Abstract
There are many contending paradigms in the study of inequality and stratification, with little dialogue across empirical, methodological and theoretical divides. To bridge some of these gaps, this article presents and analyzes a new dataset on urban wages across the world. The data provide compelling evidence that national residence has been a significant and stable force shaping wage distribution. But the data also indicate that some occupations (e.g. tradable-good producers in low-wage regions) have experienced significant upward mobility. These processes highlight the need to critically reassess how the categories of ‘skilled’ and ‘unskilled’ are mobilized to justify global inequality.
Introduction
Assessing the precise character of recent transformations in global stratification is a challenging task. Longstanding disagreements about which theories and methodological procedures are best at capturing these transformations are compounded by a lack of a basic consensus on the relevant empirical trends at hand. Data on social inequality and stratification continue to be collected primarily within national boundaries, and this, rather than theoretical considerations, still drives the level of analysis at which these fundamental social processes tend to be modeled. Almost by default, many sociologists then proceed to cast ‘globalization’ as a recent phenomenon: a new, exogenous independent variable whose significance can be assessed by including some appropriate indicators into existing, national models of stratification or mobility.
Moreover, while some concentrate in identifying the extent and direction of change in between- and within-country inequality cross-nationally, others study the changing configuration of occupations and wages within nations (e.g. as manifested in the evolution of differentials between and within skilled and unskilled workers). Overall, there is still little productive dialogue across these cleavages.
Bridging such cleavages can contribute to a better understanding of contemporary processes of social inequality, stratification, and mobility. Studies of world income tend to use very broad indicators (such as per capita national income), so there is much to gain by incorporating a more detailed understanding of how changes in world inequality might be driven by shifting wage returns to occupations across the world. On the other hand, studies of wage stratification often focus exclusively on these phenomena as they take place and are shaped within nations (e.g. by type of political regimes, the strength of organized labor, returns to education, and so forth), but, as we contend in this article, we can more productively understand wage inequality and its transformation over time (e.g. through social mobility) as being global at its very core.
The article addresses a second cleavage in the social sciences. One paradigm considers inequality and stratification to reflect primarily the independent achievement of individuals (for example, through human capital attainment in the study of wage inequality or stratification within nations) or nations (for example, through industrialization and/or appropriate state policies in the study of global inequality). According to this paradigm, the advancement of modernization opens greater opportunities for convergence and upward social mobility (for individuals and/or nations), and inequalities based on criteria other than achievement are either of marginal importance or remnants from the past that have been withering away since the Industrial Revolution. To simplify, we refer to this approach as the modernizing paradigm.
A very different paradigm considers inequality and stratification to result from social relations that benefit some people at the expense of others. Again, broadly speaking, what we will label here as the critical paradigm argues that modernization and capitalism, subversive to some forms of exploitation and exclusion, give rise to new forms that perpetuate social inequalities and patterns of stratification based heavily on the ascribed characteristics of individuals and groups. For example, in looking at the income gap that developed between poor and wealthy nations through most of the 20th century, scholars adopting the modernizing perspective would characterize it as a temporary phenomena caused by the uneven cross-national pace of industrialization, likely to inevitably give way to growing convergence as poor nations take-off into modernization (Firebaugh, 2003; Rostow, 1960). In contrast, scholars within the critical paradigm would contend that the development of wealthy nations entailed the impoverishment of other nations, leading to growing or persistent inequalities (Chase-Dunn, 1975; Frank, 1967).
We continue to find these contending paradigms within more recent studies of both global inequality and within-nation occupational and wage stratification (we summarize this literature in Figure 1). In the study of global inequality, the main focus in recent years has been on the impact of the high rates of economic growth of China and India over the past 20 years. 1 For some policy-makers and social science scholars, the decline in world inequality experienced over the last 20 years has been very significant (e.g., Firebaugh, 2003; Firebaugh and Goesling, 2004; Sala-I-Martin, 2006). Moreover, these scholars echo the modernizing perspective by interpreting this decline as providing clear evidence that national incomes are destined to converge as markets and industrialization spread across the world.
Stylized representation of prevailing paradigms and fields of inquiry.
