Within-Group Earnings Inequality in Cross-National Perspective

Within-Group Earnings Inequality in Cross-National Perspective Abstract In this research I assess within-group inequality—earnings inequality occurring among otherwise similar individuals based on observed characteristics—in a cross-national comparative perspective. While scholarly interest in the within-group portion of inequality has grown over the past 25 years, virtually all studies focus on the US case. The current research shifts focus by assessing within-group inequality in a cross-national comparative study. I do so by constructing a unique data set of country-level measures of within- and between-group inequality for annual market earnings using Luxembourg Income Study (LIS) microdata from 1.36 million full-time prime-age male and female workers nested in 143 country-years, drawn from 28 countries spanning 40 years. I then document and describe basic between-country and longitudinal trends in the relationship between total inequality and within-group inequality. I find that in nearly all countries in the LIS, within-group inequality is the primary driver of levels and trends in inequality. As inequality increases, so too does the relative importance of within-group inequality. However, substantial cross-national heterogeneity based in labour market institutions and employment protection legislation is found. Theoretical and substantive implications are discussed. The nature of individual earnings inequality growth provides a fundamental challenge to sociological thinking. Many high-income countries have experienced rising inequality in recent decades (Alderson and Nielsen, 2002; OECD, 2015). In the United States, inequality has not only risen in absolute levels but also changed regarding important distributional properties. Most earnings inequality growth since the 1970s has occurred among workers who are otherwise similar along sociodemographic, human capital, and occupational characteristics typically studied by sociologists, that is, within-group inequality (Levy and Murnane, 1992; McCall, 2000; Lemieux, 2006; Autor, Katz and Kearney, 2008, Leicht, 2008; Western and Bloome, 2009; Mouw and Kalleberg, 2010; Western and Rosenfeld, 2011; Cheng, 2014; Zhou, 2014; MacLean and Kleykamp, 2016; Xie, Killewald and Near, 2016; Leicht, 2016; Liao, 2016; VanHeuvelen, 2018). While scholars have increasingly placed within-group inequality at the centre of theoretical and analytical attention, many questions about the nature of this inequality dimension remain unanswered. Research on within-group inequality overwhelmingly focuses on the US case (but see Williams, 2012). While within-group inequality has played a central role in American inequality trends and been used to make theoretical sense of the contemporary upswing of inequality, it remains unknown whether its importance is a unique feature of the American labour market, among similar Liberal welfare regimes (Esping-Andersen, 1999), or whether within-group inequality’s contribution to inequality growth applies generally across high-income countries. Cross-national comparison provides an ideal opportunity to better understand the relationship between within-group inequality and total inequality change. The macrocomparative stratification literature illustrates the benefits of assessing inequality across countries with diverse inequality legacies, labour market institutions, policy systems, and sociodemographic compositions (Bollen, Entwisle and Alderson, 1993; Alderson and Nielsen, 2002; Bradley et al., 2003; Kenworthy, 2004; Thelen, 2014; Kollmeyer, 2015). I apply these insights to the study of within-group inequality and ask the following: How general is the importance of within-group inequality to inequality change, and how do previous explanations of skill development and deinstitutionalization fare in a cross-national context? This study uses 10 waves of Luxembourg Income Study (Luxembourg Income Study Database, 2017) microdata from 768,549 full-time prime-age male and 591,008 full-time prime-age female workers to construct a unique data set of 285 macrolevel within-group inequality distributions in an unbalanced sample of 28 countries spanning 40 years. Within-group inequality is computed from identical variance function regression (VFR) models on identical sets of microlevel grouping variables used in previous studies of market earnings, yielding the most extensive collection of country-level within-group inequality observations used in a study to date. Two main conclusions are drawn. First, total inequality levels, and inequality growth, are primarily due to within-group inequality. Second, despite the general importance of within-group inequality, there exists substantial heterogeneity across institutional contexts. Liberal and East European regimes have higher levels and contributions of within-group inequality compared with continental European and Nordic countries. Such variation is largely attributable to labour market institutional and policy arrangements, but not human capital differences. This study has important implications. First, it clarifies the relationship between within-group inequality and individual market earnings inequality growth, revealing that most earnings inequality change in high-income countries has occurred through the within-group component. Second, it sheds light on the relationship between total inequality and within-group inequality, showing that institutional and policy differences produce sizable variation in the levels and relative importance of within-group inequality. Third, it highlights the heterogeneity of the sources of inequality across country contexts, even those with the same absolute level of inequality. Two countries with similar levels of total inequality may have substantially different types of inequality, which can be revealed by focusing on within-group inequality. Background Within-group inequality (henceforth WGI)—the variance of individual earnings net of sociodemographic, human capital, and occupational characteristics that sociologists and economists typically study—has been the subject of scholarly attention for over 25 years (Levy and Murnane, 1992; Juhn, Murphy and Pierce, 1993; Card and Dinardo, 2002; Lemieux, 2006; Autor, Katz and Kearney, 2008; Western and Bloome, 2009; Zhou, 2014; VanHeuvelen, 2018). Suppose one could ‘level’ average pay differences between those attaining secondary and tertiary levels of education, workers of different experience levels, and across different industries and occupational categories, so that averages across these social categories were indistinguishable from zero. Inequality would necessarily decline. Yet beginning in the early 1990s, scholars observed that even under these conditions—and even when applied across all the usual suspects of worker and job characteristics observable to the analyst—upwards of ⅔ of inequality levels and 70% of inequality change over time would remain (Juhn, Murphy and Pierce, 1993; Autor, Katz and Kearney, 2008; Liu and Grusky, 2013; but see Mouw and Kalleberg, 2010).1 These patterns motivated the following question: Why would inequality grow independently of observable worker characteristics? The answer to this question has led to distinct interpretations of the nature of the contemporary upswing of inequality. Interpretations of empirical patterns and substantive meaning of WGI have been used to adjudicate across major theories of the inequality upswing.2 Explanations of WGI fall into three theoretical camps. First, some argue it reflects returns to skills and talents observable to employers, but not analysts (Juhn, Murphy and Pierce, 1993; Autor, Katz and Kearney, 2008; Goldin and Katz, 2008; Mouw and Kalleberg, 2010). Technological innovation has complemented the job duties of highly skilled workers, largely concentrated among professional and managerial occupations, allowing them to enhance productivity and manage increasingly complex organizations. WGI thus reflects growth in economic returns to highly skilled, highly educated workers in cognitively demanding occupations who have increased their productivity and, thus, pay. Similarly, experienced workers tend to have higher WGI due to uneven skill development, on-the-job training, and occupational mobility (Lemieux, 2006). Simply put, more educated and experienced workforces should have higher WGI. Second is deinstitutionalization, or the decline of institutional protection and stability of pay due to broadly shared changes in labour market policies, pay-setting institutions, and reduction of internal, firm-level job security, mobility, full-time employment, and benefits (McCall, 2000; Western et al., 2012; Western and Rosenfeld, 2011). In the United States, attention focuses on (1) declining internal labour markets and the fissuring of employees across an increasingly complex organizational landscape (Bidwell, 2011; Weil, 2014) and (2) falling union density and political power. Union decline, pronounced in the United States but broadly experienced across high-income countries (Visser and Checchi, 2009), has resulted in well-documented declines in wage attainment and increases in wage inequality and has also been shown to increase WGI (Western and Rosenfeld, 2011; Jaumotte and Buitron, 2015; Kristal and Cohen, 2016). This perspective conceptualizes WGI as an indicator of insecurity, or the risk of economic loss in the face of unpredictability (Western et al., 2012) and flexibility, the ease and ability of management to create insecurity via wage adjustments and job termination. In total, labour market institutional arrangements and policy systems that increase labour security and decrease flexibility should decrease WGI. Third is methodological. Some scholars explicitly assume WGI to indicate luck and free will (Jencks et al., 1972),3 random error (Cheng, 2014), or omitted variables, such as occupational differentiation (Mouw and Kalleberg, 2010).4 From this viewpoint, WGI is beyond the scope of analytical focus, or else indicates the need for a more precise microlevel model.5 Given its theoretical importance, sociologists have developed methodologies situating WGI in the centre of analysis (Western and Bloome, 2009; Cheng, 2014; Zhou, 2014; Liao, 2016), elaborated theories of WGI causes (McCall, 2000; Western et al., 2012; Western and Bloome, 2009; Cheng, 2014; Zhou, 2014), and called for further examination of dimensions of inequality beyond one-number measures of overall inequality such as the Gini coefficient and between-group gaps (Leicht, 2008, 2016). However, virtually all studies have restricted focus to a single-country study, predominantly the United States (but see Zhou, 2014 for the Chinese case and Williams, 2012 for the British case). Extending WGI research to a cross-national sample is useful for at least two reasons. First, analysis of WGI can reveal variation in the distributional properties of inequality. Many cross-national inequality studies implicitly assume that country-level inequality measures at the same level imply equivalent ‘types’ of inequality. Yet a focus on the relationship between total inequality and WGI can reveal meaningful variation across countries with similar levels of inequality. The substantive reality of inequality may vary markedly depending on how tightly wage setting is conducted in relation to human capital and occupational characteristics. Second, cross-national analyses have routinely yielded insights into basic features of inequality. For example, scholars have shown how tax and transfer policies lead to cross-national variation in redistribution (Gornick and Milanovic, 2015) and post-fisc inequality (Bradley et al., 2003). Economists have discovered substantial heterogeneity in historical inequality trajectories between English-speaking nations, continental Europe, and Nordics (Atkinson, Piketty and Saez, 2011). Similarly, sociologists and political scientists have documented the wide variation of social policies and labour market institutions that create norms and regulations for how wage setting and adjustments occur (Bradley et al., 2003; Kenworthy, 2004; Gallie, 2007; Barbieri, 2009). Cross-national research reveals general features of stratification processes not fully reducible to idiosyncratic features of a single country context (Blau and Kahn, 2005; Brady, Fullerton and Cross, 2010). Such analytical approaches have not been incorporated into the study of WGI. Yet to properly understand this component of inequality, one must situate it in a comparative perspective. The cross-national level provides a logical location to empirically assess WGI theories. For example, Blau and Kahn (2005) showed that the variation of test scores, a proxy for a workforce’s skill distribution, poorly explained between-country inequality differences, although others find returns to education and skill to be important across high-income countries (DiPrete, 2007; Goos and Manning, 2007). A skills return explanation suggests that countries with more educated and older workforces should have higher WGI. Regarding deinstitutionalization, Esping-Andersen’s regime typology provides a canonical account of the variation of market dynamics across country institutional contexts (1999). Liberal regimes—largely defined by residual states, means tested support, low labour protection, and fragmented bargaining—tend to have higher inequality than Conservative regimes—defined by male breadwinner insurance systems, occupational status distinctions, low mobility, more centralized wage setting coordination, and insider/outside cleavages—and Social Democratic regimes—defined by universal insurance systems, equal opportunity, broad systems of coordination, and low inequality (Esping-Andersen, 1999; Mandel and Shalev, 2009; Sachweh and Olafsdottir, 2012). Insofar as WGI represents returns to skill, insecurity, flexibility, and/or luck, then WGI should be more consequential in Liberal regimes than either Conservative or Social Democratic ones. Numerous theoretical traditions move beyond regime categorization, identifying specific methods of coordination, labour market institutional arrangements, or social policies that create different forms of wage setting, insecurity, flexibility, and, thus, inequality (Kenworthy, 2004; Gallie, 2007; Gebel and Giesecke, 2011; Streeck, 2011). Two such factors are relevant to WGI deinstitutionalization theories. First, countries vary in labour union membership, bargaining coverage, and centralization of wage setting coordination. In addition to the importance of union power discussed earlier (Western and Rosenfeld, 2011; Jaumotte and Buitron, 2015), centralized coordination provides standards of wage setting across firms, decreases interfirm wage setting via discretion and occupation mobility, and encourages industry- and firm-specific skill development, which in turn influences predictable wage attainment, employment tenure, and secure employment protection (Kenworthy, 2004; Gallie, 2007; Streeck, 2011; Thelen, 2014). Insofar as such institutional arrangements provide workers economic security, they should be negatively associated with WGI. Second, countries vary in their restrictions to labour flexibility. Long viewed as the culprit of Eurosclerosis (Barbieri, 2009), strict employment protection legislation (EPL) lowers the ability of firms to adjust wages in response to external economic shocks. Some argue for a blunt line dividing flexible, unequal, and high employment United States and inflexible, equal, high unemployment Europe. Recent studies identify flexibility as occurring across multiple dimensions (Barbieri, 2009). Countries differ in the broad or targeted application of flexibility across certain population segments. Particularly on continental Europe, changes to EPL largely targeted ‘outsider’, marginal groups, bifurcating the labour force into protected, secure, and egalitarian insiders and flexible and insecure outsiders (Gebel and Giesecke, 2011; Thelen, 2014). Thus, employment protection takes on two dimensions: the general extent of EPL reducing flexibility and contrasting EPL across contract types (Gebel and Giesecke, 2011; Barbieri and Cutuli, 2016). Greater flexibility in both cases should associate with higher WGI. Overall, cross-national variation of labour institutions and employment policies provides an ideal opportunity to test and extend deinstitutionalization theories. Data and Methods WGI is calculated from national surveys collected in the LIS, widely considered the gold standard of comparable cross-national income data. The LIS includes harmonized and nationally representative microdata on worker and job characteristics, including income, demographics, human capital, and occupation. The high quality of income measurements and rigorous harmonization of microdata undertaken by the LIS are crucial for this research: WGI can be computed on the same earnings measure using identical individual-level covariates across countries over a long period. I use all country surveys with available microdata that are either (1) high-income countries or (2) on the European continent. Samples Country samples are listed in Table 1. Following standard practice of WGI research, I conduct separate analyses by sex (Lemieux, 2006; Autor, Katz and Kearney, 2008; Western and Rosenfeld, 2011), which accounts for the different historical socioeconomic attainment trajectories and barriers of employment and wage attainment faced by men and women during the period studied (Esping-Andersen, 1999). I restrict samples to full-time workers to minimize potential bias introduced from cross-national differences in the frequency of part-time employment.6 Samples are restricted to prime-age workers aged 25–54. While more conservative than other studies using LIS microdata (Mandel and Shalev, 2009; Brady, Fullerton and Cross, 2010), this decision guards against potential confounding selection effects related to decommodification—such as educational training, family support, and retirement—which occurs unevenly between countries. Following similar studies, self-employed individuals are dropped (Autor, Katz and Kearney, 2008). Microlevel analyses are conducted using survey weights provided by the LIS. In total, samples include 768,549 male and 591,008 female workers. Microdata are nested in an unbalanced sample of 143 country-year inequality observations per sex, in 28 countries spanning 40 years.7 The country-year is the unit of analysis below. Table 1. Country samples Countrya  Years  Country observations  Male observations  Female observations  Austria (AT)  1994, 1997, 2000, 2004, 2013  5  7,183  3,324  Australia (AU)  1985, 1989, 2008, 2010  4  16,268  8,519  Belgium (BE)  1997, 2000  2  2,528  1,353  Canada (CA)  1987, 1991, 1994, 1997, 1998, 2000, 2004, 2007, 2010  9  81,391  64,549  Switzerland (CH)  2007, 2010, 2013  3  6,408  2,890  Czech Republic (CZ)  1996, 2004, 2007, 2010, 2013  5  21,912  19,675  Germany (DE)  1984, 1989, 1994, 2000, 2004, 2007, 2010, 2013  8  25,376  13,179  Denmark (DK)  2004, 2007, 2010, 2013  4  72,092  66,923  Estonia (EE)  2004, 2007, 2010, 2013  4  6,612  6,796  Spain (ES)  2000, 2004, 2007, 2010, 2013  5  18,670  12,829  Finland (FI)  1987, 1991, 1995, 2000, 2004, 2007, 2010, 2013  8  24,257  24,154  France (FR)  2005, 2010  2  4,076  3,023  Greece (GR)  1995, 2000, 2004, 2007, 2010, 2013  6  8,253  5,659  Hungary (HU)  1991, 1994, 1999, 2005  4  1,983  2,013  Ireland (IE)  1994, 1995, 2000, 2004, 2007, 2010  6  6,393  3,974  Israel (IL)  1979, 1986, 1992, 1997, 2001, 2005  6  10,756  5,951  Iceland (IS)  2004, 2007, 2010  3  3,521  2,904  Italy (IT)  2004, 2008, 2010  3  6,939  4,467  Lithuania (LT)  2010, 2013  2  2,570  2,932  Luxembourg (LU)  1997, 2000, 2004, 2007, 2010, 2013  6  10,122  4,885  Netherlands (NL)  1990, 2004, 2007, 2010, 2013  5  7,916  2,298  Poland (PL)  2007, 2010, 2013  3  39,525  34,317  Russia (RU)  2004, 2007, 2010, 2013  4  5,471  6,705  Slovenia (SI)  1999, 2004, 2007, 2010, 2012  5  8,395  8,303  Slovakia (SK)  2004, 2007, 2010, 2013  4  9,236  9,551  Taiwan (TW)  1981, 1986, 1991, 1995, 1997, 2000, 2005, 2007, 2010, 2013  10  71,221  47,701  United Kingdom (UK)  1986, 1999, 2004, 2007, 2010, 2013  6  38,208  24,726  United States (US)  1974, 1979, 1986, 1991, 1994, 1997, 2000, 2004, 2007, 2010, 2013  11  251,267  197,408  Total     143  768,549  591,008  Countrya  Years  Country observations  Male observations  Female observations  Austria (AT)  1994, 1997, 2000, 2004, 2013  5  7,183  3,324  Australia (AU)  1985, 1989, 2008, 2010  4  16,268  8,519  Belgium (BE)  1997, 2000  2  2,528  1,353  Canada (CA)  1987, 1991, 1994, 1997, 1998, 2000, 2004, 2007, 2010  9  81,391  64,549  Switzerland (CH)  2007, 2010, 2013  3  6,408  2,890  Czech Republic (CZ)  1996, 2004, 2007, 2010, 2013  5  21,912  19,675  Germany (DE)  1984, 1989, 1994, 2000, 2004, 2007, 2010, 2013  8  25,376  13,179  Denmark (DK)  2004, 2007, 2010, 2013  4  72,092  66,923  Estonia (EE)  2004, 2007, 2010, 2013  4  6,612  6,796  Spain (ES)  2000, 2004, 2007, 2010, 2013  5  18,670  12,829  Finland (FI)  1987, 1991, 1995, 2000, 2004, 2007, 2010, 2013  8  24,257  24,154  France (FR)  2005, 2010  2  4,076  3,023  Greece (GR)  1995, 2000, 2004, 2007, 2010, 2013  6  8,253  5,659  Hungary (HU)  1991, 1994, 1999, 2005  4  1,983  2,013  Ireland (IE)  1994, 1995, 2000, 2004, 2007, 2010  6  6,393  3,974  Israel (IL)  1979, 1986, 1992, 1997, 2001, 2005  6  10,756  5,951  Iceland (IS)  2004, 2007, 2010  3  3,521  2,904  Italy (IT)  2004, 2008, 2010  3  6,939  4,467  Lithuania (LT)  2010, 2013  2  2,570  2,932  Luxembourg (LU)  1997, 2000, 2004, 2007, 2010, 2013  6  10,122  4,885  Netherlands (NL)  1990, 2004, 2007, 2010, 2013  5  7,916  2,298  Poland (PL)  2007, 2010, 2013  3  39,525  34,317  Russia (RU)  2004, 2007, 2010, 2013  4  5,471  6,705  Slovenia (SI)  1999, 2004, 2007, 2010, 2012  5  8,395  8,303  Slovakia (SK)  2004, 2007, 2010, 2013  4  9,236  9,551  Taiwan (TW)  1981, 1986, 1991, 1995, 1997, 2000, 2005, 2007, 2010, 2013  10  71,221  47,701  United Kingdom (UK)  1986, 1999, 2004, 2007, 2010, 2013  6  38,208  24,726  United States (US)  1974, 1979, 1986, 1991, 1994, 1997, 2000, 2004, 2007, 2010, 2013  11  251,267  197,408  Total     143  768,549  591,008  a Welfare regime categorization is listed in Table 2. Table 1. Country samples Countrya  Years  Country observations  Male observations  Female observations  Austria (AT)  1994, 1997, 2000, 2004, 2013  5  7,183  3,324  Australia (AU)  1985, 1989, 2008, 2010  4  16,268  8,519  Belgium (BE)  1997, 2000  2  2,528  1,353  Canada (CA)  1987, 1991, 1994, 1997, 1998, 2000, 2004, 2007, 2010  9  81,391  64,549  Switzerland (CH)  2007, 2010, 2013  3  6,408  2,890  Czech Republic (CZ)  1996, 2004, 2007, 2010, 2013  5  21,912  19,675  Germany (DE)  1984, 1989, 1994, 2000, 2004, 2007, 2010, 2013  8  25,376  13,179  Denmark (DK)  2004, 2007, 2010, 2013  4  72,092  66,923  Estonia (EE)  2004, 2007, 2010, 2013  4  6,612  6,796  Spain (ES)  2000, 2004, 2007, 2010, 2013  5  18,670  12,829  Finland (FI)  1987, 1991, 1995, 2000, 2004, 2007, 2010, 2013  8  24,257  24,154  France (FR)  2005, 2010  2  4,076  3,023  Greece (GR)  1995, 2000, 2004, 2007, 2010, 2013  6  8,253  5,659  Hungary (HU)  1991, 1994, 1999, 2005  4  1,983  2,013  Ireland (IE)  1994, 1995, 2000, 2004, 2007, 2010  6  6,393  3,974  Israel (IL)  1979, 1986, 1992, 1997, 2001, 2005  6  10,756  5,951  Iceland (IS)  2004, 2007, 2010  3  3,521  2,904  Italy (IT)  2004, 2008, 2010  3  6,939  4,467  Lithuania (LT)  2010, 2013  2  2,570  2,932  Luxembourg (LU)  1997, 2000, 2004, 2007, 2010, 2013  6  10,122  4,885  Netherlands (NL)  1990, 2004, 2007, 2010, 2013  5  7,916  2,298  Poland (PL)  2007, 2010, 2013  3  39,525  34,317  Russia (RU)  2004, 2007, 2010, 2013  4  5,471  6,705  Slovenia (SI)  1999, 2004, 2007, 2010, 2012  5  8,395  8,303  Slovakia (SK)  2004, 2007, 2010, 2013  4  9,236  9,551  Taiwan (TW)  1981, 1986, 1991, 1995, 1997, 2000, 2005, 2007, 2010, 2013  10  71,221  47,701  United Kingdom (UK)  1986, 1999, 2004, 2007, 2010, 2013  6  38,208  24,726  United States (US)  1974, 1979, 1986, 1991, 1994, 1997, 2000, 2004, 2007, 2010, 2013  11  251,267  197,408  Total     143  768,549  591,008  Countrya  Years  Country observations  Male observations  Female observations  Austria (AT)  1994, 1997, 2000, 2004, 2013  5  7,183  3,324  Australia (AU)  1985, 1989, 2008, 2010  4  16,268  8,519  Belgium (BE)  1997, 2000  2  2,528  1,353  Canada (CA)  1987, 1991, 1994, 1997, 1998, 2000, 2004, 2007, 2010  9  81,391  64,549  Switzerland (CH)  2007, 2010, 2013  3  6,408  2,890  Czech Republic (CZ)  1996, 2004, 2007, 2010, 2013  5  21,912  19,675  Germany (DE)  1984, 1989, 1994, 2000, 2004, 2007, 2010, 2013  8  25,376  13,179  Denmark (DK)  2004, 2007, 2010, 2013  4  72,092  66,923  Estonia (EE)  2004, 2007, 2010, 2013  4  6,612  6,796  Spain (ES)  2000, 2004, 2007, 2010, 2013  5  18,670  12,829  Finland (FI)  1987, 1991, 1995, 2000, 2004, 2007, 2010, 