Portal de Periódicos da CAPES

The welfare effects of occupational segregation by gender and race: Differences across US Regions

2020; Elsevier BV; Volume: 99; Issue: 6 Linguagem: Inglês

10.1111/pirs.12551

1435-5957

Olga Alonso‐Villar, Coral del Río,

Gender, Labor, and Family Dynamics

Using tools rooted in welfare economics, this paper explores the social welfare loss that arises from occupational segregation by gender and race in the US at the regional level. After controlling for characteristics, the losses are lower in the Northeast than in the South and West according to a wide range of indicators, including those that take into account the relative size of disadvantaged groups (incidence), the magnitude of their losses (intensity), and the inequality among those groups. The West has the highest (conditional) losses, although the intensity of the phenomenon barely differs from that in the South or Midwest. Este artículo utiliza herramientas arraigadas en la economía del bienestar para estudiar la pérdida de bienestar social que surge de la segregación ocupacional por género y raza en los Estados Unidos a nivel regional. Después de controlar las características, las pérdidas son menores en el Nordeste que en el Sur y el Oeste, de acuerdo con una amplia gama de indicadores, incluidos los que tienen en cuenta el tamaño relativo de los grupos desfavorecidos (incidencia), la magnitud de sus pérdidas (intensidad) y la desigualdad entre grupos. El Oeste tiene las pérdidas más altas (condicionales), aunque la intensidad del fenómeno apenas difiere de la del Sur o la del Medio Oeste. 本稿では、福祉経済学に根付くツールを用いて、地域レベルでの米国における性別および人種的な職業分離から生じる社会福祉的な損失を探索する。特性を調整した後、不利な立場にある集団の相対的な大きさ (発生率) 、その損失の大きさ (強度)、およびその集団間における不平等を考慮したものを含む様々な指標によると、損失は南部および西部よりも北東部で低い。西部は最も (条件付きの) 損失が大きいが、この現象の強度は南部や中西部とはほとんど変わらない。 Gender and race/ethnicity are important traits that help explain why individuals hold the jobs they do (Branch, 2007; Reskin & Bielby, 2005). Men of racial minorities and women tend to be concentrated in occupations characterized by lower wages and opportunities (Blau & Winkler, 2018; Del Río & Alonso-Villar, 2015; Kaufman, 2010). However, the mechanism of segregation is complex: members of a group may benefit from one characteristic (e.g., gender) but be harmed by the other (e.g., race). The intersection of gender and race creates new categories with their own identities (Browne & Misra, 2003; Darity, Hamilton, & Stewart, 2015), a matter that has received little inquiry in the segregation literature, which has been focused mainly on segregation by either gender or race. To illustrate the magnitude of the phenomenon in the United States, Figure 1 shows the Ip segregation index (Silber, 1992), which is a generalization of the dissimilarity index to a multigroup case. 1 By using a detailed occupational classification (387 categories) and 12 gender–race/ethnicity groups, we find that during the period 2008–2012, 28% of workers would have had to change occupations to eliminate gender–race/ethnicity segregation, a percentage quite similar to that observed in 1980 (roughly 30%). When using a broad classification (7 categories) instead, segregation is much lower (17%), which shows the importance of using detailed occupational classifications to capture the phenomenon in its entirety. Occupational segregation is a mechanism that generates economic inequalities among groups (Blau & Winkler, 2018; Mouw & Kalleberg, 2010) and plays a significant role in explaining the pay gaps of men of racial minorities and women (Blau & Kahn, 2017; Cotter, Hermsen, & Vanneman, 2003; Cunningham & Zalokar, 1992; Kaufman, 2010; Petersen & Morgan, 1995). 2 To the extent that wage inequality among occupations increases over time, occupational segregation perpetuates those inequalities. Figure 2 shows the evolution of total wage inequality in the US (measured with the popular Theil0 inequality index), which is decomposed into two terms: one that shows the wage inequality that exists in the economy due to differences among individuals who work in the same occupation (within component) and another that shows the wage inequality that arises from the fact that occupations pay differently (between component). The chart reveals that wage disparities within occupations (within component) remained stagnant over the last three decades. Wage inequality increased mainly because wage differentials among occupations in 2008–2012 were almost double they were in 1980 (see between component). There is considerable body of literature that explains this between-occupation polarization based on a variety of factors: skill-biased technological change, the growth of low-skill service jobs, changes in the composition of the labour force, increasing globalization and delocalization processes, differences in the institutional mechanisms of social closure by occupations, and differences in returns to education (Acemoglu, 2002; Author & Dorn, 2013; Mouw & Kalleberg, 2010). However, we know little about how this