Others following the critical paradigm question the extent to which there is empirical evidence for a decline in world income inequalities. For example, Milanovic (2005) notes that while inequality between countries declined slightly over the last two decades of the 20th century, this decline is smaller if we take into consideration growing regional disparities within China, and disappears altogether if China is excluded from the sample; more importantly, Milanovic contends that global inequality (combining data on between- and within-country inequality) has not undergone any significant decline (for similar results see Schultz, 1997; Wade, 2004, 2008). As a whole, moreover, these critics are rather skeptical about the extent to which policy-makers in China and India will be able to sustain high rates of growth in the future, and on the likelihood that other low-income nations will be able to attain similarly high rates of growth.
The study of wage inequality and occupational stratification has taken a different direction. Here, while a critical perspective on occupational mobility continues to emphasize the extent to which ascription (e.g. family background or the type of father’s employment) continues to shape one’s eventual standing in the wage structure, there is a broader and more generalized consensus that such arrangements are converging towards achievement. Instead of those broader debates, the literature on occupational stratification has focused on the divergent patterns of wage inequality observed in high-income nations: a significant rise in inequality characterized some, such as the United States and Great Britain, over the last decades, while much of continental Europe experienced little or no rise in inequality (e.g. Acemoglu, 2002, 2003; Aghion, 2002; Alderson and Nielsen, 2002; Autor et al., 2006; Blau and Kahn, 1996; Freeman, 1993; Gottschalk and Smeeding, 1997; Gustafsson and Johansson, 1999; Harrison and Bluestone, 1988; Lemieux, 2008; Morris and Western, 1999; Piketty and Saez, 2003, 2006).
Within this literature, centered in economics but with occasional contributions from sociology as well, researchers often seek to explain a rising wage gap between skilled and unskilled workers. Some emphasize the impact of technological change generating rising demand for skills, thereby promoting the wage gap in question (e.g. Acemoglu, 2002; Autor et al., 2006). Others attribute greater weight to changes in the relative strength of wage-setting institutions, such as shifts in the relative importance of organized labor, deunionization and/or the welfare state (e.g. Blau and Kahn, 1996; Freeman, 1993; Morris and Western, 1999; Western, 1995). And some studies seek to combine both explanations, for example arguing that ‘institutional wage compression in Europe makes firms more willing to adopt technologies complementary to unskilled workers, making technical change less skill biased there’ (Acemoglu, 2003: 124). More recently, however, studies within this literature question whether the rise of wage inequality in countries such as the United States has been as sustained as frequently assumed, indicating instead that this rise was concentrated in the 1980s (Lemieux, 2008). Whether or not this is indeed the case is still a matter of debate awaiting further empirical study.
But either explicitly or implicitly, discussions of changes in between-country inequality often implicate the issue of changing patterns of differentiation between skilled and unskilled labor. The general assumption, even at the common sense level of popular perception, is that the rapid growth of countries such as China and India in recent decades, and/or the inherent decline in the income gap between wealthy and poor nations, are associated with the transfer of less skilled jobs from high- to low-income nations.
Many explain such trends through the Heckscher-Ohlin (H-O) trade model. The model predicts that as barriers fall and trade rises, different regions of the world will increase their export of goods produced using those factors that are relatively abundant locally (that is to say, relatively cheap). Eventually, through this process, factor prices (including wages) should converge across regions. In its simplest form, this approach envisions a global economy with two regions, each producing a specific type of good – one labor intensive and the other capital intensive. The region with the higher labor to capital ratio will export labor-intensive goods. This will increase the demand for labor within labor-intensive countries while undermining the demand for labor within the capital-intensive country (by offering cheaper labor-intensive goods), leading wages to converge between countries.
Scholars generally extend the H-O model to assume that poor countries have abundant low-skill labor and wealthy countries abundant high-skill labor; under such conditions, increased trade will compress the skill premium (by increasing the demand for low-skill labor) within the poorer countries and expand it (by increasing the relative demand for high-skill labor) within the wealthier ones (Kremer and Maskin, 2006). Such shifts might be induced directly through trade, or more indirectly by the incentives that growing trade might generate for intensifying the pace for technological change (hence increasing the relative demand for skilled labor) within wealthier countries. But generally, as adapted into the modernizing paradigm, the H-O model is assumed to predict an eventual convergence on the wages of poor and wealthy countries overall.