2013  8  24,257  24,154  France (FR)  2005, 2010  2  4,076  3,023  Greece (GR)  1995, 2000, 2004, 2007, 2010, 2013  6  8,253  5,659  Hungary (HU)  1991, 1994, 1999, 2005  4  1,983  2,013  Ireland (IE)  1994, 1995, 2000, 2004, 2007, 2010  6  6,393  3,974  Israel (IL)  1979, 1986, 1992, 1997, 2001, 2005  6  10,756  5,951  Iceland (IS)  2004, 2007, 2010  3  3,521  2,904  Italy (IT)  2004, 2008, 2010  3  6,939  4,467  Lithuania (LT)  2010, 2013  2  2,570  2,932  Luxembourg (LU)  1997, 2000, 2004, 2007, 2010, 2013  6  10,122  4,885  Netherlands (NL)  1990, 2004, 2007, 2010, 2013  5  7,916  2,298  Poland (PL)  2007, 2010, 2013  3  39,525  34,317  Russia (RU)  2004, 2007, 2010, 2013  4  5,471  6,705  Slovenia (SI)  1999, 2004, 2007, 2010, 2012  5  8,395  8,303  Slovakia (SK)  2004, 2007, 2010, 2013  4  9,236  9,551  Taiwan (TW)  1981, 1986, 1991, 1995, 1997, 2000, 2005, 2007, 2010, 2013  10  71,221  47,701  United Kingdom (UK)  1986, 1999, 2004, 2007, 2010, 2013  6  38,208  24,726  United States (US)  1974, 1979, 1986, 1991, 1994, 1997, 2000, 2004, 2007, 2010, 2013  11  251,267  197,408  Total     143  768,549  591,008  a Welfare regime categorization is listed in Table 2. Individual Earnings I assess logged annual earnings: monetary and non-monetary payments received in counterpart for dependent employment.8 This measure excludes self-employment, capital, and transfer incomes, avoiding confounding country differences in state institutions, capital markets, and self-employment opportunities. This inequality measure reflects market earnings before taxes and transfers. I apply the logic of LIS household income top- and bottom-coding strategies to this measure of individual-level earnings. Low earnings are bottom-coded at one percent of the mean, and high values are top-coded at 10 times the median income, separately by country and survey. I use logged earnings following previous WGI studies. However, this decision likely create a conservative account of actually existing inequality. I analyse individuals rather than households to align with most previous WGI studies. Many WGI studies examine individual hourly wages weighted by working hours (Autor, Katz and Kearney, 2008; VanHeuvelen, 2018), but necessary microdata for hourly wages are not widely available.9,10 Individual Grouping Variables I separate logged annual earnings variance into between- and within-group components using eight human capital, occupational, and sociodemographic characteristics commonly used in WGI research and widely available in the LIS high-income and European samples. Analyses are conducted separately by sex. I measure education following Brady, Fullerton and Cross (2010). Categories are based on the International Standard Classification of Education (ISCED) and include three groups: low (less than secondary education: Levels 0, 1, and 2), medium (secondary or some tertiary education: Levels 3 and 4), and high (completed tertiary education or more: Levels 5 and 6).11Potential experience is the respondent’s age minus potential years of education minus six.12 I measure potential experience two ways. First is a categorical measure differentiating 0–9, 10–19, 20–29, and 30 or more years (Autor, Katz and Kearney, 2008). Second is a continuous measure and its squared term interacted with education categories.13Industry includes (1) agriculture, forestry, and fishing; (2) mining and quarrying, manufacturing, and utilities; (3) construction; (4) wholesale and retail trade, repair, hotels and restaurants; (5) transport, storage, and communications; (6) financial intermediation; (7) real estate, renting and business activities; (8) public administration, education, health and social work; and (9) other community, social/personal services, activities of household, and extra-territorial.14Occupation includes 10 International Standard Classification of Occupations (ISCO) categories: (1) managers; (2) professionals; (3) technicians and associate professionals; (4) clerical support workers; (5) service and sales workers; (6) skilled agricultural, forestry, and fishery workers; (7) craft and related trades workers; (8) plant and machine operators, and assemblers; (9) elementary occupations and; (10) armed forces occupations.15,16,17 For family status, I measure the respondent's partnership status, whether they have young children in the household, and the number of children in their household. Methods WGI is computed using VFR models. VFRs estimate between and within portions of an outcome’s variance (for extended discussion, see Western and Bloome, 2009). The first portion of the VFR is a linear regression of logged income yi on variables Wi (education, potential experience, industry, occupation, partnership status, and young children in household):   yi=βWi+ϵi (1) The second portion of the VFR is a gamma regression with a log-link function estimated on the squared residuals, ε2, predicted from equation (1). This second portion estimates the systematic component of residuals occurring among observed characteristics:   log  ϵ^i2=πWi (2) Predicted values from equation (2) are used as weights to re-estimate equation (1). Squared residuals are recomputed and equation (2) is re-estimated. The process reiterates until model parameters stabilize.18 These equations separate total earnings variance into a between and a within component, which together necessarily sum to the overall variance of the outcome. Following standard practice, VFRs are estimated separately by year (Lemieux, 2006; Autor, Katz and Kearney, 2008; Western and Rosenfeld, 2011). I assume countries have distinct earning regimes and so estimate VFRs by country (Hauser and Xie, 2005). In total, I compute 285 country-year distributions for both WGI and BGI in main analyses.19 These macrolevel distributions are used as the main dependent variable in all analyses below. In addition to using absolute levels of WGI, I also use the relative proportion of WGI to total inequality, which is simply the ratio of these two measures, multiplied by 100. This measure indicates the relative importance of WGI to total inequality in a particular country-year.20 Country Variables To descriptively assess the relative variation of WGI across countries, I include three sets of country-year measures. The first is human capital attainment. These measure (i) country-specific averages of the educational categories drawn from the LIS microdata samples discussed above, (ii) mean potential experience, and (iii) the variance of potential experience. I also compute the percentage of prime-aged workers who completed tertiary education and educational attainment heterogeneity (Moller, Alderson and Nielsen, 2009) using the Barro–Lee Educational Attainment Data (Barro and Lee, 2010).21 Second, I sort country-years into five regime types frequently used to identify general institutional differences across state policies and labour market institutional arrangements: Liberal, Continental European,22 Nordic, Eastern European, and Taiwan (Table 2 lists categorization). Third, I include three scales measuring labour market institutional arrangements and employment policies that apply the deinstitutionalization emphasis of insecurity and flexibility to the cross-national level. The first combines information on trade union membership, wage setting coordination, and adjusted bargaining coverage. The second scale measures the general strictness of EPL for regular and temporary workers. The third measures the gap in EPL between regular and temporary workers (Barbieri and Cutuli, 2016).23Table 2 includes country-level descriptive statistics. Table 2. Descriptive statistics Variable  Mean  SD  Minimum  Maximum  Male sample           Within-group inequality (WGI)  0.224  0.106  0.064  0.576   Relative contribution of WGI to total inequality  68.640  9.421  40.008  86.601   Logged variance of annual earnings  0.320  0.129  0.116  0.709   Percent high education category (ISCED 5 and 6, LIS)  0.311  0.115  0.089  0.618   Potential experience, mean (LIS)  21.103  1.220  18.693  24.952   Potential experience, standard deviation (LIS)  8.938  0.373  8.087  9.872   Percent tertiary educational attainment (Barro–Lee)  0.176  0.071  0.063  0.337   Educational heterogeneity (Barro–Lee)  0.925  0.130  0.509  1.096  Female sample           Within-group inequality  0.241  0.112  0.049  0.546   Relative contribution of WGI to total inequality  70.174  9.380  41.151  96.342   Logged variance of annual earnings  0.333  0.135  0.079  0.689   Percent high education category (ISCED 5 and 6, LIS)  0.381  0.147  0.098  0.703   Potential experience, mean (LIS)  20.426  1.678  15.874  25.282   Potential experience, standard deviation (LIS)  9.221  0.436  8.091  10.306   Percent tertiary educational attainment (Barro–Lee)  0.165  0.076  0.022  0.342   Educational heterogeneity (Barro–Lee)  0.918  0.115  0.511  1.098  Labour market institutions, employment policies           Liberal regimea  0.25         Continental European regime  0.36         Nordic regime  0.10         Eastern European regime  0.22         Taiwan  0.07            Labour coordinationa  0.00  1.442  −2.516  2.922    Wage setting coordination  2.71  1.305  1  5    Trade union membership  34.32  19.99  6.41  95.16    Adjusted bargaining coverage  57.11  26.28  10.70  100   EPL: Total strictnessb  0.00  1.237  −2.252  2.514   EPL: Regular/temporary employment protection gapb  0.00  0.685  −1.947  1.668    EPL: Regular contracts  2.036  0.812  0.257  3.306    EPL: Temporary contracts  1.498  1.241  0.025  5  Male observations: 143 Female observations: 142  Variable  Mean  SD  Minimum  Maximum  Male sample           Within-group inequality (WGI)  0.224  0.106  0.064  0.576   Relative contribution of WGI to total inequality  68.640  9.421  40.008  86.601   Logged variance of annual earnings  0.320  0.129  0.116  0.709   Percent high education category (ISCED 5 and 6, LIS)  0.311  0.115  0.089  0.618   Potential experience, mean (LIS)  21.103  1.220  18.693  24.952   Potential experience, standard deviation (LIS)  8.938  0.373  8.087  9.872   Percent tertiary educational attainment (Barro–Lee)  0.176  0.071  0.063  0.337   Educational heterogeneity (Barro–Lee)  0.925  0.130  0.509  1.096  Female sample           Within-group inequality  0.241  0.112  0.049  0.546   Relative contribution of WGI to total inequality  70.174  9.380  41.151  96.342   Logged variance of annual earnings  0.333  0.135  0.079  0.689   Percent high education category (ISCED 5 and 6, LIS)  0.381  0.147  0.098  0.703   Potential experience, mean (LIS)  20.426  1.678  15.874  25.282   Potential experience, standard deviation (LIS)  9.221  0.436  8.091  10.306   Percent tertiary educational attainment (Barro–Lee)  0.165  0.076  0.022  0.342   Educational heterogeneity (Barro–Lee)  0.918  0.115  0.511  1.098  Labour market institutions, employment policies           Liberal regimea  0.25         Continental European regime  0.36         Nordic regime  0.10         Eastern European regime  0.22         Taiwan  0.07            Labour coordinationa  0.00  1.442  −2.516  2.922    Wage setting coordination  2.71  1.305  1  5    Trade union membership  34.32  19.99  6.41  95.16    Adjusted bargaining coverage  57.11  26.28  10.70  100   EPL: Total strictnessb  0.00  1.237  −2.252  2.514   EPL: Regular/temporary employment protection gapb  0.00  0.685  −1.947  1.668    EPL: Regular contracts  2.036  0.812  0.257  3.306    EPL: Temporary contracts  1.498  1.241  0.025  5  Male observations: 143 Female observations: 142  Israel-1986 dropped from female sample due to small sample size. EPL measures not available in Taiwan (n = 133 men, 132 women). a Liberal: Australia, Canada, Ireland, Great Britain, and United States. Continental European: combination of Conservative—Austria, Belgium, Switzerland, Germany, France, Israel (Mandel and Shalev 2009), Luxembourg, and The Netherlands—and Mediterranean—Greece, Italy, and Spain—regimes. Nordic: Denmark, Finland, and Iceland. East European: Czech Republic, Estonia, Hungary, Lithuania, Poland, Russia, Slovenia, and Slovak Republic. b Scaled created from indented items below. See online Appendix for details. Table 2. Descriptive statistics Variable  Mean  SD  Minimum  Maximum  Male sample           Within-group inequality (WGI)  0.224  0.106  0.064  0.576   Relative contribution of WGI to total inequality  68.640  9.421  40.008  86.601   Logged variance of annual earnings  0.320  0.129  0.116  0.709   Percent high education category (ISCED 5 and 6, LIS)  0.311  0.115  0.089  0.618   Potential experience, mean (LIS)  21.103  1.220  18.693  24.952   Potential experience, standard deviation (LIS)  8.938  0.373  8.087  9.872   Percent tertiary educational attainment (Barro–Lee)  0.176  0.071  0.063  0.337   Educational heterogeneity (Barro–Lee)  0.925  0.130  0.509  1.096  Female sample           Within-group inequality  0.241  0.112  0.049  0.546   Relative contribution of WGI to total inequality  70.174  9.380  41.151  96.342   Logged variance of annual earnings  0.333  0.135  0.079  0.689   Percent high education category (ISCED 5 and 6, LIS)  0.381  0.147  0.098  0.703   Potential experience, mean (LIS)  20.426  1.678  15.874  25.282   Potential experience, standard deviation (LIS)  9.221  0.436  8.091  10.306   Percent tertiary educational attainment (Barro–Lee)  0.165  0.076  0.022  0.342   Educational heterogeneity (Barro–Lee)  0.918  0.115  0.511  1.098  Labour market institutions, employment policies           Liberal regimea  0.25         Continental European regime  0.36         Nordic regime  0.10         Eastern European regime  0.22         Taiwan  0.07            Labour coordinationa  0.00  1.442  −2.516  2.922    Wage setting coordination  2.71  1.305  1  5    Trade union membership  34.32  19.99  6.41  95.16    Adjusted bargaining coverage  57.11  26.28  10.70  100   EPL: Total strictnessb  0.00  1.237  −2.252  2.514   EPL: Regular/temporary employment protection gapb  0.00  0.685  −1.947  1.668    EPL: Regular contracts  2.036  0.812  0.257  3.306    EPL: Temporary contracts  1.498  1.241  0.025  5  Male observations: 143 Female observations: 142  Variable  Mean  SD  Minimum  Maximum  Male sample           Within-group inequality (WGI)  0.224  0.106  0.064  0.576   Relative contribution of WGI to total inequality  68.640  9.421  40.008  86.601   Logged variance of annual earnings  0.320  0.129  0.116  0.709   Percent high education category (ISCED 5 and 6, LIS)  0.311  0.115  0.089  0.618   Potential experience, mean (LIS)  21.103  1.220  18.693  24.952   Potential experience, standard deviation (LIS)  8.938  0.373  8.087  9.872   Percent tertiary educational attainment (Barro–Lee)  0.176  0.071  0.063  0.337   Educational heterogeneity (Barro–Lee)  0.925  0.130  0.509  1.096  Female sample           Within-group inequality  0.241  0.112  0.049  0.546   Relative contribution of WGI to total inequality  70.174  9.380  41.151  96.342   Logged variance of annual earnings  0.333  0.135  0.079  0.689   Percent high education category (ISCED 5 and 6, LIS)  0.381  0.147  0.098  0.703   Potential experience, mean (LIS)  20.426  1.678  15.874  25.282   Potential experience, standard deviation (LIS)  9.221  0.436  8.091  10.306   Percent tertiary educational attainment (Barro–Lee)  0.165  0.076  0.022  0.342   Educational heterogeneity (Barro–Lee)  0.918  0.115  0.511  1.098  Labour market institutions, employment policies           Liberal regimea  0.25         Continental European regime  0.36         Nordic regime  0.10         Eastern European regime  0.22         Taiwan  0.07            Labour coordinationa  0.00  1.442  −2.516  2.922    Wage setting coordination  2.71  1.305  1  5    Trade union membership  34.32  19.99  6.41  95.16    Adjusted bargaining coverage  57.11  26.28  10.70  100   EPL: Total strictnessb  0.00  1.237  −2.252  2.514   EPL: Regular/temporary employment protection gapb  0.00  0.685  −1.947  1.668    EPL: Regular contracts  2.036  0.812  0.257  3.306    EPL: Temporary contracts  1.498  1.241  0.025  5  Male observations: 143 Female observations: 142  Israel-1986 dropped from female sample due to small sample size. EPL measures not available in Taiwan (n = 133 men, 132 women). a Liberal: Australia, Canada, Ireland, Great Britain, and United States. Continental European: combination of Conservative—Austria, Belgium, Switzerland, Germany, France, Israel (Mandel and Shalev 2009), Luxembourg, and The Netherlands—and Mediterranean—Greece, Italy, and Spain—regimes. Nordic: Denmark, Finland, and Iceland. East European: Czech Republic, Estonia, Hungary, Lithuania, Poland, Russia, Slovenia, and Slovak Republic. b Scaled created from indented items below. See online Appendix for details. Results How Has WGI Grown in Rich Countries? No previous research has assessed WGI cross-nationally using high-quality, comparable measurements. This study therefore provides a general, descriptive assessment of the contribution of WGI to LIS total market earnings inequality. How widely does WGI vary across countries, and what is the relative contribution of WGI to total earnings inequality? Figure 1 shows the relationship between the absolute level of annual earnings WGI (x-axes) and the percentage of total earnings inequality attributable to WGI (y-axes). The panels show results by sex. Figure 1. View largeDownload slide Relationship between absolute level of WGI and relative contribution of WGI to total inequality Note: Country codes are listed in Table 1. Figure 1. View largeDownload slide Relationship between absolute level of WGI and relative contribution of WGI to total inequality Note: Country codes are listed in Table 1. I observe substantial variation in both absolute levels of WGI and the relative contribution of WGI to total inequality. Absolute levels of WGI range by a factor of 7 for men (maximum = 0.57, minimum = 0.07) and 10 for women (maximum = 0.56, minimum = 0.05). For both male and female samples, Denmark, Italy, and Luxembourg tend to have lower absolute levels of WGI, whereas the United States, Canada, Russia, and Lithuania tend to have higher absolute levels. This finding simply reflects the wide range of inequality across high-income countries that has drawn scholars to the cross-national level. Such cross-national variability exists among WGI as well. WGI varies not only in its absolute levels but also in its relative contribution to earnings inequality. In some countries—the Slovak Republic, Australia, Austria, and Canada, for example—WGI accounts for upwards of 85% of total earnings inequality. In other words, standard microlevel grouping categories—education, work experience, and occupational and demographic characteristics—do little to explain market earnings inequality in these countries. In other countries—Luxembourg and Taiwan, for example—WGI is relatively less important for inequality trends, accounting for 40%–60% of total inequality. Put differently, I observe substantial variation in the contribution of the exact same microlevel models across country contexts. What is the relationship between levels of WGI and its relative contribution to total inequality trends? The results in Figure 1 suggest a positive association. For both men and women, WGI tends to be more important in country-years where absolute levels of WGI are higher. The correlations between these dimensions of WGI are 0.60 for men and 0.56 for women (both P < 0.001, two-tailed tests). I find similar positive associations between the relative contribution of WGI and total earnings inequality for both samples, but with smaller correlations (around 0.37 for both samples). In total, Figure 1 provides preliminary evidence that WGI typically plays a larger role for inequality where inequality levels are higher. These findings help explain why WGI has been central to US inequality debates (Lemieux, 2006; Autor, Katz and Kearney, 2008). Along with the variation noted in the above paragraph, two points of similarity across country contexts are notable. First, most individual earnings inequality in the LIS data series—the gold standard of macrocomparative stratification research—is WGI. In only 53 of the 285 country-year observations (18.5%) does WGI contribute less than 60% to total inequality. For comparison, WGI contributes over ⅔ of total earnings inequality in 163 country-years (57% of sample) and over ¾ of total inequality in 23% of the sample (65 country-years). Put simply, most cross-national prime-age market earnings inequality that researchers have studied in high-income, postindustrial societies occurs in the within-group component of inequality. Second, supplemental analyses show similarity of measurements from Figure 1, and their rank orderings, between male and female samples. Male WGI tends to be high where female WGI is also, and male WGI tends to be relatively important in country-years where female WGI is relatively important. The correlations in rank orderings between sexes are 0.72 for WGI percentage and 0.65 for absolute ranks. The relative similarities across samples cast doubt on the argument that WGI represents simple randomness: Why would randomness be so orderly across worker samples within country-year pairs? Does the United States have atypical WGI? Yes and no. On the one hand, the United States has large absolute levels of WGI, as I observe US country-year observations clustering on the right side of the x-axis in both panels. This finding reinforces previous research showing the United States to have particularly high levels of inequality, inclusive of inequality both pre- and post-tax and transfer redistribution (Gornick and Milanovic, 2015). This fact is simply reflected in the within-component of inequality. On the other hand, the United States appears typical regarding WGI’s relative contribution to total male inequality. For example, the contribution of WGI to total male earnings is similar across American, Germany, and Danish contexts, around 70%. The three countries differ in their absolute levels of WGI. Yet WGI is relatively more important to total earnings inequality in some countries—the Slovak Republic, Canada, and Australia—and is relatively less important in others—Finland, Greece, and Hungary, for example. In total, I observe US WGI to be atypical in its high levels for both men and women, and to be typical in its relative contribution to male earnings inequality. Figure 2 shifts focus from levels to change in inequality within countries over time. It shows the contribution of WGI change and BGI change to total changes in earnings inequality (left panels, e.g. change between us74 and us13) and two adjacent country-year observations (right panels, e.g. change between us74 and us79). X-axes represent change in total logged earnings variance between periods. Y-axes represent change attributable to WGI and BGI. Markers include country codes (Table 1) and are prefixed ‘W’ for WGI change and ‘B’ for BGI change. Lines show the simple linear fit between either WGI (solid) or BGI (dashed) change and total inequality change. A steeper association between total inequality and either WGI or BGI change indicates a greater relative importance of change in one of these components to total inequality change. Figure 2. View largeDownload slide Scatterplot and slope of WGI change and BGI change against total inequality change Note: Country codes listed in Table 1. Figure 2. View largeDownload slide Scatterplot and slope of WGI change and BGI change against total inequality change Note: Country codes listed in Table 1. Figure 2 clearly shows that inequality change is due primarily to WGI change. The correlations of WGI change and total inequality change range from 0.94–0.97, whereas the equivalent correlations for BGI are lower, 0.60–0.74. In addition, I tested the x-standardized regression coefficients of total inequality change on WGI and BGI change in simple linear regression models with robust standard errors. The coefficients for WGI change are significantly larger than the coefficients for BGI change across all samples from Figure 2, with the WGI slope 100%–300% larger compared with the BGI slope (P < 0.05, two-tailed tests, in all tests conducted).24Table A1 verifies these results by estimating the association between the relative contribution of WGI and total inequality with and without country- and year-fixed effects. Notably, trends in Figures 1 and 2 might be conservative due to the logged outcome variable analysed. In total, I conclude that in addition to its importance to overall levels of earnings inequality, change in WGI is primarily responsible for change in total inequality in high-income countries. As inequality grows in high-income countries, WGI tends to become increasingly important. No previous research has revealed this basic empirical finding, yet it is critical for understanding recent inequality trends. To provide a substantive illustration of above results, Figure 3 displays trajectories of male WGI, BGI, and total logged earnings variance