ASRXXX10.1177/0003122413481351American Sociological ReviewKristal 4813512013

American Sociological Review 78(3) 361­–389 The Capitalist Machine: © American Sociological Association 2013 DOI: 10.1177/0003122413481351 Computerization, Workers’ http://asr.sagepub.com Power, and the Decline in Labor’s Share within U.S. Industries

Tali Kristala

Abstract This article addresses an important trend in contemporary income inequality—a decline in labor’s share of national income and a rise in capitalists’ profits share. Since the late 1970s, labor’s share declined by 6 percent across the U.S. private sector. As I will show, this overall decline was due to a large decline (5 to 14 percent) in , , and transportation combined with an increase, albeit small (2 to 5 percent), in labor’s share within finance and services industries. To explain the overall decline and the diverse trends across industries, I argue that the main factor leading to the decline in labor’s share was the erosion in workers’ positional power, and this erosion was partly an outcome of class- biased technological change, namely computerization that favored employers over most employees. I data from several sources to test for the independent effects of workers’ positional power indicators (i.e., unionization, capital concentration, import penetration, and unemployment) and the direct and indirect effects of computer technology on changes in labor’s share within 43 nonagricultural private industries and 451 manufacturing industries between 1969 and 2007. Results from error correction models with fixed-effect estimators support the study’s arguments.

Keywords computerization, income inequality, labor unions, labor’s share

Capitalists’ profits play a crucial role in the modern capitalist . To fill this process of social stratification. Yet inequality lacuna in inequality research, I analyze research largely neglects the dynamics of income inequality between capitalists’ profits national between capital- and workers’ income in U.S. industries over ists’ profits and workers’ compensation, and the past four decades, a period in which focuses overwhelmingly on distributional income inequality surged. issues within workers. Even studies on income inequality between social classes or aUniversity of on top income shares tend to identify the capitalist class as a subset of the self- Corresponding Author: Tali Kristal, Department of Sociology and employed. This approach ignores the fact that Anthropology, University of Haifa, Haifa 31905, , not individual owners, Israel dominate production for private profit in E-mail: [email protected] 362 American Sociological Review 78(3)

Challenging the long-standing economic This article makes a further contribution to assumption regarding the constancy of labor’s the study of income inequality by clarifying share of national income, which Keynes the question regarding the mechanisms (1939:49) called “a bit of a miracle,” recent through which computerization affects studies show that over , workers and in­equality. Economic studies assume that the capitalists do not benefit similarly from the negative relations between computers and fruits of economic growth. Across rich coun- labor’s share are an outcome of a single tries, labor’s share increased in the aftermath mechanism, specifically the increase in of World War II. Similar to the case with earn- machine productivity relative to workers’ ings inequality, however, the past three dec- productivity. Consequently, previous studies ades have seen a reverse long-term trend have (1) overlooked the structurally antago- toward increasing inequality between capital- nistic social relations between capitalists and ists’ profits and workers’ compensation (Blan- workers and (2) not resolved the puzzle, chard 1997; Kristal 2010). The main argument which this study reveals, regarding the decline for rising income inequality, put by of labor’s share only in some industries (con- , is that computerization increases struction, manufacturing, and transportation), the productivity of machines and skilled despite the massive flow of computer tech- workers; through the invisible hand of the nologies across all industries. To redress these market, this has led to rising inequality shortcomings, I argue for an additional mech- between capitalists and workers as well as anism whereby the diffusion of computer among workers (Acemoglu 1998, 2002). technology across workplaces has translated Sociologists, by contrast, tend to emphasize into a decline in labor’s share through exacer- social and power relations as driving inequal- bated union decline; therefore, the term ity, both among workers (Alderson and “class-biased technological change” may best Nielsen 2002; Kristal and Cohen 2007, 2012; describe the relations between computeriza- Moller, Alderson, and Nielsen 2009; Moller tion and labor’s share. et al. 2003; Sakamoto and Kim 2010; Volscho I will first describe labor’s share at the and Kelly 2012; Western and Rosenfeld 2011) aggregate country level and its diverse trends and between capitalists and workers (Korpi across industrial sectors over the postwar 2002; Kristal 2010, 2013; Lin and Tomaskovic- period. The downward trend in labor’s share Devey forthcoming). is very evident in European countries but In this article I draw on stratification theo- relatively moderate in the United States, lead- ries that stress power relations in the study of ing to speculation on whether the United income inequality to explain inequality States is an exceptional case (Dew-Becker between capitalists’ profits and workers’ com- and Gordon 2005). To disclose the dynamics pensation. I argue that the degree of income of labor’s share in the United States, I employ inequality is primarily a function of classes’ several operational measures for labor’s share positional power, and that both sides utilize at the aggregate country level, demonstrating their relative strength to bargain over a larger that since the late 1970s labor’s share declined slice of the national income pie. The longitu- by 6 percent over the entire U.S. private sec- dinal data on U.S. industries that I use in this tor. I then provide a first description of labor’s study allow for a fruitful contribution to the share across broad industries over the post- debate over the causes of rising inequality. World War II period. My basic assumption is These data make it possible to conduct a first that measuring labor’s share in the aggregate empirical test for the effects of computer probably masks important shifts technologies and indicators for classes’ posi- among sectors and industries, which may tional power (i.e., unionization, unemploy- either offset or amplify changes in the overall ment, capital concentration, and import size of labor’s share. In fact, based on indus- penetration) on labor’s share. try data, Solow (1958:619) argued that the Kristal 363 long-standing neoclassical economic assump- financial and nonfinancial firms, interest, and tion regarding the constancy of labor’s share rent. By taking into account all income is at least partially a “mirage.” The first part sources, national accounts data allow us to of the article reinforces Solow’s proposition measure income inequality between aggre- by showing a clear and large decline in labor’s gate categories of the working and capitalist share within construction, manufacturing, and classes. transportation industries, and an increase, In a stylized Marxian manner, I define albeit small, within finance and services. capitalists as people who own and control the I then introduce the class positional power capital used in production, and workers as all approach to labor’s share and explain how employees excluded from such ownership computerization and the erosion of workers’ and control. This leads to measuring income bargaining power relate to the decline in inequality between workers and capitalists by labor’s share. In the first part of the findings, the respective shares of national income I analyze the effects of indicators for techno- going to labor (, , and fringe logical change and workers’ bargaining power benefits) versus capital (gross firms’ profits, on changes in labor’s share in 43 nonagricul- interest, dividends, and rent).1 Using national tural private industries and 451 manufactur- accounts data to measure income inequality ing industries between 1969 and 2007. In the between capitalists and workers likely con- second part of the findings I go on to test ceals differentiations and divisions within whether computerization affects wages as it classes. Moreover, methodological and con- affects labor’s share. Specifically, I test ceptual difficulties are associated with using whether computerization increased skilled national accounts data to assess the amount of workers’ wages more than less-skilled - capital income obtained by workers or the ers’ wages and more than capitalists’ profits, part that derived from financial profits. Nev- as claimed by the skill-biased technological ertheless, national accounts data clearly por- change (SBTC) hypothesis for rising tray a central dimension of inequality in the inequality. I conclude that over the past 30 polarized class relations of capitalism. years, (1) institutional changes contributed One possible criticism of analyzing the more to rising inequality by eroding most distribution of national product between capi- workers’ bargaining power and (2) one mech- talists and workers is the popular notion that anism through which computerization in today’s world there is no longer any simple decreased labor’s share (and increased wage correspondence between classes of people inequality) was class-biased technological and sources of income. Some argue that we change, which favored capitalists and high- can no longer identify the with skilled workers while eroding most rank-and- the receipt of wages and the capitalist class file workers’ bargaining power. with the receipt of profits. A person may work for IBM, for example, and own some shares in the company as well. Indeed some work- Measuring Labor’s Share ers, mainly top executives, obtain not only Stratification research usually focuses on wages but also capital income in the form of inequality in wages and salaries, largely dividends, interest on deposits, or rent from neglecting the idea that capitalists’ profits second homes.2 Yet previous studies show play a crucial role in the process of social that including top earners with the working stratification. According to national accounts class does not bias the analysis (Kristal 2010, data, wages and salaries account for only 2013). Additionally, workers’ share of total about half of the total income generated in the capital income is relatively minor. Based on economy (see Figure 1A). A large and increas- the Federal Reserve Board’s Survey of Con- ing share of national income is in the form of Finances, I estimated that only 8 per- capital income, including gross profits of cent of total capital income in 2007 went to 364 American Sociological Review 78(3)

A. GDP by Income Sources

100% Gross corporate profits 90%

80% Rent Interest 70% Non-employees 60% Fringe benefits

50% % GD P 40%

30% Wage and

20%

10%

0% 1948 1953 1958 1963 1968 1973 1978 1983 1988 1993 1998 2003 2008

B. Labor's Share of GDPa 80%

75%

70%

65%

% GD P 60%

employees' share - enre economy 55% labor's share* - enre economy labor's share** - private industries labor's share*** - private industries 50%

45% 1948 1953 1958 1963 1968 1973 1978 1983 1988 1993 1998 2003 2008

Figure 1. Shares of Different Sources of Income in U.S. National Income, 1948 to 2008 Source: Bureau of Economic Analysis (BEA) National Income and Product Accounts Tables (n.d.). aEmployees’ share is measured as the percentage of GDP (at basic ) that goes to compensate employees (wage, salary, and fringe benefits). Labor’s share is measured by dividing employees and self-employed labor income by GDP. In the first series (labor’s share*), I estimated the labor income of the self-employed by allocating two-thirds of proprietors’ income to labor earnings and one-third to capital income. In the second series (labor’s share**), I calculated the labor income of the self-employed by multiplying the number of self-employed workers by the average wages of wage and salary workers. Finally, in labor’s share***, I estimated the labor income of the self-employed according to the average wages in their industry. Kristal 365 in the form of dividends or interest the number of self-employed by the average from deposits. Almost all capital income, wages per employee (labor’s share**). therefore, is made up of gross profits from Because most of the self-employed are con- financial and nonfinancial corporations. centrated in agriculture and construction A further issue with respect to how closely industries, where average wages are lower labor and capital incomes are associated with than in other industries, I re-estimated labor class division in contemporary capitalism is income of the self-employed according to the financialization, which means that accumula- average wages in their industry (labor’s tion is now increasingly accomplished through share***).4 The last series best describes, in financial channels, rather than through trade my opinion, the distribution of national and commodity production, largely reflecting income between workers and capitalists. a growth in interest income (Epstein and Jay- Three well-known stylized facts are evi- adev 2005; Krippner 2005; Tomaskovic- dent in Figure 1B. First, labor’s share Devey and Lin 2011). The fact that financial increased gradually from the end of World profit-making depends more on rates of return War II until the late 1960s. Second, labor’s in financial markets and less on extraction of share has declined by almost six percentage surplus value from labor, as in the case of points since the early 1970s. Third, in the production profits, might lead one to assume short term, labor’s share decreased with rapid that financial profits are class-neutral. Yet the economic growth, high rates of unemploy- evidence runs counter to this assumption, ment, and rising prices (Raffalovich, Leicht, revealing that capitalists are the main benefi- and Wallace 1992). Yet we should bear in ciaries of financialization. We know that mind that labor’s share in the aggregate econ- nearly all yields from financial assets accrue omy is a weighted average of its respective to capitalist owners of one kind or another, in shares in the various industrial sectors. In the particular to bankers with the rise of real next section, I disaggregate labor’s share by interest rates. In nonfinance industries, too, industries to better understand the overall Lin and Tomaskovic-Devey (forthcoming) trend of decline. come to the same conclusion. By allocating total capital income between financial profits (interest, dividends, and capital gains) and Labor’s Share across production profits (gross profits from sales of Industrial Sectors goods and services) based on IRS data, Lin Figure 2 presents labor’s share for eight broad and Tomaskovic-Devey find that an increase private industrial sectors. Measuring labor’s in the ratio of financial to production profits share across industries reveals that the trend led to a decline in labor’s share. Hence, the of decline in transportation began in the late fact that national accounts data include finan- 1940s and intensified in the 1970s. Since the cial and production profits in overall capital mid-1970s, labor’s share decreased by 14 income does not nullify or diminish the percentage points in manufacturing, 10 per- advantage of using these data to portray a centage points in transportation, and five in central dimension of inequality between construction. During the same years, agricul- workers and capitalists.3 ture, FIRE (i.e., finance, , and real Figure 1B displays labor’s share (wages, estate), and services industries saw an oppo- salaries, and fringe benefits) of national site trend, in which labor’s share moderately income for the extended period between 1948 increased by two to five percentage points.5 and 2008. As is commonly done (Gollin The downward trend in some industries and 2002; Krueger 1999), I estimated the labor the upward trend in others sums to a decline portion of self-employed income by allocat- of six percentage points in the overall distri- ing two-thirds of proprietors’ income to labor bution of labor’s share. Thus, the current earnings (labor’s share*) or by multiplying debate over whether labor’s share in the 366 American Sociological Review 78(3)