Fundamentally, the H-O model highlights the global labor market as a wage setting institution, and stresses that the experience of workers is shaped by global forces of supply and demand. Certainly, it is possible to try to explain occupational stratification within nations by focusing solely on wage-setting institutions that are national in scope, but a more productive understanding might be gained by acknowledging that even the most ‘national’ wage-setting institutions are embedded in a global market and are more accurately understood in this world-economic context. If global or national arrangements function to insulate/isolate privileged workers from global competition, ‘national’ wage-setting institutions are simultaneously ‘global’ at their very core.
By providing more detailed insights into the changing configuration of global labor markets, the study of wage inequalities and occupational stratification might contribute to a better theoretical understanding of the complex trends observed more broadly in world income inequalities. However, studies on occupational changes and wage inequality for the most part continue to focus primarily on the experience of high-income nations, and the intersections of this literature with the existing research of world income inequalities regretfully are few.
In this article we argue that both lines of inquiry, global inequality and occupational stratification, can be enriched through such an interaction. For the study of world inequality, the research on occupational stratification can contribute a more nuanced and detailed understanding of how changes in institutional arrangements and labor markets might be reshaping social processes of stratification and mobility at a more aggregate level. On the other hand, bringing issues of between-country inequality to bear on the discussion of changing labor markets can help overcome the constraints inherent to studying these markets and occupational stratification only as they take place within high-income nations. For example, the persistence of ascriptive exclusion in shaping wage stratification only becomes more salient when examining global labor markets. Ultimately, in our view, the development of such dialogues might lead us to recast the study of world inequality and occupational stratification as parts of a single field of inquiry: the study of wage inequality and stratification as processes that always have been global in their very essence (that is, not simply the interaction of two otherwise ‘autonomous’ processes involving ‘national wage stratification’ on the one hand and ‘global inequality’ on the other).
To this end, we present and explore new data on wage inequalities that are simultaneously more detailed but global and longitudinal. We draw on these data to argue that global inequality and occupational stratification both pivot around institutions that define and identify skill as simultaneously ‘national’ and ‘global’ processes. It is the supply and demand for skills that determine wages, but skill is difficult to assess, especially in non-routinized activities. To add rigor to an otherwise fluid concept, a host of national institutions exist to explicitly disseminate, identify and allocate skill, and researchers all too often confuse this operationalization of skill with actual technical capacity. Skill-identification is further complicated on a global scale because skill-managing institutions are overwhelmingly national. Assessing skill across a global labor market is a difficult task open to misinterpretation, misunderstanding and manipulation. This article makes modest steps towards this end, with the ultimate aim of demonstrating that how patterns of social stratification and mobility entail simultaneously the ‘national’ and the ‘global’.
Data and methodology
Ideally, to arrive at a more precise mapping of how stratification and mobility changed for the world as a whole over a given period of time, we would use data that provide information on within-country distributions (say, between the skilled and the unskilled, producers of tradable and non-tradable goods, men and women) while allowing an assessment of how these within-country population segments are performing relative to similar population segments of other countries. Such a mapping would enable us to assess how changes in the relative position of various population segments vis-à-vis one another (e.g. different occupations within a country, or similar occupations across nations) might be accompanied by shifts in wage differentials and the geographical distribution of the ‘skilled’ and the ‘unskilled’, and to estimate over time what have been the implications for changing returns to various strategies of social mobility (e.g. returns to skill and/or education).
But the scarcity of adequate comparable data and the theoretical assumptions hitherto guiding research on stratification and mobility continue to constrain efforts to construct such a mapping. In fact, empirical constraints are intractably linked to prevailing theoretical assumptions. Many conceive social stratification and mobility as taking place primarily within national boundaries, and these assumptions became deeply entrenched in data collection as the latter developed over the last century. Researchers draw data on inequality from national surveys of individuals and/or households, collected primarily by national statistical agencies for the purpose of shaping policies at a national level.
A new dataset on global wages
This article presents a set of disaggregated wage data that, while modest, moves us toward a global mapping of stratification and mobility. We consider that making available such a database itself provides an important contribution to the literature, as the available data on social stratification and mobility (e.g. returns to skill and education) is generally limited to a small number of high-income nations or methodologically inconsistent across countries. Our mapping begins to help us develop comparative estimates on wages for various population segments in a large number of high-, middle-, and low-income nations. Moreover, our data allow us to estimate how these returns have changed over time, and provide some insights into how changes in one set of nations might be related to shifts affecting other populations within our sample.
