Additional Controls 

Outcome: % women and children 



```text
                                      Sweden, 1864–1895                    United States, 1870–1880

                                (1)       (2)       (3)       (4)       (5)       (6)        (7)        (8)
Firm size (ln)               5.972∗∗∗  5.819∗∗∗  6.180∗∗∗  6.043∗∗∗  4.288∗∗∗  4.570∗∗∗   4.549∗∗∗   4.973∗∗∗

                             (0.694)   (0.652)    (0.691)   (0.647)   (0.366)   (0.282)   (0.384)    (0.306)
Steam power (=1)                                 -2.392∗∗  -2.329∗∗                       -2.794∗∗∗  -3.013∗∗∗

                                                 (0.960)    (0.944)                       (0.968)    (0.827)
Water power (=1)                                  0.962      0.426                        -4.417∗∗   -6.851∗∗∗

                                                 (1.190)    (1.213)                       (2.105)    (1.269)

Firm FE                        Yes       Yes       Yes        Yes       Yes       Yes       Yes        Yes
Year FE                        Yes       Yes       Yes        Yes       Yes       Yes       Yes        Yes
County × year FE                No       Yes        No        Yes       No        Yes        No        Yes
Industry × year FE              No       Yes        No        Yes       No        Yes        No        Yes
Firm controls × year FE         No       Yes        No        Yes       No        Yes        No        Yes
Mean outcome                  25.34     25.35      25.34     25.35     14.16     14.46     14.16      14.46
Observations                  17787     17744     17787     17744     15676      15234     15676      15234
Firms                          2518      2517      2518      2517      7838      7617       7838       7617
```

_Notes:_ Firm-level OLS regressions for linked Swedish manufacturing firms observed in 1864–1895 and linked U.S. firms observed in 1870 and 1880. The dependent variable is the percentage of workers who are women or children. In the Swedish power-controlled specifications (columns 3–4), the steam and water indicators switch to one when adoption of the respective technology likely began, whereas in the U.S. specifications (columns 7–8) they capture reported power use in each census year. Firm controls are baseline ln firm size, an indicator for corporate ownership, and the baseline percentage of women and children, each interacted with year fixed effects where indicated. County corresponds to _län_ in Sweden and county in the United States. Industry corresponds to the harmonized set of 20 detailed industry groups. Standard errors clustered at the firm level are reported in parentheses.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1.

<!-- PDF page 59 -->

Table E2: Panel OLS Regressions: Full-Time Firms Outcome: ln FTE annual wages per worker 



```text
Sample:                           All      No power        All            All
                                  (1)          (2)         (3)            (4)

                                      ∗∗∗          ∗∗∗         ∗∗∗            ∗∗∗
Firm size (ln)                 -0.121      -0.132       -0.136        -0.143
                                (0.012)     (0.017)      (0.013)       (0.013)
                                                               ∗∗∗           ∗∗∗
Steam power (=1)                                        0.115          0.121
                                                         (0.024)       (0.025)
Water power (=1)                                          0.068         0.039
                                                         (0.048)       (0.049)

Firm FE                           Yes          Yes         Yes            Yes
Year FE                           Yes          Yes         Yes            Yes
County × year FE                  No           No          No             Yes
Industry × year FE                No           No          No             Yes
Firm controls × year FE           No           No          No             Yes
Mean outcome                      6.00        5.95         6.00          6.01
Observations                     12156        7054        12156         11750
Firms                            6078         3527        6078           5875
```

_Notes:_ Firm-level OLS regressions for linked U.S. firms observed in 1870 and 1880 that operated full-time in both years and had valid wage inputs. The dependent variable is the log of full-time-equivalent annual wages per worker. The no-power sample consists of firms that used neither steam nor water power in either year. Baseline firm controls, interacted with year fixed effects, include ln firm size, an indicator for corporate ownership, and the baseline percentage of women and children. Industry corresponds to the harmonized set of 20 detailed industry groups. Standard errors clustered at the firm level are reported in parentheses.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1.

<!-- PDF page 60 -->

Table E3: Panel OLS Regressions: Higher-Quality Links Outcome: ln FTE annual wages per worker 



```text
Sample:                           All      No power        All            All
                                  (1)          (2)         (3)            (4)

                                      ∗∗∗          ∗∗∗         ∗∗∗            ∗∗∗
Firm size (ln)                 -0.211      -0.192       -0.219        -0.240
                                (0.018)     (0.023)      (0.019)       (0.019)
                                                               ∗∗∗           ∗∗∗
Steam power (=1)                                        0.110          0.134
                                                         (0.031)       (0.030)
Water power (=1)                                          0.071         0.065
                                                         (0.054)       (0.054)

Firm FE                           Yes          Yes         Yes            Yes
Year FE                           Yes          Yes         Yes            Yes
County × year FE                  No           No          No             Yes
Industry × year FE                No           No          No             Yes
Firm controls × year FE           No           No          No             Yes
Mean outcome                      5.96        5.93         5.96          5.98
Observations                     9678         4434        9678           9186
Firms                            4839         2217        4839           4593
```

_Notes:_ Firm-level OLS regressions for linked U.S. firms observed in 1870 and 1880 with valid wage inputs and a composite match score of at least 0.99. The dependent variable is the log of full-time-equivalent annual wages per worker. The no-power sample consists of firms that used neither steam nor water power in either year. Baseline firm controls, interacted with year fixed effects, include ln firm size, an indicator for corporate ownership, and the baseline percentage of women and children. Industry corresponds to the harmonized set of 20 detailed industry groups. Standard errors clustered at the firm level are reported in parentheses.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1.