100% 100% Agriculture Mining 90% 90% 80% 80% 70% 70% 60% 60% 50% 50% Labor's Share 40% 40% 30% 30% 20% 20% 1948 1958 1968 1978 1988 19982008 1948 1958 1968 1978 1988 1998 2008 100% 100% Construction Manufacturing 90% 90% 80% 80% 70% 70% 60% 60% 50% 50% Labor's Share 40% 40% 30% 30% 20% 20% 1948 1958 1968 1978 1988 19982008 1948 1958 1968 1978 1988 1998 2008 100% 100% Transportation and Utilities Trade 90% 90% 80% 80% 70% 70% 60% 60% 50% 50% Labor's Share 40% 40% 30% 30% 20% 20% 1948 1958 1968 1978 1988 19982008 1948 1958 1968 1978 1988 1998 2008 100% 100% FIRE Services 90% 90% 80% 80% 70% 70% 60% 60% 50% 50% Labor's Share 40% 40% 30% 30% 20% 20% 1948 1958 1968 1978 1988 19982008 1948 1958 1968 1978 1988 1998 2008

Figure 2. Labor’s Share in Private Industriesa (gray line) and by Industrial Sectorb (black line), 1948 to 2007 Source: BEA Industry Economic Accounts (n.d.). aExcluding government, service-sector aggregates with substantial government (e.g., healthcare and educational services), and private households. bIt is not possible to provide comparable time-series trends by industrial sectors for the entire 1948 to 2007 period. Starting in 1997, the Census Bureau shifted to a new industry classification structure, the North American Industry Classification System (NAICS), which replaced the 1987 Standard Industrial Classification (SIC) system. Recently, the Bureau of Economic Analysis published industry data according to the NAICS classification back to 1987, which makes it possible to estimate labor’s share for broad industrial sectors over two long periods: 1948 to 1997 (SIC in the solid line) and 1987 to 2008 (NAICS in the broken line). Kristal 367

1.01

1

.99

.98 e in

Shar .97 r's ange Ch

Labo .96

.95 observed change in labor’s share

.94 expected change due to within industries expected change due to industrial distribution .93 1974 1978 1982 1986 1990 1994 1998 2002 2006

Figure 3. Decomposition of the Change in Labor’s Share between 1974 and 2007 (1974 = 1) Source: BEA Industry Economic Accounts (n.d.). Note: I calculated the expected change in labor’s share due to changes within industries by holding constant the industrial distribution. Specifically, for each industrial sector I weighted labor’s share in each year by its product share in 1974 and then summed the results for all industries. I calculated the expected change in labor’s share due to changes in the industrial distribution by holding constant labor’s share within industries. Specifically, for each industrial sector I weighted its product share in each year by the level of labor’s share in 1974 and then summed the results for all industries.