To move towards such an objective, we map occupational inequalities by drawing on the wage data available through the periodical publications of UBS (formerly the Union Bank of Switzerland). Since 1971, UBS has published a survey of prices and salaries every three years, designed to provide clients of the bank accurate international price and wage comparisons with greater specificity than the widely available price indices. At each site, with the aid of branches of the bank, the data were collected by UBS using at least two independent, unrelated organizations in each site (UBS, 1982: 3; 1985: 3). These data allow us to reconstruct, for over three dozen cities across the world, average wages, benefits, working hours and vacation days for over a dozen occupational categories (ranging from construction laborers and unskilled female factory workers, to bus drivers and primary school teachers, to managers and engineers).
The survey began in 1970, with the first results published in UBS (1971). The number of cities included in the survey changed over time, gradually growing from the 31 cities included in the 1970 survey to 73 in 2009. Except for one (Beirut), most of the 31 cities surveyed in 1970 remained included in 2009: most of these (23 cities, or 74.2% of the sample) were in Japan, Australia, Canada, the United States, and Western Europe. But included in 1970 were five Latin American cities and one each from Africa, Asia, and the Middle East. By the 2009 survey, as the number of sampled cities expanded, those located in Japan, Australia, Canada, the United States, and Western Europe grew to 32, but they now represent 43.8% of the sample, giving way to greater representation of cities in Eastern Europe (12), Asia (13), the Middle East (six), Africa (two) and Latin America (eight). The cities sampled in the exercises reported in this article are located in countries that together account for roughly 70 percent of the world’s population (over the 1982–2009 period, ranging from 66.1% in 1991 to 76.4% in 2003). The biggest deficiency in global coverage in the data is China (Shanghai and Beijing do not show up in the sample until 1997 and 2009, respectively). To partially compensate for this deficiency, some analyses draw on a modified sample in which we estimate past wages for these Chinese cities.
There were also changes in the number of occupations covered by the survey over time, but this dimension of the survey remained more constant after the late 1970s. The original 1970 survey included only five occupations (automobile mechanics, bank tellers, bus drivers, primary school teachers, and secretaries). Departments heads were added in 1973 and building laborers, female factory workers, and skilled industrial workers in 1976, and cooks, engineers, and saleswomen in 1979. These 12 occupations are tracked consistently to the present. Recently, product managers were added to the list in 2003, and call center agents in 2006.
From the very beginning, each of the occupational categories being surveyed was clearly defined. For example, ‘Primary School Teachers’ were defined in 1982 as individuals who have ‘taught in the public school system (not in private schools) for about 10 years, about 35 years old, married, no children’ (UBS, 1982: 34). In the same report, a female textile worker was an ‘Unskilled or semi-skilled operator in a medium-sized plant, about 25 years old, single’, and electrical engineers were ‘Employed by an industrial firm in the machinery or electrical equipment industry, electrical power company or similar; completed university studies (college, technical institute or institute of higher technical education) with at least 5 years of practical experience; about 35 years old, married, no children’. These definitions did not change substantially from survey to survey, though adjustments were made to reflect the new technological requirements.
The data thus gives us access to wage information for a large sample of city and occupation combinations (henceforth, COCs) (e.g. Chicago School Teachers, or Mexico City Bus Drivers). To compensate for the addition of new cities over time (the sampled COCs rises from 558 in 1982 to 1019 in 2009), we make use of two subsamples when drawing inferences about wage trends over time. First, we look at changes between 1982 and 2009 using a sample of 42 cities and 498 COCs that are represented in both years. Four of those cities are not represented in the intervening years, so we use a sample of 38 cities and 444 COCs when looking at wage trends through the period. 2
We use the reported gross wage data for each occupational category to derive an estimate of hourly wages, attained by dividing the pay per week by the average number of working hours (also reported in the data). For this particular exercise, we have chosen not to use the average vacation days reported by UBS: as opposed to the information on working hours, this item is reported with less consistency, and for some occupations (such as teachers) it is not reported at all.
Reported wages in the data represent an average wage for workers across several companies – the criteria used to sample wages are much more precise and homogenous than what can be usually found in national surveys. Because of the precision of these definitions, we can gain some confidence that wage differences between engineers in Chicago and Mumbai, for example, are not primarily because Chicago engineers have stronger educational credentials or more experience. As noted earlier, our data do not directly assess technical capacity (but neither do any other data currently in existence): our data do not allow us to prove that skill levels are identical across sites, but we can be confident that workers across the nations in question have comparable skill credentials.