<!-- PDF page 61 -->

## **F Alternative Measures: Skills and Organizational Change** 

This section uses linked employer-employee data for the Swedish printing industry assembled by Burnette and Stanfors (2020) to (i) document that other measures of worker skills beyond the share of women and children correlate with firm size; and (ii) provide more direct evidence on organizational differences across the firm size distribution. The printing industry is a useful setting because women and children made up a high share of employment in this industry in both Europe and the United States (Online Appendix Figure B2). The data are based on the 1902/03 Swedish Board of Commerce survey of printing and allied trades, which collected separate employer and employee data. I aggregate workers to firms and restrict the sample to firms with 2–1,000 workers as in the main analysis, leaving about 7,900 workers in 321 firms. 

### **F.1 Firm Size and Skills in the Printing Industry** 

Although women and children constitute a historically relevant measure of unskilled workers, it is an open question whether their employment correlates with other measures of worker skills that are not available in the historical censuses used in the main analysis. 

Using the linked employer-employee data for the Swedish printing industry, I perform two tests. First, I examine whether the share of women and children is correlated with alternative measures of worker skills. Second, I ask whether larger firms employ relatively more unskilled workers using these alternative measures. 

I construct the percent of workers who are women or children, where children are workers younger than 18 to align with the Swedish census data used in the main analysis. I then use two alternative skill measures: the percent of workers in unskilled occupations and the percent with any secondary education. For each measure, I calculate the firm-level percentage among all workers and separately among men aged 18 or older. I code unskilled occupations from the original occupation strings and detailed occupation groups and separate recognized craft, supervisory, and skilled printing jobs from unskilled work.<sup>13</sup> 

Figure F1 provides both tests and reports outcomes for all workers and, separately, for men aged 18 or older. Panels A and B show that firms with more women and children have substantially higher shares of workers in unskilled occupations and a lower share of workers with a secondary education. The association is similar, though naturally somewhat weaker, 

> 13In the context of the printing industry it is relatively straightforward to delineate skilled and unskilled jobs: Compositors, typesetters, skilled pressmen, stereotypers, lithographers, photoengravers, full bookbinders, and supervisors are classified as skilled; by contrast, press feeders and takers-off, composition-room support workers, simple press workers, typefounding and lithographic support workers, simple bookbinding and finishing workers, packing, cleaning, carrying, and general labor are classified as unskilled.

<!-- PDF page 62 -->

when considering the share of adult men in unskilled occupations and the share of adult men who lack a secondary education, which suggests that the association is not purely driven by compositional differences due to the employment of women and children. Panels C and D compare the same skill measures across the firm size distribution showing that larger firms have higher unskilled shares both overall and among adult men, while the association with the share of workers with a secondary education is negative. 

Taken together, these results suggest that the share of women and children is a useful proxy for unskilled labor in the nineteenth and early twentieth centuries. Not only did women and children constitute an important source of unskilled labor and feature prominently in contemporary debates over deskilling, but the measure is also strongly correlated with alternative measures of worker skill based on education and occupational tasks.

<!-- PDF page 63 -->

<!-- Start of picture text -->
All workers: Beta (s.e.) = 0.811 (0.056) All workers: Beta (s.e.) = -0.093 (0.017)<br>Adult men: Beta (s.e.) = 0.497 (0.056) Adult men: Beta (s.e.) = -0.040 (0.026)<br>80 15<br>60<br>10<br>40<br>5<br>20<br>0 0<br>0 20 40 60 80 100 0 20 40 60 80 100<br>% women and children % women and children<br>All workers Adult men All workers Adult men<br>(a) Women and Children vs. Unskilled Jobs (b) Women and Children vs. Education<br>All workers: Beta (s.e.) = 8.347 (1.186) All workers: Beta (s.e.) = -0.630 (0.543)<br>Adult men: Beta (s.e.) = 5.050 (1.069) Adult men: Beta (s.e.) = -0.380 (0.795)<br>60 20<br>15<br>40<br>10<br>20<br>5<br>0 0<br>2 5 10 20 50 100 200 500 1000 2 5 10 20 50 100 200 500 1000<br>ln Firm size (workers) ln Firm size (workers)<br>All workers Adult men All workers Adult men<br>(c) Firm Size and Unskilled Jobs (d) Firm Size and Worker Education<br>% in unskilled occupations<br>% with any secondary education<br>% in unskilled occupations<br>% with any secondary education<br><!-- End of picture text -->