polarization has impacted the various gender–race groups and the magnitude of the associated aggregated welfare losses. This literature gap is especially evident at the subnational level because, when accounting for women and men of various races, most scholarship on occupational segregation in the US has been carried out at the national level. However, in different regions, groups may face labour markets that feature different industrial structures, demographic compositions, and education levels—factors that may facilitate or hinder the integration of some groups into the labour market (Patrick, Stephens, & Weinstein, 2016; Reid, Adelman, & Jaret, 2007). Furthermore, cultural and social stereotypes are not homogenous nationwide; whites tend to hold more conservative attitudes toward race and gender in the South than elsewhere (Charles, Guryan, & Pan, 2018; Kuklinski, Cobb, & Gilens, 1997). Significant regional differences also exist in political gender equality indicators, with the Northeast outscoring the Midwest, West, and South (Di Noia, 2002). The incorporation of gender–race groups into the economy and the patterns of labour market segmentation are not independent of the institutional, social, and economic environment surrounding labour markets (Odland & Ellis, 1998; Peck, 1996). This paper aims to explore the economic consequences of occupational segregation in the US at a regional level by considering 387 occupational categories and using an intersectional framework that distinguishes among 12 gender–race/ethnicity groups. The analysis draws on the 1980, 1990, and 2000 decennial censuses, as well as the 2008–2012 five-year sample of the American Community Survey (ACS), which enables us to explore the evolution of the phenomenon over a 30-year period. We study subnational variation across the four census regions: Northeast, Midwest, South, and West. Although these regions comprise labour markets under different state authorities, they have a long tradition in comparative statistical analyses—because they group states based on historical, demographic, and economic characteristics—and are large enough to account for sizable samples of women and men of six races/ethnicities, which is especially convenient when using detailed occupational classifications. To provide additional insight at a finer geographic scale, we also include evidence of the phenomenon for the 25 states in which the size sample is large enough to differentiate among at least five races/ethnicities. 3 To measure the consequences of segregation in each region, we use the tools proposed by Alonso-Villar and Del Río (2017a) and Del Río and Alonso-Villar (2015, 2018), which are rooted in the welfare economics and deprivation/poverty literature. This enables us to check whether, when accounting for not only the occupational achievements of the 12 demographic groups but also the occupational inequalities existing within them, the social welfare losses are larger in the South than elsewhere. Another advantage of this approach is that when aggregating the situation of the 12 groups, the losses derived from the concentration of deprived groups in low-paying occupations are not offset by the gains of those groups concentrated in high-paying occupations. Furthermore, we explore the causes of observed interregional disparities in social welfare losses using the propensity score procedure proposed by DiNardo, Fortin, and Lemieux (1996), as adapted by Gradín (2013) and Gradín, Del Río, and Alonso-Villar (2015). To do this, we build a counterfactual economy in which no regional differences exist in terms of gender–race composition, education levels, immigration profile, or industrial structure, and check whether regional disparities in welfare losses remain the same. This paper departs from most studies on segregation in several ways. First, we address segregation using an intersectional framework. Second, we explore spatial disparities in social welfare losses while accounting for detailed demographic groups. Third, we address the economic consequences of segregation using a social welfare function approach, which allows us to account for not only the average occupational achievements of each group (i.e., the average wage of the occupations in which the group works), but also the occupational inequality within the group. Fourth, we account for spatial differences in demographics (gender, race, education, English proficiency, and years of US residence) and industrial structures that may explain those regional disparities. The contribution of each explanatory factor is obtained using the Shapley decomposition, which is independent of the sequence in which the factors are introduced, thus improving the procedures usually employed in the wage gap literature. Our analysis indicates that regional disparities in social welfare losses associated with segregation by gender and race/ethnicity have increased in the US since 1980. During the period 2008–2012, the monetary losses associated with segregation in