The data discriminate between gross and net wages. For the exercise presented in this article, we have chosen to focus on gross or ‘pre-fisc’ wage data (wages before taxes). This decision is driven primarily by the focus of this article, which is to assess longitudinal changes in wage costs comparatively across the world and not the post-facto redistributive efforts of the state (Lindert, 2000). But in addition, a focus on net wages would require that some attention be paid to differences in the type of services (such as social security or health benefits) received from the state – this involves complicated questions we are addressing in other iterations of our research.
The article presents data on hourly wages in both current and constant dollars (the latter adjusts for changes due to inflation, converting the UBS current wages to constant 2009 US dollars using the consumer price index adjustment factors (CPI-U) from the Bureau of Labor Statistics). Our focus is on the operation of the global labor market: the supply, demand and intervening institutional arrangements that shape nominal wages. Some might argue that such an exercise distorts actual wage differentials by overlooking differences in Purchasing Power Parities (PPPs). We indeed do not look at PPP-adjusted wages in this article. In part this is because we explore the issue of differences and inequality in consumption in other parts of our research. We treat this topic separately because the appropriate measurement of PPPs involves variables and indicators that are very complex and often problematic, in ways that are usually ignored by many social science scholars that make a frequent but uncritical use of such measures (in fact, as suggested in Korzeniewicz et al., 2004, most economists and social science scholars using PPPs have little idea how these data were collected or why they are inappropriate for longitudinal studies). Most importantly, however, the choice of appropriate variables and indicators should be guided by the theoretical and empirical questions at hand, rather than by academic fashion or convenience. For the purpose of this article – to assess the extent and evolution of nominal wage inequality across the world – we are confident that we are focusing the appropriate indicator. We certainly expect that the price data provided by UBS will allow productive future work exploring the impact of changing patterns in nominal and real wages and the relative command over consumption across the world, but this constitutes an intellectual exercise that is distinct from the one undertaken in this article.
Estimates from the UBS data are consistent with those reported in traditional studies of global inequality. For example, find that wage dispersions within countries, as identified through the UBS data, are highly correlated with broader measures of national inequality. The correlation between city Ginis calculated from the UBS data and national measures is .609, and that value increases to .836 when we use only Ginis from countries in which more than 75 percent of the population is urban (a reasonable adjustment considering it is urban-only data; World Bank, 2011). The correlation between the city average wage from the UBS data and the national Gross National Income per capita is .935 for 2009.
We should note that the UBS sample misses important sources of wage variance. First, COC wages are averages, and therefore miss the variance within these groups, including exceptionally high wages for the richest workers, and changes in the relative size of occupational groups. But in those analyses where this limitation might be significant, our results are consistent with those reported in similar studies. When we are estimating the impact of skill and other individual characteristics on wages, a lack of within-COC variance does not bias our results. Second, the UBS data cover only urban workers in large cities. We explicitly recognize that our results reflect only wage dynamics for urban workers. We, too, wish the coverage was broader in some areas, but, alas, that all inclusive, globally and longitudinally consistent dataset does not exist.
In short, the UBS data are unlike any other. The quality that most sets these data apart from other sources is that they are truly longitudinal as well as global. Newer efforts to collect global data on income or wages, or standardize existing national data for international comparison, cannot replicate the 40 years of remarkably consistent coverage available here. This is key for our effort to assess not only the relative standing of workers around the world but to trace how discrete populations fared relative to one another over time and how these trends might reflect national and international patterns in the distributions of technology and trade. Such an exercise allows us to begin mapping how the relative returns to particular occupations changed over time; the occupations that are characterized by lesser or greater global convergence in their returns; and whether changes in relative wages might be traced to specific processes of upward and/or downward mobility (e.g. returns to education, and/or national economic growth).
Analytic approach
In the next section, we use this unique data set to analyze trends and patterns in global wages. First, we construct a global hierarchy of COCs. We locate COCs by occupation and city and track their mobility over time, exploring the relationship between these characteristics of the global distribution of wages. Then, we attempt to more directly explain the global distribution of wages and wage trends, focusing on worker location, skill level, and whether they are involved in the production of tradable goods. Finally, we explore the intersections of models of national and global wage inequality, and discuss these findings as they relate to contending wage models.