Figure F1: Women and Children vs. Alternative Skill Measures 

_Notes:_ This figure displays binned scatterplots for Swedish printing firms using linked employer-employee data from Burnette and Stanfors (2020). The unit of observation is the firm, and the sample is restricted to firms with 2–1,000 workers. Panels A and B plot the percent of workers in unskilled occupations and the percent with any secondary education against the percent of workers who are women or children. Panels C and D plot the same outcomes against ln firm size, with x-axis labels reported in worker counts. In each panel, filled circles and solid fitted lines report outcomes for all workers; hollow green diamonds and dashed fitted lines report outcomes for men aged 18 or older. All panels use 30 bins. Points in Panels A and B are firm-size-weighted binned means; best-fit lines and reported regression statistics use the same firm-size weights. Panels C and D use unweighted firm-level binned means and unweighted firm-level regressions.

<!-- PDF page 64 -->

### **F.2 Firm Size and Organizational Change in the Printing Industry** 

As discussed in Section 4.3, firm size may be associated with three broad organizational changes: a greater division of labor and finer decomposition of tasks, a more tightly regulated work regime and factory discipline, and more developed managerial hierarchies. The survey of the printing industry describes how the organizational structure of large and small firms differed in such dimensions: 

“Before the printing trade became a factory industry during the second half of the nineteenth century, there were only two groups of workers, compositors and pressmen. Although this specialization had existed since the seventeenth century [. . . ], these groups do not appear to have always been clearly separated; instead, the same worker was commonly skilled in both composition and printing. There was no division of labor: each job was assigned to a worker, who completed it alone or with a companion ( _spåne_ ). The same is still true today in small printing offices, where the proprietor usually works at the cases and is confined to small, occasional orders. In large printing offices, by contrast, the division of labor has brought about a far-reaching specialization [...]”<sup>14</sup> 

I here use different measures of occupational structure, working time and scheduling, and supervision to examine how they vary across large and small printing firms. Table F1 presents estimates from firm-level OLS regressions where I regress measures of organizational structure on log employment for establishments with 2–1,000 workers, controlling for whether a firm is located in one of the three largest cities. 

Panel A measures occupational diversity and task decomposition by using normalized entropy across detailed occupational groups, the number of occupations represented in a firm, the ratio of support to craft workers, and the shares of apprentices and compositors, which captures differences in the division of labor (Kommerskollegium, 1909, pp. 59–60). Larger establishments had substantially more differentiated occupational structures and a greater number of occupations represented, as well as a greater number of support relative to craft workers. At the same time, the share of apprentices is lower, while the estimated 

> 14The Swedish original reads: “Innan boktryckerinäringen under senare hälften af 1800-talet öfvergick till fabriksindustri, funnos endast två grupper af arbetare inom densamma, sättare och tryckare, men ehuru denna specialisering förekom redan på 1600-talet [. . . ] synas dock ej heller dessa grupper alltid hafva varit klart åtskilda, utan var det snarare vanligt, att samma arbetare var kunnig i både sättning och tryckning. Ingen fördelning af arbetet ägde rum, utan hvarje särskildt arbete utlämnades till en arbetare, som ensam eller tillsammans med någon kamrat (»spåne«) fullbordade detsamma. Samma är förhållandet ännu i dag vid de små tryckerierna, där arbetsgifvaren vanligen själf står vid kasten och är hänvisad till smärre, tillfälliga beställningar. I de stora tryckerierna däremot har arbetsfördelningen medfört en långt drifven specialisering [...]” (Kommerskollegium, 1909, p. 189).

<!-- PDF page 65 -->

association with the share of compositors is negative but statistically insignificant. Broadly, the pattern is consistent with larger establishments combining a wider set of specialized occupations with relatively more unskilled auxiliary work. 

Panel B measures the work regime using establishment means of worker-reported usual weekly hours and annual overtime hours, together with indicators for whether any worker reports shift, night, or Sunday work. These measures capture both the duration and scheduling of work, including arrangements outside ordinary daytime hours. A doubling of employment is associated with additional overtime hours per year and increases in the probability that some worker reports shift, night, or Sunday work. The association with usual weekly hours is small and statistically insignificant.<sup>15</sup> 

Panel C measures supervision and employment stability using indicators for any recorded foreman and any _Faktor_ , the number of recorded foremen, workers per recorded foreman among establishments with one, and average worker tenure.<sup>16</sup> The supervision measures proxy for supervisory roles and their span of control, while tenure captures the stability of the employment relationship. The estimates show that larger establishments more often employed foremen and a _Faktor_ . Among establishments with a foreman, firm size is also associated with a greater number of workers per foreman, while average tenure also increases with firm size. 

Taken together, the results show that larger printing establishments differed systematically in terms of occupational diversity, work regime, and supervision. 

> 15The report indicates that shift, night, and Sunday work was concentrated in newspaper printing, so the associations may partly reflect differences in establishments’ production mix (Kommerskollegium, 1909, pp. 106–108). 