the West are estimated at 5.6% of all earnings in the region, whereas such losses in the South, Northeast, and Midwest are 4.9%, 4.5%, and 4.2%, respectively (in 1980, the losses ranged from 5.1% to 5.6%). Around half of these spatial disparities persist after controlling for regional characteristics—racial composition and, to a lesser extent, immigration profile being the most important factors. Conditional welfare losses due to segregation are lower in the Northeast than in the South and West according to a wide range of indicators, including those that account for the relative size of disadvantaged groups (incidence), the magnitude of their losses (intensity), and the inequality among those groups. The West has the highest conditional welfare losses, although the phenomenon's intensity barely differs from that in the Midwest and South. The paper is structured as follows. Section 2 provides a background on the role that occupational segregation plays in explaining the wage gap. Section 3 introduces the data and methods. Section 4 estimates the welfare losses that each region experienced during the last three decades due to occupational segregation by gender and race/ethnicity. Section 5 tries to explain those regional disparities. Section 6 contains the paper's main conclusions. Notwithstanding the large salary discrepancies that exist between women and men working in the same occupations (Goldin, 2014), segregation plays an important and increasing role in explaining the wage gap. In 2010, occupational segregation accounted for 32% of the gender wage gap, a much larger percentage than in 1980, when it was 11% (Blau & Kahn, 2017). In the case of college graduates, Goldin (2014) found that occupations explain between 30% and 42% of the gender wage gap, depending on the method used. This is a high share, but it is lower than the within-occupation gap. As is widely documented, gender segregation decreased during the second half of the 20th century, 4 mainly due to the entry of new cohorts of women with higher educational achievements than their predecessors into the workforce (Blau, Brummund, & Liu, 2013) and as a result of political pressure for gender equality that became a force in the 1970s, yet essentially halted just two decades later (Tomaskovic-Devey et al., 2006). Between 1970 and 2015, the percentage of women working in predominantly male occupations (e.g., architects, chemists, dentists, industrial engineers, lawyers) increased considerably, though women still lagged behind in most science, technology, engineering, and mathematics (STEM) fields and men entered formerly female occupations to a much lesser extent (Blau & Winkler, 2018). In 2010, four out of five women (respectively 5 out of 10 men) still worked in occupations with at least 75% of female (respectively male) employment (Hegewisch, Willians, & Henderson, 2011). As with gender, occupational segregation also plays an important role in explaining the black–white wage gap and may account for 20% of that gap (Grodsky & Pager, 2001) or more (Kaufman, 2010). Segregation between blacks and non-blacks decreased in the second half of the 20th century, but segregation between Hispanics and non-Hispanics increased (Queneau, 2009). In any case, race/ethnicity does not affect women and men equally: racial segregation is higher for men than it is for women (Alonso-Villar, Del Río, & Gradín, 2012; Spriggs & Williams, 1996). Furthermore, segregation by gender does not affect all racial/ethnic groups in the same way: it is higher for Hispanics and lower for Asians than it is for other groups (Hegewisch, Liepmann, Hayes, & Hartmann, 2010). In an intersectional framework in which gender and race/ethnicity are considered jointly, occupations play an important role. Working in feminized occupations and in local labour markets with high levels of gender segregation negatively impacts the wages of white, Hispanic, Asian, and, especially, African American women (Cotter et al., 2003). Among individuals with bachelors' degrees, occupational segregation appears to explain at least half of the wage disadvantage of white, black, and Hispanic women and black men relative to the average wage of high-skilled workers (Del Río & Alonso-Villar, 2015). Roughly half of the wage advantage of white men also comes from their occupational sorting (the occupational advantage is even more intense in the case of Asian men). This indicates that differences in the groups' occupational sorting help to explain intergroup earning differentials. Scholarship on occupational segregation by gender or race has been mainly undertaken at the national level in the US, although a few studies have reported important spatial disparities (Alonso-Villar & Del Río, 2017b; Gradín et al., 2015; Lorence, 1992). However, these studies did not account for the racial/ethnic diversity of women and men (i.e., they did not distinguish among 6 races/ethnicities), which prevented them from revealing the full extent of labour inequality. This paper aims to address those spatial inequalities using a comprehensive intersectional approach that distinguishes among men and women of various races/ethnicities. To undertake such an analysis using a detailed occupational classification, the study is conducted at a 4-region level to ensure enough observations for all groups in each location, although we provide further information for some states. We use the US decennial censuses (covering 1980, 1990, and 2000) and the 2008–12 five-year sample of the American Community Survey—which replaced the census long form after 2000 and reports data on occupation. The dataset, which offers harmonized information, was provided by the Integrated Public Use Microdata Series (IPUMS-USA; Ruggles et al., 2010). 