Findings
Throughout the period considered in this article, wages vary widely between cities located in high- and low-income nations. We can highlight this point with a few striking examples. Figure 2 charts average wages (in current and constant dollars, with upper and lower bounds) from our data in New York City and Mumbai from 1982 to 2009. Throughout the period, the average hourly wage in New York based on the 12 surveyed occupations is more than 10 times higher than in Mumbai. The gap between cities is so large that it dwarfs opportunities for mobility within India. For example, department heads received the highest wages in Mumbai throughout the period, but their average wage in 2009 was less than half the lowest average wage (earned by car mechanics) in New York. While the wage of Mumbai department heads between 1982 and 2009 increased at a respectable 3.66 percent per year in 2009 dollars, they would need another 107 years at that rate of growth to catch up to the nominal wage level of New York department heads in 1982.
Mean wage with upper and lower bounds*, New York, Mumbai, 1982 to 2009.
Mumbai is not the only city to have its wages dwarfed by those in New York. The lowest surveyed wage in New York City in 1982 was higher than the highest wage in seven cities (of 47). New York City’s lowest wage in 2009 was higher than the highest wage in 18 (of 73) cities, and in cities in every continent but Australia. Figure 3 illustrates in greater detail the interaction between location and occupation. This figure combines the average hourly wage (shown in both log and linear scales) for 12 occupations in four cities (Mumbai, Buenos Aires, Madrid and New York City). As illustrated by the figure, in 1982 all 12 occupations in Mumbai had average hourly wages below the lowest paid occupations in Madrid and New York City. The average wage of a building laborer in Madrid ($2.56 in 1982 US dollars) was 89.6 percent higher than that of a department head in Mumbai ($1.35), while the wage of a New York female factory worker ($5.34) were 295.6 percent higher. Engineers in Mumbai ($1.18), Buenos Aires ($7.64), and Madrid ($9.07) had lower average wages in 1982 than building laborers in the New York ($10.00). On the other hand, in the city located in a middle-income country, Buenos Aires, the lowest wages in 1982 overlapped with the top of the distribution in Mumbai, while the highest wages were comparable with their counterparts in Madrid, and overlapped with the lowest wages in New York City.
Wages of city/occupation combinations (COCs) for selected cities, 1982.
Such differences in relative wage distributions should not be surprising. In 1982, the GNI per capita in the United States was 2.5 times greater than in Spain, 5.4 times greater than in Argentina, and more than 47 times greater than in India (World Bank, 2011). However, the extent of these wage disparities challenges common sense expectations that those employed in the better-paid occupations of poor countries should rank higher in the global distribution of wages (and, conversely, that those employed in the lower-paid occupations of wealthy countries should rank lower in that distribution). In other words, focusing on global patterns of wage inequality provides a different and alternative standpoint from which to think about social stratification.
Mapping the global distribution of wages
To further illustrate what mapping global social stratification ideally would entail, Figure 4 charts a kernel density plot of global wages (thick line) on top of a (simplified) box plot for 1982 and 2009. For the sake of comparison, wages have been indexed as a proportion of the highest measured wage in each time period. For reference, we divide the observed range of hourly wages into 10 equal intervals, defined by the percentage of each wage to the global maximum. As illustrated, in 1982 all 12 occupations in Mumbai, located in a low-income country, fell below 10 percent of the global maximum or in the first interval. On the other hand, in a wealthy nation such as the United States, New York City wages range from the second (10–20% of the global maximum) to the ninth interval (80–90%). The relative wage differences in India are roughly as large as in the United States, but from a global perspective, the absolute wage differences within India are insignificant.
Kernel density plot of global wages, 1982 and 2009.
We use kernel density estimation to approximate the global distribution of wages represented by our sample. Consistent with national wage and income distributions, most workers are clustered at the lower end. In both periods, most workers wages fall in the first interval (<10% of the global maximum). In 1982, there were 166 COCs in the lowest wage interval and 149 COCs in the top seven intervals combined, with the top two intervals having only two representatives each. The distribution in 2009 is generally flatter, with a significant clustering in the third interval (20–30% of the global maximum). Like national wages, most COCs fall in the lower intervals with fewer scattered in the higher-wage intervals; in this case, however, the long upper tail of the distribution is populated by both high and low-wage occupations from high-income cities while workers from low-income cities cluster at the bottom.