in the United States, Canada, Germany, Finland, and Luxembourg. I examined equivalent figures for all countries included in analyses. These five countries were selected because they represent the range of inequality patterns.25 Panels show trends for WGI (left), BGI (centre), and total income (right). Figure 3. View largeDownload slide Male WGI, BGI, and total inequality of logged annual earnings over time in five countries Note: Country codes listed in Table 1. Figure 3. View largeDownload slide Male WGI, BGI, and total inequality of logged annual earnings over time in five countries Note: Country codes listed in Table 1. Markers indicate LIS waves. The trajectory of American WGI (hollow circles) generally follow patterns from previous studies (Lemieux, 2006; Autor, Katz and Kearney, 2008; Western and Rosenfeld, 2011). WGI rose rapidly through the 1980s, stabilized for the next two decades, and began to rise again at the end of the 2000s. As suggested in Figures 1 and 2, the divergent trajectories and levels of inequality in these five countries are qualitatively due to WGI. Although the United States and Luxembourg have higher BGI than Germany, Finland, and Canada, overall BGI does little to distinguish country inequality trends. Turning to WGI, Canadian and American patterns closely resemble one another. The difference in total inequality between the two countries stems primarily from higher American BGI. German WGI spiked after reunification in 1990 and grew slowly afterwards. Finnish WGI remained unchanged for nearly two decades, then doubled in the mid-2000s. Luxembourg WGI increased slightly, but its importance pales in comparison with BGI, which doubled over the period of study to end up resembling American BGI. While WGI increased in all five countries, the nature of WGI change varied widely, and it is primarily these WGI trends that distinguish the trends in total inequality in the rightmost panel. From the many findings of the descriptive analyses in this section, I draw one general conclusion: Within-group inequality primarily drives cross-national differences in prime-age market earnings inequality among high-income countries.26 I observe WGI to be central to inequality levels, trends in inequality change, and substantive inter-country inequality patterns. Although US studies have shown WGI to be important for total inequality trends, it was unknown whether these American inequality dynamics were unique, or whether WGI was more generally important for contemporary trends of rising inequality among high-income countries. My results point to the latter. WGI, Skill, and Deinstitutionalization Thus far, results suggest a general importance of WGI for total inequality. I next assess how WGI varies along cross-national differences in human capital attainment and labour market institutional arrangements. Results are shown in Tables 3 and 4. Table 3. Education and welfare regime variation in WGI levels and relative importance   WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  (9)  (10)  (11)  (12)  Welfare regime (Liberal = reference)                           Continental Europe  −8.025***  −8.878***  −8.021***  −5.340**  −6.859***  −4.801*  −0.108***  −0.080***  −0.069***  −0.102***  −0.063**  −0.061**  (1.826)  (2.058)  (1.999)  (1.899)  (1.957)  (2.062)  (0.018)  (0.019)  (0.018)  (0.021)  (0.023)  (0.019)   Nordic  −6.080**  −7.945**  −6.392**  −7.809***  −8.857**  −7.779**  −0.151***  −0.110***  −0.127***  −0.201***  −0.172***  −0.230***  (2.151)  (2.563)  (2.358)  (1.974)  (2.736)  (2.581)  (0.025)  (0.028)  (0.025)  (0.026)  (0.033)  (0.030)   Eastern Europe  2.011  1.180  2.258  −1.301  −3.223  −1.387  −0.046+  0.039  0.037  −0.060*  −0.060+  −0.039  (1.779)  (1.954)  (2.030)  (1.984)  (2.674)  (2.807)  (0.027)  (0.028)  (0.029)  (0.027)  (0.032)  (0.029)   Taiwan  −21.006***  −23.775***  −20.422***  −24.123***  −26.995***  −22.003***  −0.183***  −0.072**  −0.183***  −0.186***  −0.128***  −0.136***  (1.351)  (2.543)  (1.558)  (2.050)  (2.858)  (3.060)  (0.016)  (0.024)  (0.016)  (0.018)  (0.027)  (0.018)  Completed tertiary education    −23.115+      −17.854      0.621***      0.569***      (13.712)      (11.772)      (0.139)      (0.144)    Educational heterogeneity    4.315      −4.950      0.105      −0.184*      (6.551)      (8.857)      (0.072)      (0.080)    Percent secondary education      3.453      12.499      −0.065      0.101+      (6.363)      (10.391)      (0.065)      (0.060)  Percent in tertiary education      4.187      12.944      0.401***      0.528***      (9.098)      (12.616)      (0.091)      (0.095)  Average potential experience    −0.925  −0.324    −0.411  0.009    −0.009  −0.006    0.005  0.019**    (0.662)  (0.616)    (0.650)  (0.694)    (0.005)  (0.007)    (0.006)  (0.006)  Potential experience SD    −2.583  −2.081    −0.111  −0.082    −0.028  −0.043*    −0.001  −0.001    (1.670)  (1.676)    (0.093)  (0.090)    (0.017)  (0.018)    (0.001)  (0.001)  N  143  143  143  142  142  142  143  143  143  142  142  142    WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  (9)  (10)  (11)  (12)  Welfare regime (Liberal = reference)                           Continental Europe  −8.025***  −8.878***  −8.021***  −5.340**  −6.859***  −4.801*  −0.108***  −0.080***  −0.069***  −0.102***  −0.063**  −0.061**  (1.826)  (2.058)  (1.999)  (1.899)  (1.957)  (2.062)  (0.018)  (0.019)  (0.018)  (0.021)  (0.023)  (0.019)   Nordic  −6.080**  −7.945**  −6.392**  −7.809***  −8.857**  −7.779**  −0.151***  −0.110***  −0.127***  −0.201***  −0.172***  −0.230***  (2.151)  (2.563)  (2.358)  (1.974)  (2.736)  (2.581)  (0.025)  (0.028)  (0.025)  (0.026)  (0.033)  (0.030)   Eastern Europe  2.011  1.180  2.258  −1.301  −3.223  −1.387  −0.046+  0.039  0.037  −0.060*  −0.060+  −0.039  (1.779)  (1.954)  (2.030)  (1.984)  (2.674)  (2.807)  (0.027)  (0.028)  (0.029)  (0.027)  (0.032)  (0.029)   Taiwan  −21.006***  −23.775***  −20.422***  −24.123***  −26.995***  −22.003***  −0.183***  −0.072**  −0.183***  −0.186***  −0.128***  −0.136***  (1.351)  (2.543)  (1.558)  (2.050)  (2.858)  (3.060)  (0.016)  (0.024)  (0.016)  (0.018)  (0.027)  (0.018)  Completed tertiary education    −23.115+      −17.854      0.621***      0.569***      (13.712)      (11.772)      (0.139)      (0.144)    Educational heterogeneity    4.315      −4.950      0.105      −0.184*      (6.551)      (8.857)      (0.072)      (0.080)    Percent secondary education      3.453      12.499      −0.065      0.101+      (6.363)      (10.391)      (0.065)      (0.060)  Percent in tertiary education      4.187      12.944      0.401***      0.528***      (9.098)      (12.616)      (0.091)      (0.095)  Average potential experience    −0.925  −0.324    −0.411  0.009    −0.009  −0.006    0.005  0.019**    (0.662)  (0.616)    (0.650)  (0.694)    (0.005)  (0.007)    (0.006)  (0.006)  Potential experience SD    −2.583  −2.081    −0.111  −0.082    −0.028  −0.043*    −0.001  −0.001    (1.670)  (1.676)    (0.093)  (0.090)    (0.017)  (0.018)    (0.001)  (0.001)  N  143  143  143  142  142  142  143  143  143  142  142  142  Robust standard errors in parentheses. +  P < 0.10, * P < 0.05, ** P < 0.01, *** P < 0.001, two-tailed test.  Models include continuous measure of year. ‘Completed tertiary education’ and ‘Educational heterogeneity’ computed from Barro–Lee database individuals aged 25–55, separately by sex. Additional education and potential experience measures computed from LIS samples. Table 3. Education and welfare regime variation in WGI levels and relative importance   WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  (9)  (10)  (11)  (12)  Welfare regime (Liberal = reference)                           Continental Europe  −8.025***  −8.878***  −8.021***  −5.340**  −6.859***  −4.801*  −0.108***  −0.080***  −0.069***  −0.102***  −0.063**  −0.061**  (1.826)  (2.058)  (1.999)  (1.899)  (1.957)  (2.062)  (0.018)  (0.019)  (0.018)  (0.021)  (0.023)  (0.019)   Nordic  −6.080**  −7.945**  −6.392**  −7.809***  −8.857**  −7.779**  −0.151***  −0.110***  −0.127***  −0.201***  −0.172***  −0.230***  (2.151)  (2.563)  (2.358)  (1.974)  (2.736)  (2.581)  (0.025)  (0.028)  (0.025)  (0.026)  (0.033)  (0.030)   Eastern Europe  2.011  1.180  2.258  −1.301  −3.223  −1.387  −0.046+  0.039  0.037  −0.060*  −0.060+  −0.039  (1.779)  (1.954)  (2.030)  (1.984)  (2.674)  (2.807)  (0.027)  (0.028)  (0.029)  (0.027)  (0.032)  (0.029)   Taiwan  −21.006***  −23.775***  −20.422***  −24.123***  −26.995***  −22.003***  −0.183***  −0.072**  −0.183***  −0.186***  −0.128***  −0.136***  (1.351)  (2.543)  (1.558)  (2.050)  (2.858)  (3.060)  (0.016)  (0.024)  (0.016)  (0.018)  (0.027)  (0.018)  Completed tertiary education    −23.115+      −17.854      0.621***      0.569***      (13.712)      (11.772)      (0.139)      (0.144)    Educational heterogeneity    4.315      −4.950      0.105      −0.184*      (6.551)      (8.857)      (0.072)      (0.080)    Percent secondary education      3.453      12.499      −0.065      0.101+      (6.363)      (10.391)      (0.065)      (0.060)  Percent in tertiary education      4.187      12.944      0.401***      0.528***      (9.098)      (12.616)      (0.091)      (0.095)  Average potential experience    −0.925  −0.324    −0.411  0.009    −0.009  −0.006    0.005  0.019**    (0.662)  (0.616)    (0.650)  (0.694)    (0.005)  (0.007)    (0.006)  (0.006)  Potential experience SD    −2.583  −2.081    −0.111  −0.082    −0.028  −0.043*    −0.001  −0.001    (1.670)  (1.676)    (0.093)  (0.090)    (0.017)  (0.018)    (0.001)  (0.001)  N  143  143  143  142  142  142  143  143  143  142  142  142    WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  (9)  (10)  (11)  (12)  Welfare regime (Liberal = reference)                           Continental Europe  −8.025***  −8.878***  −8.021***  −5.340**  −6.859***  −4.801*  −0.108***  −0.080***  −0.069***  −0.102***  −0.063**  −0.061**  (1.826)  (2.058)  (1.999)  (1.899)  (1.957)  (2.062)  (0.018)  (0.019)  (0.018)  (0.021)  (0.023)  (0.019)   Nordic  −6.080**  −7.945**  −6.392**  −7.809***  −8.857**  −7.779**  −0.151***  −0.110***  −0.127***  −0.201***  −0.172***  −0.230***  (2.151)  (2.563)  (2.358)  (1.974)  (2.736)  (2.581)  (0.025)  (0.028)  (0.025)  (0.026)  (0.033)  (0.030)   Eastern Europe  2.011  1.180  2.258  −1.301  −3.223  −1.387  −0.046+  0.039  0.037  −0.060*  −0.060+  −0.039  (1.779)  (1.954)  (2.030)  (1.984)  (2.674)  (2.807)  (0.027)  (0.028)  (0.029)  (0.027)  (0.032)  (0.029)   Taiwan  −21.006***  −23.775***  −20.422***  −24.123***  −26.995***  −22.003***  −0.183***  −0.072**  −0.183***  −0.186***  −0.128***  −0.136***  (1.351)  (2.543)  (1.558)  (2.050)  (2.858)  (3.060)  (0.016)  (0.024)  (0.016)  (0.018)  (0.027)  (0.018)  Completed tertiary education    −23.115+      −17.854      0.621***      0.569***      (13.712)      (11.772)      (0.139)      (0.144)    Educational heterogeneity    4.315      −4.950      0.105      −0.184*      (6.551)      (8.857)      (0.072)      (0.080)    Percent secondary education      3.453      12.499      −0.065      0.101+      (6.363)      (10.391)      (0.065)      (0.060)  Percent in tertiary education      4.187      12.944      0.401***      0.528***      (9.098)      (12.616)      (0.091)      (0.095)  Average potential experience    −0.925  −0.324    −0.411  0.009    −0.009  −0.006    0.005  0.019**    (0.662)  (0.616)    (0.650)  (0.694)    (0.005)  (0.007)    (0.006)  (0.006)  Potential experience SD    −2.583  −2.081    −0.111  −0.082    −0.028  −0.043*    −0.001  −0.001    (1.670)  (1.676)    (0.093)  (0.090)    (0.017)  (0.018)    (0.001)  (0.001)  N  143  143  143  142  142  142  143  143  143  142  142  142  Robust standard errors in parentheses. +  P < 0.10, * P < 0.05, ** P < 0.01, *** P < 0.001, two-tailed test.  Models include continuous measure of year. ‘Completed tertiary education’ and ‘Educational heterogeneity’ computed from Barro–Lee database individuals aged 25–55, separately by sex. Additional education and potential experience measures computed from LIS samples. Table 4. Labour coordination, EPL, and welfare regime variation in WGI levels and relative importance   WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (7)  (8)  (9)  (4)  (5)  (6)  (10)  (11)  (12)  Labour coordination  −1.673**  −1.671**  −0.773  −0.830  −0.300  1.011  −0.034***  −0.037***  −0.031***  −0.037***  −0.034***  −0.024**  (0.498)  (0.602)  (0.852)  (0.539)  (0.654)  (1.070)  (0.006)  (0.006)  (0.007)  (0.007)  (0.007)  (0.009)  General EPL  −1.780*  −2.531**  −3.771**  −1.993**  −2.652***  −4.065***  −0.025***  −0.015*  −0.037**  −0.021***  −0.018*  −0.040**  (0.725)  (0.816)  (1.321)  (0.623)  (0.615)  (1.171)  (0.006)  (0.007)  (0.013)  (0.006)  (0.007)  (0.012)  Regular − temporary EPL difference  3.251**  1.561  −0.310  3.981***  3.050*  2.227  0.002  0.022+  −0.003  0.001  0.008  0.000  (1.067)  (1.171)  (1.136)  (1.131)  (1.243)  (1.484)  (0.011)  (0.012)  (0.014)  (0.012)  (0.013)  (0.016)  Welfare regime (Liberal = reference)                           Continental Europe      1.448      2.717      0.052      0.066+      (3.687)      (3.415)      (0.038)      (0.035)   Nordic      −0.072      −5.109      0.041      −0.019      (4.034)      (4.740)      (0.040)      (0.047)   Eastern Europe      8.609*      3.245      0.121***      0.047      (3.458)      (4.225)      (0.035)      (0.042)  Completed tertiary education    −45.627**  −48.614**    −9.380  −4.782    0.297  0.228    0.264  0.330+    (16.360)  (16.531)    (13.848)  (14.689)    (0.186)  (0.198)    (0.189)  (0.192)  Educational heterogeneity    0.619  11.575+    −14.446  −15.212    0.111+  0.234***    −0.054  −0.088    (6.893)  (6.343)    (8.777)  (9.178)    (0.065)  (0.068)    (0.088)  (0.088)  Average potential experience    −0.891  −1.269+    −1.009+  −0.814    −0.006  −0.011*    −0.005  −0.003    (0.685)  (0.653)    (0.517)  (0.627)    (0.006)  (0.005)    (0.005)  (0.006)  Potential experience SD    −0.327  −1.633    −0.136  −0.164+    −0.015  −0.023    −0.001  −0.001    (1.817)  (1.811)    (0.088)  (0.091)    (0.017)  (0.018)    (0.001)  (0.001)  N  133  133  133  132  132  132  133  133  133  132  132  132    WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (7)  (8)  (9)  (4)  (5)  (6)  (10)  (11)  (12)  Labour coordination  −1.673**  −1.671**  −0.773  −0.830  −0.300  1.011  −0.034***  −0.037***  −0.031***  −0.037***  −0.034***  −0.024**  (0.498)  (0.602)  (0.852)  (0.539)  (0.654)  (1.070)  (0.006)  (0.006)  (0.007)  (0.007)  (0.007)  (0.009)  General EPL  −1.780*  −2.531**  −3.771**  −1.993**  −2.652***  −4.065***  −0.025***  −0.015*  −0.037**  −0.021***  −0.018*  −0.040**  (0.725)  (0.816)  (1.321)  (0.623)  (0.615)  (1.171)  (0.006)  (0.007)  (0.013)  (0.006)  (0.007)  (0.012)  Regular − temporary EPL difference  3.251**  1.561  −0.310  3.981***  3.050*  2.227  0.002  0.022+  −0.003  0.001  0.008  0.000  (1.067)  (1.171)  (1.136)  (1.131)  (1.243)  (1.484)  (0.011)  (0.012)  (0.014)  (0.012)  (0.013)  (0.016)  Welfare regime (Liberal = reference)                           Continental Europe      1.448      2.717      0.052      0.066+      (3.687)      (3.415)      (0.038)      (0.035)   Nordic      −0.072      −5.109      0.041      −0.019      (4.034)      (4.740)      (0.040)      (0.047)   Eastern Europe      8.609*      3.245      0.121***      0.047      (3.458)      (4.225)      (0.035)      (0.042)  Completed tertiary education    −45.627**  −48.614**    −9.380  −4.782    0.297  0.228    0.264  0.330+    (16.360)  (16.531)    (13.848)  (14.689)    (0.186)  (0.198)    (0.189)  (0.192)  Educational heterogeneity    0.619  11.575+    −14.446  −15.212    0.111+  0.234***    −0.054  −0.088    (6.893)  (6.343)    (8.777)  (9.178)    (0.065)  (0.068)    (0.088)  (0.088)  Average potential experience    −0.891  −1.269+    −1.009+  −0.814    −0.006  −0.011*    −0.005  −0.003    (0.685)  (0.653)    (0.517)  (0.627)    (0.006)  (0.005)    (0.005)  (0.006)  Potential experience SD    −0.327  −1.633    −0.136  −0.164+    −0.015  −0.023    −0.001  −0.001    (1.817)  (1.811)    (0.088)  (0.091)    (0.017)  (0.018)    (0.001)  (0.001)  N  133  133  133  132  132  132  133  133  133  132  132  132  Robust standard errors in parentheses. +  P < 0.10, * P < 0.05, ** P < 0.01, *** P < 0.001, two-tailed test.  Models include continuous measure of year. ‘Completed tertiary education’ and ‘Educational heterogeneity’ computed from Barro–Lee database individuals aged 25–55, separately by sex. Potential experience measures computed from LIS samples. See Table 2 and online Appendix for description of Labour coordination and EPL scales.  Taiwan EPL values missing. Table A4 replicates results with only Labour coordination scale. Table 4. Labour coordination, EPL, and welfare regime variation in WGI levels and relative importance   WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (7)  (8)  (9)  (4)  (5)  (6)  (10)  (11)  (12)  Labour coordination  −1.673**  −1.671**  −0.773  −0.830  −0.300  1.011  −0.034***  −0.037***  −0.031***  −0.037***  −0.034***  −0.024**  (0.498)  (0.602)  (0.852)  (0.539)  (0.654)  (1.070)  (0.006)  (0.006)  (0.007)  (0.007)  (0.007)  (0.009)  General EPL  −1.780*  −2.531**  −3.771**  −1.993**  −2.652***  −4.065***  −0.025***  −0.015*  −0.037**  −0.021***  −0.018*  −0.040**  (0.725)  (0.816)  (1.321)  (0.623)  (0.615)  (1.171)  (0.006)  (0.007)  (0.013)  (0.006)  (0.007)  (0.012)  Regular − temporary EPL difference  3.251**  1.561  −0.310  3.981***  3.050*  2.227  0.002  0.022+  −0.003  0.001  0.008  0.000  (1.067)  (1.171)  (1.136)  (1.131)  (1.243)  (1.484)  (0.011)  (0.012)  (0.014)  (0.012)  (0.013)  (0.016)  Welfare regime (Liberal = reference)                           Continental Europe      1.448      2.717      0.052      0.066+      (3.687)      (3.415)      (0.038)      (0.035)   Nordic      −0.072      −5.109      0.041      −0.019      (4.034)      (4.740)      (0.040)      (0.047)   Eastern Europe      8.609*      3.245      0.121***      0.047      (3.458)      (4.225)      (0.035)      (0.042)  Completed tertiary education    −45.627**  −48.614**    −9.380  −4.782    0.297  0.228    0.264  0.330+    (16.360)  (16.531)    (13.848)  (14.689)    (0.186)  (0.198)    (0.189)  (0.192)  Educational heterogeneity    0.619  11.575+    −14.446  −15.212    0.111+  0.234***    −0.054  −0.088    (6.893)  (6.343)    (8.777)  (9.178)    (0.065)  (0.068)    (0.088)  (0.088)  Average potential experience    −0.891  −1.269+    −1.009+  −0.814    −0.006  −0.011*    −0.005  −0.003    (0.685)  (0.653)    (0.517)  (0.627)    (0.006)  (0.005)    (0.005)  (0.006)  Potential experience SD    −0.327  −1.633    −0.136  −0.164+    −0.015  −0.023    −0.001  −0.001    (1.817)  (1.811)    (0.088)  (0.091)    (0.017)  (0.018)    (0.001)  (0.001)  N  133  133  133  132  132  132  133  133  133  132  132  132    WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (7)  (8)  (9)  (4)  (5)  (6)  (10)  (11)  (12)  Labour coordination  −1.673**  −1.671**  −0.773  −0.830  −0.300  1.011  −0.034***  −0.037***  −0.031***  −0.037***  −0.034***  −0.024**  (0.498)  (0.602)  (0.852)  (0.539)  (0.654)  (1.070)  (0.006)  (0.006)  (0.007)  (0.007)  (0.007)  (0.009)  General EPL  −1.780*  −2.531**  −3.771**  −1.993**  −2.652***  −4.065***  −0.025***  −0.015*  −0.037**  −0.021***  −0.018*  −0.040**  (0.725)  (0.816)  (1.321)  (0.623)  (0.615)  (1.171)  (0.006)  (0.007)  (0.013)  (0.006)  (0.007)  (0.012)  Regular − temporary EPL difference  3.251**  1.561  −0.310  3.981***  3.050*  2.227  0.002  0.022+  −0.003  0.001  0.008  0.000  (1.067)  (1.171)  (1.136)  (1.131)  (1.243)  (1.484)  (0.011)  (0.012)  (0.014)  (0.012)  (0.013)  (0.016)  Welfare regime (Liberal = reference)                           Continental Europe      1.448      2.717      0.052      0.066+      (3.687)      (3.415)      (0.038)      (0.035)   Nordic      −0.072      −5.109      0.041      −0.019      (4.034)      (4.740)      (0.040)      (0.047)   Eastern Europe      8.609*      3.245      0.121***      0.047      (3.458)      (4.225)      (0.035)      (0.042)  Completed tertiary education    −45.627**  −48.614**    −9.380  −4.782    0.297  0.228    0.264  0.330+    (16.360)  (16.531)    (13.848)  (14.689)    (0.186)  (0.198)    (0.189)  (0.192)  Educational heterogeneity    0.619  11.575+    −14.446  −15.212    0.111+  0.234***    −0.054  −0.088    (6.893)  (6.343)    (8.777)  (9.178)    (0.065)  (0.068)    (0.088)  (0.088)  Average potential experience    −0.891  −1.269+    −1.009+  −0.814    −0.006  −0.011*    −0.005  −0.003    (0.685)  (0.653)    (0.517)  (0.627)    (0.006)  (0.005)    (0.005)  (0.006)  Potential experience SD    −0.327  −1.633    −0.136  −0.164+    −0.015  −0.023    −0.001  −0.001    (1.817)  (1.811)    (0.088)  (0.091)    (0.017)  (0.018)    (0.001)  (0.001)  N  133  133  133  132  132  132  133  133  133  132  132  132  Robust standard errors in parentheses. +  P < 0.10, * P < 0.05, ** P < 0.01, *** P < 0.001, two-tailed test.  Models include continuous measure of year. ‘Completed tertiary education’ and ‘Educational heterogeneity’ computed from Barro–Lee database individuals aged 25–55, separately by sex. Potential experience measures computed from LIS samples. See Table 2 and online Appendix for description of Labour coordination and EPL scales.  Taiwan EPL values missing. Table A4 replicates results with only Labour coordination scale. Across results, several main conclusions can be drawn. For men and women, the relative contribution of WGI to total inequality varies significantly across welfare regimes. Continental European, Nordic countries, and Taiwan differ significantly from Liberal and East European regimes in their lower relative proportion of WGI (men: 8%, 6%, and 21% lower, women: 5%, 8%, 24%, all P < 0.01). Results are replicated across models with education and potential experience information, and in sensitivity analyses including overall earnings inequality (Table A5). Regime differences are thus not reducible to differences in human capital attainment or levels of total earning variance. Models 7–12 show absolute WGI levels. Continental European, Nordic, and Taiwanese country-years have lower absolute levels of WGI compared to Liberal and East European country-years. Table 4 assesses WGI in relation to country-specific labour market institutional configurations and employment protection laws. Results largely replicate those from Table 3 but reveal an important source of inter- and intra-regime variation. Countries with stronger labour unions, more universal bargaining coverage, and more centralized wage bargaining have both lower levels of WGI and lower proportions of WGI contributing to total inequality. Similarly, stricter EPL is associated with lower WGI and its relative importance, while a gap between regular and temporary contract EPL associates with higher relative female WGI. A standard deviation increase in labour coordination reduces the relative contribution of WGI by about 2.5 percentage points for men, and a standard deviation increase in EPL strictness reduces WGI’s relative contribution by about 3 percentage points for both sexes. For female results, while both institutional arrangements and employment protection laws influence total inequality levels, only employment protection laws reduce the relative proportion of WGI, although both are significant and negatively signed when included separately. Notably, Models 3, 9, 6, and 12 show that regime differences are near fully attributable to labour market institutions and EPL. Results for educational attainment merit special focus. While central to many US studies of WGI, associations between educational attainment and WGI are at best mixed when examined cross-nationally. A positive and significant association between tertiary educational attainment and the relative contribution of WGI is detectable in simple models without controls. Yet the association is removed, and in some cases reversed, with the addition of institutional variables. No consistent conclusion is detectable across sensitivity tests. These contrasting results are suggestive that deinstitutionalization fares better to explain WGI in the cross-national context than returns to skill. In total, results from Tables 3 and 4 provide an important addendum to general conclusions. Although WGI tends to be central to inequality across high-income countries, the magnitude of this fact is channelled through a country’s institutional context. Stronger labour protections tend to reduce not only absolute inequality levels but also the distributional properties of inequality. Conclusion This research assessed cross-national macrolevel within-group earnings inequality, asking the questions: How general is the importance of within-group inequality to inequality change, and how do previous explanations of skill and deinstitutionalization fare in a cross-national context? I applied the logic of decomposing individual market earnings inequality into within-group (WGI) and between-group (BGI) components, predominantly done in the American context, to