5 The 5-year sample covers 6.9 million workers and includes the two years before and after 2010. The number of workers in the decennial censuses ranges from 5 million in 1980 to 6.4 million in 2000. We distinguish the Northeast, Midwest, South, and West census regions. With respect to the occupational breakdown, we use the consistent long-term classification provided by IPUMS-USA, which is based on the 1990 Census Bureau classification and accounts for 387 job titles. We use a detailed classification of occupations because otherwise differences among demographic groups within broad categories of occupations would not be captured and so the measurement of segregation, and its economic consequences, would be underestimated (as Figure 1 illustrates). The wage of each occupation is proxied by the average hourly wage, which is estimated based on reported wages and number of hours worked—after trimming the tails of the hourly wage distribution to prevent outliers from skewing the average (for this we eliminate all workers whose wages are either zero, below the 1st percentile or above the 99th percentile of positive values in that occupation). We consider the 12 mutually exclusive groups of workers that result from combining gender with six racial/ethnic groups: the four major single-race groups not of Hispanic origin (which we label as whites, African Americans, Asians, and Native Americans); Hispanics irrespective of race (all labelled as Hispanics); and "other races" (non-Hispanics that self-report some other race or more than one race). 6 To quantify a region's social welfare loss, we follow two steps. First, using the index proposed by Alonso-Villar and Del Río (2017a), we quantify the differential between the well-being each gender–race/ethnicity group in that region derives from its occupational sorting and the well-being the group would derive if it were evenly distributed across occupations. 7 We also calculate these losses (gains) in monetary terms using the index developed by Del Río and Alonso-Villar (2015), which has a very intuitive interpretation although it implies disregarding within-group inequalities. These two indices are positive when the group tends to fill highly paid occupations, negative when the opposite holds, and are equal to zero when the group has no segregation or all occupations have the same wage. Second, we aggregate the well-being losses of the groups (in each region) via the approach developed in Del Río and Alonso-Villar (2018). This method is similar to the one followed in the literature on deprivation and poverty, since a group's well-being loss can be viewed as a shortfall with respect to the case of no segregation. To interpret this index, note that is positive (respectively negative) when the group is overrepresented (respectively underrepresented) in high-paying occupations and underrepresented (respectively overrepresented) in low-paying ones. This is so because any occupation j in which the group is overrepresented contributes positively to the index if and only if that occupation's wage is higher than the average wage . The value of depends not only on the group's earnings but also on the within-group inequality that arises from the fact that some group's members may work in low-paying occupations and others in high-paying ones. This inequality aversion, which diminishes the well-being of groups with larger discrepancies in the occupational achievements of their members, is embodied in the concavity of the ln function. The difference between these two indices is that assumes inequality aversion whereas Γg does not. Inequality aversion is a convenient property when the group has members with very different occupational achievements, as is the case of Asian women and men. The approach just described allows us to transcend the mere measurement of unevenness, on which most segregation analyses focus, to address the economic consequences of that unevenness, which is where the main problem lies. The above tool is insufficient for determining the welfare loss of an entire region due to segregation. The reason is that some groups may derive gains—while other groups endure losses—stemming from their occupational sorting. One way of dealing with this issue would be to calculate the average