> 16The _Faktor_ indicator follows the report’s occupational classification, which lists _Faktorer_ together with foremen who also participated in production (Kommerskollegium, 1909, p. 59). Foremen who did not directly participate in the work were classified as management staff and did not complete the worker questionnaire (Kommerskollegium, 1909, p. 55).

<!-- PDF page 66 -->

Table F1: Firm Size and Organization 

Panel A. Task decomposition 



```text
                      Occ. entropy        # occupations      Support/craft     Apprentices (%)      Compositors (%)
                            (1)                  (2)                (3)                (4)                  (5)
Firm size (ln)           0.091∗∗∗             2.825∗∗∗           0.195∗∗∗           -2.959∗∗              -2.567

                          (0.008)              (0.146)           (0.072)             (1.253)              (1.767)
Big-city control            Yes                  Yes               Yes                 Yes                  Yes
Mean outcome               0.427                5.831             0.693              29.485               49.604
Number of frims             320                  320               311                 321                  321

                                                         Panel B. Work regime

                      Weekly hours        OT hours/year Any shift (=1)          Any night (=1)      Any Sunday (=1)
                            (6)                  (7)                (8)                (9)                 (10)
Firm size (ln)             0.267              7.241∗∗∗           0.080∗∗∗           0.095∗∗∗             0.071∗∗∗

                          (0.257)              (1.312)           (0.019)             (0.021)              (0.019)
Big-city control            Yes                  Yes               Yes                 Yes                  Yes
Mean outcome              56.761               18.512             0.097               0.121                0.078
Number of firms             320                  321               321                 321                  321

                                          Panel C. Managerial hierarchy and formalization

                   Any foreman (=1) Any factor (=1)            # foremen       Workers/foreman         Mean tenure
                           (11)                 (12)               (13)               (14)                 (15)
Firm size (ln)           0.128∗∗∗              0.041∗            0.421∗∗∗           20.034∗∗∗            0.952∗∗∗

                          (0.024)              (0.021)           (0.059)             (2.795)              (0.166)
Big-city control            Yes                  Yes               Yes                 Yes                  Yes
Mean outcome               0.545                0.212             0.738              18.090                5.086
Number of firms             321                  321               321                 175                  320
```

_Notes:_ Establishment-level OLS regressions using worker-level returns from Swedish printing and alliedtrades establishments surveyed in 1902/03. The sample includes establishments with 2–1,000 workers. Each column regresses the indicated organizational outcome on ln establishment employment and includes a bigcity control, which equals one for establishments in Stockholm, Göteborg, or Malmö. Entropy is normalized Shannon entropy across 26 detailed occupation groups and is divided by ln(26). Occupations is the number of represented detailed groups; both outcomes exclude the establishment without a valid detailed occupation code. Support/craft is the number of auxiliary and simple-task workers—such as composing-room assistants, press feeders, handling and finishing workers, lithographic assistants, and general labor—divided by the skilled craft core, including typesetters, pressmen, stereotypers, type founders, engravers, lithographers, bookbinders, and gilders; establishments without a craft worker are excluded from that column. Apprentices and compositors are percentages of the roster; compositors belong to groups 2–4. Weekly hours and annual overtime hours are establishment means. The shift, night, and Sunday indicators equal one if any worker reports the corresponding arrangement. Any foreman indicates at least one respondent coded as a foreman, foremen is the number of such respondents, and workers per foreman is defined only for establishments with at least one. Any factor equals one if any worker’s reported occupation contains the title _Faktor_ . The survey lists _Faktorer_ with foremen among workers who also participated in production; foremen who did not participate directly in production did not complete employee returns. Mean tenure is measured in years with the current employer. Industry and survey year are fixed by construction. The estimates describe cross-sectional associations. Robust standard errors are reported in parentheses.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05, _∗ p <_ 0 _._ 1.

<!-- PDF page 67 -->

## **G Firm Size and Deskilling in Brickmaking** 

Brickmaking is a useful case study for examining the relationship between firm size and the employment of unskilled labor. In particular, bricks were more standardized than many manufactured goods and the key production steps were shared across product types—for example, common bricks, facing bricks, and paving bricks—and in both large and small firms. For ordinary handmade bricks, the basic sequence of production also changed less than in many other industries over the nineteenth century, even as clay preparation, kiln design, and drying arrangements saw gradual improvement (Hounsell, 2022). Clay was first extracted, weathered, and prepared by working it into a plastic consistency, either by hand or using a pug mill (Dobson, 1850). The prepared clay was then hand-moulded into individual bricks by a moulder, who pressed the clay into a wetted or sanded wooden mold. An offbearer carried the filled mold to a drying area, where the green bricks were left to harden before being stacked and fired in a clamp or kiln. The moulder was the most skilled worker in the process, but depended on a team of assistants to supply clay, remove filled molds, and manage the drying and firing stages (Dobson, 1850; Hounsell, 2022). 