a sample of full-time prime-age workers in LIS microdata from 28 countries spanning 40 years. Results from this research help clarify the nature of the contemporary rise of individual market earnings inequality in high-income countries. The major descriptive finding of this research is that cross-national differences in prime-age individual market earnings inequality largely stem from within-group differences. Simply put, if one could ‘level’ major occupational, human capital, and sociodemographic characteristics which sociologists typically study in relation to earnings inequality, cross-national patterns of earnings inequality and their changes over time would remain similar. This basic empirical fact was hitherto unknown. Nor were there a priori reasons to suspect that the importance of WGI would be broadly shared across national and historical contexts of high-income countries. My findings therefore provide an important contribution to the understanding of the distributional properties underlying contemporary patterns of earnings inequality in high-income countries. The past 40 years have undergone changes to not only inequality levels but also inequality forms. At the same time, results reveal heterogeneity in the relative importance of WGI across countries in different institutional contexts. Continental European and Nordic countries have significantly lower levels of WGI and proportions of inequality attributable to WGI than Liberal and East European countries, independent of inequality levels and human capital composition. Differences are largely attributable to variation in labour market institutions and employment legislation. These results suggest that a deinstitutionalization explanation of WGI does well in the cross-national context, especially when considered alongside the mixed evidence found for human capital composition. On the one hand, these findings make sense. They follow related stratification literature documenting the importance of labour market institutional configurations across high-income countries (Esping-Andersen, 1999; Mandel and Shalev, 2009). And previous research has documented the varied ways that institutional differences translate into inequality and wage attainment (Gangl, 2004). WGI is another inequality dimension aligning with this research tradition. On the other hand, these findings extend comparative stratification research. Through their influence on flexibility and security, institutional and policy contexts affect not only redistribution and inequality levels but also basic distributional properties of inequality, resulting in variation of forms of inequality across countries (Thelen, 2014). Future research is needed to extend and verify the precise institutional and policy mechanisms that influence the relationship between inequality, WGI, and BGI. Generally, findings highlight the utility of moving beyond one-number summaries and average wage gaps to fully understand the influence of institutional and policy variation on inequality. While results do not support skill return explanations, future research on the specific relationships between WGI, deinstitutionalization, and skill return is needed. A straightforward reading of results is that such explanations generalize poorly beyond the US case, a conclusion that would reflect research by Blau and Kahn (2005). Alternatively, inconsistent results might reflect heterogeneity in the association between WGI and skill returns across institutional arrangements. Its explanatory power may be dependent on deinstitutionalization, suggesting that regime- and/or country-specific studies are needed. More research is needed to assess where, when, and why skill returns are manifest as WGI. Results are suggestive of difficulties faced by actors who focus on closing between-group earnings gaps (Leicht, 2008). WGI and BGI are positively correlated, while BGI is relatively less consequential to total inequality in high-inequality observations (Figure A1). Focus on closing specific between-group gaps while ignoring broader forces that increase within-group earnings inequality might inevitably prove to be unproductive. While group differences in pay tend to be higher in high-inequality observations, they appear to be of secondary importance for general inequality trends. Overall, these results contribute to recent calls for sociologists to complement focus on between-group earnings gaps with other dimensions of inequality (Leicht, 2008, 2016). Findings are drawn from a conservative method of computing WGI across country contexts, with samples restricted to those strongly attached to the labour market. These sampling decisions were made to ensure that results did not simply reflect variation in state intervention across country contexts. It is likely that the inclusion of broader age ranges and employment statuses would bolster results, as would an assessment of earnings after taxes and transfers. Such analyses are necessary to further develop comparative knowledge of WGI. A natural next step for future research is to further examine macrolevel associations between WGI and country-level characteristics. Do alternative measures of skill (such as test scores), social welfare policies, leftist politics, globalization, or demographic shifts better explain variation in WGI than factors used here? These questions are beyond the scope of the current project, as they necessarily build upon the empirical and theoretical extension of WGI to cross-national inequality. However, with the importance of WGI for cross-national inequality documented, future research would do well to more fully unpack the macrolevel causes of this dimension of inequality. This article has important limitations. Perhaps most importantly, the LIS does not provide fine-grained microlevel occupational information. Studies of occupational polarization and wage inequality typically sort individuals into at least 300 occupational groups, or even 300 occupation-by-300 industry groups.27 To what extent are results biased from omitting such fine-grained Gemeinschaft occupational communities (Liu and Grusky, 2013)? The current research cannot say. However, if such an omission were to level WGI differences across countries, this result itself would be a valuable piece of knowledge. For occupational contrasts to run counter to findings of this study, occupational polarization would need to be of greater relative importance in Liberal regimes than in continental Europe, and among high-inequality country-years compared with low-inequality ones. Such a finding would reinforce a deinstitutionalization explanation of WGI, as the importance of fine-grained occupational differentiation would then be secondary to cross-national variation of labour policies and institutions that compress pay differences between workers. Regardless, the current research underscores the importance of continued research of microclass occupations in relation to inequality forms and labour market institutional arrangements. Similarly, how might compositional changes across observable characteristics be responsible for results? The use of counterfactual reweighting methodologies is beyond the scope of the current project (Lemieux, 2006, Western and Rosenfeld, 2011). However, an assessment of the change in WGI in relation to reweighted population compositions held constant over time, and held constant between countries, provides a promising next step for research. More generally, the assessment of WGI in a cross-national context highlights the central importance of a seemingly simple question: What is a group? Does it reflect a substantively meaningful concept that can be similarly assessed across country and historical contexts? Cross-national research has shown the widely variable and context-specific nature of groups based on educational and occupational attainment, for example (Bol and Weeden, 2014). Phenotypic race is an anchor of group membership in the United States, whereas linguistic and immigrant identities play important roles in some European countries, like Belgium and Sweden. In sensitivity analyses, I found that results were robust against alternative modelling decisions at the microlevel with different conceptual approaches to the idea of group. I replicated WGI measures including immigrant status. Results were substantively the same, and correlations of WGI measures with and without immigrant groups were over 0.97.28 Pragmatism guided this project’s definition of group. Definitions were based on previous US studies and data availability. Yet theoretical attention to the concept of group as it relates to economic inequality is clearly needed. WGI is central to cross-national patterns of inequality and of growing importance to high-income countries. Yet substantial institution-based heterogeneity exists. This study provides insights into the basic distributional properties of inequality in high-income countries and reveals the promises of examining the relationship between total inequality and WGI in a cross-national perspective. Tom VanHeuvelen is Assistant Professor of Sociology at the University of Illinois at Urbana-Champaign. Current research interests comprise inequality and stratification, cross-national and comparative sociology, and the sociology of development. His work has been published in the American Journal of Sociology, Social Forces, Social Science Research, and the Oxford University Press. Footnotes 1 Studies diverge in how a ‘group’ is defined (VanHeuvelen, 2018), typically using some combination of human capital, occupation, and demographic characteristics. I use all available microlevel characteristics in the LIS used in previous research. 2 See the online appendix, and VanHeuvelen (2018), for discussion of theoretical meaning of WGI. 3 Luck could also reflect the discretion and discrimination held by a worker’s employer. Future work assessing employer-based discretion, WGI, and BGI is needed. 4 The relationship between an omitted microlevel variable and WGI is unclear. Consider the loss of a union job. Western and Rosenfeld’s results (2011) suggest this would increase WGI. Yet an omitted variable approach suggests a more precise microlevel variable, like personality or skill, is needed for the subsequent wage attainment model. 5 This camp frequently argues that micro-class occupations or jobs represent the bulk of WGI. This critique has been used to reinforce both skill- and deinstitutionalization-based explanations of inequality (Goos and Manning, 2007; Williams, 2012). 6 This restriction also assists with overcoming difficulties associated with unavailable microlevel data, such as union coverage, fine-grained occupation, and annual work hours, all of which may influence the rate of individuals in part-time employment. Note too that these results should provide a conservative estimate of cross-national variation in WGI, as individuals most sensitive to labour market flexibility are excluded from samples. 7 Female results for il86 do not converge due to a small sample size (n = 140). This observation is excluded from analyses, lowering the female sample to 142. 8 The LIS labels it paid employment labour income. 9 I replicate main analyses controlling for average annual work hours from OECD data and reach the same conclusions. 10 See the online Appendix for a comparison and discussion of US wage and earnings inequality. 11 Early samples from the United Kingdom measuring education in years are transformed to align with the categorical measure. 12 Actual experience is not widely available. Countries have different typical starting ages of education. WGI and BGI measurements are unaffected by adjusting measurements to country-specific starting ages. 13 These two measures of human capital follow the logic of Autor, Katz and Kearney (2008) and allow for complex nonlinear patterns of WGI across the age distribution while using relatively small samples. 14 All categories are unavailable in at94, at97, at00, gr95, gr00, and il79. I instead use three-category industry variables: (1) agriculture, (2) industry, and (3) services. Results are substantively similar excluding these cases. 15 Many WGI studies include fine-grained occupations, or industry-by-occupation contrasts (Goos and Manning, 2007; Williams, 2012). Such information is unavailable in the LIS. Implications are discussed in the Conclusion. 16 Three samples, au08, au10, and uk86, do not have 10 category ISCO codes. I therefore use available eight (au) and nine (uk) category country-specific occupational information. Results are unaffected if excluding these cases. 17 All results are replicated using a three-category occupation measure (available upon request). Correlation across WGI measures using 3- or 10-occupation code is 0.99. 18 Individual-level coefficients are not the main focus of this study. Stata do-files that reproduce individual VFR models through the LIS job submission system are available at www.tomvanheuvelen.com. 19 IL86 has a small sample of female workers and so VFR models cannot converge. This sample is dropped from analyses. 20 This resembles the R2 of the microlevel model. I prefer measures following the VFR for reasons outlined by VanHeuvelen (2018). 21 Education and experience at the individual level parse earnings into within and between components. Lemieux (2006) showed that WGI is higher among more educated and experienced workers. Country-level variables account for compositional differences across countries. 22 This category combines Conservative and Mediterranean regimes due to small Mediterranean samples. Results are similar if Conservative and Mediterranean countries are separated. 23 Additional information is included in the online Appendix. 24 Comparison using Bayesian information criterion scores favoured simple regression models predicting total inequality using WGI instead of BGI change. 25 Replication of Figure 3 using female samples (online Appendix) results in similar conclusions. 26 Results in the online Appendix show WGI (levels and proportions) and various measures of post-fisc household income inequality to be significantly and positively correlated. 27 It is unlikely that cross-national variation fully reduces to omitted occupation variables. For example, I conducted supplementary analyses using US Census microdata and estimated models with and without 82,000 occupation-by-industry groups in 2010, in addition to standard education-by-work experience interactions. Including these 82,000 occupational groups decreased the relative contribution of WGI in this data set from 65% to 55%. 28 Race and ethnic information is available in few samples. Supplementary Data Supplementary data are available at ESR online. Acknowledgements The author thanks Arthur S. Alderson, Clem Brooks, Andrew Halpern-Manners, Tim Liao, Monica McDermott, Patricia McManus, and Jane VanHeuvelen for their helpful comments and discussion. Funding This work has been supported by a National Science Foundation dissertation improvement grant (no. 1519186). References Alderson A. S., Nielsen F. ( 2002). Globalization and the great U-turn: income inequality trends in 16 OECD countries. American Journal of Sociology , 107, 1244– 1299. Google Scholar CrossRef Search ADS   Atkinson A., Piketty T., Saez E. ( 2011). Top incomes in the long run of history. Journal of Economic Literature , 49, 3– 71. Google Scholar CrossRef Search ADS   Autor D., Katz L., Kearney M. ( 2008). Trends in US wage inequality: revising the revisionists. Review of Economics and Statistics , 90, 300– 323. Google Scholar CrossRef Search ADS   Barbieri P. ( 2009). Flexible employment and inequality in Europe. European Sociological Review , 25, 621– 628. Google Scholar CrossRef Search ADS   Barbieri P., Cutuli G. ( 2016). Employment protection legislation, labour market dualism, and inequality in Europe. European Sociological Review , 32, 501– 516. Google Scholar CrossRef Search ADS   Barro R., Lee J. ( 2010). A new dataset of educational attainment in the world, 1950-2010. Journal of Development Economics , 104, 184– 198. Google Scholar CrossRef Search ADS   Bidwell M. ( 2011). Paying more to get less: the effects of external hiring versus internal mobility. Administrative Science Quarterly , 56, 369– 407. Google Scholar CrossRef Search ADS   Blau F., Kahn L. ( 2005). Do cognitive test scores explain higher US wage inequality? Review of Economics and Statistics , 87, 184– 193. Google Scholar CrossRef Search ADS   Bollen K. A., Entwisle B., Alderson A. S. ( 1993). Macrocomparative research methods. Annual Review of Sociology , 19, 321– 351. Google Scholar CrossRef Search ADS   Bol T., Weeden K. ( 2014). Occupational closure and wage inequality in Germany and the United Kingdom. European Sociological Review , 31, 354– 369. Google Scholar CrossRef Search ADS   Bradley D. et al.   ( 2003). Distribution and redistribution in postindustrial democracies. World Politics , 55, 193– 228. Google Scholar CrossRef Search ADS   Brady D., Fullerton A., Cross J. M. ( 2010). More than just nickels and dimes: a cross-national analysis of working poverty in affluent democracies. Social Problems , 57, 559– 585. Google Scholar CrossRef Search ADS PubMed  Card D., Dinardo J. ( 2002). Skill-biased technological change and rising wage inequality: some problems and puzzles. Journal of Labor Economics , 20, 733– 783. Google Scholar CrossRef Search ADS   Cheng S. ( 2014). A life course trajectory framework for understanding the intracohort pattern of wage inequality. American Journal of Sociology , 120, 633– 700. Google Scholar CrossRef Search ADS   DiPrete T. ( 2007). What has sociology to contribute to the study of inequality trends? A historical and comparative perspective. American Behavioral Scientist , 50, 603– 618. Google Scholar CrossRef Search ADS   Esping-Andersen G. ( 1999). Social Foundations of Postindustrial Economies . Oxford, UK: Oxford University Press. Google Scholar CrossRef Search ADS   Gallie D. ( 2007). Production regimes and the quality of employment in Europe. Annual Review of Sociology , 33, 85– 104. Google Scholar CrossRef Search ADS   Gangl M. ( 2004). Welfare states and the scar effects of unemployment: a comparative analysis of the United States and West Germany. American Journal of Sociology , 109, 1319– 1364. Google Scholar CrossRef Search ADS   Gebel M., Giesecke J. ( 2011). Labor market flexibility and inequality: the changing skill-based temporary employment and unemployment risks in Europe. Social Forces , 90, 17– 39. Google Scholar CrossRef Search ADS   Goldin C., Katz L. ( 2008). The Race between Education and Technology . Cambridge: Belknap Press of Harvard University Press. Goos M., Manning A. ( 2007). Lousy and lovely jobs: the rising polarization of work in Britain. The Review of Economics and Statistics , 89, 118– 133. Google Scholar CrossRef Search ADS   Gornick J., Milanovic B. ( 2015). Income inequality in the United States in cross-national perspective: redistribution revisited. LIS Center Research Brief , 1– 7. Hauser S., Xie Y. ( 2005). Temporal and regional variation in earnings inequality: urban China in transition between 1988 and 1995. Social Science Research , 34, 44– 79. Google Scholar CrossRef Search ADS   Jaumotte F., Buitron C. ( 2015). Inequality and Labor Market Institutions. IMF Staff Discussion Note . International Monetary Fund, Washington, DC. Jencks C. et al.   ( 1972). Inequality: A Reassessment of the Effect of Family and Schooling in America . New York: Harper. Juhn C., Murphy K., Pierce B. ( 1993). Wage inequality and the rise in returns to skill. Journal of Political Economy , 101, 410– 442. Google Scholar CrossRef Search ADS   Kenworthy L. ( 2004). Egalitarian Capitalism: Jobs, Incomes, and Growth in Affluent Countries . New York: Russell Sage Foundation. Kollmeyer C. ( 2015). Globalization and income inequality: how public sector spending moderates this relationship in affluent countries. International Journal of Comparative Sociology , 56, 3– 28. Google Scholar CrossRef Search ADS   Kristal T., Cohen Y. ( 2016). The causes of rising wage inequality: the race between institutions and technology. Socio-Economic Review , 15, 187– 212. Leicht K. ( 2008). Broken down by race and gender? Sociological explanations of new sources of earnings inequality. Annual Review of Sociology , 34, 237– 255. Google Scholar CrossRef Search ADS   Leicht K. ( 2016). Getting serious about inequality. Sociological Quarterly , 57, 211– 231. Google Scholar CrossRef Search ADS   Lemieux T. ( 2006). Increasing residual wage inequality: composition effects, noisy data, or rising demand for skill? American Economic Review , 96, 461– 498. Google Scholar CrossRef Search ADS   Levy F., Murnane R. ( 1992). United States earnings levels and earnings inequality: a review of recent trends and proposed explanations. Journal of Economic Literature , 30, 1333– 1381. Liao T. ( 2016). Evaluating distributional differences in income inequality. Socius , 2, 1– 14. Liu Y., Grusky D. ( 2013). The payoff to skill in the third industrial revolution. American Journal of Sociology , 118, 1330– 1374. Google Scholar CrossRef Search ADS   Luxembourg Income Study Database ( 2017). http://www.lisdatacenter.org (various countries; May–September 2017). Luxembourg: LIS. McCall L. ( 2000). Explaining levels of within-group wage inequality in US labor markets. Demography , 37, 415– 430. Google Scholar CrossRef Search ADS PubMed  MacLean A., Kleykamp M. ( 2016). Income inequality and the veteran experience. The Annals of the American Academy of Political and Social Science , 663, 99– 116. Google Scholar CrossRef Search ADS   Mandel H., Shalev M. ( 2009). How welfare states shape the gender pay gap: a theoretical and comparative analysis. Social Forces , 87, 1873– 1911. Google Scholar CrossRef Search ADS   Moller S., Alderson A. S., Nielsen F. ( 2009). Changing patterns of income inequality in US counties, 1970–2000. American Journal of Sociology , 114, 1037– 1101. Google Scholar CrossRef Search ADS   Mouw T., Kalleberg A. ( 2010). Occupations and the structure of wage inequality in the United States, 1980s to 2000s. American Sociological Review , 75, 402– 431. Google Scholar CrossRef Search ADS   Organisation for Economic Co-operation and Development (OECD) ( 2015). It Together: Why Less Inequality Benefits All . Paris: OECD Publishing. Sachweh P., Olafsdottir S. ( 2012). The welfare state and equality? Stratification realities and aspirations in three welfare regimes. European Sociological Review , 28, 149– 168. Google Scholar CrossRef Search ADS   Streeck W. ( 2011). E Pluribus Unum? Varieties and commonalities of capitalism. In Granovetter M., Swedberg R. (Eds.), The Sociology of Economic Life , 3rd edn. New York, NY: Westview, pp. 419– 455. Thelen K. ( 2014). Varieties of Liberalization and the New Politics of Social Solidarity . Cambridge, UK: Cambridge University Press. Google Scholar CrossRef Search ADS   VanHeuvelen T. ( 2018). Recovering the missing middle: a mesocomparative analysis of within-group inequality, 1970-2011. American Journal of Sociology , 123, 1064– 1116. Google Scholar CrossRef Search ADS   Visser J., Checchi D. ( 2009). Inequality and the labour market: unions. In Salverda W., Nolan B.,, Smeeding T. (Eds.), The Oxford Handbook of Economic Inequality . Oxford, UK: Oxford University Press. Weil D. ( 2014). The Fissured Workplace: Why Work Became so Bad for so Many and What Can Be Done to Improve It . Cambridge, MA: Harvard University Press. Google Scholar CrossRef Search ADS   Western B., Bloome D. ( 2009). Variance function regressions for studying inequality. Sociological Methodology , 39, 293– 326. Google Scholar CrossRef Search ADS   Western B., Rosenfeld J. ( 2011). Unions, norms, and the rise in U.S. wage inequality. American Sociological Review , 76, 513– 537. Google Scholar CrossRef Search ADS   Western B. et al.   ( 2012). Economic insecurity and social stratification. Annual Review of Sociology , 38, 341– 359. Google Scholar CrossRef Search ADS   Williams M. ( 2012). Occupations and British wage inequality, 1970s–2000s. European Sociological Review , 29, 841– 857. Google Scholar CrossRef Search ADS   Xie Y., Killewald A., Near C. ( 2016). Between- and within-occupation inequality: the case of high-status professions. The Annals of the American Academy of Political and Social Science , 663, 53– 79. Google Scholar CrossRef Search ADS PubMed  Zhou X. ( 2014). Increasing returns to education, changing labor force structure, and the rise of earnings inequality in Urban China, 1996–2010. Social Forces , 93, 429– 455. Google Scholar CrossRef Search ADS   © The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png European Sociological Review Oxford University Press

Within-Group Earnings Inequality in Cross-National Perspective

Loading next page...
 
/lp/ou_press/within-group-earnings-inequality-in-cross-national-perspective-hde8GHi0lV
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
0266-7215
eISSN
1468-2672
D.O.I.