Tracking global mobility patterns
Contending paradigms in the study of world inequality would offer differing interpretations of these patterns. In the modernizing paradigm, disparities between countries in wages and/or income are not surprising, as they reflect the uneven achievement of nations in attaining or promoting certain attributes (ranging from human capital attainment, to productivity, to appropriate state policies for the promotion of modernization). In the critical paradigm, the persistence of inequalities is attributed to the nature of relations between wealthy and poor countries that promote growth in the former and constrain it in the latter (such as the exploitation of human and natural resources, the uneven exchange of manufactures for raw materials, or the support of by supranational organizations for the implementation of austerity policies and debt repayment in peripheral countries). In the 1980s, each of these approaches also would have differed on whether to expect wage and income inequalities to converge (modernization paradigm) or continue diverging (critical paradigm) in the future.
So how much change has there been in the ranking of COCs over the last 30 years? Figure 5 uses a scatter plot to compare the relative standing of each city/occupation combination in 1982 (x-axis) and 2009 (y-axis) using the same reference intervals as in Figure 4. COCs that were upwardly mobile fall to the right of the diagonal (solid line), and the change in standing relative to the global maximum is equal to the vertical distance from the diagonal. The dashed line is a LOWESS (locally weighted scatterplot smoothing) curve. Functionally, it is a locally weighted regression estimate of the relationship between wages in 1982 and 2009. We use it as an estimate of typical wage growth (again the vertical distance between the dashed LOWESS curve and the solid diagonal) for segments of the wage distribution. LOWESS estimation requires densely sampled data so estimates in the lightly populated upper tail of the distribution – notably the large downturn – are not robust.
COC wages in 1982 and 2009.
Thus, in the figure, the top wage earners in New York City (department heads, identified with a large ‘O’) were among those with the highest hourly wages in 1982 and 2009: their relative standing did not change over this period as a whole, as they remained in the second highest income interval. At the other end of the spectrum, all 12 COCs from Mumbai (represented with an ‘x’) remained in the bottom interval between 1982 and 2009. While the average wage in current dollars increased in Mumbai over the period, the absolute wage growth was insignificant on a global scale.
Overall, there was considerable stability in the relative standing of the 498 COCs included in Figure 5. The rank order correlation for COCs from 1982 and 2009 is .848; the typical COC moved 52 (of 498) spots. (We would expect the typical COC to move 166 spots if rank order in 2009 were independent of 1982.) COCs in the second through fifth intervals were the most mobile over the period, changing an average of 70.7, 73.5, 71.6 and 61.5 spots, respectively. The bottom end was the most static – the typical COC from the bottom decile moved up or down only 23.6 spots. Consistent with the stability in rank order, COC wages in 1982 are a very good predictor of wages 27 years later. Wages in 1982 explain 76.3 percent of the variance in wages in 2009, leaving only 27.7 percent of the variation to be explained by measurement error and economic developments within and between nations over the last 30 years.
The data indicate that while there were some significant changes in the shape of the global distribution of COCs, the relative standing of workers showed considerable stability. Looking at the LOWESS curve, typical wage growth was fastest in the second, third, and fourth intervals. Relative upward mobility of those in the second quartile reduced the skew of the distribution of wages from 1.28 in 1982 to .84 in 2009; the typical worker is now closer to the sample median, as is illustrated by the kernel density estimates in Figure 4. COCs in the bottom three intervals experienced significantly more wage growth in relative terms compared to the other seven intervals (36.1% versus 3.7%), but the absolute impact on wages was relatively small for the poorest COCs. Despite experiencing average wage growth of 36.7 percent in current dollars over the 27-year period, 72 percent of those COCs with wages less than 10 percent of the global maximum in 1982 were still below that mark in 2009.
Some COCs experienced more significant mobility, but they were a minority of our sample: only 49 (just under 10%) of COCs moved up more than 20 percentage points relative to the global maximum over the period. And these big movers were not randomly distribution. All but five of the 98 came from Europe. Fourteen (or one in three) department heads moved at least 20 percentage points, but only one each from bus drivers, female sales assistants and female factory workers was as successful. In short, the data provide support to the notion that the 1982–2009 period as a whole was characterized by stability in the relative position of workers, despite what might be interpreted as a strengthening of the global middle class, largely because the gaps between rich and poor are so great in absolute terms.