Among non-mechanized yards, the key difference between small and large brickworks was organizational rather than technological. Small brickyards were typically operated by a single moulder and a few helpers, often family members. Large urban brickworks employed the same hand-moulding techniques but organized production around gangs—teams of roughly six workers, including women and children alongside adult men (Hounsell, 2022). Women typically served the moulder by preparing and transporting clay to the moulding table, while children performed ancillary tasks such as edging green bricks, carrying bricks to the drying area, and assisting at the kiln. Among brickmakers in Russia, Sweden, and the United States that did not use inanimate power, larger firms employed substantially higher shares of women and children in all three countries as shown in Online Appendix Figure G2 and Table G1. 

Although brick-moulding patents began to appear in the United States by the early 1850s, and steam-powered pug mills gradually replaced horse-driven ones, the diffusion of steampowered machinery in brickmaking was slow and uneven through the end of the nineteenth century (Hounsell, 2022). Online Appendix Figure G1 shows that this pattern is visible in the census data. In Russia, Sweden, and the United States, brickmakers were more likely than other manufacturers to operate without inanimate power and less likely to use steam. Several forces kept the vast majority of brickyards reliant on hand methods: handmade bricks were considered superior in quality and commanded a higher price than machinemade bricks (United States Bureau of Labor, 1899), while the high fixed costs associated

<!-- PDF page 68 -->

with steam adoption may have further limited mechanization to the largest firms (Atack et al., 2008). The result was an industry in which the large non-mechanized brickyard—not the steam-powered factory—was the typical setting for the employment of women and children and where the expansion in scale rested on a greater division of labor and more developed supervision, not on fundamentally different production technology.

<!-- PDF page 69 -->

<!-- Start of picture text -->
1.0<br>0.8<br>0.6<br>0.4<br>0.2<br>No Power<br>Water<br>Steam<br>0.0<br>Russia Sweden U.S. Russia Sweden U.S.<br>Other Brick<br>Share of firms<br><!-- End of picture text -->

Figure G1: Power Use Among Brickmakers and Other Firms 

_Notes:_ This figure displays the share of firms using no inanimate power, water power, and steam power among brickmakers and all other manufacturing firms in Russia, Sweden, and the United States. Each bar sums to one. The steam category includes firms using steam, with or without water power; the water category is limited to firms using water but not steam.

<!-- PDF page 70 -->

<!-- Start of picture text -->
20<br>15<br>10<br>5<br>0 (1) 1 (3) 2 (7) 3 (20) 4 (55) 5 (148)<br>ln firm size (workers)<br>(a) Russia<br>35<br>30<br>25<br>20<br>15<br>0 (1) 1 (3) 2 (7) 3 (20) 4 (55) 5 (148)<br>ln firm size (workers)<br>(b) Sweden<br>16<br>14<br>12<br>10<br>8<br>6<br>0 (1) 1 (3) 2 (7) 3 (20) 4 (55) 5 (148)<br>ln firm size (workers)<br>(c) United States<br>% women and children<br>% women and children<br>% women and children<br><!-- End of picture text -->

Figure G2: Firm Size and Unskilled Labor: Brickmakers 

_Notes:_ This figure displays binned scatterplots of the percentage of women and children in total employment against ln firm size for brickmaking firms that use neither steam nor water power in Russia (Panel A), Sweden (Panel B), and the United States (Panel C). The Sweden and U.S. panels residualize the outcome and ln firm size using year fixed effects; the Russia panel is a single cross section. Each panel uses 10 bins; best-fit lines are estimated on the corresponding underlying data.

<!-- PDF page 71 -->

Table G1: Non-Powered Brickmakers 

Outcome: % women and children 



```text
                     Russia (1894) Sweden (1864–1895) U.S. (1870–1880)
                           (1)                  (2)                   (3)

Firm size (ln)           4.982∗∗∗            5.100∗∗∗               0.945∗∗

                         (1.158)             (0.568)                (0.448)
Region × year FE           Yes                 Yes                    Yes

Mean outcome              11.82               22.45                  10.48
Firms                      565                 3598                  4840
```

_Notes:_ Firm-level OLS regressions for brickmaking firms that use neither steam nor water power. The dependent variable is the percentage of women and children in total employment. The Russia specification includes region fixed effects; the Sweden and U.S. specifications include year _×_ region fixed effects. Region corresponds to city/ _uezd_ , district ( _fögderi_ ), and county for Russia, Sweden, and the United States, respectively. Robust standard errors are reported in parentheses.<sup>_∗∗∗_</sup> _p <_ 0 _._ 01,<sup>_∗∗_</sup> _p <_ 0 _._ 05,<sup>_∗_</sup> _p <_ 0 _._ 1.