10.1093/esr/jcy011
Publisher site
See Article on Publisher Site

Abstract

Abstract In this research I assess within-group inequality—earnings inequality occurring among otherwise similar individuals based on observed characteristics—in a cross-national comparative perspective. While scholarly interest in the within-group portion of inequality has grown over the past 25 years, virtually all studies focus on the US case. The current research shifts focus by assessing within-group inequality in a cross-national comparative study. I do so by constructing a unique data set of country-level measures of within- and between-group inequality for annual market earnings using Luxembourg Income Study (LIS) microdata from 1.36 million full-time prime-age male and female workers nested in 143 country-years, drawn from 28 countries spanning 40 years. I then document and describe basic between-country and longitudinal trends in the relationship between total inequality and within-group inequality. I find that in nearly all countries in the LIS, within-group inequality is the primary driver of levels and trends in inequality. As inequality increases, so too does the relative importance of within-group inequality. However, substantial cross-national heterogeneity based in labour market institutions and employment protection legislation is found. Theoretical and substantive implications are discussed. The nature of individual earnings inequality growth provides a fundamental challenge to sociological thinking. Many high-income countries have experienced rising inequality in recent decades (Alderson and Nielsen, 2002; OECD, 2015). In the United States, inequality has not only risen in absolute levels but also changed regarding important distributional properties. Most earnings inequality growth since the 1970s has occurred among workers who are otherwise similar along sociodemographic, human capital, and occupational characteristics typically studied by sociologists, that is, within-group inequality (Levy and Murnane, 1992; McCall, 2000; Lemieux, 2006; Autor, Katz and Kearney, 2008, Leicht, 2008; Western and Bloome, 2009; Mouw and Kalleberg, 2010; Western and Rosenfeld, 2011; Cheng, 2014; Zhou, 2014; MacLean and Kleykamp, 2016; Xie, Killewald and Near, 2016; Leicht, 2016; Liao, 2016; VanHeuvelen, 2018). While scholars have increasingly placed within-group inequality at the centre of theoretical and analytical attention, many questions about the nature of this inequality dimension remain unanswered. Research on within-group inequality overwhelmingly focuses on the US case (but see Williams, 2012). While within-group inequality has played a central role in American inequality trends and been used to make theoretical sense of the contemporary upswing of inequality, it remains unknown whether its importance is a unique feature of the American labour market, among similar Liberal welfare regimes (Esping-Andersen, 1999), or whether within-group inequality’s contribution to inequality growth applies generally across high-income countries. Cross-national comparison provides an ideal opportunity to better understand the relationship between within-group inequality and total inequality change. The macrocomparative stratification literature illustrates the benefits of assessing inequality across countries with diverse inequality legacies, labour market institutions, policy systems, and sociodemographic compositions (Bollen, Entwisle and Alderson, 1993; Alderson and Nielsen, 2002; Bradley et al., 2003; Kenworthy, 2004; Thelen, 2014; Kollmeyer, 2015). I apply these insights to the study of within-group inequality and ask the following: How general is the importance of within-group inequality to inequality change, and how do previous explanations of skill development and deinstitutionalization fare in a cross-national context? This study uses 10 waves of Luxembourg Income Study (Luxembourg Income Study Database, 2017) microdata from 768,549 full-time prime-age male and 591,008 full-time prime-age female workers to construct a unique data set of 285 macrolevel within-group inequality distributions in an unbalanced sample of 28 countries spanning 40 years. Within-group inequality is computed from identical variance function regression (VFR) models on identical sets of microlevel grouping variables used in previous studies of market earnings, yielding the most extensive collection of country-level within-group inequality observations used in a study to date. Two main conclusions are drawn. First, total inequality levels, and inequality growth, are primarily due to within-group inequality. Second, despite the general importance of within-group inequality, there exists substantial heterogeneity across institutional contexts. Liberal and East European regimes have higher levels and contributions of within-group inequality compared with continental European and Nordic countries. Such variation is largely attributable to labour market institutional and policy arrangements, but not human capital differences. This study has important implications. First, it clarifies the relationship between within-group inequality and individual market earnings inequality growth, revealing that most earnings inequality change in high-income countries has occurred through the within-group component. Second, it sheds light on the relationship between total inequality and within-group inequality, showing that institutional and policy differences produce sizable variation in the levels and relative importance of within-group inequality. Third, it highlights the heterogeneity of the sources of inequality across country contexts, even those with the same absolute level of inequality. Two countries with similar levels of total inequality may have substantially different types of inequality, which can be revealed by focusing on within-group inequality. Background Within-group inequality (henceforth WGI)—the variance of individual earnings net of sociodemographic, human capital, and occupational characteristics that sociologists and economists typically study—has been the subject of scholarly attention for over 25 years (Levy and Murnane, 1992; Juhn, Murphy and Pierce, 1993; Card and Dinardo, 2002; Lemieux, 2006; Autor, Katz and Kearney, 2008; Western and Bloome, 2009; Zhou, 2014; VanHeuvelen, 2018). Suppose one could ‘level’ average pay differences between those attaining secondary and tertiary levels of education, workers of different experience levels, and across different industries and occupational categories, so that averages across these social categories were indistinguishable from zero. Inequality would necessarily decline. Yet beginning in the early 1990s, scholars observed that even under these conditions—and even when applied across all the usual suspects of worker and job characteristics observable to the analyst—upwards of ⅔ of inequality levels and 70% of inequality change over time would remain (Juhn, Murphy and Pierce, 1993; Autor, Katz and Kearney, 2008; Liu and Grusky, 2013; but see Mouw and Kalleberg, 2010).1 These patterns motivated the following question: Why would inequality grow independently of observable worker characteristics? The answer to this question has led to distinct interpretations of the nature of the contemporary upswing of inequality. Interpretations of empirical patterns and substantive meaning of WGI have been used to adjudicate across major theories of the inequality upswing.2 Explanations of WGI fall into three theoretical camps. First, some argue it reflects returns to skills and talents observable to employers, but not analysts (Juhn, Murphy and Pierce, 1993; Autor, Katz and Kearney, 2008; Goldin and Katz, 2008; Mouw and Kalleberg, 2010). Technological innovation has complemented the job duties of highly skilled workers, largely concentrated among professional and managerial occupations, allowing them to enhance productivity and manage increasingly complex organizations. WGI thus reflects growth in economic returns to highly skilled, highly educated workers in cognitively demanding occupations who have increased their productivity and, thus, pay. Similarly, experienced workers tend to have higher WGI due to uneven skill development, on-the-job training, and occupational mobility (Lemieux, 2006). Simply put, more educated and experienced workforces should have higher WGI. Second is deinstitutionalization, or the decline of institutional protection and stability of pay due to broadly shared changes in labour market policies, pay-setting institutions, and reduction of internal, firm-level job security, mobility, full-time employment, and benefits (McCall, 2000; Western et al., 2012; Western and Rosenfeld, 2011). In the United States, attention focuses on (1) declining internal labour markets and the fissuring of employees across an increasingly complex organizational landscape (Bidwell, 2011; Weil, 2014) and (2) falling union density and political power. Union decline, pronounced in the United States but broadly experienced across high-income countries (Visser and Checchi, 2009), has resulted in well-documented declines in wage attainment and increases in wage inequality and has also been shown to increase WGI (Western and Rosenfeld, 2011; Jaumotte and Buitron, 2015; Kristal and Cohen, 2016). This perspective conceptualizes WGI as an indicator of insecurity, or the risk of economic loss in the face of unpredictability (Western et al., 2012) and flexibility, the ease and ability of management to create insecurity via wage adjustments and job termination. In total, labour market institutional arrangements and policy systems that increase labour security and decrease flexibility should decrease WGI. Third is methodological. Some scholars explicitly assume WGI to indicate luck and free will (Jencks et al., 1972),3 random error (Cheng, 2014), or omitted variables, such as occupational differentiation (Mouw and Kalleberg, 2010).4 From this viewpoint, WGI is beyond the scope of analytical focus, or else indicates the need for a more precise microlevel model.5 Given its theoretical importance, sociologists have developed methodologies situating WGI in the centre of analysis (Western and Bloome, 2009; Cheng, 2014; Zhou, 2014; Liao, 2016), elaborated theories of WGI causes (McCall, 2000; Western et al., 2012; Western and Bloome, 2009; Cheng, 2014; Zhou, 2014), and called for further examination of dimensions of inequality beyond one-number measures of overall inequality such as the Gini coefficient and between-group gaps (Leicht, 2008, 2016). However, virtually all studies have restricted focus to a single-country study, predominantly the United States (but see Zhou, 2014 for the Chinese case and Williams, 2012 for the British case). Extending WGI research to a cross-national sample is useful for at least two reasons. First, analysis of WGI can reveal variation in the distributional properties of inequality. Many cross-national inequality studies implicitly assume that country-level inequality measures at the same level imply equivalent ‘types’ of inequality. Yet a focus on the relationship between total inequality and WGI can reveal meaningful variation across countries with similar levels of inequality. The substantive reality of inequality may vary markedly depending on how tightly wage setting is conducted in relation to human capital and occupational characteristics. Second, cross-national analyses have routinely yielded insights into basic features of inequality. For example, scholars have shown how tax and transfer policies lead to cross-national variation in redistribution (Gornick and Milanovic, 2015) and post-fisc inequality (Bradley et al., 2003). Economists have discovered substantial heterogeneity in historical inequality trajectories between English-speaking nations, continental Europe, and Nordics (Atkinson, Piketty and Saez, 2011). Similarly, sociologists and political scientists have documented the wide variation of social policies and labour market institutions that create norms and regulations for how wage setting and adjustments occur (Bradley et al., 2003; Kenworthy, 2004; Gallie, 2007; Barbieri, 2009). Cross-national research reveals general features of stratification processes not fully reducible to idiosyncratic features of a single country context (Blau and Kahn, 2005; Brady, Fullerton and Cross, 2010). Such analytical approaches have not been incorporated into the study of WGI. Yet to properly understand this component of inequality, one must situate it in a comparative perspective. The cross-national level provides a logical location to empirically assess WGI theories. For example, Blau and Kahn (2005) showed that the variation of test scores, a proxy for a workforce’s skill distribution, poorly explained between-country inequality differences, although others find returns to education and skill to be important across high-income countries (DiPrete, 2007; Goos and Manning, 2007). A skills return explanation suggests that countries with more educated and older workforces should have higher WGI. Regarding deinstitutionalization, Esping-Andersen’s regime typology provides a canonical account of the variation of market dynamics across country institutional contexts (1999). Liberal regimes—largely defined by residual states, means tested support, low labour protection, and fragmented bargaining—tend to have higher inequality than Conservative regimes—defined by male breadwinner insurance systems, occupational status distinctions, low mobility, more centralized wage setting coordination, and insider/outside cleavages—and Social Democratic regimes—defined by universal insurance systems, equal opportunity, broad systems of coordination, and low inequality (Esping-Andersen, 1999; Mandel and Shalev, 2009; Sachweh and Olafsdottir, 2012). Insofar as WGI represents returns to skill, insecurity, flexibility, and/or luck, then WGI should be more consequential in Liberal regimes than either Conservative or Social Democratic ones. Numerous theoretical traditions move beyond regime categorization, identifying specific methods of coordination, labour market institutional arrangements, or social policies that create different forms of wage setting, insecurity, flexibility, and, thus, inequality (Kenworthy, 2004; Gallie, 2007; Gebel and Giesecke, 2011; Streeck, 2011). Two such factors are relevant to WGI deinstitutionalization theories. First, countries vary in labour union membership, bargaining coverage, and centralization of wage setting coordination. In addition to the importance of union power discussed earlier (Western and Rosenfeld, 2011; Jaumotte and Buitron, 2015), centralized coordination provides standards of wage setting across firms, decreases interfirm wage setting via discretion and occupation mobility, and encourages industry- and firm-specific skill development, which in turn influences predictable wage attainment, employment tenure, and secure employment protection (Kenworthy, 2004; Gallie, 2007; Streeck, 2011; Thelen, 2014). Insofar as such institutional arrangements provide workers economic security, they should be negatively associated with WGI. Second, countries vary in their restrictions to labour flexibility. Long viewed as the culprit of Eurosclerosis (Barbieri, 2009), strict employment protection legislation (EPL) lowers the ability of firms to adjust wages in response to external economic shocks. Some argue for a blunt line dividing flexible, unequal, and high employment United States and inflexible, equal, high unemployment Europe. Recent studies identify flexibility as occurring across multiple dimensions (Barbieri, 2009). Countries differ in the broad or targeted application of flexibility across certain population segments. Particularly on continental Europe, changes to EPL largely targeted ‘outsider’, marginal groups, bifurcating the labour force into protected, secure, and egalitarian insiders and flexible and insecure outsiders (Gebel and Giesecke, 2011; Thelen, 2014). Thus, employment protection takes on two dimensions: the general extent of EPL reducing flexibility and contrasting EPL across contract types (Gebel and Giesecke, 2011; Barbieri and Cutuli, 2016). Greater flexibility in both cases should associate with higher WGI. Overall, cross-national variation of labour institutions and employment policies provides an ideal opportunity to test and extend deinstitutionalization theories. Data and Methods WGI is calculated from national surveys collected in the LIS, widely considered the gold standard of comparable cross-national income data. The LIS includes harmonized and nationally representative microdata on worker and job characteristics, including income, demographics, human capital, and occupation. The high quality of income measurements and rigorous harmonization of microdata undertaken by the LIS are crucial for this research: WGI can be computed on the same earnings measure using identical individual-level covariates across countries over a long period. I use all country surveys with available microdata that are either (1) high-income countries or (2) on the European continent. Samples Country samples are listed in Table 1. Following standard practice of WGI research, I conduct separate analyses by sex (Lemieux, 2006; Autor, Katz and Kearney, 2008; Western and Rosenfeld, 2011), which accounts for the different historical socioeconomic attainment trajectories and barriers of employment and wage attainment faced by men and women during the period studied (Esping-Andersen, 1999). I restrict samples to full-time workers to minimize potential bias introduced from cross-national differences in the frequency of part-time employment.6 Samples are restricted to prime-age workers aged 25–54. While more conservative than other studies using LIS microdata (Mandel and Shalev, 2009; Brady, Fullerton and Cross, 2010), this decision guards against potential confounding selection effects related to decommodification—such as educational training, family support, and retirement—which occurs unevenly between countries. Following similar studies, self-employed individuals are dropped (Autor, Katz and Kearney, 2008). Microlevel analyses are conducted using survey weights provided by the LIS. In total, samples include 768,549 male and 591,008 female workers. Microdata are nested in an unbalanced sample of 143 country-year inequality observations per sex, in 28 countries spanning 40 years.7 The country-year is the unit of analysis below. Table 1. Country samples Countrya  Years  Country observations  Male observations  Female observations  Austria (AT)  1994, 1997, 2000, 2004, 2013  5  7,183  3,324  Australia (AU)  1985, 1989, 2008, 2010  4  16,268  8,519  Belgium (BE)  1997, 2000  2  2,528  1,353  Canada (CA)  1987, 1991, 1994, 1997, 1998, 2000, 2004, 2007, 2010  9  81,391  64,549  Switzerland (CH)  2007, 2010, 2013  3  6,408  2,890  Czech Republic (CZ)  1996, 2004, 2007, 2010, 2013  5  21,912  19,675  Germany (DE)  1984, 1989, 1994, 2000, 2004, 2007, 2010, 2013  8  25,376  13,179  Denmark (DK)  2004, 2007, 2010, 2013  4  72,092  66,923  Estonia (EE)  2004, 2007, 2010, 2013  4  6,612  6,796  Spain (ES)  2000, 2004, 2007, 2010, 2013  5  18,670  12,829  Finland (FI)  1987, 1991, 1995, 2000, 2004, 2007, 2010, 2013  8  24,257  24,154  France (FR)  2005, 2010  2  4,076  3,023  Greece (GR)  1995, 2000, 2004, 2007, 2010, 2013  6  8,253  5,659  Hungary (HU)  1991, 1994, 1999, 2005  4  1,983  2,013  Ireland (IE)  1994, 1995, 2000, 2004, 2007, 2010  6  6,393  3,974  Israel (IL)  1979, 1986, 1992, 1997, 2001, 2005  6  10,756  5,951  Iceland (IS)  2004, 2007, 2010  3  3,521  2,904  Italy (IT)  2004, 2008, 2010  3  6,939  4,467  Lithuania (LT)  2010, 2013  2  2,570  2,932  Luxembourg (LU)  1997, 2000, 2004, 2007, 2010, 2013  6  10,122  4,885  Netherlands (NL)  1990, 2004, 2007, 2010, 2013  5  7,916  2,298  Poland (PL)  2007, 2010, 2013  3  39,525  34,317  Russia (RU)  2004, 2007, 2010, 2013  4  5,471  6,705  Slovenia (SI)  1999, 2004, 2007, 2010, 2012  5  8,395  8,303  Slovakia (SK)  2004, 2007, 2010, 2013  4  9,236  9,551  Taiwan (TW)  1981, 1986, 1991, 1995, 1997, 2000, 2005, 2007, 2010, 2013  10  71,221  47,701  United Kingdom (UK)  1986, 1999, 2004, 2007, 2010, 2013  6  38,208  24,726  United States (US)  1974, 1979, 1986, 1991, 1994, 1997, 2000, 2004, 2007, 2010, 2013  11  251,267  197,408  Total     143  768,549  591,008  Countrya  Years  Country observations  Male observations  Female observations  Austria (AT)  1994, 1997, 2000, 2004, 2013  5  7,183  3,324  Australia (AU)  1985, 1989, 2008, 2010  4  16,268  8,519  Belgium (BE)  1997, 2000  2  2,528  1,353  Canada (CA)  1987, 1991, 1994, 1997, 1998, 2000, 2004, 2007, 2010  9  81,391  64,549  Switzerland (CH)  2007, 2010, 2013  3  6,408  2,890  Czech Republic (CZ)  1996, 2004, 2007, 2010, 2013  5  21,912  19,675  Germany (DE)  1984, 1989, 1994, 2000, 2004, 2007, 2010, 2013  8  25,376  13,179  Denmark (DK)  2004, 2007, 2010, 2013  4  72,092  66,923  Estonia (EE)  2004, 2007, 2010, 2013  4  6,612  6,796  Spain (ES)  2000, 2004, 2007, 2010, 2013  5  18,670  12,829  Finland (FI)  1987, 1991, 1995, 2000, 2004, 2007, 2010, 2013  8  24,257  24,154  France (FR)  2005, 2010  2  4,076  3,023  Greece (GR)  1995, 2000, 2004, 2007, 2010, 2013  6  8,253  5,659  Hungary (HU)  1991, 1994, 1999, 2005  4  1,983  2,013  Ireland (IE)  1994, 1995, 2000, 2004, 2007, 2010  6  6,393  3,974  Israel (IL)  1979, 1986, 1992, 1997, 2001, 2005  6  10,756  5,951  Iceland (IS)  2004, 2007, 2010  3  3,521  2,904  Italy (IT)  2004, 2008, 2010  3  6,939  4,467  Lithuania (LT)  2010, 2013  2  2,570  2,932  Luxembourg (LU)  1997, 2000, 2004, 2007, 2010, 2013  6  10,122  4,885  Netherlands (NL)  1990, 2004, 2007, 2010, 2013  5  7,916  2,298  Poland (PL)  2007, 2010, 2013  3  39,525  34,317  Russia (RU)  2004, 2007, 2010, 2013  4  5,471  6,705  Slovenia (SI)  1999, 2004, 2007, 2010, 2012  5  8,395  8,303  Slovakia (SK)  2004, 2007, 2010, 2013  4  9,236  9,551  Taiwan (TW)  1981, 1986, 1991, 1995, 1997, 2000, 2005, 2007, 2010, 2013  10  71,221  47,701  United Kingdom (UK)  1986, 1999, 2004, 2007, 2010, 2013  6  38,208  24,726  United States (US)  1974, 1979, 1986, 1991, 1994, 1997, 2000, 2004, 2007, 2010, 2013  11  251,267  197,408  Total     143  768,549  591,008  a Welfare regime categorization is listed in Table 2. Table 1. Country samples Countrya  Years  Country observations  Male observations  Female observations  Austria (AT)  1994, 1997, 2000, 2004, 2013  5  7,183  3,324  Australia (AU)  1985, 1989, 2008, 2010  4  16,268  8,519  Belgium (BE)  1997, 2000  2  2,528  1,353  Canada (CA)  1987, 1991, 1994, 1997, 1998, 2000, 2004, 2007, 2010  9  81,391  64,549  Switzerland (CH)  2007, 2010, 2013  3  6,408  2,890  Czech Republic (CZ)  1996, 2004, 2007, 2010, 2013  5  21,912  19,675  Germany (DE)  1984, 1989, 1994, 2000, 2004, 2007, 2010, 2013  8  25,376  13,179  Denmark (DK)  2004, 2007, 2010, 2013  4  72,092  66,923  Estonia (EE)  2004, 2007, 2010, 2013  4  6,612  6,796  Spain (ES)  2000, 2004, 2007, 2010, 2013  5  18,670  12,829  Finland (FI)  1987, 1991, 1995, 2000, 2004, 2007, 2010, 2013  8  24,257  24,154  France (FR)  2005, 2010  2  4,076  3,023  Greece (GR)  1995, 2000, 2004, 2007, 2010, 2013  6  8,253  5,659  Hungary (HU)  1991, 1994, 1999, 2005  4  1,983  2,013  Ireland (IE)  1994, 1995, 2000, 2004, 2007, 2010  6  6,393  3,974  Israel (IL)  1979, 1986, 1992, 1997, 2001, 2005  6  10,756  5,951  Iceland (IS)  2004, 2007, 2010  3  3,521  2,904  Italy (IT)  2004, 2008, 2010  3  6,939  4,467  Lithuania (LT)  2010, 2013  2  2,570  2,932  Luxembourg (LU)  1997, 2000, 2004, 2007, 2010, 2013  6  10,122  4,885  Netherlands (NL)  1990, 2004, 2007, 2010, 2013  5  7,916  2,298  Poland (PL)  2007, 2010, 2013  3  39,525  34,317  Russia (RU)  2004, 2007, 2010, 2013  4  5,471  6,705  Slovenia (SI)  1999, 2004, 2007, 2010, 2012  5  8,395  8,303  Slovakia (SK)  2004, 2007, 2010, 2013  4  9,236  9,551  Taiwan (TW)  1981, 1986, 1991, 1995, 1997, 2000, 2005, 2007, 2010, 2013  10  71,221  47,701  United Kingdom (UK)  1986, 1999, 2004, 2007, 2010, 2013  6  38,208  24,726  United States (US)  1974, 1979, 1986, 1991, 1994, 1997, 2000, 2004, 2007, 2010, 2013  11  251,267  197,408  Total     143  768,549  591,008  Countrya  Years  Country observations  Male observations  Female observations  Austria (AT)  1994, 1997, 2000, 2004, 2013  5  7,183  3,324  Australia (AU)  1985, 1989, 2008, 2010  4  16,268  8,519  Belgium (BE)  1997, 2000  2  2,528  1,353  Canada (CA)  1987, 1991, 1994, 1997, 1998, 2000, 2004, 2007, 2010  9  81,391  64,549  Switzerland (CH)  2007, 2010, 2013  3  6,408  2,890  Czech Republic (CZ)  1996, 2004, 2007, 2010, 2013  5  21,912  19,675  Germany (DE)  1984, 1989, 1994, 2000, 2004, 2007, 2010, 2013  8  25,376  13,179  Denmark (DK)  2004, 2007, 2010, 2013  4  72,092  66,923  Estonia (EE)  2004, 2007, 2010, 2013  4  6,612  6,796  Spain (ES)  2000, 2004, 2007, 2010, 2013  5  18,670  12,829  Finland (FI)  1987, 1991, 1995, 2000, 2004, 2007, 2010, 2013  