well-being losses or gains of the groups involved. However, this approach presumes that advantaged groups' gains offset disadvantaged groups' losses of the same magnitude—an assumption that would be called into question by those people who are inequality averse. A more suitable way of quantifying a region's social welfare loss resulting from the occupational sorting of its demographic groups is to use, as proposed by Del Río and Alonso-Villar (2018), a framework similar to the one employed in the literature on deprivation and poverty. The abscissa value at which the curve becomes horizontal, denoted by h, represents the incidence of the phenomenon—namely, the population share that the groups with well-being losses account for. The maximum height of the curve conveys the problem's intensity (i.e., the total cumulative losses of the groups divided by T). Finally, the curvature of the WLAS curve between the origin and point h illustrates the inequality that exits among disadvantaged groups (i.e., those with well-being losses). 8 These curves are a powerful tool because, when one curve dominates another (i.e., when the former is never above the latter and is below it at some point) then we can conclude that the social welfare loss in the first situation is lower than that in the second according to a wide range of indices that satisfy basic properties commonly accepted in the literature on poverty and deprivation. When α > 1, these indices are consistent with the dominance criterion defined by the WLAS curves. It follows that, when a curve dominates another, we can ensure that with any of these indices the social welfare losses would be lower in the economy represented by the former curve. When no domination exists between the two curves (i.e., if the curves cross) the outcome can change depending on which index is used. Note that index FGT0 (which represents the proportion of individuals belonging to disadvantaged groups, i.e., h) and index FGT1 (which measures the well-being losses of the disadvantaged groups divided by T, i.e., the height of W) are not consistent with the WLAS dominance criterion. Nevertheless, our empirical analysis employs both the FGT0 and FGT1 indices because they allow measuring the incidence and intensity of the phenomenon separately. Our analysis relies also on the FGT2 index, which combines the three dimensions of the phenomenon—its incidence, intensity, and inequality among deprived groups—at the same time. We begin the analysis by seeing whether there exist significant differences in the regional social welfare losses associated with the occupational sorting of the gender–race/ethnicity groups that work in each of them. After examining the data at the end of our period of analysis (ACS 2008–12, 5-year sample), we will analyse the trends observed since 1980 (based on the decennial censuses). To obtain an overview of the problem's intensity, we first calculate the regions' losses in monetary terms (without accounting for within- and between-group inequalities). We find that the West outscores the other regions. Losses in the West represent 5.6% of the region's earnings, which is higher than in the South (4.9%), Northeast (4.5%), and Midwest (4.2%). 9 How do the regions rank when intra- and inter-group inequalities are also accounted for? Figure 4 (and Table A2 in the Appendix) highlights the welfare losses based on the FGT2 index, which accounts for not only the intensity of the phenomenon (the average loss per worker), but also its incidence (the percentage of workers who belong to groups with losses) and the inequality among deprived groups. 10 The map shows that the problem is more severe in the West and that the Midwest has the lowest losses in the country. To shed light on which groups are behind the regions' losses, Figure 5 shows the index for each group in each region. With the exception of white women (and men from "other races"), the groups with well-being losses associated with their occupational sorting are the same in all regions. Hispanic women and men have the largest losses, especially in the Northeast and West. African American women and Native American women come next in the ranking, with losses above those of their male counterparts. Like their female peers, African American men fare better in the West whereas Native American men and women have better occupational sorting in the South. White women only experience (small) gains in the West, and their largest losses occur in the Midwest. The occupational achievements of white men are also larger in the West than elsewhere. The analysis indicates that white women and men tend to be better off in places with more racial diversity. This is consistent with previous findings for these two groups at the metropolitan level (Alonso-Villar & Del Río, 2017b) and with theories of labour segmentation and queues according to which, when applying to a job, individuals are ranked by their gender and race (Kaufman, 2010; Reskin & Ross, 1990). Asian populations, especially men, seem to fare better in the Midwest and South, regions in which they have a lower presence. 