<!-- PDF page 72 -->

## **H Linking U.S. Firms** 

**Linkage Process** I link manufacturing firms observed in 1870 and 1880 using a conservative probabilistic matching procedure similar in spirit to that used in linking population censuses (Abramitzky et al., 2021). I first clean and standardize firm names by lowercasing the reported names, removing punctuation, expanding common abbreviations, and dropping legal suffixes and other generic firm-name terms. I then compare firms only within blocks defined by county, the 13-group linkage industry classification defined in Hornbeck et al. (2024), and the first letter of the standardized firm name. For each potential pair, I compute a bigram string similarity score based on the reported firm name following Wasi and Blasnik (2014), where I compare overlapping two-character sequences in the standardized names and measure the degree of overlap ranging from 0 (no shared bigrams) to 1 (identical strings). I convert this name score into a composite match score and retain the best-scoring 1880 candidate for each 1870 firm if the composite score is at least 0.95. 

The linkage procedure is intentionally conservative and is sensitive to ownership changes that alter the recorded firm name, such as the addition of partners or other changes in ownership. As a result, the procedure may miss some true matches and therefore generate false negatives. But these restrictions also reduce the risk of false positives that would result from making a more liberal match on changing firm names. I view the tradeoff that I might match relatively fewer firms that underwent more substantial ownership changes and changed industries between 1870 and 1880 as appropriate given that I mainly want to track firms that change in terms of size and technology use, but that otherwise remain broadly similar in dimensions that might also affect the employment of women and children. 

**Linkage Rates** Among the 113,718 U.S. firms observed in 1870 in the restricted analysis sample, 113,296 have usable firm names and are eligible to enter the linkage procedure. I link 13,744 of these firms to an 1880 record, or about 12.1 percent of the eligible sample. We lack reliable data on firm survival in the nineteenth century, which makes it challenging to benchmark this linkage rate. However, the BLS reports that the survival rate of manufacturing establishments today is less than 50 percent over a 10-year period, suggesting that I may link a relatively larger number of surviving firms (Bureau of Labor Statistics, U.S. Department of Labor, 2024). Note that the sample in the main analysis further restricts the sample to firms that have not adopted steam or water by 1870, which accounts for the smaller number of firms included in the main analysis.

<!-- PDF page 73 -->

Table H1: Selection into the Linked U.S. Panel 

All 1870 Eligible Linked Eligible, not linked Difference (linked _−_ not linked) p-value 



```text
Firm size (ln)               1.524   1.524  1.856       1.478              0.377 ( 0.38 SD)       0.000
Any women or children (=1)   0.213   0.213  0.273       0.205              0.068 ( 0.17 SD)       0.000
% women and children        10.177  10.185  12.025      9.931              2.094 ( 0.09 SD)       0.000
Steam power (=1)             0.236   0.236  0.263       0.232              0.030 ( 0.07 SD)       0.000
Water power (=1)             0.164   0.164  0.186       0.161              0.024 ( 0.07 SD)       0.000
ln(HP per worker + 1)        0.586   0.585  0.607       0.583              0.024 ( 0.03 SD)       0.004
Urban (=1)                   0.416   0.417  0.596       0.392              0.204 ( 0.41 SD)       0.000
Firm name length (characters)13.68713.728   14.026      13.686             0.340 ( 0.08 SD)       0.000
Food & beverages (=1)        0.162   0.162  0.148       0.164             -0.016 ( -0.04 SD)      0.000
Textiles (=1)                0.179   0.179  0.187       0.178              0.009 ( 0.02 SD)       0.011
Leather (=1)                 0.053   0.053  0.055       0.053              0.003 ( 0.01 SD)       0.170
Wood (=1)                    0.254   0.254  0.243       0.256             -0.013 ( -0.03 SD)      0.001
Paper & printing (=1)        0.024   0.024  0.032       0.022              0.010 ( 0.06 SD)       0.000
Glass, stone, etc. (=1)      0.052   0.053  0.063       0.051              0.012 ( 0.05 SD)       0.000
Chemicals (=1)               0.030   0.030  0.029       0.030             -0.001 ( -0.01 SD)      0.356
Machinery & metals (=1)      0.245   0.245  0.243       0.246             -0.003 ( -0.01 SD)      0.437
Firms                       113718  113296  13744       99552
```

_Notes:_ This table compares firms observed in the 1870 Census of Manufactures that enter the linkage procedure to firms that are successfully linked to 1880. Columns 1–4 report means for all 1870 firms in the restricted analysis sample, firms with usable firm names after the cleaning and standardization steps used in the linkage procedure, firms linked to an 1880 record, and linkage-eligible firms that are not linked, respectively. Column 5 reports the difference between columns 3 and 4, with the difference expressed as a share of the eligible-sample standard deviation in parentheses. Column 6 reports the p-value from a test of equality of means between linked and eligible-but-not-linked firms. The 1870 sample contains 113,718 firms, of which 113,296 are linkage-eligible and 13,744 are linked to 1880, or 12.1 percent of the eligible sample. The firm-name-length row is based on firms with nonmissing reported firm names. 