8  24,257  24,154  France (FR)  2005, 2010  2  4,076  3,023  Greece (GR)  1995, 2000, 2004, 2007, 2010, 2013  6  8,253  5,659  Hungary (HU)  1991, 1994, 1999, 2005  4  1,983  2,013  Ireland (IE)  1994, 1995, 2000, 2004, 2007, 2010  6  6,393  3,974  Israel (IL)  1979, 1986, 1992, 1997, 2001, 2005  6  10,756  5,951  Iceland (IS)  2004, 2007, 2010  3  3,521  2,904  Italy (IT)  2004, 2008, 2010  3  6,939  4,467  Lithuania (LT)  2010, 2013  2  2,570  2,932  Luxembourg (LU)  1997, 2000, 2004, 2007, 2010, 2013  6  10,122  4,885  Netherlands (NL)  1990, 2004, 2007, 2010, 2013  5  7,916  2,298  Poland (PL)  2007, 2010, 2013  3  39,525  34,317  Russia (RU)  2004, 2007, 2010, 2013  4  5,471  6,705  Slovenia (SI)  1999, 2004, 2007, 2010, 2012  5  8,395  8,303  Slovakia (SK)  2004, 2007, 2010, 2013  4  9,236  9,551  Taiwan (TW)  1981, 1986, 1991, 1995, 1997, 2000, 2005, 2007, 2010, 2013  10  71,221  47,701  United Kingdom (UK)  1986, 1999, 2004, 2007, 2010, 2013  6  38,208  24,726  United States (US)  1974, 1979, 1986, 1991, 1994, 1997, 2000, 2004, 2007, 2010, 2013  11  251,267  197,408  Total     143  768,549  591,008  a Welfare regime categorization is listed in Table 2. Individual Earnings I assess logged annual earnings: monetary and non-monetary payments received in counterpart for dependent employment.8 This measure excludes self-employment, capital, and transfer incomes, avoiding confounding country differences in state institutions, capital markets, and self-employment opportunities. This inequality measure reflects market earnings before taxes and transfers. I apply the logic of LIS household income top- and bottom-coding strategies to this measure of individual-level earnings. Low earnings are bottom-coded at one percent of the mean, and high values are top-coded at 10 times the median income, separately by country and survey. I use logged earnings following previous WGI studies. However, this decision likely create a conservative account of actually existing inequality. I analyse individuals rather than households to align with most previous WGI studies. Many WGI studies examine individual hourly wages weighted by working hours (Autor, Katz and Kearney, 2008; VanHeuvelen, 2018), but necessary microdata for hourly wages are not widely available.9,10 Individual Grouping Variables I separate logged annual earnings variance into between- and within-group components using eight human capital, occupational, and sociodemographic characteristics commonly used in WGI research and widely available in the LIS high-income and European samples. Analyses are conducted separately by sex. I measure education following Brady, Fullerton and Cross (2010). Categories are based on the International Standard Classification of Education (ISCED) and include three groups: low (less than secondary education: Levels 0, 1, and 2), medium (secondary or some tertiary education: Levels 3 and 4), and high (completed tertiary education or more: Levels 5 and 6).11Potential experience is the respondent’s age minus potential years of education minus six.12 I measure potential experience two ways. First is a categorical measure differentiating 0–9, 10–19, 20–29, and 30 or more years (Autor, Katz and Kearney, 2008). Second is a continuous measure and its squared term interacted with education categories.13Industry includes (1) agriculture, forestry, and fishing; (2) mining and quarrying, manufacturing, and utilities; (3) construction; (4) wholesale and retail trade, repair, hotels and restaurants; (5) transport, storage, and communications; (6) financial intermediation; (7) real estate, renting and business activities; (8) public administration, education, health and social work; and (9) other community, social/personal services, activities of household, and extra-territorial.14Occupation includes 10 International Standard Classification of Occupations (ISCO) categories: (1) managers; (2) professionals; (3) technicians and associate professionals; (4) clerical support workers; (5) service and sales workers; (6) skilled agricultural, forestry, and fishery workers; (7) craft and related trades workers; (8) plant and machine operators, and assemblers; (9) elementary occupations and; (10) armed forces occupations.15,16,17 For family status, I measure the respondent's partnership status, whether they have young children in the household, and the number of children in their household. Methods WGI is computed using VFR models. VFRs estimate between and within portions of an outcome’s variance (for extended discussion, see Western and Bloome, 2009). The first portion of the VFR is a linear regression of logged income yi on variables Wi (education, potential experience, industry, occupation, partnership status, and young children in household):   yi=βWi+ϵi (1) The second portion of the VFR is a gamma regression with a log-link function estimated on the squared residuals, ε2, predicted from equation (1). This second portion estimates the systematic component of residuals occurring among observed characteristics:   log  ϵ^i2=πWi (2) Predicted values from equation (2) are used as weights to re-estimate equation (1). Squared residuals are recomputed and equation (2) is re-estimated. The process reiterates until model parameters stabilize.18 These equations separate total earnings variance into a between and a within component, which together necessarily sum to the overall variance of the outcome. Following standard practice, VFRs are estimated separately by year (Lemieux, 2006; Autor, Katz and Kearney, 2008; Western and Rosenfeld, 2011). I assume countries have distinct earning regimes and so estimate VFRs by country (Hauser and Xie, 2005). In total, I compute 285 country-year distributions for both WGI and BGI in main analyses.19 These macrolevel distributions are used as the main dependent variable in all analyses below. In addition to using absolute levels of WGI, I also use the relative proportion of WGI to total inequality, which is simply the ratio of these two measures, multiplied by 100. This measure indicates the relative importance of WGI to total inequality in a particular country-year.20 Country Variables To descriptively assess the relative variation of WGI across countries, I include three sets of country-year measures. The first is human capital attainment. These measure (i) country-specific averages of the educational categories drawn from the LIS microdata samples discussed above, (ii) mean potential experience, and (iii) the variance of potential experience. I also compute the percentage of prime-aged workers who completed tertiary education and educational attainment heterogeneity (Moller, Alderson and Nielsen, 2009) using the Barro–Lee Educational Attainment Data (Barro and Lee, 2010).21 Second, I sort country-years into five regime types frequently used to identify general institutional differences across state policies and labour market institutional arrangements: Liberal, Continental European,22 Nordic, Eastern European, and Taiwan (Table 2 lists categorization). Third, I include three scales measuring labour market institutional arrangements and employment policies that apply the deinstitutionalization emphasis of insecurity and flexibility to the cross-national level. The first combines information on trade union membership, wage setting coordination, and adjusted bargaining coverage. The second scale measures the general strictness of EPL for regular and temporary workers. The third measures the gap in EPL between regular and temporary workers (Barbieri and Cutuli, 2016).23Table 2 includes country-level descriptive statistics. Table 2. Descriptive statistics Variable  Mean  SD  Minimum  Maximum  Male sample           Within-group inequality (WGI)  0.224  0.106  0.064  0.576   Relative contribution of WGI to total inequality  68.640  9.421  40.008  86.601   Logged variance of annual earnings  0.320  0.129  0.116  0.709   Percent high education category (ISCED 5 and 6, LIS)  0.311  0.115  0.089  0.618   Potential experience, mean (LIS)  21.103  1.220  18.693  24.952   Potential experience, standard deviation (LIS)  8.938  0.373  8.087  9.872   Percent tertiary educational attainment (Barro–Lee)  0.176  0.071  0.063  0.337   Educational heterogeneity (Barro–Lee)  0.925  0.130  0.509  1.096  Female sample           Within-group inequality  0.241  0.112  0.049  0.546   Relative contribution of WGI to total inequality  70.174  9.380  41.151  96.342   Logged variance of annual earnings  0.333  0.135  0.079  0.689   Percent high education category (ISCED 5 and 6, LIS)  0.381  0.147  0.098  0.703   Potential experience, mean (LIS)  20.426  1.678  15.874  25.282   Potential experience, standard deviation (LIS)  9.221  0.436  8.091  10.306   Percent tertiary educational attainment (Barro–Lee)  0.165  0.076  0.022  0.342   Educational heterogeneity (Barro–Lee)  0.918  0.115  0.511  1.098  Labour market institutions, employment policies           Liberal regimea  0.25         Continental European regime  0.36         Nordic regime  0.10         Eastern European regime  0.22         Taiwan  0.07            Labour coordinationa  0.00  1.442  −2.516  2.922    Wage setting coordination  2.71  1.305  1  5    Trade union membership  34.32  19.99  6.41  95.16    Adjusted bargaining coverage  57.11  26.28  10.70  100   EPL: Total strictnessb  0.00  1.237  −2.252  2.514   EPL: Regular/temporary employment protection gapb  0.00  0.685  −1.947  1.668    EPL: Regular contracts  2.036  0.812  0.257  3.306    EPL: Temporary contracts  1.498  1.241  0.025  5  Male observations: 143 Female observations: 142  Variable  Mean  SD  Minimum  Maximum  Male sample           Within-group inequality (WGI)  0.224  0.106  0.064  0.576   Relative contribution of WGI to total inequality  68.640  9.421  40.008  86.601   Logged variance of annual earnings  0.320  0.129  0.116  0.709   Percent high education category (ISCED 5 and 6, LIS)  0.311  0.115  0.089  0.618   Potential experience, mean (LIS)  21.103  1.220  18.693  24.952   Potential experience, standard deviation (LIS)  8.938  0.373  8.087  9.872   Percent tertiary educational attainment (Barro–Lee)  0.176  0.071  0.063  0.337   Educational heterogeneity (Barro–Lee)  0.925  0.130  0.509  1.096  Female sample           Within-group inequality  0.241  0.112  0.049  0.546   Relative contribution of WGI to total inequality  70.174  9.380  41.151  96.342   Logged variance of annual earnings  0.333  0.135  0.079  0.689   Percent high education category (ISCED 5 and 6, LIS)  0.381  0.147  0.098  0.703   Potential experience, mean (LIS)  20.426  1.678  15.874  25.282   Potential experience, standard deviation (LIS)  9.221  0.436  8.091  10.306   Percent tertiary educational attainment (Barro–Lee)  0.165  0.076  0.022  0.342   Educational heterogeneity (Barro–Lee)  0.918  0.115  0.511  1.098  Labour market institutions, employment policies           Liberal regimea  0.25         Continental European regime  0.36         Nordic regime  0.10         Eastern European regime  0.22         Taiwan  0.07            Labour coordinationa  0.00  1.442  −2.516  2.922    Wage setting coordination  2.71  1.305  1  5    Trade union membership  34.32  19.99  6.41  95.16    Adjusted bargaining coverage  57.11  26.28  10.70  100   EPL: Total strictnessb  0.00  1.237  −2.252  2.514   EPL: Regular/temporary employment protection gapb  0.00  0.685  −1.947  1.668    EPL: Regular contracts  2.036  0.812  0.257  3.306    EPL: Temporary contracts  1.498  1.241  0.025  5  Male observations: 143 Female observations: 142  Israel-1986 dropped from female sample due to small sample size. EPL measures not available in Taiwan (n = 133 men, 132 women). a Liberal: Australia, Canada, Ireland, Great Britain, and United States. Continental European: combination of Conservative—Austria, Belgium, Switzerland, Germany, France, Israel (Mandel and Shalev 2009), Luxembourg, and The Netherlands—and Mediterranean—Greece, Italy, and Spain—regimes. Nordic: Denmark, Finland, and Iceland. East European: Czech Republic, Estonia, Hungary, Lithuania, Poland, Russia, Slovenia, and Slovak Republic. b Scaled created from indented items below. See online Appendix for details. Table 2. Descriptive statistics Variable  Mean  SD  Minimum  Maximum  Male sample           Within-group inequality (WGI)  0.224  0.106  0.064  0.576   Relative contribution of WGI to total inequality  68.640  9.421  40.008  86.601   Logged variance of annual earnings  0.320  0.129  0.116  0.709   Percent high education category (ISCED 5 and 6, LIS)  0.311  0.115  0.089  0.618   Potential experience, mean (LIS)  21.103  1.220  18.693  24.952   Potential experience, standard deviation (LIS)  8.938  0.373  8.087  9.872   Percent tertiary educational attainment (Barro–Lee)  0.176  0.071  0.063  0.337   Educational heterogeneity (Barro–Lee)  0.925  0.130  0.509  1.096  Female sample           Within-group inequality  0.241  0.112  0.049  0.546   Relative contribution of WGI to total inequality  70.174  9.380  41.151  96.342   Logged variance of annual earnings  0.333  0.135  0.079  0.689   Percent high education category (ISCED 5 and 6, LIS)  0.381  0.147  0.098  0.703   Potential experience, mean (LIS)  20.426  1.678  15.874  25.282   Potential experience, standard deviation (LIS)  9.221  0.436  8.091  10.306   Percent tertiary educational attainment (Barro–Lee)  0.165  0.076  0.022  0.342   Educational heterogeneity (Barro–Lee)  0.918  0.115  0.511  1.098  Labour market institutions, employment policies           Liberal regimea  0.25         Continental European regime  0.36         Nordic regime  0.10         Eastern European regime  0.22         Taiwan  0.07            Labour coordinationa  0.00  1.442  −2.516  2.922    Wage setting coordination  2.71  1.305  1  5    Trade union membership  34.32  19.99  6.41  95.16    Adjusted bargaining coverage  57.11  26.28  10.70  100   EPL: Total strictnessb  0.00  1.237  −2.252  2.514   EPL: Regular/temporary employment protection gapb  0.00  0.685  −1.947  1.668    EPL: Regular contracts  2.036  0.812  0.257  3.306    EPL: Temporary contracts  1.498  1.241  0.025  5  Male observations: 143 Female observations: 142  Variable  Mean  SD  Minimum  Maximum  Male sample           Within-group inequality (WGI)  0.224  0.106  0.064  0.576   Relative contribution of WGI to total inequality  68.640  9.421  40.008  86.601   Logged variance of annual earnings  0.320  0.129  0.116  0.709   Percent high education category (ISCED 5 and 6, LIS)  0.311  0.115  0.089  0.618   Potential experience, mean (LIS)  21.103  1.220  18.693  24.952   Potential experience, standard deviation (LIS)  8.938  0.373  8.087  9.872   Percent tertiary educational attainment (Barro–Lee)  0.176  0.071  0.063  0.337   Educational heterogeneity (Barro–Lee)  0.925  0.130  0.509  1.096  Female sample           Within-group inequality  0.241  0.112  0.049  0.546   Relative contribution of WGI to total inequality  70.174  9.380  41.151  96.342   Logged variance of annual earnings  0.333  0.135  0.079  0.689   Percent high education category (ISCED 5 and 6, LIS)  0.381  0.147  0.098  0.703   Potential experience, mean (LIS)  20.426  1.678  15.874  25.282   Potential experience, standard deviation (LIS)  9.221  0.436  8.091  10.306   Percent tertiary educational attainment (Barro–Lee)  0.165  0.076  0.022  0.342   Educational heterogeneity (Barro–Lee)  0.918  0.115  0.511  1.098  Labour market institutions, employment policies           Liberal regimea  0.25         Continental European regime  0.36         Nordic regime  0.10         Eastern European regime  0.22         Taiwan  0.07            Labour coordinationa  0.00  1.442  −2.516  2.922    Wage setting coordination  2.71  1.305  1  5    Trade union membership  34.32  19.99  6.41  95.16    Adjusted bargaining coverage  57.11  26.28  10.70  100   EPL: Total strictnessb  0.00  1.237  −2.252  2.514   EPL: Regular/temporary employment protection gapb  0.00  0.685  −1.947  1.668    EPL: Regular contracts  2.036  0.812  0.257  3.306    EPL: Temporary contracts  1.498  1.241  0.025  5  Male observations: 143 Female observations: 142  Israel-1986 dropped from female sample due to small sample size. EPL measures not available in Taiwan (n = 133 men, 132 women). a Liberal: Australia, Canada, Ireland, Great Britain, and United States. Continental European: combination of Conservative—Austria, Belgium, Switzerland, Germany, France, Israel (Mandel and Shalev 2009), Luxembourg, and The Netherlands—and Mediterranean—Greece, Italy, and Spain—regimes. Nordic: Denmark, Finland, and Iceland. East European: Czech Republic, Estonia, Hungary, Lithuania, Poland, Russia, Slovenia, and Slovak Republic. b Scaled created from indented items below. See online Appendix for details. Results How Has WGI Grown in Rich Countries? No previous research has assessed WGI cross-nationally using high-quality, comparable measurements. This study therefore provides a general, descriptive assessment of the contribution of WGI to LIS total market earnings inequality. How widely does WGI vary across countries, and what is the relative contribution of WGI to total earnings inequality? Figure 1 shows the relationship between the absolute level of annual earnings WGI (x-axes) and the percentage of total earnings inequality attributable to WGI (y-axes). The panels show results by sex. Figure 1. View largeDownload slide Relationship between absolute level of WGI and relative contribution of WGI to total inequality Note: Country codes are listed in Table 1. Figure 1. View largeDownload slide Relationship between absolute level of WGI and relative contribution of WGI to total inequality Note: Country codes are listed in Table 1. I observe substantial variation in both absolute levels of WGI and the relative contribution of WGI to total inequality. Absolute levels of WGI range by a factor of 7 for men (maximum = 0.57, minimum = 0.07) and 10 for women (maximum = 0.56, minimum = 0.05). For both male and female samples, Denmark, Italy, and Luxembourg tend to have lower absolute levels of WGI, whereas the United States, Canada, Russia, and Lithuania tend to have higher absolute levels. This finding simply reflects the wide range of inequality across high-income countries that has drawn scholars to the cross-national level. Such cross-national variability exists among WGI as well. WGI varies not only in its absolute levels but also in its relative contribution to earnings inequality. In some countries—the Slovak Republic, Australia, Austria, and Canada, for example—WGI accounts for upwards of 85% of total earnings inequality. In other words, standard microlevel grouping categories—education, work experience, and occupational and demographic characteristics—do little to explain market earnings inequality in these countries. In other countries—Luxembourg and Taiwan, for example—WGI is relatively less important for inequality trends, accounting for 40%–60% of total inequality. Put differently, I observe substantial variation in the contribution of the exact same microlevel models across country contexts. What is the relationship between levels of WGI and its relative contribution to total inequality trends? The results in Figure 1 suggest a positive association. For both men and women, WGI tends to be more important in country-years where absolute levels of WGI are higher. The correlations between these dimensions of WGI are 0.60 for men and 0.56 for women (both P < 0.001, two-tailed tests). I find similar positive associations between the relative contribution of WGI and total earnings inequality for both samples, but with smaller correlations (around 0.37 for both samples). In total, Figure 1 provides preliminary evidence that WGI typically plays a larger role for inequality where inequality levels are higher. These findings help explain why WGI has been central to US inequality debates (Lemieux, 2006; Autor, Katz and Kearney, 2008). Along with the variation noted in the above paragraph, two points of similarity across country contexts are notable. First, most individual earnings inequality in the LIS data series—the gold standard of macrocomparative stratification research—is WGI. In only 53 of the 285 country-year observations (18.5%) does WGI contribute less than 60% to total inequality. For comparison, WGI contributes over ⅔ of total earnings inequality in 163 country-years (57% of sample) and over ¾ of total inequality in 23% of the sample (65 country-years). Put simply, most cross-national prime-age market earnings inequality that researchers have studied in high-income, postindustrial societies occurs in the within-group component of inequality. Second, supplemental analyses show similarity of measurements from Figure 1, and their rank orderings, between male and female samples. Male WGI tends to be high where female WGI is also, and male WGI tends to be relatively important in country-years where female WGI is relatively important. The correlations in rank orderings between sexes are 0.72 for WGI percentage and 0.65 for absolute ranks. The relative similarities across samples cast doubt on the argument that WGI represents simple randomness: Why would randomness be so orderly across worker samples within country-year pairs? Does the United States have atypical WGI? Yes and no. On the one hand, the United States has large absolute levels of WGI, as I observe US country-year observations clustering on the right side of the x-axis in both panels. This finding reinforces previous research showing the United States to have particularly high levels of inequality, inclusive of inequality both pre- and post-tax and transfer redistribution (Gornick and Milanovic, 2015). This fact is simply reflected in the within-component of inequality. On the other hand, the United States appears typical regarding WGI’s relative contribution to total male inequality. For example, the contribution of WGI to total male earnings is similar across American, Germany, and Danish contexts, around 70%. The three countries differ in their absolute levels of WGI. Yet WGI is relatively more important to total earnings inequality in some countries—the Slovak Republic, Canada, and Australia—and is relatively less important in others—Finland, Greece, and Hungary, for example. In total, I observe US WGI to be atypical in its high levels for both men and women, and to be typical in its relative contribution to male earnings inequality. Figure 2 shifts focus from levels to change in inequality within countries over time. It shows the contribution of WGI change and BGI change to total changes in earnings inequality (left panels, e.g. change between us74 and us13) and two adjacent country-year observations (right panels, e.g. change between us74 and us79). X-axes represent change in total logged earnings variance between periods. Y-axes represent change attributable to WGI and BGI. Markers include country codes (Table 1) and are prefixed ‘W’ for WGI change and ‘B’ for BGI change. Lines show the simple linear fit between either WGI (solid) or BGI (dashed) change and total inequality change. A steeper association between total inequality and either WGI or BGI change indicates a greater relative importance of change in one of these components to total inequality change. Figure 2. View largeDownload slide Scatterplot and slope of WGI change and BGI change against total inequality change Note: Country codes listed in Table 1. Figure 2. View largeDownload slide Scatterplot and slope of WGI change and BGI change against total inequality change Note: Country codes listed in Table 1. Figure 2 clearly shows that inequality change is due primarily to WGI change. The correlations of WGI change and total inequality change range from 0.94–0.97, whereas the equivalent correlations for BGI are lower, 0.60–0.74. In addition, I tested the x-standardized regression coefficients of total inequality change on WGI and BGI change in simple linear regression models with robust standard errors. The coefficients for WGI change are significantly larger than the coefficients for BGI change across all samples from Figure 2, with the WGI slope 100%–300% larger compared with the BGI slope (P < 0.05, two-tailed tests, in all tests conducted).24Table A1 verifies these results by estimating the association between the relative contribution of WGI and total inequality with and without country- and year-fixed effects. Notably, trends in Figures 1 and 2 might be conservative due to the logged outcome variable analysed. In total, I conclude that in addition to its importance to overall levels of earnings inequality, change in WGI is primarily responsible for change in total inequality in high-income countries. As inequality grows in high-income countries, WGI tends to become increasingly important. No previous research has revealed this basic empirical finding, yet it is critical for understanding recent inequality trends. To provide a substantive illustration of above results, Figure 3 displays trajectories of male