United States has moderately declined or presents results from this accounting exercise stayed constant over the past decades over- and shows that if only the industrial distribu- looks critical evidence. Measuring labor’s tion had changed then labor’s share would share across broad industrial sectors reveals have fallen by only two percentage points. that the core industries experienced a large However, if only labor’s share within indus- decline in workers’ share of production out- tries had changed, then labor’s share in the put and a rise in capitalists’ share. entire economy would have fallen by six Based on the variance in absolute levels of percentage points, which is the observed total labor’s share between industries (high in con- decline in labor’s share. This being the case, struction, manufacturing, trade, and services, it is evident that most of the observed decline but low in agriculture, mining, and FIRE), in labor’s share was due to changes within one might presume that a broad shift in the industries rather than to shifts in the size of economy’s sectoral composition, in particular industries. the decline in manufacturing employment and the financialization of the U.S. economy, could induce an aggregation bias in the aggre- Explaining The Dynamics gate labor’s share. I assess this structural of Labor’s Share explanation for the decline in labor’s share by Classes’ Positional Power and Income asking the following counterfactual question: Distribution If the within-sector labor’s share had remained constant over time while the industrial distri- Mainstream neoclassical economic theory bution was allowed to change, by how much conceives of labor market processes as out- would labor’s share have declined? Figure 3 comes of free exchange in the competitive 368 American Sociological Review 78(3) market, but sociologists have pointed out that income shares (summarized below under social and power relations between labor mar- factor-biased technological change), I argue ket actors are crucial to the nature of the mar- for additional mechanisms that link technol- ket and, more importantly, to its outcomes. ogy use to classes’ positional power in the The study of how power relations determine labor process and hence to their relative income includes extensive examination of the income shares (summarized below under relative power of positions—empty places in class-biased technological change). In par- the social structure—and their related material ticular, I argue that computerization is one rewards. Some of these conceptions focus on cause of organized labor’s decline, its influ- class relations (Wright and Perrone 1977) or ence channeled through (1) downsizing of occupational groups’ power (Grusky and unionized manufacturing jobs, (2) increased Sørensen 1998; Weeden 2002); others focus intensity of management anti-union actions, on employers’ versus employees’ bargaining and (3) skill polarization of the workforce power (Esping-Andersen 1985; Hicks 1999; that undermines worker . Korpi 1983; Stephens 1979). To be sure, arguing for a negative relation Following these studies, my basic assump- between computerization and labor’s share in tion is that the degree of income inequality is the first instance contradicts an abundance of largely a function of the power relations that evidence documenting a strong correlation constrain and regulate the process of income between adoption of computer-based technolo- acquisition and distribution. Specifically, I gies and wages of college-educated labor argue that the income distribution process (Autor, Katz, and Krueger 1998; Berman, between capitalists and workers is primarily a Bound, and Griliches 1994; Krueger 1993). It function of classes’ positional power, and that may also challenge arguments that computer both sides utilize their relative strength to bar- technology’s complementariness with human gain over a larger slice of the income pie. I capital (Acemoglu 1998) up-skills some com- assume, first, that classes’ positional power puter professionals and engineers (Vallas and results from the historically specific distribu- Beck 1996), or the idea that computers enhance tion of and powers over the production access to labor-market information and serve as process (Wright 1979), which may vary with, a signal of competence (DiMaggio and Bon- among other things, changes in ownership ikowski 2008), thereby increasing individual structure, prolongation of unemployment, and earnings. Yet while previous studies focus on production technology. The second component computer technology’s positive effect on indi- of classes’ positional power takes into account vidual workers’ earnings, I am interested in what Burawoy (1985) termed the “politics of computer technology’s effect on workers’ production,” stressing that workers’ positional aggregate income relative to capitalists’ income. power depends on the effectiveness of political struggle against the power relations within Factor-biased technological change. production, such as union struggles over better The factor-biased technological change (here- wages and work conditions. after FBTC) argument (occasionally named capital-biased technological change) suggests that new computer technologies are not fac- Computer Technologies tor-neutral: they benefit physical capital (i.e., The general hypothesis regarding computer- machine and equipment) productivity more ization holds that there is a negative empirical than labor productivity. In turn, this has association between new computer technolo- sparked a faster rise in capital income than in gies and labor’s share.6 Whereas most work in labor income (Acemoglu 2002, 2003; Bento- focuses on mechanisms that link lila and Saint-Paul 2003; Blanchard 1997). technology use to workers and physical capi- The FBTC argument has two related hypoth- tal productivity and hence to their relative eses. First, new technologies enjoy a relative Kristal 369 complementariness with physical capital, performed manually by blue-collar, mostly meaning that due to computer technologies, unionized workers, thus downsizing many machines and other equipment have become unionized manufacturing jobs. Even in union- much more productive than workers. Second, ized workplaces where technological change this complementariness with physical capital is implemented in agreement with the union, has prompted firms to gradually reduce their workers often lose out; for example, follow- demand for labor. A demand shift favoring ing union-backed plant modernization at a machines over workers in the production pro- automobile assembly plant, cess can result in (1) firms using more production workers experienced a sharp machines and equipment for tasks previously decline in employment when about a third of performed by workers to maximize produc- them lost their jobs (Milkman 1995). tivity, thereby decreasing labor costs and The second plausible mechanism is that labor’s share of income from an industry’s management’s greater control due to the com- product,7 or (2) firms maintaining the same puter revolution empowered employers and level of production mix (i.e., the quantity of management, allowing them to use more legal fixed assets of plant and equipment relative to and illegal anti-union tactics, such as illegal the amount of workers), thereby keeping discharge of union activists, surveillance of labor costs constant while the overall income union leaders, mandatory captive-audience pie increases due to rising productivity of meetings with top management, and refusal to capital, which in turn leads to a decline in negotiate a (Bronfen- labor’s share. brenner 2009). Previous studies show that computer technologies have enhanced Class-biased technological change. employers’ superior position in the labor pro- Factor biases in technological change clearly cess by augmenting their “technocratic con- could affect income distribution, but other trol” (Burris 1993; Wallace and Brady 2001), biases may have been more important. As an a system in which employers and managers additional explanation, I advance a class- have the flexibility and coordinating features biased technological change (hereafter necessary to facilitate work (Burris 1998; CBTC) argument that also predicts negative Crowley et al. 2010; Vallas 1993; Zuboff relations between computer technology and 1988).8 One outcome of employers’ superior labor’s share. My argument differs from the position in the labor process, I argue, is the FBTC argument by shifting the spotlight from evolution in sophistication and intensity of factors productivity to classes’ positional their anti-union tactics, designed to intensely power in the labor process. I expect that monitor and punish union activity. computer-based technologies are not class- An additional mechanism links computer neutral but embody essential characteristics technology to skill polarization of the work- that favor capitalists (and high-skilled work- force, which undermines established workers’ ers), while eroding most rank-and-file work- solidarity, thereby reducing the likelihood of ers’ bargaining power. In particular, I argue working-class cohesion and . that computerization has reduced labor’s Studies show that new computer technologies share indirectly through its role in reducing have had highly polarizing effects on the work- unionization. This is in contrast to the FBTC force: skilled workers experienced up-skilling, hypothesis that computerization has reduced while many production workers underwent labor’s share directly. de-skilling (Burris 1998; Vallas and Beck Why might computer technologies have 1996). This skill polarization deepened divi- led to a decline in labor unions? The first sions among workers and most likely sapped plausible mechanism is that of the social and organizational bases on which the production process prompted firms to uti- workers’ collective resistance might grow. lize computer equipment in tasks previously U.S. workers’ skill polarization with the influx 370 American Sociological Review 78(3) of Information and Communication Technolo- widespread among private-sector production gies (ICT) has thus undermined workplace workers, in particular in the manufacturing, relations as the source of worker solidarity and construction, mining, and transportation thereby weakened the labor movement. industries, and emerged The CBTC hypothesis that computeriza- as the industrial workplace norm. Although tion’s effect on labor’s share is channeled the U.S. labor movement is generally charac- through unionization may solve the puzzle as terized as rather than social to why the diffusion of computer technolo- movement unionism, especially after defeat of gies across all industries led to a decline in the “red” unions in the early years of the Cold labor’s share only in construction, manufac- War (Stepan-Norris and Zeitlin 2003), U.S. turing, and transportation. If computer tech- labor unions have had a significant effect on nologies’ effect on labor’s share is partly workers’ well-being. In fact, studies show that channeled through the erosion of labor unions, U.S. trade unions increased not only wages as I argue, then industries where unionization and fringe benefits but also labor’s share of was relatively high in the 1960s and 1970s, national income, at least until the 1970s (Hen- such as manufacturing, transportation, and ley 1987; Kalleberg et al. 1984; Macpherson construction, should have experienced a sig- 1990; Rubin 1986; Wallace et al. 1999). nificant decline in labor’s share. On the other In the past three decades the social contract hand, in industries where unionization was between capital, labor, and the state has been always low, such as finance, trade, and ser- broken, most likely affecting the dynamics of vices, we should find only a weak direct labor’s share. has been in decline effect of computers on labor’s share, and the since reaching its peak in the mid-1950s. In the channeled effect should be marginal. private sector, union density has dropped from one-in-four wage and salary workers being union members in the early 1970s to below Workers’ Relative Bargaining Power one-in-thirteen today (Western and Rosenfeld Previous studies stress that the decrease in 2011). Unions declined as jobs shifted from workers’ bargaining power is the main poten- unionized, core industries to less unionized, tial explanation for the current decline in service industries (Farber and Western 2001). labor’s share. Studies show that the more pow- Unions also found themselves under relentless erful and integrated are working-class organi- attack from employers using legal and illegal zations, the better able they are to counteract anti-union tactics (Bronfenbrenner 2009), the capitalists and shift the distribution of rents anti-union Reagan administration (Tope and from firms to workers (Kalleberg, Wallace, Jacobs 2009), and labor legislation that had and Raffalovich 1984; Kristal 2010, 2013; powerful, negative implications for the labor Rubin 1986; Wallace, Leicht, and Raffalovich movement (Jacobs and Dixon 2006; Wallace, 1999). Evidence points to a substantial rise in Rubin, and Smith 1988). Although some labor’s share during the 1950s and 1960s due unions have recently countered the organizing to workers having gained organizational power trend of the 1980s and pursued industry-wide in the economic and political spheres. Since organizing (Voss and Sherman 2000), private- then, labor’s share has declined in all rich sector density grew only very recently, in 2007 countries, as labor unions and labor-affiliated (Southworth and Stepan-Norris 2009). fell on lean and workers Unionization is important for the dynamics left without a strong collective voice to of labor’s share, but it does not fully capture confront employers (Kristal 2010). workers’ relative bargaining power. What is In the United States, organiza- missing here is capitalists’ power, that is, their tions that empower workers’ militancy are ability to exert control over product and labor particularly important to working-class power. markets as well as the production process During the 1940s and 1950s, unionization was itself. Capitalists’ monopoly power, in particular, Kristal 371 generally augments corporate capitalists’ mar- workers (Alderson and Nielsen 2002), and ket power and enables them to increase their reorients manufacturing activity toward capital- profits through their control over pricing mech- intensive plants (Bernard, Jensen, and Schott anisms and to gain monopoly rents in the form 2006). Therefore, although importing manu- of political influence (Jacobs 1988). As politi- factured goods from less-developed countries cal Kalecki (1938) noted, an increase increases the economy’s income, it does not in the degree of monopoly power might result translate into a rise in average earnings and in a reduced share of national income accruing thus decreases labor’s share (Kristal 2010). to wage-earners. This long-standing argument is supported by only little empirical evidence, from the printing industry between 1946 and Data, Variables, and 1978 (Kalleberg at al. 1984) and manufacturing Method industries in 1972 (Henley 1987). Data and Variables I suggest that, all else being equal, augmenta- tion of capitalists’ power within industries may I tested the effect of computer technology and increase average workers’ compensation due to workers’ bargaining power factors on labor’s mechanisms such as an internal labor market and share using longitudinal data on U.S. indus- labor unions (Kalleberg and Van Buren 1996), tries. I used a pooled cross-sectional time- but it will decrease workers’ share of industry series design (i.e., yearly observations for income relative to capitalists’ profits. The sim- each industry) to test the study’s arguments. plest way for employers to decrease product The combined industry-year datasets include market competition is to purchase other firms, as 43 comparable (two-digit) industries that was frequently done in the merger waves of the cover the entire nonagricultural private sec- late 1960s and mid-1980s (Stearns and Allan tor. Due to the major change in the industry 1996). Although economy-wide concentration classification structure in 1997, I have one levels in the private sector have not increased dataset for the years 1969 to 1997 and another since the 1960s (White 2002), it might be the dataset for the years 1988 to 2007. I also col- case that aggregate economy data mask impor- lected data only for manufacturing industries, tant shifts among industries that may offset and this third database includes data on 451 changes in capital concentration. We have (four-digit) manufacturing industries for the seen, for example, a pattern of increasing con- years 1977 to 2002. It is impossible to ana- centration in the automobile, airline, petroleum lyze earlier data because information on production, motion-picture distribution, micro- unionization by two-digit industry is avail- computer, steel, tire, and wine industries, to able only from 1968 onward, and data on name just a few. computer investments by four-digit industry The final component of workers’ relative is available only from 1977. bargaining power is more global and relates to Analyses are based on data drawn from sev- U.S. trade with low-wage countries. As U.S. eral governmental and census publications on trade barriers have fallen in recent years, low- U.S. industries. I combined data on labor’s share wage countries like China and India have and the magnitude and composition of each begun exporting to the United States many of industry’s capital investments from the Bureau the more labor-intensive products (e.g., t-shirts of Economic Analysis (BEA) Industry Eco- and sneakers) formerly produced domestically. nomic Accounts data and the Annual Survey of This import penetration places U.S. workers in Manufactures (ASM), with data on unionization direct competition with lower-paid workers in and unemployment from Current Population developing countries. Competition curbs Survey (CPS) samples and the Bureau of Labor workers’ bargaining power, brings down the Statistics (BLS), data on capital concentration wages of the least-skilled U.S. workers (Wood from the Census of Manufacturing (CM), and 1994), increases earnings inequality among data on import penetration from Schott (2010). A 372 American Sociological Review 78(3) full description of these data sources can be estimates of parameters when industry effects found in the Data Appendix. are arbitrarily correlated with measured I followed previous studies and measured explanatory variables (Halaby 2004). By labor’s share by dividing labor income by an applying fixed-effects estimators, the models industry’s value added (Gollin 2002; Krueger focus on within-industry variation over time, 1999). Value added is net of indirect taxes and and coefficients represent a cross-industry is allocated as either labor income or capital average of the longitudinal effect. This esti- income. Labor income includes compensa- mation strategy is most appropriate to the tion to employees and self-employed indi- current study because (1) the overall decline viduals’ imputed income, based on the in labor’s share was due to a decline within average wage in their industry; capital income industries, and (2) the study’s arguments for a includes the self-employed’s residual income positive effect of unions on labor’s share and and firms’ profits. I employed a simple meas- a negative effect for computer investments, ure of computer technology by measuring real unemployment, capital concentration, and investments in computers and software as a import penetration apply to dynamics within share of total investments. Although the BEA all included industries, whereas the diverse and Census industry data do not directly trends across industries are explained by dif- measure the kind of technology implemented ferent levels of the independent variables, in the production process, I assumed that mainly unionization. when firms invest in computing equipment To estimate the long- and short-run effects they are most likely to use it at different of indicators for computer technology and stages of the production process. I measured workers’ bargaining power on labor’s share, I union density by dividing the number of analyzed single-equation error correction union members in each industry by the num- models (ECMs)10 that can accommodate sta- ber of wage and salary workers. Unemploy- tionary and nonstationary variables, given ment is measured by dividing the number of that the errors are stationary (Beck and Katz unemployed in each industry by the number 2011; De Boef and Keele 2008).11 Indeed, no of employed and unemployed persons.9 Capi- statistical testing is required to see that the tal concentration data are available only for variables observed annually for relatively manufacturing industries from the CM, and short periods trend over time and do not reach the measure consists of the ratio of sales by equilibrium. There are very few cycles in the four largest firms to the total volume of labor’s share, unionization, or computer sales in each industry. I used import data by investments over the past 40 years, and the industry and country to measure imports in data series seem to be integrated (i.e., nonsta- manufacturing industries originating in low- tionary).12 The fact that the data series are wage countries. I measured import penetra- nonstationary does not rule out a long-run tion by imports from low-wage countries as a equilibrium relationship. It may be the case share from industry’s value added. Table 1 that the data series are cointegrated; that is, shows descriptive statistics for all variables. the dependent and independent variables maintain a long-run error correction relation- ship (Engle and Granger 1987). To test Method whether the data series are cointegrated, I I analyzed the determinants of labor’s share in performed the standard two-step cointegra- time-series cross-sectional dynamic specifi- tion test by regressing Y on X (in levels) and cation (a lagged dependent variable is then testing whether the residual is stationary. included among the predictors) by fixed- Based on the results (see Table A1 in the effects estimators. Fixed-effects estimators, Appendix), we can reject the null of no cointe- which exploit within-industry variation as a gration for almost all variables in all datasets, means of purging unit heterogeneity, make it concluding there are long-run relationships possible to obtain unbiased and consistent between the variables. Only the null of no Kristal 373

Table 1. Descriptive Statistics of Relevant Variables Sector Private Industries Manufacturing Industries