**Evaluating the Linked Firms** Table H1 compares the linked firms to the eligible firms that are not linked, showing that the linked firms are not a random subset of firms. Relative to firms that are not linked, linked firms are larger and more urban. The largest standardized differences are 0.41 standard deviations for urban location and 0.38 standard deviations for firm size. The estimate for firm size implies that linked firms are about 46 percent larger at baseline, while the 20.4 percentage point difference in urban location is about 52 percent of the mean among eligible firms that are not linked. Linked firms are also more likely to employ women or children, although the difference in the employment share of women and children is relatively small at 0.09 standard deviations. Differences in steam use, water use, and horsepower per worker are smaller in standardized terms. Industry composition is broadly similar across the two groups, with all standardized differences below 0.06 standard deviations. 

Taken together, these patterns are consistent with the conservative firm name-based linkage that is more likely to recover firms that survive, remain in the same industry, and have stable ownership identity in 1870 and 1880. I further examine robustness to linkage quality in the main text.

<!-- PDF page 74 -->

## **References** 

- **Abramitzky, Ran, Leah Boustan, Katherine Eriksson, James Feigenbaum, and Santiago Pérez** , “Automated Linking of Historical Data,” _Journal of Economic Literature_ , 2021, _59_ (3), 865–918. 

- **Atack, Jeremy, Fred Bateman, and Robert A. Margo** , “Part-Year Operation in Nineteenth-Century American Manufacturing: Evidence from the 1870 and 1880 Censuses,” _The Journal of Economic History_ , 2002, _62_ (3), 792–809. 

   - **, , and** , “Steam Power, Establishment Size, and Labor Productivity Growth in Nineteenth-Century American Manufacturing,” _Explorations in Economic History_ , 2008, _45_ (2), 185–198. 

- **Bairoch, Paul** , “International Industrialization Levels from 1750 to 1980,” _Journal of European Economic History_ , 1982, _11_ (2), 269–333. 

   - **and Gary Goertz** , “Factors of Urbanisation in the Nineteenth Century Developed 

   - Countries: A Descriptive and Econometric Analysis,” _Urban Studies_ , 1986, _23_ (4), 285– 305. 

- **Baten, Joerg** , “Numeracy (Total),” IISH Dataverse 2015. 

   - **and Mathias Blum** , “Height,” IISH Dataverse 2015. 

- **Berger, Thor** , “Engines of Liberation? Technology and the Decline of Child Labor,” 2026. Mimeo. 

   - , “Limits of Legislation: Did Laws Eliminate Child Labor?,” 2026. Mimeo. 

   - **and Vinzent Ostermeyer** , “Firm Survival and the Rise of the Factory,” _The Economic_ 

   - _History Review_ , 2025, _78_ (1), 62–86. 

   - **and** , “Institutional Innovation and the Adoption of New Technologies: The Case of 

   - Steam,” _The Journal of Economic History_ , 2025, _85_ (3), 844–873. 

- **Bolt, Jutta and Jan Luiten van Zanden** , “Maddison-Style Estimates of the Evolution of the World Economy: A New 2023 Update,” _Journal of Economic Surveys_ , 2025, _39_ (2), 631–671. 

- **Bureau of Labor Statistics, U.S. Department of Labor** , “34.7 Percent of Business Establishments Born in 2013 Were Still Operating in 2023,” The Economics Daily 2024. Visited June 22, 2026.

<!-- PDF page 75 -->

- **Burnette, Joyce and Maria Stanfors** , “Understanding the Gender Gap Further: The Case of Turn-of-the-Century Swedish Compositors,” _The Journal of Economic History_ , 2020, _80_ (1), 175–206. 

- **Coppedge, Michael, John Gerring, Carl Henrik Knutsen, Staffan I. Lindberg, Jan Teorell, David Altman, Fabio Angiolillo, Michael Bernhard, Agnes Cornell, M. Steven Fish, Linnea Fox, Lisa Gastaldi, Haakon Gjerløw, Adam Glynn, Ana Good God, Allen Hicken, Katrin Kinzelbach, Joshua Krusell, Kyle L. Marquardt, Kelly McMann, Valeriya Mechkova, Juraj Medzihorsky, Anja Neundorf, Pamela Paxton, Daniel Pemstein, Josefine Pernes, Johannes von Römer, Brigitte Seim, Rachel Sigman, Svend-Erik Skaaning, Jeffrey Staton, Aksel Sundström, Marcus Tannenberg, Eitan Tzelgov, Yi ting Wang, Tore Wig, Steven Wilson, and Daniel Ziblatt** , “V-Dem [Country-Year/Country-Date] Dataset v16,” Varieties of Democracy (V-Dem) Project 2026. 

- **de Chaisemartin, Clément and Xavier D’Haultfœuille** , “Difference-in-Differences Estimators of Intertemporal Treatment Effects,” _The Review of Economics and Statistics_ , 2026, _108_ (4), 863–880. 

- **Dobson, Edward** , _A Rudimentary Treatise on the Manufacture of Bricks and Tiles: Containing an Outline of the Principles of Brickmaking_ , London: John Weale, 1850. 

- **Donaldson, Dave and Richard Hornbeck** , “Railroads and American Economic Growth: A “Market Access” Approach,” _The Quarterly Journal of Economics_ , 2016, _131_ (2), 799– 858. 