11 To provide a sense of the spatial discrepancies within regions, Figure 6 shows an estimate of each state's FGT2 index, accounting for more or fewer races/ethnicities depending on the case (we do not have enough observations for all incumbent female and male groups in all states). 12 We can only distinguish among six races/ethnicities or five (which results from subsuming either Native Americans or Asians within "other races") in 25 states. 13 In the remaining 26 states, which are less racially diverse, we can only account for 2 racial groups (whites and non-whites), 3, or 4. The map also reports the weight each state represents in terms of regional employment. Western states, especially California (the state with the highest employment by far), Arizona, Colorado, and Washington tend have the largest FGT2 values in the country. In contrast, the values in Midwestern states are among the lowest (Illinois has the highest value). In the South, Texas stands out as the state with the largest losses, which are lower than those in California. Finally, the losses in the Northeast come mainly from New Jersey and, to a lesser extent, New York. Figure 7 reveals that the WLAS (social welfare loss associated with segregation) curve of the Midwest dominates the others (i.e., it is below than or equal to those of the other regions). This means that the Midwest has the country's lowest social welfare losses not only for the FGT2 index (as shown earlier) but for a wide range of indices (in particular, all FGTα indices for which α > 1). The WLAS curve of the Northeast indicates social welfare losses that are only slightly greater than those in the Midwest, at least in terms of intensity and incidence. Yet the WLAS curve of the Northeast exhibits a much greater curvature than that of the Midwest, which suggests that the difference between these regions are mainly the result of larger discrepancies in well-being losses among deprived groups in the Northeast than in the Midwest. Figure 7 also shows that the West's WLAS is clearly dominated by that for the other regions, which implies that social welfare losses are the greatest in this region according to many indices. Observe that, in the West, the population share belonging to groups with losses (incidence) is substantially lower than in the other regions (32% vs. more than 50%). The reason of this is that the West is the only region where white women had gains associated with their occupational sorting (Figure 5). The West being dominated occurs because there the phenomenon's intensity far exceeded that in the other regions. Finally, we remark that a ranking between the South and the Northeast is not possible because the curves intersect. In fact, the intensity is clearly higher in the South whereas the FGT2 index is higher in the Northeast, as shown earlier. Figure 8, which plots the WLAS for all regions in 1980, reveals a considerably different scenario. Here, all curves cross, so we are unable to determine which regions are better-off or worse-off. If we ignore all groups except those with the largest well-being losses (which account for 20% of the population), the four regions are virtually indistinguishable. Only when the cumulative percentage of such workers exceeds 20% does the curve of the Midwest start to deviate from those for the other regions, which is indicative of a more severe problem in that region in terms of intensity. The chart also shows that, in 1980, the curves of the other three regions differed little from each other. When expressed in monetary terms, rather than in terms of well-being, the losses in the Midwest represent 5.6% of the regions' wages, whereas the losses in the other regions ranged between 5.1% and 5.3%. Figures 9, 10, and 11 illustrate the evolution of the FGT2, FGT0, and FGT1 indices, respectively. The Midwest, which is less racially diverse than the other regions (see Table A1 in the Appendix), improved its relative position in terms of the FGT2 index (Figure 9), at least in part, due to the remarkable reduction in the intensity of the phenomenon (Figure 11). Note that white women in the Midwest accounted for a larger (and increasing) share of workers than in the other regions, so how they fared has an important effect on the region's losses. In 1980, the greatest well-being losses for white women did, in fact, occur in the Midwest (see Figures A1-A4 in the Appendix). Although this pattern has remained stable over time, the actual amount of these losses has decreased considerably since 1980. On the other hand, the relatively small group of Asian women experienced notable occupational advances in the period as well. They were experiencing well-being gains (rather than losses) by the 1990s. However, the experience of male groups in the Midwest does not conform to reductions in the FGT2 index (Figure A2). African Americans and Hispanics worsened over the period, as they shifted from having small we