N industries 43 43 451a 393 Years 1969 to 1997 1988 to 2007 1978 to 2002 1978 to 2002 Dependent Variables Labor’s Share (%) Mean (SD) 68.8 (17.9) 62.2 (16.9)b 48.4 (13.8) 47.6 (13.9) Minimum–maximum 5.3–103.8 5–92.9 3.3–106.0 3.3–106.0 Mean annual change (SD) –.260 (4.09) –.186 (3.32) –.291 (5.43) –.273 (5.37) Source BEA BEA ASM ASM Skilled Wage-Bill Share (%) Mean (SD) 35.7 (16.4) 47.4 (17.6) 37.1 (12.0) 37.8 (12.1) Minimum–maximum 5.7–88.4 11.2–92.8 .0–83.1 .0–83.1 Mean annual change (SD) .818 (4.35) .711 (4.69) .243 (2.43) .255 (2.40) Source BEA, CPS BEA, CPS ASM ASM Skilled Income Share (%) Mean (SD) 19.6 (10.3) 27.4 (13.2) 17.7 (7.5) 17.8 (7.6) Minimum–maximum 1.2–64.8 2.3–72.2 .0–83.0 .0–83.0 Mean annual change (SD) .101 (3.92) .359 (3.26) .015 (2.63) .024 (2.58) Source BEA, CPS BEA, CPS ASM ASM Unskilled Income Share (%) Mean (SD) 45.6 (16.4) 31.6 (15.3) 30.9 (11.1) 30.1 (11.1) Minimum–maximum 3.0–88.7 1.7–70.1 2.1–100 2.1–100 Mean annual change (SD) –.294 (4.51) –.552 (3.14) –.303 (3.76) –.293 (3.68) Source BEA, CPS BEA, CPS ASM ASM

Independent Variables Ratio of Computer Investment to Total Investment (%) Mean (SD) 6.3 (8.2) 18.4 (13.9) 6.2 (7.4) 6.2 (7.2) Minimum–maximum 0–51.0 .4–66.0 0–117.2 0–103.3 Mean annual change (SD) .552 (1.32) .415 (2.23) .288 (3.98) .293 (3.67) Source BEA BEA ASM ASM Unionization (%) Mean (SD) 24.2 (18.0) 15.6 (14.6) 21.5 (12.4) 21.3 (12.1) Minimum–maximum 0–83.9 .2–80.5 0–100 0–100 Mean annual change (SD) –.440 (2.40) –.407 (1.72) –.806 (4.56) –.803 (4.62) Source CPS, BLS CPS, BLS CPS CPS Unemployment (%) Mean (SD) 6.3 (3.6) 5.5 (2.9) 7.1 (4.6) 7.0 (4.6) Minimum–maximum .2–26.4 .1–21.4 0–56.2 0–56.2 Mean annual change (SD) .056 (2.6) –.114 (2.19) .001 (4.5) –.001 (4.6) Source CPS CPS CPS CPS Capital Concentration (%) Mean (SD) 39.2 (20.5) 39.6 (20.5) Minimum–maximum 2–100 2–100 Mean annual change (SD) .211 (1.47) .203 (1.47) Source CM CM Import Penetration (%) Mean (SD) 20.2 (141.5) Minimum–maximum 0–5,984 Mean annual change (SD) 3.01 (41.9) Source Schott, ASM

N 1,247 860 11,269 9,819

Note: BEA = Bureau of Economic Analysis; ASM = Annual Survey of Manufacture; CPS = Current Population Surveys; BLS = Bureau of Labor Statistics; CM = Census of Manufacturing. aData are not available for eight manufacturing industries that changed classification in 1996. bLabor’s share for 1988 to 2007 was calculated without taking into account the move of self-employed to wage and salary employment due to lack of consistent data on self-employed and wages. This may have biased results for the service sector. 374 American Sociological Review 78(3) cointegration between computer investments, Empirical Analyses Of union density, and labor’s share for the years Labor’s Share Dynamics 1988 to 2007 cannot be rejected. This sug- within Industries gests we need to be more careful with inter- pretation of the coefficients of the lagged Using the methods described in the preceding levels for these two variables during the years section, I modeled the change in labor’s share 1988 to 2007. Because these coefficients are within industries as a function of short-term not statistically significant, it does not affect changes (i.e., first differences) and long-term the reading of the findings. levels (i.e., lagged values) of industry-level The ECMs’ parameterization has the measures of workers’ bargaining power and advantage of explicitly modeling both short- computer utilization. To control for year-spe- and long-run effects on labor’s share, provid- cific economy-wide shocks, the models ing easily interpretable estimates of these include a dummy variable equal to 1 for parameters, which makes these models par- recession years (1969 to 1970, 1973 to 1975, ticularly appropriate in the context of this 1980 to 1982, 1990 to 1991, 2001, and 2007), study. For example, the ECMs make it possi- which captures effects of periodic expansion ble to estimate two effects of unionization on and contraction of output. The rationale is labor’s share: one that occurs immediately that during recession, labor’s share should with a decline in union density, and another increase in the short-term because massive that is dispersed across future time periods cuts in wages and employment are usually with the erosion of labor unions. I therefore limited by institutional constraints, and firms’ specify the time-series cross-section variant profits are the first to be negatively affected of the single-equation error correction model (Raffalovich et al. 1992). for the dynamic relationships: I also tested for the effect of labor-affili- , ∆∆labor ss_ hareX=+α β ated government on labor’s share in the U.S. it,,01it context by including a dummy variable taking , −−ββ21(()labor ss_ hareXit,,−−31it + εit, on a value of 1 in years with Democratic presidents and 0 when Republicans held the In this model, current changes in labor’s presidency. Results were not statistically sig- share (measured in first difference, i.e., Y – nificant (data not shown). Table 2 shows t Y ) are a function of both short-term changes results for two-digit private sector industries t–1 (i.e., first differences) in the independent var- (Models 1 through 4), two-digit core indus- iables and their long-term levels. Specifically, tries (Models 5 through 8), and four-digit b1 captures any short-term effects on labor’s manufacturing industries (Models 9 through share, and long-term effects are captured by 12). To illustrate the dynamic pattern of rela- b3. The long-term effect occurs at a rate dic- tions, Figure 4 plots their lag distributions for tated by the value of b2 that captures the rate the core (Models 5 and 6) and manufacturing of return to equilibrium. By dividing the coef- (Models 11 and 12) industries. The lag distri- ficient of the lagged-level variables by the bution, presented by the comparable semi- coefficient of the lagged labor’s share, we get standardized coefficients, is the amount by the long-term multiplier that represents the which labor’s share changed each year, total long- and short-term effect on labor’s expressed in percentage points, in response to share for a one-point increase in the inde- an increase in one standard deviation of the pendent variable. In all models, estimates are independent variable.13 weighted by industry size to make sure results Overall, I found empirical support for my are not biased by small industries represent- argument that higher levels of workers’ bar- ing only a small fraction of the total product. gaining power redistribute income toward the ** ** ** ** ** ** ** ** 12 .120 .063 .064 .024 .054 .193 Yes Yes (.020) (.002) (.021) (.019) (.019) (.009) (.244) (.023) (.029) (.039) (.028) (.074) –.003 –.014 1.730 –.034 –.017 –.277 –.114 1.99 393 9,819 ** ** ** ** ** ** 1978 to 2002 11 .081 .024 .020 .034 .170 Yes Yes (.002) (.020) (.023) (.008) (.258) (.024) (.020) (.036) (.031) (.071) –.004 –.019 2.006 –.096 –.231 –.115 2.00 ** ** ** ** ** ** 10 Manufacturing .121 .068 .066 .053 .190 Yes Yes (.019) (.020) (.019) (.019) (.229) (.021) (.028) (.036) (.028) (.070) –.017 1.671 –.032 –.010 –.287 –.108 1.99 451 11,269 ** ** ** ** ** 9 1978 to 2002 .086 .019 .043 .167 Yes Yes (.020) (.023) (.242) (.021) (.019) (.033) (.031) (.067) –.023 1.958 –.094 –.240 –.114 2.00 ** ** 8 .031 .213 .881 .068 .018 .174 Yes Yes (.052) (.112) (.554) (.128) (.123) (.053) (.083) (.065) –.108 –.044 –.249 1.91 480 24 ** ** 7 1988 to 2007 .215 .895 .173 Yes Yes (.114) (.538) (.128) (.057) (.088) (.061) –.106 –.003 –.039 –.244 1.91 ** ** ** ** 6 ∆ Labor’s Share ∆ Labor’s .119 .061 .200 Yes Yes Core Industries (.135) (.026) (.326) (.116) (.049) (.087) (.032) (.093) 1.223 –.299 –.030 –.115 –.242 –.014 1.96 812 28 ** ** ** 5 1969 to 1997 .111 .193 Yes Yes (.137) (.351) (.120) (.049) (.076) (.037) 1.370 –.348 –.076 –.128 –.227 1.97 ** ** 4 .049 .265 .028 .035 .021 .174 Yes Yes (.080) (.217) (.043) (.039) (.018) (.065) (.037) (.093) –.154 –.248 –.009 1.84 860 43 ** ** 3 1988 to 2007 .054 .257 .023 .013 .173 Yes Yes (.080) (.218) (.043) (.019) (.068) (.035) –.145 –.246 1.88 ** ** ** 2 .069 .677 .083 .014 .169 Yes Yes (.106) (.256) (.067) (.025) (.015) (.064) (.025) (.078) –.051 –.108 –.235 –.020 1.87 43 1,247 ** ** ** Nonagricultural Private Sector 1 1969 to 1997 .065 .796 .156 Yes Yes (.106) (.271) (.066) (.018) (.057) (.029) –.077 –.005 –.119 –.211 1.89 ( t –1) ( t –1) ( t –1) ( t –1) ( t –1) ( t –1) p < .05 (two-tailed test). 2 ∆ Unemployment N of industries Sector Years Recession (dummy) . Unstandardized Coefficients from Single Equation ECM, Dependent Variable Is Annual Change in Labor’s Share Is Annual Change in Labor’s 2 . Unstandardized Coefficients from Single Equation ECM, Dependent Variable Table Dependent Variable ∆ Computer investments Unemployment Model Union density ∆ Import penetration Import penetration Labor’s share Labor’s Computer investments ∆ Union density ∆ Capital concentration Constant Modified DW Capital concentration Each column represents a pooled regression of changes in labor’s share. Table entries are OLS estimates. Robust standard errors in parentheses heteroskedasticity and share. Table Note : Each column represents a pooled regression of changes in labor’s autocorrelation consistent. Estimates are weighted by mean industry share of total value added over the years. ∆ indicates annual change in variable. Core industries include construction, manufacturing, and transportation. ** Industry dummies N R

375 376 American Sociological Review 78(3)

Core Industries 1969 to 1997 Manufacturing Industries 1978 to 2002 Computers without a Control for Unions Computers without a Control for Unions 1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0 0 1234 0 1 2 34 -0.5 -0.5

Change in Labor's Shar e -1.0 -1.0 Time Periods Time Periods

Computers with a Control for Unions Computers with a Control for Unions 1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0 0 1234 01234 -0.5 -0.5

Change in Labor's Share -1.0 -1.0 Time Periods Time Periods

Union Density Union Density 1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0 01234 01234 -0.5 -0.5