WGI, BGI, and total logged earnings variance in the United States, Canada, Germany, Finland, and Luxembourg. I examined equivalent figures for all countries included in analyses. These five countries were selected because they represent the range of inequality patterns.25 Panels show trends for WGI (left), BGI (centre), and total income (right). Figure 3. View largeDownload slide Male WGI, BGI, and total inequality of logged annual earnings over time in five countries Note: Country codes listed in Table 1. Figure 3. View largeDownload slide Male WGI, BGI, and total inequality of logged annual earnings over time in five countries Note: Country codes listed in Table 1. Markers indicate LIS waves. The trajectory of American WGI (hollow circles) generally follow patterns from previous studies (Lemieux, 2006; Autor, Katz and Kearney, 2008; Western and Rosenfeld, 2011). WGI rose rapidly through the 1980s, stabilized for the next two decades, and began to rise again at the end of the 2000s. As suggested in Figures 1 and 2, the divergent trajectories and levels of inequality in these five countries are qualitatively due to WGI. Although the United States and Luxembourg have higher BGI than Germany, Finland, and Canada, overall BGI does little to distinguish country inequality trends. Turning to WGI, Canadian and American patterns closely resemble one another. The difference in total inequality between the two countries stems primarily from higher American BGI. German WGI spiked after reunification in 1990 and grew slowly afterwards. Finnish WGI remained unchanged for nearly two decades, then doubled in the mid-2000s. Luxembourg WGI increased slightly, but its importance pales in comparison with BGI, which doubled over the period of study to end up resembling American BGI. While WGI increased in all five countries, the nature of WGI change varied widely, and it is primarily these WGI trends that distinguish the trends in total inequality in the rightmost panel. From the many findings of the descriptive analyses in this section, I draw one general conclusion: Within-group inequality primarily drives cross-national differences in prime-age market earnings inequality among high-income countries.26 I observe WGI to be central to inequality levels, trends in inequality change, and substantive inter-country inequality patterns. Although US studies have shown WGI to be important for total inequality trends, it was unknown whether these American inequality dynamics were unique, or whether WGI was more generally important for contemporary trends of rising inequality among high-income countries. My results point to the latter. WGI, Skill, and Deinstitutionalization Thus far, results suggest a general importance of WGI for total inequality. I next assess how WGI varies along cross-national differences in human capital attainment and labour market institutional arrangements. Results are shown in Tables 3 and 4. Table 3. Education and welfare regime variation in WGI levels and relative importance   WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  (9)  (10)  (11)  (12)  Welfare regime (Liberal = reference)                           Continental Europe  −8.025***  −8.878***  −8.021***  −5.340**  −6.859***  −4.801*  −0.108***  −0.080***  −0.069***  −0.102***  −0.063**  −0.061**  (1.826)  (2.058)  (1.999)  (1.899)  (1.957)  (2.062)  (0.018)  (0.019)  (0.018)  (0.021)  (0.023)  (0.019)   Nordic  −6.080**  −7.945**  −6.392**  −7.809***  −8.857**  −7.779**  −0.151***  −0.110***  −0.127***  −0.201***  −0.172***  −0.230***  (2.151)  (2.563)  (2.358)  (1.974)  (2.736)  (2.581)  (0.025)  (0.028)  (0.025)  (0.026)  (0.033)  (0.030)   Eastern Europe  2.011  1.180  2.258  −1.301  −3.223  −1.387  −0.046+  0.039  0.037  −0.060*  −0.060+  −0.039  (1.779)  (1.954)  (2.030)  (1.984)  (2.674)  (2.807)  (0.027)  (0.028)  (0.029)  (0.027)  (0.032)  (0.029)   Taiwan  −21.006***  −23.775***  −20.422***  −24.123***  −26.995***  −22.003***  −0.183***  −0.072**  −0.183***  −0.186***  −0.128***  −0.136***  (1.351)  (2.543)  (1.558)  (2.050)  (2.858)  (3.060)  (0.016)  (0.024)  (0.016)  (0.018)  (0.027)  (0.018)  Completed tertiary education    −23.115+      −17.854      0.621***      0.569***      (13.712)      (11.772)      (0.139)      (0.144)    Educational heterogeneity    4.315      −4.950      0.105      −0.184*      (6.551)      (8.857)      (0.072)      (0.080)    Percent secondary education      3.453      12.499      −0.065      0.101+      (6.363)      (10.391)      (0.065)      (0.060)  Percent in tertiary education      4.187      12.944      0.401***      0.528***      (9.098)      (12.616)      (0.091)      (0.095)  Average potential experience    −0.925  −0.324    −0.411  0.009    −0.009  −0.006    0.005  0.019**    (0.662)  (0.616)    (0.650)  (0.694)    (0.005)  (0.007)    (0.006)  (0.006)  Potential experience SD    −2.583  −2.081    −0.111  −0.082    −0.028  −0.043*    −0.001  −0.001    (1.670)  (1.676)    (0.093)  (0.090)    (0.017)  (0.018)    (0.001)  (0.001)  N  143  143  143  142  142  142  143  143  143  142  142  142    WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  (9)  (10)  (11)  (12)  Welfare regime (Liberal = reference)                           Continental Europe  −8.025***  −8.878***  −8.021***  −5.340**  −6.859***  −4.801*  −0.108***  −0.080***  −0.069***  −0.102***  −0.063**  −0.061**  (1.826)  (2.058)  (1.999)  (1.899)  (1.957)  (2.062)  (0.018)  (0.019)  (0.018)  (0.021)  (0.023)  (0.019)   Nordic  −6.080**  −7.945**  −6.392**  −7.809***  −8.857**  −7.779**  −0.151***  −0.110***  −0.127***  −0.201***  −0.172***  −0.230***  (2.151)  (2.563)  (2.358)  (1.974)  (2.736)  (2.581)  (0.025)  (0.028)  (0.025)  (0.026)  (0.033)  (0.030)   Eastern Europe  2.011  1.180  2.258  −1.301  −3.223  −1.387  −0.046+  0.039  0.037  −0.060*  −0.060+  −0.039  (1.779)  (1.954)  (2.030)  (1.984)  (2.674)  (2.807)  (0.027)  (0.028)  (0.029)  (0.027)  (0.032)  (0.029)   Taiwan  −21.006***  −23.775***  −20.422***  −24.123***  −26.995***  −22.003***  −0.183***  −0.072**  −0.183***  −0.186***  −0.128***  −0.136***  (1.351)  (2.543)  (1.558)  (2.050)  (2.858)  (3.060)  (0.016)  (0.024)  (0.016)  (0.018)  (0.027)  (0.018)  Completed tertiary education    −23.115+      −17.854      0.621***      0.569***      (13.712)      (11.772)      (0.139)      (0.144)    Educational heterogeneity    4.315      −4.950      0.105      −0.184*      (6.551)      (8.857)      (0.072)      (0.080)    Percent secondary education      3.453      12.499      −0.065      0.101+      (6.363)      (10.391)      (0.065)      (0.060)  Percent in tertiary education      4.187      12.944      0.401***      0.528***      (9.098)      (12.616)      (0.091)      (0.095)  Average potential experience    −0.925  −0.324    −0.411  0.009    −0.009  −0.006    0.005  0.019**    (0.662)  (0.616)    (0.650)  (0.694)    (0.005)  (0.007)    (0.006)  (0.006)  Potential experience SD    −2.583  −2.081    −0.111  −0.082    −0.028  −0.043*    −0.001  −0.001    (1.670)  (1.676)    (0.093)  (0.090)    (0.017)  (0.018)    (0.001)  (0.001)  N  143  143  143  142  142  142  143  143  143  142  142  142  Robust standard errors in parentheses. +  P < 0.10, * P < 0.05, ** P < 0.01, *** P < 0.001, two-tailed test.  Models include continuous measure of year. ‘Completed tertiary education’ and ‘Educational heterogeneity’ computed from Barro–Lee database individuals aged 25–55, separately by sex. Additional education and potential experience measures computed from LIS samples. Table 3. Education and welfare regime variation in WGI levels and relative importance   WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  (9)  (10)  (11)  (12)  Welfare regime (Liberal = reference)                           Continental Europe  −8.025***  −8.878***  −8.021***  −5.340**  −6.859***  −4.801*  −0.108***  −0.080***  −0.069***  −0.102***  −0.063**  −0.061**  (1.826)  (2.058)  (1.999)  (1.899)  (1.957)  (2.062)  (0.018)  (0.019)  (0.018)  (0.021)  (0.023)  (0.019)   Nordic  −6.080**  −7.945**  −6.392**  −7.809***  −8.857**  −7.779**  −0.151***  −0.110***  −0.127***  −0.201***  −0.172***  −0.230***  (2.151)  (2.563)  (2.358)  (1.974)  (2.736)  (2.581)  (0.025)  (0.028)  (0.025)  (0.026)  (0.033)  (0.030)   Eastern Europe  2.011  1.180  2.258  −1.301  −3.223  −1.387  −0.046+  0.039  0.037  −0.060*  −0.060+  −0.039  (1.779)  (1.954)  (2.030)  (1.984)  (2.674)  (2.807)  (0.027)  (0.028)  (0.029)  (0.027)  (0.032)  (0.029)   Taiwan  −21.006***  −23.775***  −20.422***  −24.123***  −26.995***  −22.003***  −0.183***  −0.072**  −0.183***  −0.186***  −0.128***  −0.136***  (1.351)  (2.543)  (1.558)  (2.050)  (2.858)  (3.060)  (0.016)  (0.024)  (0.016)  (0.018)  (0.027)  (0.018)  Completed tertiary education    −23.115+      −17.854      0.621***      0.569***      (13.712)      (11.772)      (0.139)      (0.144)    Educational heterogeneity    4.315      −4.950      0.105      −0.184*      (6.551)      (8.857)      (0.072)      (0.080)    Percent secondary education      3.453      12.499      −0.065      0.101+      (6.363)      (10.391)      (0.065)      (0.060)  Percent in tertiary education      4.187      12.944      0.401***      0.528***      (9.098)      (12.616)      (0.091)      (0.095)  Average potential experience    −0.925  −0.324    −0.411  0.009    −0.009  −0.006    0.005  0.019**    (0.662)  (0.616)    (0.650)  (0.694)    (0.005)  (0.007)    (0.006)  (0.006)  Potential experience SD    −2.583  −2.081    −0.111  −0.082    −0.028  −0.043*    −0.001  −0.001    (1.670)  (1.676)    (0.093)  (0.090)    (0.017)  (0.018)    (0.001)  (0.001)  N  143  143  143  142  142  142  143  143  143  142  142  142    WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  (9)  (10)  (11)  (12)  Welfare regime (Liberal = reference)                           Continental Europe  −8.025***  −8.878***  −8.021***  −5.340**  −6.859***  −4.801*  −0.108***  −0.080***  −0.069***  −0.102***  −0.063**  −0.061**  (1.826)  (2.058)  (1.999)  (1.899)  (1.957)  (2.062)  (0.018)  (0.019)  (0.018)  (0.021)  (0.023)  (0.019)   Nordic  −6.080**  −7.945**  −6.392**  −7.809***  −8.857**  −7.779**  −0.151***  −0.110***  −0.127***  −0.201***  −0.172***  −0.230***  (2.151)  (2.563)  (2.358)  (1.974)  (2.736)  (2.581)  (0.025)  (0.028)  (0.025)  (0.026)  (0.033)  (0.030)   Eastern Europe  2.011  1.180  2.258  −1.301  −3.223  −1.387  −0.046+  0.039  0.037  −0.060*  −0.060+  −0.039  (1.779)  (1.954)  (2.030)  (1.984)  (2.674)  (2.807)  (0.027)  (0.028)  (0.029)  (0.027)  (0.032)  (0.029)   Taiwan  −21.006***  −23.775***  −20.422***  −24.123***  −26.995***  −22.003***  −0.183***  −0.072**  −0.183***  −0.186***  −0.128***  −0.136***  (1.351)  (2.543)  (1.558)  (2.050)  (2.858)  (3.060)  (0.016)  (0.024)  (0.016)  (0.018)  (0.027)  (0.018)  Completed tertiary education    −23.115+      −17.854      0.621***      0.569***      (13.712)      (11.772)      (0.139)      (0.144)    Educational heterogeneity    4.315      −4.950      0.105      −0.184*      (6.551)      (8.857)      (0.072)      (0.080)    Percent secondary education      3.453      12.499      −0.065      0.101+      (6.363)      (10.391)      (0.065)      (0.060)  Percent in tertiary education      4.187      12.944      0.401***      0.528***      (9.098)      (12.616)      (0.091)      (0.095)  Average potential experience    −0.925  −0.324    −0.411  0.009    −0.009  −0.006    0.005  0.019**    (0.662)  (0.616)    (0.650)  (0.694)    (0.005)  (0.007)    (0.006)  (0.006)  Potential experience SD    −2.583  −2.081    −0.111  −0.082    −0.028  −0.043*    −0.001  −0.001    (1.670)  (1.676)    (0.093)  (0.090)    (0.017)  (0.018)    (0.001)  (0.001)  N  143  143  143  142  142  142  143  143  143  142  142  142  Robust standard errors in parentheses. +  P < 0.10, * P < 0.05, ** P < 0.01, *** P < 0.001, two-tailed test.  Models include continuous measure of year. ‘Completed tertiary education’ and ‘Educational heterogeneity’ computed from Barro–Lee database individuals aged 25–55, separately by sex. Additional education and potential experience measures computed from LIS samples. Table 4. Labour coordination, EPL, and welfare regime variation in WGI levels and relative importance   WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (7)  (8)  (9)  (4)  (5)  (6)  (10)  (11)  (12)  Labour coordination  −1.673**  −1.671**  −0.773  −0.830  −0.300  1.011  −0.034***  −0.037***  −0.031***  −0.037***  −0.034***  −0.024**  (0.498)  (0.602)  (0.852)  (0.539)  (0.654)  (1.070)  (0.006)  (0.006)  (0.007)  (0.007)  (0.007)  (0.009)  General EPL  −1.780*  −2.531**  −3.771**  −1.993**  −2.652***  −4.065***  −0.025***  −0.015*  −0.037**  −0.021***  −0.018*  −0.040**  (0.725)  (0.816)  (1.321)  (0.623)  (0.615)  (1.171)  (0.006)  (0.007)  (0.013)  (0.006)  (0.007)  (0.012)  Regular − temporary EPL difference  3.251**  1.561  −0.310  3.981***  3.050*  2.227  0.002  0.022+  −0.003  0.001  0.008  0.000  (1.067)  (1.171)  (1.136)  (1.131)  (1.243)  (1.484)  (0.011)  (0.012)  (0.014)  (0.012)  (0.013)  (0.016)  Welfare regime (Liberal = reference)                           Continental Europe      1.448      2.717      0.052      0.066+      (3.687)      (3.415)      (0.038)      (0.035)   Nordic      −0.072      −5.109      0.041      −0.019      (4.034)      (4.740)      (0.040)      (0.047)   Eastern Europe      8.609*      3.245      0.121***      0.047      (3.458)      (4.225)      (0.035)      (0.042)  Completed tertiary education    −45.627**  −48.614**    −9.380  −4.782    0.297  0.228    0.264  0.330+    (16.360)  (16.531)    (13.848)  (14.689)    (0.186)  (0.198)    (0.189)  (0.192)  Educational heterogeneity    0.619  11.575+    −14.446  −15.212    0.111+  0.234***    −0.054  −0.088    (6.893)  (6.343)    (8.777)  (9.178)    (0.065)  (0.068)    (0.088)  (0.088)  Average potential experience    −0.891  −1.269+    −1.009+  −0.814    −0.006  −0.011*    −0.005  −0.003    (0.685)  (0.653)    (0.517)  (0.627)    (0.006)  (0.005)    (0.005)  (0.006)  Potential experience SD    −0.327  −1.633    −0.136  −0.164+    −0.015  −0.023    −0.001  −0.001    (1.817)  (1.811)    (0.088)  (0.091)    (0.017)  (0.018)    (0.001)  (0.001)  N  133  133  133  132  132  132  133  133  133  132  132  132    WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (7)  (8)  (9)  (4)  (5)  (6)  (10)  (11)  (12)  Labour coordination  −1.673**  −1.671**  −0.773  −0.830  −0.300  1.011  −0.034***  −0.037***  −0.031***  −0.037***  −0.034***  −0.024**  (0.498)  (0.602)  (0.852)  (0.539)  (0.654)  (1.070)  (0.006)  (0.006)  (0.007)  (0.007)  (0.007)  (0.009)  General EPL  −1.780*  −2.531**  −3.771**  −1.993**  −2.652***  −4.065***  −0.025***  −0.015*  −0.037**  −0.021***  −0.018*  −0.040**  (0.725)  (0.816)  (1.321)  (0.623)  (0.615)  (1.171)  (0.006)  (0.007)  (0.013)  (0.006)  (0.007)  (0.012)  Regular − temporary EPL difference  3.251**  1.561  −0.310  3.981***  3.050*  2.227  0.002  0.022+  −0.003  0.001  0.008  0.000  (1.067)  (1.171)  (1.136)  (1.131)  (1.243)  (1.484)  (0.011)  (0.012)  (0.014)  (0.012)  (0.013)  (0.016)  Welfare regime (Liberal = reference)                           Continental Europe      1.448      2.717      0.052      0.066+      (3.687)      (3.415)      (0.038)      (0.035)   Nordic      −0.072      −5.109      0.041      −0.019      (4.034)      (4.740)      (0.040)      (0.047)   Eastern Europe      8.609*      3.245      0.121***      0.047      (3.458)      (4.225)      (0.035)      (0.042)  Completed tertiary education    −45.627**  −48.614**    −9.380  −4.782    0.297  0.228    0.264  0.330+    (16.360)  (16.531)    (13.848)  (14.689)    (0.186)  (0.198)    (0.189)  (0.192)  Educational heterogeneity    0.619  11.575+    −14.446  −15.212    0.111+  0.234***    −0.054  −0.088    (6.893)  (6.343)    (8.777)  (9.178)    (0.065)  (0.068)    (0.088)  (0.088)  Average potential experience    −0.891  −1.269+    −1.009+  −0.814    −0.006  −0.011*    −0.005  −0.003    (0.685)  (0.653)    (0.517)  (0.627)    (0.006)  (0.005)    (0.005)  (0.006)  Potential experience SD    −0.327  −1.633    −0.136  −0.164+    −0.015  −0.023    −0.001  −0.001    (1.817)  (1.811)    (0.088)  (0.091)    (0.017)  (0.018)    (0.001)  (0.001)  N  133  133  133  132  132  132  133  133  133  132  132  132  Robust standard errors in parentheses. +  P < 0.10, * P < 0.05, ** P < 0.01, *** P < 0.001, two-tailed test.  Models include continuous measure of year. ‘Completed tertiary education’ and ‘Educational heterogeneity’ computed from Barro–Lee database individuals aged 25–55, separately by sex. Potential experience measures computed from LIS samples. See Table 2 and online Appendix for description of Labour coordination and EPL scales.  Taiwan EPL values missing. Table A4 replicates results with only Labour coordination scale. Table 4. Labour coordination, EPL, and welfare regime variation in WGI levels and relative importance   WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (7)  (8)  (9)  (4)  (5)  (6)  (10)  (11)  (12)  Labour coordination  −1.673**  −1.671**  −0.773  −0.830  −0.300  1.011  −0.034***  −0.037***  −0.031***  −0.037***  −0.034***  −0.024**  (0.498)  (0.602)  (0.852)  (0.539)  (0.654)  (1.070)  (0.006)  (0.006)  (0.007)  (0.007)  (0.007)  (0.009)  General EPL  −1.780*  −2.531**  −3.771**  −1.993**  −2.652***  −4.065***  −0.025***  −0.015*  −0.037**  −0.021***  −0.018*  −0.040**  (0.725)  (0.816)  (1.321)  (0.623)  (0.615)  (1.171)  (0.006)  (0.007)  (0.013)  (0.006)  (0.007)  (0.012)  Regular − temporary EPL difference  3.251**  1.561  −0.310  3.981***  3.050*  2.227  0.002  0.022+  −0.003  0.001  0.008  0.000  (1.067)  (1.171)  (1.136)  (1.131)  (1.243)  (1.484)  (0.011)  (0.012)  (0.014)  (0.012)  (0.013)  (0.016)  Welfare regime (Liberal = reference)                           Continental Europe      1.448      2.717      0.052      0.066+      (3.687)      (3.415)      (0.038)      (0.035)   Nordic      −0.072      −5.109      0.041      −0.019      (4.034)      (4.740)      (0.040)      (0.047)   Eastern Europe      8.609*      3.245      0.121***      0.047      (3.458)      (4.225)      (0.035)      (0.042)  Completed tertiary education    −45.627**  −48.614**    −9.380  −4.782    0.297  0.228    0.264  0.330+    (16.360)  (16.531)    (13.848)  (14.689)    (0.186)  (0.198)    (0.189)  (0.192)  Educational heterogeneity    0.619  11.575+    −14.446  −15.212    0.111+  0.234***    −0.054  −0.088    (6.893)  (6.343)    (8.777)  (9.178)    (0.065)  (0.068)    (0.088)  (0.088)  Average potential experience    −0.891  −1.269+    −1.009+  −0.814    −0.006  −0.011*    −0.005  −0.003    (0.685)  (0.653)    (0.517)  (0.627)    (0.006)  (0.005)    (0.005)  (0.006)  Potential experience SD    −0.327  −1.633    −0.136  −0.164+    −0.015  −0.023    −0.001  −0.001    (1.817)  (1.811)    (0.088)  (0.091)    (0.017)  (0.018)    (0.001)  (0.001)  N  133  133  133  132  132  132  133  133  133  132  132  132    WGI relative contribution to total inequality   WGI absolute levels   Men   Women   Men   Women   (1)  (2)  (3)  (7)  (8)  (9)  (4)  (5)  (6)  (10)  (11)  (12)  Labour coordination  −1.673**  −1.671**  −0.773  −0.830  −0.300  1.011  −0.034***  −0.037***  −0.031***  −0.037***  −0.034***  −0.024**  (0.498)  (0.602)  (0.852)  (0.539)  (0.654)  (1.070)  (0.006)  (0.006)  (0.007)  (0.007)  (0.007)  (0.009)  General EPL  −1.780*  −2.531**  −3.771**  −1.993**  −2.652***  −4.065***  −0.025***  −0.015*  −0.037**  −0.021***  −0.018*  −0.040**  (0.725)  (0.816)  (1.321)  (0.623)  (0.615)  (1.171)  (0.006)  (0.007)  (0.013)  (0.006)  (0.007)  (0.012)  Regular − temporary EPL difference  3.251**  1.561  −0.310  3.981***  3.050*  2.227  0.002  0.022+  −0.003  0.001  0.008  0.000  (1.067)  (1.171)  (1.136)  (1.131)  (1.243)  (1.484)  (0.011)  (0.012)  (0.014)  (0.012)  (0.013)  (0.016)  Welfare regime (Liberal = reference)                           Continental Europe      1.448      2.717      0.052      0.066+      (3.687)      (3.415)      (0.038)      (0.035)   Nordic      −0.072      −5.109      0.041      −0.019      (4.034)      (4.740)      (0.040)      (0.047)   Eastern Europe      8.609*      3.245      0.121***      0.047      (3.458)      (4.225)      (0.035)      (0.042)  Completed tertiary education    −45.627**  −48.614**    −9.380  −4.782    0.297  0.228    0.264  0.330+    (16.360)  (16.531)    (13.848)  (14.689)    (0.186)  (0.198)    (0.189)  (0.192)  Educational heterogeneity    0.619  11.575+    −14.446  −15.212    0.111+  0.234***    −0.054  −0.088    (6.893)  (6.343)    (8.777)  (9.178)    (0.065)  (0.068)    (0.088)  (0.088)  Average potential experience    −0.891  −1.269+    −1.009+  −0.814    −0.006  −0.011*    −0.005  −0.003    (0.685)  (0.653)    (0.517)  (0.627)    (0.006)  (0.005)    (0.005)  (0.006)  Potential experience SD    −0.327  −1.633    −0.136  −0.164+    −0.015  −0.023    −0.001  −0.001    (1.817)  (1.811)    (0.088)  (0.091)    (0.017)  (0.018)    (0.001)  (0.001)  N  133  133  133  132  132  132  133  133  133  132  132  132  Robust standard errors in parentheses. +  P < 0.10, * P < 0.05, ** P < 0.01, *** P < 0.001, two-tailed test.  Models include continuous measure of year. ‘Completed tertiary education’ and ‘Educational heterogeneity’ computed from Barro–Lee database individuals aged 25–55, separately by sex. Potential experience measures computed from LIS samples. See Table 2 and online Appendix for description of Labour coordination and EPL scales.  