- **Gorshkov, Boris B.** , _Russia’s Factory Children: State, Society, and Law, 1800–1917_ , Pittsburgh, PA: University of Pittsburgh Press, 2009. 

- **Gozen, Ruveyda, Richard Hornbeck, Anders Humlum, and Martin Rotemberg** , “Historical Differences in Female-Owned Manufacturing Establishments: The United States, 1850–1880,” _AEA Papers and Proceedings_ , 2025, _115_ , 483–488. 

- **Gregg, Amanda** , “Cliometrics, the Russian Empire, and the Soviet Union,” 2023. Mimeo. 

   - **and Tamar Matiashvili** , “Modernization in Progress: Part-Year Operation, Mechaniza- 

   - tion, and Labor Force Composition in Late Imperial Russia,” _The Journal of Economic History_ , 2022, _82_ (4), 1143–1182. 

- **Gregg, Amanda G.** , “Factory Productivity and the Concession System of Incorporation in Late Imperial Russia, 1894–1908,” _American Economic Review_ , 2020, _110_ (2), 401–427.

<!-- PDF page 76 -->

- **Haines, Michael R.** , “Historical, Demographic, Economic, and Social Data: The United States, 1790–2002,” Inter-university Consortium for Political and Social Research [distributor] 2010. Dataset. Version V3. Published May 21, 2010. 

- **Heckscher, Eli F.** , _An Economic History of Sweden_ , 2 ed., Cambridge, MA: Harvard University Press, 1954. 

- **Hornbeck, Richard and Martin Rotemberg** , “Growth Off the Rails: Aggregate Productivity Growth in Distorted Economies,” _Journal of Political Economy_ , 2024, _132_ (11), 3547–3602. 

   - **, Shanon Hsuan-Ming Hsu, Anders Humlum, and Martin Rotemberg** , “Gaining Steam: Incumbent Lock-in and Entrant Leapfrogging,” Working Paper 32384, National Bureau of Economic Research 2024. 

- **Hounsell, Peter** , _Bricks of Victorian London: A Social and Economic History_ , University of Hertfordshire Press, 2022. 

- **Hounshell, David A.** , _From the American System to Mass Production, 1800–1932: The Development of Manufacturing Technology in the United States_ , Baltimore: Johns Hopkins University Press, 1984. 

- **Kommerskollegium** , “Undersökning af Tryckerier och Därmed Förenad Handtering i Sverige,” Arbetsstatistik A:7, K. Kommerskollegii Afdelning för Arbetsstatistik, Stockholm 1909. 

- **Ljungberg, Jonas and Lennart Schön** , “Domestic Markets and International Integration: Paths to Industrialisation in the Nordic Countries,” _Scandinavian Economic History Review_ , 2013, _61_ (2), 101–121. 

- **Margo, Robert A.** , “Economies of Scale in Nineteenth-Century American Manufacturing Revisited: A Resolution of the Entrepreneurial Labor Input Problem,” in William J. Collins and Robert A. Margo, eds., _Enterprising America: Businesses, Banks, and Credit Markets in Historical Perspective_ , University of Chicago Press, 2015, pp. 215–244. 

- **Markevich, Andrei and Ekaterina Zhuravskaya** , “The Economic Effects of the Abolition of Serfdom: Evidence from the Russian Empire,” _American Economic Review_ , 2018, _108_ (4–5), 1074–1117. 

- **Moehling, Carolyn M.** , “State Child Labor Laws and the Decline of Child Labor,” _Explorations in Economic History_ , 1999, _36_ (1), 72–106.

<!-- PDF page 77 -->

- **Sandberg, Lars G.** , “The Case of the Impoverished Sophisticate: Human Capital and Swedish Economic Growth before World War I,” _The Journal of Economic History_ , 1979, _39_ (1), 225–241. 

**Schön, Lennart** , _An Economic History of Modern Sweden_ , London: Routledge, 2012. 

- **Sokoloff, Kenneth L.** , “Was the Transition from the Artisanal Shop to the Nonmechanized Factory Associated with Gains in Efficiency? Evidence from the U.S. Manufacturing Censuses of 1820 and 1850,” _Explorations in Economic History_ , 1984, _21_ (4), 351–382. 

- **United States Bureau of Labor** , _Thirteenth Annual Report of the Commissioner of Labor, 1898: Hand and Machine Labor_ , Vol. 1, Washington, DC: Government Printing Office, 1899. 

- **van Leeuwen, Bas and Jieli van Leeuwen-Li** , “Average Years of Education,” IISH Dataverse 2015. 

- **Wasi, Nattavudh and Michael Blasnik** , “reclink2: Stata module for probabilistic record linkage,” 2014. Statistical Software Components (SSC), Boston College. 

- **Wright, Gavin** , “The Origins of American Industrial Success, 1879–1940,” _American Economic Review_ , 1990, _80_ (4), 651–668. 

- **Zhuravskaya, Ekaterina, Sergei Guriev, and Andrei Markevich** , “New Russian Economic History,” _Journal of Economic Literature_ , 2024, _62_ (1), 47–114.