Change in Labor's Share -1.0 -1.0 Time Periods Time Periods

Figure 4. Estimated Lag Distributions for Change in Labor’s Share Within-Industries Note: Unstandardized regression coefficient multiplied by the sample standard deviation of the independent variable. This represents the change in Y over time after an increase of one standard deviation in X, in original units of Y. working class. Results reveal that the decline (from 30 to 16) from 1969 to 1997 decreased in unionization, the rise in unemployment (in labor’s share by five percentage points, which the entire private sector), and the importation was the overall decline in labor’s share during of goods from less-developed countries all these years. Also in manufacturing, union curbed the bargaining power of many work- regression explains almost the entire decline ers over the past decades and led to a signifi- in labor’s share. The 20-perctange-point cant decline in labor’s share. Union density, decline in union density in manufacturing in particular, had a robust, long-term positive industries from 1978 to 2002 depressed effect on changes in labor’s share in all mod- labor’s share by 8.4 percentage points. els except for the period 1988 to 2007, and an As expected, I found a negative long-run additional short-term positive effect in manu- effect for unemployment on changes in facturing. The effect of union regression on labor’s share in private sector industries that the decline in labor’s share was substantial. explains the overall decline of two percentage The 14-perctange-point decline in union den- points in labor’s share between 1988 and sity in nonagricultural private industries 2007. This negative effect of unemployment Kristal 377 on labor’s share indicates that an increase in effect (according to CBTC) through unioniza- labor’s reserve army diminished labor’s bar- tion on labor’s share. Results in Tables 2 and gaining power over wages and benefits, 3 mainly support the CBTC theory of an indi- thereby causing a decline in labor’s share. Yet, rect effect, but there is also some support for contrary to my hypothesis, when I analyzed the the FBTC argument. more unionized core and manufacturing indus- First, and most important, although the dif- tries between 1988 and 2007, I found only a fusion of computer technology occurred in all positive short-term effect for unemployment industries, I found a significant negative on changes in labor’s share, suggesting that effect for computer investments on changes workers in the nonunionized sector bore the in labor’s share only within core (Models 5 brunt of increasing unemployment. and 6 in Table 2) and manufacturing (Models I also found that importation of manufac- 9 and 11 in Table 2) industries. This finding tured goods from less-developed countries suggests a channeled effect for computeriza- increased labor’s share in manufacturing tion, because labor unions were relatively industries in the short term, but decreased its strong in core industries until the 1970s, share in the long term. The unexpected posi- while labor unions’ power in other private tive short-run effect of import penetration on industries, such as trade, FIRE, and services labor’s share might be due to short-run industries, was at no time substantial. Second, employment growth (Kristal 2010). In the with addition of union density to the model, long-run, however, import penetration the effect of computers is significantly attenu- increases the size of the national income pie ated in core industries (Models 5 and 6 in without increasing wages, which leads to a Table 2) and becomes positive in manufacturing decline in labor’s share. Overall, the increase industries (Models 9 through 12 in Table 2).14 in import penetration from less than one per- Table 3 provides additional support for the centage point in 1978 to more than 18 per- CBTC argument by re-estimating Models 1 centage points in 2002 decreased labor’s and 3 in Table 2 with an interaction dummy share by .2 percentage points, most likely due for unionization, where 1 denotes industries to a in low-skilled workers’ earn- with a decline in union density of more than ings and benefits. Finally, in most models for five percentage points and 0 denotes all other manufacturing industries I did not find the industries (mainly trade, FIRE, and services expected negative coefficient for the effect of industries where unionization was always capital concentration on labor’s share. It may low). Results show that computers had a be the case that capital concentration’s posi- negative effect on labor’s share only in indus- tive effect on capitalists’ profits was offset by tries where unionization declined and only its positive effect on workers’ compensation, when workers retained some organizational in particular due to labor unions that thrived power. in monopolistic industries where employers could pass on higher labor costs to customers. The findings for negative effect of com- Computerization and puter technology on labor’s share support the Income Distribution general argument of the FBTC and CBTC All in all, I found empirical support for both theories that utilization of computer technolo- of this study’s arguments regarding the causes gies across workplaces decreased aggregate of the decline in labor’s share. I consistently workers’ compensation as a share of an indus- found a positive relation between indicators try’s income, although only in core industries for workers’ relative bargaining power for the years 1969 to 1997. To better under- (mainly unionization) and labor’s share, and a stand the link between computers and labor’s negative relation between computer technolo- share, I tested the validity of computers’ direct gies and labor’s share, channeled through the effect (according to FBTC) or channeled decline in unionization. At first glance, the 378 American Sociological Review 78(3)

Table 3. Unstandardized Coefficients from Single Equation ECM, Dependent Variable Is Annual Change in Labor’s Share

Dependent Variable ∆ Labor’s Share

Sector Nonagricultural Private Sector

N of Industries 43 43

Years 1969 to 1997 1988 to 2007

Model 1 2

∆ Computer investments x union declined –.322** .009 (.110) (.125) ∆ Computer investments x union constant .009 .026 (.063) (.043) Computer investments x union declined –.087 –.001 (t–1) (.049) (.048) Computer investments x union constant .008 .021 (t–1) (.014) (.017) ∆ Unemployment x union declined .136 .176 (.124) (.100) ∆ Unemployment x union constant –.170 –.163 (.116) (.137) Unemployment x union declined –.132** –.068 (t–1) (.066) (.087) Unemployment x union constant –.079 –.309** (t–1) (.112) (.130) Union declined (dummy)a –2.497** 10.007** (.877) (2.251) Recession (dummy) .839** .217 (.239) (.218) Labor’s share –.236** –.247** (t–1) (.027) (.036)

Constant Yes Yes Industry dummies Yes Yes R2 .181 .183 Modified DW 1.88 1.87 N 1,247 860

Note: Each column represents a pooled regression of changes in labor’s share. Table entries are OLS estimates. Robust standard errors in parentheses are heteroskedasticity and autocorrelation consistent. Estimates are weighted by mean industry share of total value added over the years. ∆ indicates the annual change in the variable. aUnion density declined by more than five percentage points. **p < .05 (two-tailed test). negative relation between computerization answer is simple: computers benefited capi- and labor’s share is puzzling. If computer talists’ profits more than educated workers’ technologies increase educated workers’ compensation (and, of course, more than less- wages, as the dominant skill-biased techno- skilled workers’ compensation). logical change argument claims (Acemoglu I tested this hypothesis by estimating the 1998; Autor et al. 1998), how could comput- effect of computers on skilled workers’ share of ers have led to a decline in the aggregate an industry’s wage bill (relative to less-skilled workers’ share of industry income? The workers) and their share of an industry’s income Kristal 379

(relative to less-skilled workers and capitalists). relative wages, as argued by the SBTC If computers increased skilled workers’ share hypothesis. However, the coefficient for com- of an industry’s wage bill, compared to that of puter investments is zero for skilled workers’ less-skilled workers, but did not increase skilled share of an industry’s income, compared to workers’ share of an industry’s income, com- that of less-skilled workers and capitalists pared to that of less-skilled workers and capi- (Models 5 and 6 in Table 4 and Model 5 in talists, we may conclude that computer Table 5), suggesting that computers did not technologies benefited capitalists’ profits more increase skilled workers’ wages and fringe than skilled workers’ compensation. benefits relative to capitalists’ profits. The Data on manufacturing industries from the one exception is the positive effect for com- ASM make it possible to empirically test this puter investments on skilled workers’ share of hypothesis; Table 4 shows the results. I fol- an industry’s income for the period 1988 to lowed Berman and colleagues (1994) and 2007 (Model 6 in Table 5), suggesting that Autor and colleagues (1998) in using the computerization noticeably raised skilled available information on production and non- workers’ wages in the 1990s and early 2000s. production workers as indicators for educa- tion and skill levels. Nonproduction workers include managers above the line-supervisor Conclusions and level, as well as clerical, sales, office, profes- Discussion sional, and technical workers. To analyze the This article marks advances on three fronts effect of computerization on the wage-bill regarding the recent decline in labor’s share share and income shares of skilled and less- and, more generally, enhances our under- skilled workers in all private industries, I standing of the causes of rising income drew on data from the March CPS; Table 5 inequality in the past 30 years as well as the shows the results. I used the March files from contemporary state of inequality. First, I 1969 to 2008 (covering earnings from 1968 to showed that the aggregate trend of decline in 2007) to compile a sample of annual earnings U.S. labor’s share by almost six percentage for wage/salary workers age 18 to 65 years points since the early 1970s conceals diverse who participated in the labor force on a full- trends within industrial sectors. Measuring time, full-year (FTFY) basis, defined as labor’s share across broad industrial sectors working 35-plus hours per week and 50-plus reveals that the construction, manufacturing, weeks per year. Skilled workers are defined and transportation industries saw a large according to their educational attainments decline in workers’ share of production out- and include college equivalents (college grad- put and a rise in capitalists’ share; in trade, uates plus half of those with some college); FIRE, and services industries, labor’s share unskilled workers include noncollege (or high stayed relatively constant or even increased. school) equivalents (half of those with some That labor’s share declined only in core college plus workers with 12 or fewer years unionized industries, despite the massive of schooling). flow of computer technologies across all Results generally confirm my hypothesis industries, suggests that played and suggest the diffusion of computer tech- a central role in the decline of labor’s share. nologies across workplaces was not only Second, I further developed a power rela- good for educated workers’ income, but even tions thesis, stating that the erosion in work- better for capitalists’ profits. The coefficient ers’ positional power was the main factor for computer investments is positive for leading to the decline in labor’s share. This skilled workers’ share of the wage bill, com- argument could be empirically tested with the pared to that of less-skilled workers (Models industrial data utilized in this study, because 1 and 2 in Tables 4 and 5), indicating these data make it possible to directly meas- that computers increased educated workers’ ure indicators not only for workers’ power but ** ** ** ** ** ** 6 Yes Yes 393 9,819 .027 .006 .024 .042 .008 .179 .000 .889 (.028) (.008) (.021) (.009) (.011) (.019) (.036) (.009) (.004) (.038) (.000) (.109) –.008 –.001 –.086 –.012 1.97 –.238 1978 to 2002 ** ** ** ** ** 5 Nonproduction Workers Yes Yes 451 .027 .008 .024 .044 .000 .178 .851 11,269 (.025) (.008) (.020) (.008) (.011) (.018) (.035) (.008) (.038) (.103) –.004 –.078 –.013 1.96 –.248 1978 to 2002 ** ** ** ** ** ** ** 4 Yes Yes 393 9,819 .057 .099 .019 .019 .011 .163 .763 (.017) (.013) (.010) (.013) (.013) (.034) (.023) (.006) (.012) (.019) (.156) (.001) –.011 –.024 –.033 –.022 2.06 –.242 –.004 1978 to 2002 ∆ Workers’ Share in Industry’s Income Share in Industry’s ∆ Workers’ ** ** ** ** ** Production Workers 3 Yes Yes 451 .057 .020 .099 .010 .156 .741 11,269 (.015) (.012) (.010) (.012) (.012) (.032) (.022) (.012) (.019) (.146) –.006 –.026 –.034 –.017 2.06 –.250 1978 to 2002 ** ** ** ** ** ** ** ** 2 Yes Yes 393 9,819 .021 .030 .042 .007 .135 .003 .506 (.010) (.006) (.010) (.008) (.007) (.022) (.009) (.001) (.010) (.028) (.001) (.070) –.020 –.019 –.029 –.002 –.002 1.96 –.165 1978 to 2002 ** ** ** ** ** ** ** 1 Yes Yes 451 Nonproduction Workers .016 .031 .041 .005 .132 .501 11,269 (.009) (.006) (.009) (.008) (.007) (.009) (.022) (.009) (.026) (.065) –.019 –.019 –.020 –.003 1.94 –.169 1978 to 2002 ∆ Workers’ Share in Industry’s Wage Bill Wage Share in Industry’s ∆ Workers’ ( t –1) ( t –1) ( t –1) ( t –1) ( t –1) ( t –1) p < .05 (two-tailed test). 2 Class Model . Computer and Income Distribution in Four-Digit Manufacturing Industries, Unstandardized Coefficients from Single Equation ECM; 4 . Computer and Income Distribution in Four-Digit Table Income Share Share and in Skilled Unskilled Workers’ Wage-Bill Are Annual Change in Skilled Workers’ Dependent Variables Dependent Variable ∆ Computer investments Computer investments N of Industries Union density ∆ Union density ∆ Unemployment Unemployment ∆ Capital concentration Capital concentration ∆ Import penetration Import penetration N Constant Industry dummies R Modified DW entries are OLS estimates. Robust standard errors in parentheses heteroskedasticity and share. Table Note: Each column represents a pooled regression of changes in labor’s autocorrelation consistent. Estimates are weighted by mean industry share of total value added over the years. ∆ indicates annual change in variable. ** Recession (dummy) Dependent variable Years