Taiwan EPL values missing. Table A4 replicates results with only Labour coordination scale. Across results, several main conclusions can be drawn. For men and women, the relative contribution of WGI to total inequality varies significantly across welfare regimes. Continental European, Nordic countries, and Taiwan differ significantly from Liberal and East European regimes in their lower relative proportion of WGI (men: 8%, 6%, and 21% lower, women: 5%, 8%, 24%, all P < 0.01). Results are replicated across models with education and potential experience information, and in sensitivity analyses including overall earnings inequality (Table A5). Regime differences are thus not reducible to differences in human capital attainment or levels of total earning variance. Models 7–12 show absolute WGI levels. Continental European, Nordic, and Taiwanese country-years have lower absolute levels of WGI compared to Liberal and East European country-years. Table 4 assesses WGI in relation to country-specific labour market institutional configurations and employment protection laws. Results largely replicate those from Table 3 but reveal an important source of inter- and intra-regime variation. Countries with stronger labour unions, more universal bargaining coverage, and more centralized wage bargaining have both lower levels of WGI and lower proportions of WGI contributing to total inequality. Similarly, stricter EPL is associated with lower WGI and its relative importance, while a gap between regular and temporary contract EPL associates with higher relative female WGI. A standard deviation increase in labour coordination reduces the relative contribution of WGI by about 2.5 percentage points for men, and a standard deviation increase in EPL strictness reduces WGI’s relative contribution by about 3 percentage points for both sexes. For female results, while both institutional arrangements and employment protection laws influence total inequality levels, only employment protection laws reduce the relative proportion of WGI, although both are significant and negatively signed when included separately. Notably, Models 3, 9, 6, and 12 show that regime differences are near fully attributable to labour market institutions and EPL. Results for educational attainment merit special focus. While central to many US studies of WGI, associations between educational attainment and WGI are at best mixed when examined cross-nationally. A positive and significant association between tertiary educational attainment and the relative contribution of WGI is detectable in simple models without controls. Yet the association is removed, and in some cases reversed, with the addition of institutional variables. No consistent conclusion is detectable across sensitivity tests. These contrasting results are suggestive that deinstitutionalization fares better to explain WGI in the cross-national context than returns to skill. In total, results from Tables 3 and 4 provide an important addendum to general conclusions. Although WGI tends to be central to inequality across high-income countries, the magnitude of this fact is channelled through a country’s institutional context. Stronger labour protections tend to reduce not only absolute inequality levels but also the distributional properties of inequality. Conclusion This research assessed cross-national macrolevel within-group earnings inequality, asking the questions: How general is the importance of within-group inequality to inequality change, and how do previous explanations of skill and deinstitutionalization fare in a cross-national context? I applied the logic of decomposing individual market earnings inequality into within-group (WGI) and between-group (BGI) components, predominantly done in the American context, to a sample of full-time prime-age workers in LIS microdata from 28 countries spanning 40 years. Results from this research help clarify the nature of the contemporary rise of individual market earnings inequality in high-income countries. The major descriptive finding of this research is that cross-national differences in prime-age individual market earnings inequality largely stem from within-group differences. Simply put, if one could ‘level’ major occupational, human capital, and sociodemographic characteristics which sociologists typically study in relation to earnings inequality, cross-national patterns of earnings inequality and their changes over time would remain similar. This basic empirical fact was hitherto unknown. Nor were there a priori reasons to suspect that the importance of WGI would be broadly shared across national and historical contexts of high-income countries. My findings therefore provide an important contribution to the understanding of the distributional properties underlying contemporary patterns of earnings inequality in high-income countries. The past 40 years have undergone changes to not only inequality levels but also inequality forms. At the same time, results reveal heterogeneity in the relative importance of WGI across countries in different institutional contexts. Continental European and Nordic countries have significantly lower levels of WGI and proportions of inequality attributable to WGI than Liberal and East European countries, independent of inequality levels and human capital composition. Differences are largely attributable to variation in labour market institutions and employment legislation. These results suggest that a deinstitutionalization explanation of WGI does well in the cross-national context, especially when considered alongside the mixed evidence found for human capital composition. On the one hand, these findings make sense. They follow related stratification literature documenting the importance of labour market institutional configurations across high-income countries (Esping-Andersen, 1999; Mandel and Shalev, 2009). And previous research has documented the varied ways that institutional differences translate into inequality and wage attainment (Gangl, 2004). WGI is another inequality dimension aligning with this research tradition. On the other hand, these findings extend comparative stratification research. Through their influence on flexibility and security, institutional and policy contexts affect not only redistribution and inequality levels but also basic distributional properties of inequality, resulting in variation of forms of inequality across countries (Thelen, 2014). Future research is needed to extend and verify the precise institutional and policy mechanisms that influence the relationship between inequality, WGI, and BGI. Generally, findings highlight the utility of moving beyond one-number summaries and average wage gaps to fully understand the influence of institutional and policy variation on inequality. While results do not support skill return explanations, future research on the specific relationships between WGI, deinstitutionalization, and skill return is needed. A straightforward reading of results is that such explanations generalize poorly beyond the US case, a conclusion that would reflect research by Blau and Kahn (2005). Alternatively, inconsistent results might reflect heterogeneity in the association between WGI and skill returns across institutional arrangements. Its explanatory power may be dependent on deinstitutionalization, suggesting that regime- and/or country-specific studies are needed. More research is needed to assess where, when, and why skill returns are manifest as WGI. Results are suggestive of difficulties faced by actors who focus on closing between-group earnings gaps (Leicht, 2008). WGI and BGI are positively correlated, while BGI is relatively less consequential to total inequality in high-inequality observations (Figure A1). Focus on closing specific between-group gaps while ignoring broader forces that increase within-group earnings inequality might inevitably prove to be unproductive. While group differences in pay tend to be higher in high-inequality observations, they appear to be of secondary importance for general inequality trends. Overall, these results contribute to recent calls for sociologists to complement focus on between-group earnings gaps with other dimensions of inequality (Leicht, 2008, 2016). Findings are drawn from a conservative method of computing WGI across country contexts, with samples restricted to those strongly attached to the labour market. These sampling decisions were made to ensure that results did not simply reflect variation in state intervention across country contexts. It is likely that the inclusion of broader age ranges and employment statuses would bolster results, as would an assessment of earnings after taxes and transfers. Such analyses are necessary to further develop comparative knowledge of WGI. A natural next step for future research is to further examine macrolevel associations between WGI and country-level characteristics. Do alternative measures of skill (such as test scores), social welfare policies, leftist politics, globalization, or demographic shifts better explain variation in WGI than factors used here? These questions are beyond the scope of the current project, as they necessarily build upon the empirical and theoretical extension of WGI to cross-national inequality. However, with the importance of WGI for cross-national inequality documented, future research would do well to more fully unpack the macrolevel causes of this dimension of inequality. This article has important limitations. Perhaps most importantly, the LIS does not provide fine-grained microlevel occupational information. Studies of occupational polarization and wage inequality typically sort individuals into at least 300 occupational groups, or even 300 occupation-by-300 industry groups.27 To what extent are results biased from omitting such fine-grained Gemeinschaft occupational communities (Liu and Grusky, 2013)? The current research cannot say. However, if such an omission were to level WGI differences across countries, this result itself would be a valuable piece of knowledge. For occupational contrasts to run counter to findings of this study, occupational polarization would need to be of greater relative importance in Liberal regimes than in continental Europe, and among high-inequality country-years compared with low-inequality ones. Such a finding would reinforce a deinstitutionalization explanation of WGI, as the importance of fine-grained occupational differentiation would then be secondary to cross-national variation of labour policies and institutions that compress pay differences between workers. Regardless, the current research underscores the importance of continued research of microclass occupations in relation to inequality forms and labour market institutional arrangements. Similarly, how might compositional changes across observable characteristics be responsible for results? The use of counterfactual reweighting methodologies is beyond the scope of the current project (Lemieux, 2006, Western and Rosenfeld, 2011). However, an assessment of the change in WGI in relation to reweighted population compositions held constant over time, and held constant between countries, provides a promising next step for research. More generally, the assessment of WGI in a cross-national context highlights the central importance of a seemingly simple question: What is a group? Does it reflect a substantively meaningful concept that can be similarly assessed across country and historical contexts? Cross-national research has shown the widely variable and context-specific nature of groups based on educational and occupational attainment, for example (Bol and Weeden, 2014). Phenotypic race is an anchor of group membership in the United States, whereas linguistic and immigrant identities play important roles in some European countries, like Belgium and Sweden. In sensitivity analyses, I found that results were robust against alternative modelling decisions at the microlevel with different conceptual approaches to the idea of group. I replicated WGI measures including immigrant status. Results were substantively the same, and correlations of WGI measures with and without immigrant groups were over 0.97.28 Pragmatism guided this project’s definition of group. Definitions were based on previous US studies and data availability. Yet theoretical attention to the concept of group as it relates to economic inequality is clearly needed. WGI is central to cross-national patterns of inequality and of growing importance to high-income countries. Yet substantial institution-based heterogeneity exists. This study provides insights into the basic distributional properties of inequality in high-income countries and reveals the promises of examining the relationship between total inequality and WGI in a cross-national perspective. Tom VanHeuvelen is Assistant Professor of Sociology at the University of Illinois at Urbana-Champaign. Current research interests comprise inequality and stratification, cross-national and comparative sociology, and the sociology of development. His work has been published in the American Journal of Sociology, Social Forces, Social Science Research, and the Oxford University Press. Footnotes 1 Studies diverge in how a ‘group’ is defined (VanHeuvelen, 2018), typically using some combination of human capital, occupation, and demographic characteristics. I use all available microlevel characteristics in the LIS used in previous research. 2 See the online appendix, and VanHeuvelen (2018), for discussion of theoretical meaning of WGI. 3 Luck could also reflect the discretion and discrimination held by a worker’s employer. Future work assessing employer-based discretion, WGI, and BGI is needed. 4 The relationship between an omitted microlevel variable and WGI is unclear. Consider the loss of a union job. Western and Rosenfeld’s results (2011) suggest this would increase WGI. Yet an omitted variable approach suggests a more precise microlevel variable, like personality or skill, is needed for the subsequent wage attainment model. 5 This camp frequently argues that micro-class occupations or jobs represent the bulk of WGI. This critique has been used to reinforce both skill- and deinstitutionalization-based explanations of inequality (Goos and Manning, 2007; Williams, 2012). 6 This restriction also assists with overcoming difficulties associated with unavailable microlevel data, such as union coverage, fine-grained occupation, and annual work hours, all of which may influence the rate of individuals in part-time employment. Note too that these results should provide a conservative estimate of cross-national variation in WGI, as individuals most sensitive to labour market flexibility are excluded from samples. 7 Female results for il86 do not converge due to a small sample size (n = 140). This observation is excluded from analyses, lowering the female sample to 142. 8 The LIS labels it paid employment labour income. 9 I replicate main analyses controlling for average annual work hours from OECD data and reach the same conclusions. 10 See the online Appendix for a comparison and discussion of US wage and earnings inequality. 11 Early samples from the United Kingdom measuring education in years are transformed to align with the categorical measure. 12 Actual experience is not widely available. Countries have different typical starting ages of education. WGI and BGI measurements are unaffected by adjusting measurements to country-specific starting ages. 13 These two measures of human capital follow the logic of Autor, Katz and Kearney (2008) and allow for complex nonlinear patterns of WGI across the age distribution while using relatively small samples. 14 All categories are unavailable in at94, at97, at00, gr95, gr00, and il79. I instead use three-category industry variables: (1) agriculture, (2) industry, and (3) services. Results are substantively similar excluding these cases. 15 Many WGI studies include fine-grained occupations, or industry-by-occupation contrasts (Goos and Manning, 2007; Williams, 2012). Such information is unavailable in the LIS. Implications are discussed in the Conclusion. 16 Three samples, au08, au10, and uk86, do not have 10 category ISCO codes. I therefore use available eight (au) and nine (uk) category country-specific occupational information. Results are unaffected if excluding these cases. 17 All results are replicated using a three-category occupation measure (available upon request). Correlation across WGI measures using 3- or 10-occupation code is 0.99. 18 Individual-level coefficients are not the main focus of this study. Stata do-files that reproduce individual VFR models through the LIS job submission system are available at www.tomvanheuvelen.com. 19 IL86 has a small sample of female workers and so VFR models cannot converge. This sample is dropped from analyses. 20 This resembles the R2 of the microlevel model. I prefer measures following the VFR for reasons outlined by VanHeuvelen (2018). 21 Education and experience at the individual level parse earnings into within and between components. Lemieux (2006) showed that WGI is higher among more educated and experienced workers. Country-level variables account for compositional differences across countries. 22 This category combines Conservative and Mediterranean regimes due to small Mediterranean samples. Results are similar if Conservative and Mediterranean countries are separated. 23 Additional information is included in the online Appendix. 24 Comparison using Bayesian information criterion scores favoured simple regression models predicting total inequality using WGI instead of BGI change. 25 Replication of Figure 3 using female samples (online Appendix) results in similar conclusions. 26 Results in the online Appendix show WGI (levels and proportions) and various measures of post-fisc household income inequality to be significantly and positively correlated. 27 It is unlikely that cross-national variation fully reduces to omitted occupation variables. For example, I conducted supplementary analyses using US Census microdata and estimated models with and without 82,000 occupation-by-industry groups in 2010, in addition to standard education-by-work experience interactions. Including these 82,000 occupational groups decreased the relative contribution of WGI in this data set from 65% to 55%. 28 Race and ethnic information is available in few samples. Supplementary Data Supplementary data are available at ESR online. Acknowledgements The author thanks Arthur S. Alderson, Clem Brooks, Andrew Halpern-Manners, Tim Liao, Monica McDermott, Patricia McManus, and Jane VanHeuvelen for their helpful comments and discussion. Funding This work has been supported by a National Science Foundation dissertation improvement grant (no. 1519186). References Alderson A. S., Nielsen F. ( 2002). Globalization and the great U-turn: income inequality trends in 16 OECD countries. American Journal of Sociology , 107, 1244– 1299. Google Scholar CrossRef Search ADS   Atkinson A., Piketty T., Saez E. ( 2011). Top incomes in the long run of history. Journal of Economic Literature , 49, 3– 71. Google Scholar CrossRef Search ADS   Autor D., Katz L., Kearney M. ( 2008). Trends in US wage inequality: revising the revisionists. Review of Economics and Statistics , 90, 300– 323. Google Scholar CrossRef Search ADS   Barbieri P. ( 2009). Flexible employment and inequality in Europe. European Sociological Review , 25, 621– 628. Google Scholar CrossRef Search ADS   Barbieri P., Cutuli G. ( 2016). Employment protection legislation, labour market dualism, and inequality in Europe. European Sociological Review , 32, 501– 516. Google Scholar CrossRef Search ADS   Barro R., Lee J. ( 2010). A new dataset of educational attainment in the world, 1950-2010. Journal of Development Economics , 104, 184– 198. Google Scholar CrossRef Search ADS   Bidwell M. ( 2011). Paying more to get less: the effects of external hiring versus internal mobility. Administrative Science Quarterly , 56, 369– 407. Google Scholar CrossRef Search ADS   Blau F., Kahn L. ( 2005). Do cognitive test scores explain higher US wage inequality? Review of Economics and Statistics , 87, 184– 193. Google Scholar CrossRef Search ADS   Bollen K. A., Entwisle B., Alderson A. S. ( 1993). Macrocomparative research methods. Annual Review of Sociology , 19, 321– 351. Google Scholar CrossRef Search ADS   Bol T., Weeden K. ( 2014). Occupational closure and wage inequality in Germany and the United Kingdom. European Sociological Review , 31, 354– 369. Google Scholar CrossRef Search ADS   Bradley D. et al.   ( 2003). Distribution and redistribution in postindustrial democracies. World Politics , 55, 193– 228. Google Scholar CrossRef Search ADS   Brady D., Fullerton A., Cross J. M. ( 2010). More than just nickels and dimes: a cross-national analysis of working poverty in affluent democracies. Social Problems , 57, 559– 585. Google Scholar CrossRef Search ADS PubMed  Card D., Dinardo J. ( 2002). Skill-biased technological change and rising wage inequality: some problems and puzzles. Journal of Labor Economics , 20, 733– 783. Google Scholar CrossRef Search ADS   Cheng S. ( 2014). A life course trajectory framework for understanding the intracohort pattern of wage inequality. American Journal of Sociology , 120, 633– 700. Google Scholar CrossRef Search ADS   DiPrete T. ( 2007). What has sociology to contribute to the study of inequality trends? A historical and comparative perspective. American Behavioral Scientist , 50, 603– 618. Google Scholar CrossRef Search ADS   Esping-Andersen G. ( 1999). Social Foundations of Postindustrial Economies . Oxford, UK: Oxford University Press. Google Scholar CrossRef Search ADS   Gallie D. ( 2007). Production regimes and the quality of employment in Europe. Annual Review of Sociology , 33, 85– 104. Google Scholar CrossRef Search ADS   Gangl M. ( 2004). Welfare states and the scar effects of unemployment: a comparative analysis of the United States and West Germany. American Journal of Sociology , 109, 1319– 1364. Google Scholar CrossRef Search ADS   Gebel M., Giesecke J. ( 2011). Labor market flexibility and inequality: the changing skill-based temporary employment and unemployment risks in Europe. Social Forces , 90, 17– 39. Google Scholar CrossRef Search ADS   Goldin C., Katz L. ( 2008). The Race between Education and Technology . Cambridge: Belknap Press of Harvard University Press. Goos M., Manning A. ( 2007). Lousy and lovely jobs: the rising polarization of work in Britain. The Review of Economics and Statistics , 89, 118– 133. Google Scholar CrossRef Search ADS   Gornick J., Milanovic B. ( 2015). Income inequality in the United States in cross-national perspective: redistribution revisited. LIS Center Research Brief , 1– 7. Hauser S., Xie Y. ( 2005). Temporal and regional variation in earnings inequality: urban China in transition between 1988 and 1995. Social Science Research , 34, 44– 79. Google Scholar CrossRef Search ADS   Jaumotte F., Buitron C. ( 2015). Inequality and Labor Market Institutions. IMF Staff Discussion Note . International Monetary Fund, Washington, DC. Jencks C. et al.   ( 1972). Inequality: A Reassessment of the Effect of Family and Schooling in America . New York: Harper. Juhn C., Murphy K., Pierce B. ( 1993). Wage inequality and the rise in returns to skill. Journal of Political Economy , 101, 410– 442. Google Scholar CrossRef Search ADS   Kenworthy L. ( 2004). Egalitarian Capitalism: Jobs, Incomes, and Growth in Affluent Countries . New York: Russell Sage Foundation. Kollmeyer C. ( 2015). Globalization and income inequality: how public sector spending moderates this relationship in affluent countries. International Journal of Comparative Sociology , 56, 3– 28. Google Scholar CrossRef Search ADS   Kristal T., Cohen Y. ( 2016). The causes of rising wage inequality: the race between institutions and technology. Socio-Economic Review , 15, 187– 212. Leicht K. ( 2008). Broken down by race and gender? Sociological explanations of new sources of earnings inequality. Annual Review of Sociology , 34, 237– 255. Google Scholar CrossRef Search ADS   Leicht K. ( 2016). Getting serious about inequality. Sociological Quarterly , 57, 211– 231. Google Scholar CrossRef Search ADS   Lemieux T. ( 2006). Increasing residual wage inequality: composition effects, noisy data, or rising demand for skill? American Economic Review , 96, 461– 498. Google Scholar CrossRef Search ADS   Levy F., Murnane R. ( 1992). United States earnings levels and earnings inequality: a review of recent trends and proposed explanations. Journal of Economic Literature , 30, 1333– 1381. Liao T. ( 2016). Evaluating distributional differences in income inequality. Socius , 2, 1– 14. Liu Y., Grusky D. ( 2013). The payoff to skill in the third industrial revolution. American Journal of Sociology , 118, 1330– 1374. Google Scholar CrossRef Search ADS   Luxembourg Income Study Database ( 2017). http://www.lisdatacenter.org (various countries; May–September 2017). Luxembourg: LIS. McCall L. ( 2000). Explaining levels of within-group wage inequality in US labor markets. Demography , 37, 415– 430. Google Scholar CrossRef Search ADS PubMed  MacLean A., Kleykamp M. ( 2016). Income inequality and the veteran experience. The Annals of the American Academy of Political and Social Science , 663, 99– 116. Google Scholar CrossRef Search ADS   Mandel H., Shalev M. ( 2009). How welfare states shape the gender pay gap: a theoretical and comparative analysis. Social Forces , 87, 1873– 1911. Google Scholar CrossRef Search ADS   Moller S., Alderson A. S., Nielsen F. ( 2009). Changing patterns of income inequality in US counties, 1970–2000. American Journal of Sociology , 114, 1037– 1101. Google Scholar CrossRef Search ADS   Mouw T., Kalleberg A. ( 2010). Occupations and the structure of wage inequality in the United States, 1980s to 2000s. American Sociological Review , 75, 402– 431. Google Scholar CrossRef Search ADS   Organisation for Economic Co-operation and Development (OECD) ( 2015). It Together: Why Less Inequality Benefits All . Paris: OECD Publishing. Sachweh P., Olafsdottir S. ( 2012). The welfare state and equality? Stratification realities and aspirations in three welfare regimes. European Sociological Review , 28, 149– 168. Google Scholar CrossRef Search ADS   Streeck W. ( 2011). E Pluribus Unum? Varieties and commonalities of capitalism. In Granovetter M., Swedberg R. (Eds.), The Sociology of Economic Life , 3rd edn. New York, NY: Westview, pp. 419– 455. Thelen K. ( 2014). Varieties of Liberalization and the New Politics of Social Solidarity . Cambridge, UK: Cambridge University Press. Google Scholar CrossRef Search ADS   VanHeuvelen T. ( 2018). Recovering the missing middle: a mesocomparative analysis of within-group inequality, 1970-2011. American Journal of Sociology , 123, 1064– 1116. Google Scholar CrossRef Search ADS   Visser J., Checchi D. ( 2009). Inequality and the labour market: unions. In Salverda W., Nolan B.,, Smeeding T. (Eds.), The Oxford Handbook of Economic Inequality . Oxford, UK: Oxford University Press. Weil D. ( 2014). The Fissured Workplace: Why Work Became so Bad for so Many and What Can Be Done to Improve It . Cambridge, MA: Harvard University Press. Google Scholar CrossRef Search ADS   Western B., Bloome D. ( 2009). Variance function regressions for studying inequality. Sociological Methodology , 39, 293– 326. Google Scholar CrossRef Search ADS   Western B., Rosenfeld J. ( 2011). Unions, norms, and the rise in U.S. wage inequality. American Sociological Review , 76, 513– 537. Google Scholar CrossRef Search ADS   Western B. et al.   ( 2012). Economic insecurity and social stratification. Annual Review of Sociology , 38, 341– 359. Google Scholar CrossRef Search ADS   Williams M. ( 2012). Occupations and British wage inequality, 1970s–2000s. European Sociological Review , 29, 841– 857. Google Scholar CrossRef Search ADS   Xie Y., Killewald A., Near C. ( 2016). Between- and within-occupation inequality: the case of high-status professions. The Annals of the American Academy of Political and Social Science , 663, 53– 79. Google Scholar CrossRef Search ADS PubMed  Zhou X. ( 2014). Increasing returns to education, changing labor force structure, and the rise of earnings inequality in Urban China, 1996–2010. Social Forces , 93, 429– 455. Google Scholar CrossRef Search ADS   © The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

European Sociological ReviewOxford University Press

Published: May 9, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off