380 ** ** ** 6 43 860 .082 .068 .252 Yes Yes .169 (.053) (.020) (.079) (.027) (.079) (.067) (.213) (.042) –.045 –.081 –.076 –.043 –.277 1.98 1988 to 2007 College Workers ** 5 43 1,247 .023 .065 .150 Yes Yes .141 (.142) (.022) (.047) (.024) (.099) (.088) (.260) (.068) –.184 –.055 –.008 –.123 –.235 2.02 1969 to 1997 Industry’s Income Industry’s ** ** ** ** ∆ Workers’ Share in ∆ Workers’ 4 43 860 .077 .114 .134 .017 Yes Yes .207 (.034) (.014) (.076) (.072) (.043) (.142) (.047) (.031) –.056 –.050 –.092 –.299 2.28 1988 to 2007 ** ** High School Workers 3 43 1,247 .134 .005 .078 .088 .419 Yes Yes .116 (.144) (.067) (.016) (.081) (.033) (.257) (.080) (.045) –.008 –.004 –.191 2.23 1969 to 1997 ** ** ** ** 2 43 860 .140 .160 .052 Yes Yes .236 (.071) (.059) (.179) (.120) (.075) (.545) (.089) (.088) –.240 –.069 –.282 –.357 –.415 2.04 1988 to 2007 College Workers ** ** ** ** ** ** ∆ Workers’ Share in ∆ Workers’ Industry’s Wage Bill Wage Industry’s 1 43 1,247 .123 .073 .081 .126 Yes Yes .132 (.059) (.026) (.077) (.053) (.038) (.349) (.050) (.041) –.244 –.164 –.089 –.241 2.15 1969 to 1997 ( t –1) ( t –1) ( t –1) ( t –1) p < .05 (two-tailed test). 2 Years . Computer and Income Distribution in Two-Digit Private Industries, Unstandardized Coefficients from Single Equation ECM; Dependent 5 . Computer and Income Distribution in Two-Digit Table Income Share Share and in Skilled Unskilled Workers’ Wage-Bill Are Annual Change in Skilled Workers’ Variables Dependent Variable Computer investments ∆ Computer investments Model N of Industries ∆ Union density Union density ∆ Unemployment Unemployment Recession (dummy) Dependent variable Constant Industry dummies R N Modified DW Class Each column represents a pooled regression of changes in labor’s share. Table entries are OLS estimates. Robust standard errors in parentheses heteroskedasticity and share. Table Note : Each column represents a pooled regression of changes in labor’s autocorrelation consistent. Estimates are weighted by mean industry share of total value added over the years. ∆ indicates annual change in variable. **

381 382 American Sociological Review 78(3) for computerization as well, which is the not received much attention in the labor main counterargument for the decline in movement literature and thus require further labor’s share. The empirical results strongly investigation. First, the computer revolution’s support the power relations thesis and reveal strengthening of management control may that the decline in unionization, rise in unem- have empowered employers and manage- ployment, and importation of goods from ment, allowing them to use more legal and less-developed countries all curbed workers’ illegal anti-union tactics designed to intensely bargaining power over the past decades and monitor and punish union activity. Second, led to a significant decline in labor’s share. In computer technology is linked to skill polari- particular, waning unionization, which led to zation in the workforce, which may have the erosion of rank-and-file workers’ bargain- undermined workers’ solidarity, thereby ing power, was the main force behind the reducing the likelihood of working-class decline in labor’s share. cohesion and collective action. Third, given all the evidence presented The arguments and findings of this study here, the class-biased technological change also contribute to the debate over the causes hypothesis appears to be a fruitful contribu- of rising earnings inequality in the past 30 tion for explaining inequality dynamics over years, indicating that worker disempower- the past 30 years. Whereas most work in eco- ment is behind both the decline in labor’s nomics focuses on mechanisms that link tech- share and the rise in earnings inequality. The nology use to workers and physical capital rise in earnings inequality in the United States productivity and hence to their relative and other countries is usually attributed to income shares, I argue for additional mecha- two sets of factors: most researchers view nisms that link technology use to classes’ skill-biased technological change—the com- positional power in the labor process and puterization of many workplaces that favors hence to their relative income shares. Specifi- high-skilled workers—as the main cause of cally, I contend that computer-based technol- rising wage inequality in the United States ogies embody essential characteristics that and relegate institutional factors, such as favor capitalists (and high-skilled workers), declining unionization, to a secondary role. In while eroding most rank-and-file workers’ a recent study, Western and Rosenfeld (2011) organizational power. Consequently, comput- present evidence that counters this widely erization has reduced labor’s share indirectly held view. They show that organized labor’s through its role in reducing unionization. decline and the growing stratification of Results from the empirical analysis confirm wages by education, which are indicative of the CBTC thesis and demonstrate that com- the expected outcome of computerization puterization had little effect on labor’s share according to the SBTC, have the same explan- in industries where organized labor never had atory power for rising earnings inequality. much of a presence. The evidence presented here gives an even Computerization may have exacerbated clearer-cut answer to the inequality debate. union decline as a result of several mecha- By incorporating direct measures for com- nisms. Among them, I emphasized the one puter technologies and unionization at the documented by Milkman (1995) for auto industry level, this study reveals that organ- workers. Milkman shows that automation of ized labor’s decline is the main factor that led the production process downsized many to the decline in labor’s share. Moreover, unionized manufacturing jobs by utilizing computerization contributed to the decline in computer equipment for tasks previously per- labor’s share partly through its negative formed manually by blue-collar, mostly impact on unionization, supporting a class- unionized workers. I also identified two other biased technological change argument. For plausible mechanisms through which com- that reason, this study strongly advocates puterization degrades labor unions that have using industrial data, which make it possible Kristal 383 to directly measure computerization and evidence presented here suggests it is clearly unionization, and testing the CBTC thesis in capitalists who have rarely had it so good. the context of rising earnings inequality. More generally, the analysis casts light on a defining feature of the current state of Data Appendix inequality—capitalists have grabbed the Computerization in Two-Digit lion’s share of income growth over the past Industry Data three decades. From 1948 to 1973, the hourly compensation of a typical U.S. worker grew I used data on investments in fixed assets in tandem with productivity, indicating a rela- from the Bureau of Economic Analysis (BEA) tively equal social distribution of the fruits of Industry Economic Accounts to measure economic growth and productivity gains. The computer technology at the two-digit industry state of inequality dramatically shifted in the level between 1969 and 2007. Computer tech- past three decades. Although productivity nology in the BEA data is measured by real grew 80.4 percent between 1973 and 2011, investments in computers as a share of total expanding total income, average hourly com- investments (in millions of 2008 dollars). pensation, which includes the pay of CEOs, Under computers, I included investments in increased by only 39.2 percent and—even mainframe computers, personal computers, more strikingly—the median worker’s hourly direct access storage devices, computer termi- compensation grew by just 10.7 percent nals, computer storage devices, integrated (Mishel et al. 2012). So where did the income systems, and software. growth go? As this study shows, income growth occurred mainly in income that Computerization and Capital accrues to owning capital—profits, interest, Concentration in Four-Digit dividends, and rent. As a result, workers have experienced a large and persistent reduction Manufacturing Industry Data in their share of national income. I relied on data collected by the U.S. Census Capitalists, however, are not the only Bureau on four-digit manufacturing industries social stratum that grabbed a handful of to examine possible explanations for changes income. Some workers did, too, in particular in labor’s share within U.S. manufacturing. the “working rich.” Using individual tax My main data sources were the Census returns data, Piketty and Saez (2003) docu- Bureau’s Annual Survey of Manufactures ment a dramatic growth in the top shares of (ASM) and Census of Manufactures (CM). income and wages since the 1980s. They also The ASM is a sample of about 60,000 manu- show that the dramatic growth in top income facturing establishments, carried out by the shares was primarily due to a surge in wages Census Bureau. The sample is drawn from the at the very top of the wage distribution, with CM, which is performed every five years and little growth of capital incomes held by these designed to collect data on all manufacturing rich individuals. This may suggest that these establishments in the country. The ASM col- new high-income earners, mainly CEOs and lects data on total employment, total payroll, workers employed in the finance sector, have value added, production worker wages, and not yet had time to accumulate substantial expenditures on new capital investment. The fortunes that yield capital income and there- information is reported for four-digit manu- fore have gained only slightly from the surge facturing industries. in capital income. Hence, although most In a joint effort by the National Bureau of research on rising inequality focuses on rising Economic Research (NBER) and the U.S. earnings inequality among workers and the Census Bureau’s Center for Economic Stud- rise in the top shares of income, “that may ies (CES), many of the ASM variables were be yesterday’s story” (Krugman 2012). The combined into one dataset covering the period 384 American Sociological Review 78(3)

1958 to 2005. I used the NBER dataset and month to a quarter of the sample (the outgoing added several relevant variables that are not rotation groups). Union membership is mea- part of the NBER database, based on hard sured in the standard way by the ratio of num- copies of the ASM and CM. Overall, my data- ber of union members to the total industry base contains annual information on 451 workforce. Union membership figures have four-digit manufacturing industries for the been compiled for all employed wage period 1978 through 2002. The industries are and salary workers, age 16 years and over. Not those defined in the 1987 SIC, and they cover included are employed 14- to 15-year-olds, the entire manufacturing sector. self-employed workers, and a small number of An indicator for computer technology in unpaid workers. the specified manufacturing industries is sim- All information on union membership pre- ilar to the one I used for the entire private sented so far was taken from the CPS. How- sector. The ASM asked about computer ever, these surveys did not collect information investments (“new capital expenditures on on union membership back to the late 1950s; computers and peripheral data processing to extend the series back to those years, we equipment”) in economic census years (1977, must rely on data reported by labor unions 1982, 1987, and 1992). This question was not every two years to the federal government asked in 1997 but was asked again in 2002. I and published by the Bureau of Labor Statis- imputed missing data using linear interpola- tics (BLS) in the Directory of National Unions tion. The U.S. Bureau of the Census, as part and Employee Associations. Because this is a of the periodic CM, also compiles aggregate biennial survey, data at the industry level are concentration figures for this sector. These available only for every second year, and val- data are based solely on the domestic opera- ues for intervening years must be interpo- tions of firms in the manufacturing sector. lated. These reports include information on Concentration is scored between zero and union membership by industry for only 17 one, calculated by the CM as the ratio of sales (1958 to 1964) and 28 (1968 to 1978) nonag- by the four largest firms to the total volume of ricultural private industries. I therefore sales in each industry. These data are avail­ imputed the aggregate information to the able for the economic census years (plus a more detailed industries, assuming the disag- few additional years) beginning in 1947 and gregate industries were relatively similar in continuing through the most recently avail­ terms of union density. The resulting series able census in 2007. for unionization are likely quite accurate for the 1970s but substantially less so for the preceding years. The Directory series also has Unionization a number of drawbacks as reliable informa- I collected data from the CPS for unionization tion on union membership by industrial sec- at the three-digit industry level. The first time tor. Most important, unions are thought to a union status question was asked of private inflate their membership figures to present a and public sector workers was in the March slightly exaggerated impression of their size. 1971 survey. The CPS began collecting indi- Unemployed and retired members who have vidual union membership information on a stopped paying their are often regular basis in May 1973. From 1973 to 1980, kept on the books. I adjusted the membership the May CPS administered union questions to data published every two years in the Direc- the full monthly samples (all rotation groups); tory by an estimate of the ratio of union mem- only a quarter of the sample was asked union bership in the CPS to union membership status questions in May 1981. There were no reported in the Directory. The ratios I used to union questions in the 1982 CPS. Beginning in adjust the Directory membership figures were January 1983, the CPS began asking union a simple average of the numbers for 1970, membership and coverage questions each 1972, 1974, 1976, and 1978. ** ** ** ** .000 .000 .000

Import –2.107 Penetration ** ** ** ** .000 .000 .000 Capital –2.161 Concentration ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** .000 .000 .000 .000 .000 .000 .001 .031 .000 .000 .002 .000 –2.365 –1.769 –2.059 Unemployment ** ** ** ** ** ** ** ** ** ** ** ** ** .000 .000 .000 .000 .000 .000 .002 .254 .000 .000 .000 .000 Union Density –2.500 –1.572 –2.189 ** ** ** ** ** ** ** ** ** ** ** ** ** .000 .000 .000 .000 .000 .000 .002 .362 .001 .000 .014 .003 –2.144 –1.535 –1.942 Computer Investments b ** ** ** ** ** ** a d .000 .000 .000 .000 .000 .000 Share c Labor’s Labor’s e Manufacturing industries Manufacturing industries Manufacturing industries Manufacturing industries Manufacturing industries Private industries Private industries Private industries Private industries Private industries Industries Private industries Private industries Private industries Private industries Private industries p < .05 (two-tailed test). The residual from a simple bivariate regression of Y on X. null hypothesis is that all panels contain unit roots in the residual. The null hypothesis is that all panels contain unit roots in the first-difference form. The null hypothesis is that panels contain unit roots in the first-difference form. The residual from a simple bivariate regression of Y on X. null hypothesis is that panels contain unit roots in the residual. t -bar is the mean of industry-specific Dickey-Fuller tests. The Levin-Lin-Chu (2002) test assumes that all panels have the same autoregressive parameter. The Im-Pesaran-Shin (2003) test relaxes the assumption and Note : The Levin-Lin-Chu (2002) test assumes that all panels have the same autoregressive parameter. The alternative hypothesis in the Im-Pesaran-Shin test is that fraction of panels are stationary allows each panel to have its own autoregressive parameter. is nonzero. p value from Im-Pesaran-Shin test for unit-root in first-differences p value from Im-Pesaran-Shin test for unit-root in the residual t -bar from Im-Pesaran-Shin test for unit-root in the residual p value from Levin-Lin-Chu test for unit-root in the residual 1978–02 1978–02 1978–02 1978–02 1978–02 1988–07 1988–07 1988–07 1988–07 1988–07 . Tests for Cointegration in the Time-Series Cross-Sectional Data for Cointegration in the Time-Series A1 . Tests Table Years p value from Levin-Lin-Chu test for unit-root in first-differences 1969–97 1969–97 1969–97 1969–97 1969–97 a b c d e **

385 386 American Sociological Review 78(3)

Import Penetration than those of net income; the latter are based on net incomes of firms whose allowance for depreciation is more or less arbitrary. I used import data by industry and country 2. Realized capital gains, which became key compo- from Schott (2010) to construct a measure of nents of top executive remuneration in the 1990s the share of imports in each industry originat- (Piketty and Saez 2003), are generally not counted in ing in low-wage countries (import penetra- national income accounts and therefore are not part tion). I classified a country as low-wage in of either labor’s compensation or capitalists’ profits. 3. I treat financialization as an integral part of overall year t if its per capita GDP was less than 20 capital income and consider its growing share rela- percent of U.S. per capita GDP (data on coun- tive to labor income to be determined by the ero- tries’ per capita GDP are from the Penn World sion in workers’ positional power. This is in some Table). This cutoff captures an average of 80 way different from the approach taken by Lin and countries per year. The list of countries clas- Tomaskovic-Devey (forthcoming), who examine financialization as an independent variable that sified as low wage includes China and India affects income inequality. as well as relatively small exporters such as 4. A measure of labor’s share that includes only Angola. I chose a 20 percent cutoff to classify employees’ compensation on the labor side (thereby countries as low wage because it represents treating income of the self-employed entirely as the world’s most labor-abundant cohort of capital income) is biased over time because it does not take into account the move from self-employ- countries and therefore the set of countries ment to wage and salary employment (Johnson most likely to have an effect on U.S. manu- 1954; Kravis 1959). To avoid this bias, the numera- facturing plants. Using import data by indus- tor of labor’s share here is labor income of employ- try and country from Schott (2010), I ees and self-employed. computed import penetration for 393 of 459 5. The notoriously high salaries on Wall Street may have pulled labor’s share in FIRE upward over the four-digit 1987 SIC manufacturing industries 1990s. Even so, the finance sector has the lowest lev- between 1978 and 2002. These 393 industries els of labor’s share, mainly due to the highly profit- encompass 87 percent of manufacturing able economic activity in real estate and banking. employment and 93 percent of manufacturing 6. Common examples of new computer technology value. in manufacturing include computerized industrial robots, automated inventory and parts storage, and computers for monitoring, analyzing, and control- Funding ling industrial processes. Prominent applications of computer technology in services include common This research was supported by The Israel Science desktop software, data entry and transactions pro- Foundation (ISF). cessing systems, automated teller machines, and inventory management devices and software. Acknowledgments 7. A more critical view holds that in many workplaces new technology is meant to replace workers, not I started this project in 2010 when I was Elfenworks transform work in a way that increases productivity postdoctoral scholar at the Stanford Center for the Study (Noble 1978). of Poverty and Inequality, where I benefited from the 8. Two examples given by Skott (2010) clearly illus- opportunity to present earlier versions of this article at trate how computer technology marks a shift in the the Center’s conferences and seminars. I am especially relationship between owners and workers. First, a grateful to David Grusky for insightful discussions and new “black box” installed in trucks gives a truck’s valuable comments. I also thank Yinon Cohen, the ASR owner the ability to monitor driver performance by editors, and anonymous reviewers for their comments on providing full information on the vehicle for every earlier drafts. Earlier versions of this article were pre- second. In retail, computer technology provides a sented at the spring meeting of the ISA Research way of ensuring that the money collected from cus- Committee on Social Stratification and Mobility (2010) tomers matches the money a clerk hands over to the and the 106th Annual Meeting of the American employer, thereby affecting workers’ and employ- Sociological Association (2011). ers’ relative power. 9. The CPS series are available only at the three-digit level, which made me assume that the three-digit Notes unionization and unemployment information is 1. The main reason for including capital depreciation is relevant to the distribution process at the four- that statistics of gross income are much more reliable digit level in manufacturing industries. To be sure, Kristal 387

matching three-digit level data to four-digit manu- Bernard, Andrew B., J. Jensen, and Peter K. facturing data creates a measurement error and Schott. 2006. “Survival of the Best Fit: Exposure to decreases the likelihood of finding a significant Low-Wage Countries and the (Uneven) Growth of impact of unionism and unemployment. U.S. Manufacturing Plants.” Journal of International 10. ECMs are more familiar as the general Auto-regressive Economics 68:219–37. Distributed Lag Model (ADL), also called a Partial- Blanchard, Olivier J. 1997. “The Medium Run.” Brook- Adjustment Differential Equation Model (PADEM). ings Papers on Economic Activity 2:89–141. 11. Results from unit root tests for panel data show that Bronfenbrenner, Kate. 2009. “No Holds Barred: The in all models we can reject the null hypothesis that Intensification of Employer Opposition to Organiz- panels contain unit roots in the residual. ing.” Institute, Briefing Paper No. 12. Stationary tests confirm that the variables are non- 235. stationary. Yet these variables are bounded between Burawoy, Michael. 1985. The Politics of Production. 0 and 100 percent and therefore do not perform : Verso. like integrated variables, which tend to wander far Bureau of Economic Analysis (BEA). N.d. Industry from their means so the variance of the observations Economic Accounts. Retrieved September 22, 2009 grows larger and larger over time. (http://www.bea.gov/industry/index.htm). 13. To deal with the potential endogeneity of computer Bureau of Economic Analysis (BEA). N.d. National technology (i.e., the ICT revolution was a response Income and Product Accounts Tables. Retrieved Sep- to the profit squeeze in the 1960s), I estimated mod- tember 22, 2009 (http://www.bea.gov/national/). els with a three-year lagged value for computer Burris, Beverly H. 1993. Technocracy at Work. Albany: technology. Results were substantially the same as State University of New York Press. those for one-year lagged value. Burris, Beverly H. 1998. “Computerization of the Work- 14. That the net effect of computerization raised labor’s place.” Annual Review of Sociology 24:141–57. share in manufacturing in the short-run might be Crowley, Martha, Daniel Tope, Lindsey Joyce Chamber- due to a short-term rise in skilled workers’ wages, lain, and Randy Hodson. 2010. “Neo-Taylorism at which was offset in the long-run by a larger rise in Work: Occupational Change in the Post-Fordist Era.” capitalists’ profits. Social Problems 57:421–47. De Boef, Suzanna and Luke Keele. 2008. “Taking Time Seriously.” American Journal of Political Science References 52:184–200. Acemoglu, Daron. 1998. “Why Do New Technologies Dew-Becker, Ian and Robert J. Gordon. 2005. “Where Complement Skills? Directed Technical Change and Did the Productivity Growth Go? Inflation Dynamics Wage Inequality.” Quarterly Journal of Economics and the Distribution of Income.” Brookings Papers 113:1055–1089. on Economic Activity 2:67–127. Acemoglu, Daron. 2002. “Directed Technical Change.” DiMaggio, Paul and Bart Bonikowski. 2008. “Make Review of Economic Studies 69:781–809. Money Surfing the Web? The Impact of Internet Use Acemoglu, Daron. 2003. “Labor- and Capital-Augment- on the Earnings of U.S. Workers.” American Socio- ing Technical Change.” Journal of European Eco- logical Review 73:227–50. nomic Association 1:1–37. Engle, Robert F. and Clive W. J. Granger. 1987. “Co-inte- Alderson, Arthur S. and Francois Nielsen. 2002. “Glo- gration and Error Correction: Representation, Estima- balization and the Great U-Turn: Income Inequality tion, and Testing.” Econometrica 55:251–76. Trends in 16 OECD Countries.” American Journal of Epstein, Gerald A. and Arjun Jayadev. 2005. “The Rise of Sociology 107:1244–99. Rentier Incomes in OECD Countries: Financialization, Autor, David, Lawrence F. Katz, and Alan Krueger. 1998. Central Policy and Labor Solidarity.” Pp. 46–74 “Computing Inequality: Have Computers Changed in Financialization and the World Economy, edited by the Labor Market?” Quarterly Journal of Economics G. A. Epstein. Northampton, MA: Edward Elgar. 113:1169–1214. Esping-Andersen, Gøsta. 1985. Politics against Markets. Beck, Nathaniel and Jonathan N. Katz. 2011. “Modeling Princeton, NJ: Princeton University Press. Dynamics in Time-Series–Cross-Section Political Farber, Henry S. and Bruce Western. 2001. “Account- Economy Data.” Annual Review of Political Science ing for the Decline of Unions in the Private Sector, 14:331–52. 1973–1998.” Journal of Labor Research 22:459–85. Bentolila, Samuel and Gilles Saint-Paul. 2003. “Explain- Gollin, Douglas. 2002. “Getting Income Shares Right.” ing Movements in the Labor Share.” Contributions to Journal of Political Economy 110:458–74. Macroeconomics 3:Article 9. Grusky, David B. and Jesper B. Sørensen. 1998. “Can Berman, Eli, John Bound, and Zvi Griliches. 1994. Class Analysis Be Salvaged?” American Journal of “Changes in the Demand for Skilled Labor within Sociology 103:1187–1234. U.S. Manufacturing: Evidence from the Annual Sur- Halaby, Charles N. 2004. “Panel Models in Sociological vey of Manufactures.” Quarterly Journal of Econom- Research: Theory into Practice.” Annual Review of ics 109:367–97. Sociology 30:507–544. 388 American Sociological Review 78(3)

Henley, Andrew G. 1987. “Trades Unions, Market Con- Krueger, Alan. 1993. “How Computers Have Changed centration, and Income Distribution in United States the Wage Structure: Evidence from Microdata, 1984– Manufacturing Industry.” International Journal of 1989.” Quarterly Journal of Economics 108:33–60. Industrial Organization 5:193–210. Krueger, Alan. 1999. “Measuring Labor’s Share.” Ameri- Hicks, Alexander. 1999. Social and can Economic Review 89:45–51. Capitalism: A Century of Income Security Politics. Krugman, Paul. 2012. “Robots and Robber Barons.” New Ithaca, NY: Cornell University Press. York Times, December 9th. Im, Kyung S., Hashem M. Pesaran, and Yongcheol Shin. Levin, Andrew, Chien-Fu Lin, and Chia-Shang James 2003. “Testing for Unit Roots in Heterogeneous Pan- Chu. 2002. “Unit Root Tests in Panel Data: Asymp- els.” Journal of Econometrics 115:53–74. totic and Finite-Sample .” Journal of Jacobs, David. 1988. “Corporate Economic Power and Econometrics 108:1–24. the State: A Longitudinal Assessment of Two Expla- Lin, Ken-Hou and Donald Tomaskovic-Devey. Forth- nations.” American Journal of Sociology 93:852–81. coming. “Financialization and U.S. Income Inequal- Jacobs, David and Marc Dixon. 2006. “The Politics of ity, 1970–2008.” American Journal of Sociology. Labor-Management Relations: Detecting the Condi- Macpherson, David A. 1990. “Trade Unions and Labor’s tions That Affect Changes in Right-to-Work Laws.” Share in U.S. Manufacturing Industries.” Interna- Social Problems 53:118–37. tional Journal of Industrial Organization 8:143–51. Johnson, D. Gale. 1954. “The Functional Distribution of Milkman, Ruth. 1995. Farewell to the Factory: Auto Income in the United States, 1850–1952.” Review of Workers in the Late Twentieth Century. Berkeley: Economics and Statistics 35:175–82. Press. Kalecki, Michal. 1938. “The Determinants of Distribu- Mishel, Lawrence, Josh Bivens, Elise Gould, and Heidi Shi- tion of the National Income.” Econometrica 6:97– erholz. 2012. The State of , 12th ed. 112. Ithaca, NY: Economic Policy Institute and ILR Press. Kalleberg, Arne L. and Mark E. Van Buren. 1996. “Is Moller, Stephanie, Arthur S. Alderson, and Francois Bigger Better? Explaining the Relationship between Nielsen. 2009. “Changing Patterns of Income Organization Size and Job Rewards.” American Soci- Inequality in U.S. Counties, 1970–2000.” American ological Review 61:47–66. Journal of Sociology 114:1037–1101. Kalleberg, Arne L., Michael Wallace, and Lawrence E. Raf- Moller, Stephanie, Evelyne Huber, John D. Stephens, falovich. 1984. “Accounting for Labor’s Share: Class David Bradley, and Francois Nielsen. 2003. “Deter- and Income Distribution in the Printing Industry.” minants of Relative Poverty in Advanced Capital- Industrial and Labor Relations Review 37:386–402. ist .” American Sociological Review Keynes, John M. 1939. “Relative Movements of Real 68:22–51. Wages and Output.” Economic Journal 49:34–51. Noble, David F. 1978. “Social Choice in Machine Design: Korpi, Walter. 1983. The Democratic Class Struggle. The Case of Automatically Controlled Machine London: Routledge and Kegan Paul. Tools, and a Challenge for Labor.” Politics and Soci- Korpi, Walter. 2002. “The Great Trough in Unemploy- ety 8:313–47. ment: A Long-Term View of Unemployment, Infla- Piketty, Thomas and Emmanuel Saez. 2003. “Income tion, Strikes, and the Profit/Wage Ratio.” Politics and Inequality in the United States, 1913–1998.” Quar- Society 30:365–426. terly Journal of Economics 118:1–39. Kravis, Irving. 1959. “Relative Income Shares in Fact Raffalovich, Lawrence E., Kevin T. Leicht, and Michael and Theory.” American Economic Review 49:917–49. Wallace. 1992. “Macroeconomic Structure and Krippner, R. Greta. 2005. “The Financialization of Labor’s Share of Income: United States, 1950 to the American Economy.” Socio-Economic Review 1980.” American Sociological Review 57:243–58. 3:173–208. Rubin, Beth A. 1986. “Trade Union Organization, Labor Kristal, Tali. 2010. “Good Times, Bad Times: Post- Militancy, and Labor’s Share of National Income in war Labor’s Share of National Income in Capital- the United States, 1949–1978.” Research in Social ist Democracies.” American Sociological Review Stratification and Mobility 5:223–42. 75:729–63. Sakamoto, Arthur and Changhwan Kim. 2010. “Is Rising Kristal, Tali. 2013. “Slicing the Pie: State Policy, Class Earnings Inequality Associated With Increased Exploi- Organization, Class Integration, and Labor’s Share of tation? Evidence for U.S. Manufacturing Industries, Israeli National Income.” Social Problems 60:100– 1971–1996.” Sociological Perspectives 53:19–43. 127. Schott, Peter K. 2010. U.S. Manufacturing Exports and Kristal, Tali and Yinon Cohen. 2007. “Decentralization Imports Data by SIC or NAICS Category and Part- of Collective Wage Agreements and Rising Wage ner Country, 1972 to 2005. Retrieved May 12, 2010 Inequality in Israel.” 46:613–35. (http://faculty.som.yale.edu/peterschott/sub_interna- Kristal, Tali and Yinon Cohen. 2012. The Causes of Ris- tional.htm). ing Inequality: What do Computerization and Fading Skott, Peter. 2010. “Power Biased Technical Change, Pay-Setting Institutions Do? Unpublished manu- Endogenous Mismatch and Institutional Change.” script. Department of Sociology, University of Haifa. New School Economic Review 4:50–64. Kristal 389

Solow, Robert M. 1958. “A Skeptical Note on the Con- and the Restructuring of Work at the Turn of the Cen- stancy of Relative Shares.” American Economic tury.” Pp. 101–133 in Sourcebook of Labor Markets: Review 48:618–31. Evolving Structures and Processes, edited by I. Berg Southworth, Caleb and Judith Stepan-Norris. 2009. and A. Kalleberg. New York: Plenum Press. “American Trade Unions and Data Limitations: A Wallace, Michael, Kevin T. Leicht, and Lawrence E. Raf- New Agenda for Labor Studies.” Annual Review of falovich. 1999. “Unions, Strikes, and Labor’s Share Sociology 35:297–320. of Income: A Quarterly Analysis of the United States, Stearns, Linda B. and Kenneth D. Allan. 1996. “Eco- 1949–1992.” Social Science Research 28:265–88. nomic Behavior in Institutional Environments: The Wallace, Michael, Beth A. Rubin, and Brian T. Smith. Corporate Merger Wave of the 1980s.” American 1988. “American Labor Law: Its Impact on Working Sociological Review 61:699–718. Class Militancy, 1901–1980.” Social Science History Stepan-Norris, Judith and Maurice Zeitlin. 2003. Left 12:1–29. Out: Reds and America’s Industrial Unions. New Weeden, Kim A. 2002. “Why Do Some Occupations York: Cambridge University Press. Pay More than Others? Social Closure and Earnings Stephens, John D. 1979. The Transition from Capitalism Inequality in the United States.” American Journal of to . London: Macmillan. Sociology 108:55–101. Tomaskovic-Devey, Donald and Ken-Hou Lin. 2011. Western, Bruce and Jake Rosenfeld. 2011. “Unions, “Income Dynamics, Economic Rents, and the Finan- Norms, and the Rise in U.S. Wage Inequality.” Ameri- cialization of the U.S. Economy.” American Socio- can Sociological Review 76:513–37. logical Review 76:538–59. White, Lawrence J. 2002. “Trends in Aggregate Concen- Tope, Daniel and David Jacobs. 2009. “The Politics of tration in the United States.” Journal of Economic Union Decline: The Contingent Determinants of Perspectives 16:137–60. Union Recognition and Victories.” Ameri- Wood, Adrian. 1994. North–South Trade, Employment can Sociological Review 74:842–64. and Inequality. , UK: . Vallas, Steven P. 1993. Power in the Workplace. Albany: Wright, Erik Olin. 1979. Class Structure and Income State University of New York Press. Determination. New York: Academic Press. Vallas, Steven P. and John P. Beck. 1996. “The Trans- Wright, Erik Olin and Luca Perrone. 1977. “Marxist formation of Work Revisited: The Limits of Flex- Class Categories and Income Inequality.” American ibility in American Manufacturing.” Social Problems Sociological Review 42:32–55. 43:339–61. Zuboff, Shoshana. 1988. In the Age of the Smart Machine. Volscho, Thomas W. and Nathan J. Kelly. 2012. “The New York: Basic Books. Rise of the Super-Rich: Power Resources, Taxes, Financial Markets, and the Dynamics of the Top 1 Percent, 1949 to 2008.” American Sociological Review 77:679–99. Tali Kristal is an Assistant Professor of Sociology at Voss, Kim and Rachel Sherman. 2000. “Breaking the Iron University of Haifa. Her research examines pay-setting Law of Oligarchy: Union Revitalization in the Ameri- institutions, politics, and trends in income inequality in can Labor Movement.” American Journal of Sociol- Israel, the United States, and other rich countries. Her ogy 106:303–349. current work investigates the impact of changes in pay- Wallace, Michael and David Brady. 2001. “The Next setting institutions and computerization on rising earn- Long Swing: Spatialization, Technocratic Control, ings and compensation inequality in the United States.