Regional inequalities
These patterns of mobility were unevenly distributed geographically. Western Europe enjoyed the most absolute wage growth, rank mobility, and interval mobility. Eastern European cities came in a close second, moving up 32 spots on average. The trajectory of the three cities from the United States was mixed: New York and Los Angeles averaged about half an interval of upward mobility and Chicago COCs moved down .7 intervals on average. On the other hand, African COCs saw their wages fall by a full dollar on average and 37.5 percent of African COCs fell to a lower interval while only 8.3 percent would find themselves upwardly mobile by that measure.
Table 1 shows the relative wage change (expressed in constant 2009 US dollars) for regions and occupations. Eastern Europe, Western Europe and Asia (1.20 without Tokyo) were the only regions that on the average experienced wage growth. Wages in 1982 and 2009 were down about 25 percent in both Africa and the Middle East, while Latin America and North America/Oceania experienced little change. Even in the United States, average constant wages slightly declined for the period as a whole.
Constant wages by region and occupation, 1982 and 2009
Table 1 also suggests that the structure of occupational stratification showed considerable persistence for the period as a whole – the three lowest wage occupations in 1982 still received the lowest wages in 2009, and the three highest wage occupations in 1982 were still on top in 2009. But 11 of 12 occupations experienced wage increases, and generally, lower wage occupations experienced the most growth – the correlation between wages in 1982 and wage growth between 1982 and 2009 is -.594. Wages increased 51 percent and 36 percent, respectively, for female factory workers and female sales assistants, the lowest paid occupations in 1982 and 2009; wages increased by 5 percent and 15 percent for engineers and department heads.
Analysis of variance
The implication that wages are driven primarily by geographical location is corroborated further by the data in Figure 6. Applying ANOVA, we estimate COC wages using city of residence and occupation for each of the 10 most recent sample years. The data indicate that occupation explains only about 20 percent of wage variance, while city of residence explains three or four times more, and that these patterns were fairly stable over the past three decades. 3
(In)wage variance explained (ANOVA) by city and occupation, 1982 to 2009.
Roughly speaking, this suggests that about 20 percent of the average wage of a group of similarly employed individuals is a product of the occupation they can access through their skills, education, experience, marital status, and (in some cases) sex, and 60 percent to 70 percent is a consequence of where they live (and the technology and capital they can access in that city). This is consistent with other decompositions of world inequality that suggest that between-nation disparities account for two-thirds to three-quarters of global inequality (see Korzeniewicz and Moran, 1997; Milanovic, 2005).
Measuring the China effect
But this raises an important question. How would our results change if wages from mainland China were included in our analysis? Our sample has limited data for China, which both has a very large population and experienced high rates of national economic growth. Yang et al. (2010) estimate that average real wages grew sevenfold between the early 1980s and 2007 (and tripled just between 1997 and 2007), but with an uneven distribution of this growth.
Figure 7 illustrates the magnitude of wage growth in China. The left side of the figure shows the distribution of wages in Beijing in 2006 with wages in Mumbai, Buenos Aires, Madrid and New York. The average across the 14 COCs in Beijing in 2006 is $3.18 in current US dollars, compared to $2.53 in Mumbai and $4.51 in Buenos Aires. The gap between low and high wages in Beijing is greater than in Buenos Aires and similar to Mumbai (signaling differences in the extent of inequality within each of these city distributions).
COC average wage (In scale) for select cities, 2006 and 2009.
In 2009, as indicated by the right side of Figure 7, the higher Beijing wages in our sample surpassed those in Buenos Aires, but the range is even wider than in 2006, so that the lower Beijing wages are lower than those of their counterparts in Buenos Aires. The highest paid occupations in Beijing in 2009 caught up with the lowest paid occupations in New York (as opposed to 2006, when there was no overlap). Growing inequality in Beijing and convergence between Beijing and historically richer cities illustrates how growth in China might be reducing the importance of location over time.
To estimate the impact of China on our estimates in Figure 6, we modify our basic sample in two ways. Matching COCs to industry wage growth estimates from Yang, Chen and Monarch (2010), we estimate wages in mainland China back to 1982. Second, we weight sampled COCs by the number of large cities (i.e. more than 1,000,000 inhabitants) in each country (as identified in United Nations, 2010), to assess whether results might be driven by the fact that UBS data represent a select sampling of cities (regardless of population numbers). 4
Under these new conditions, excluding China, city of residence explains 75.6 percent of the wage variance in 1982 and 80.5% of the variance in 2009, suggesting a mild increase in the importance of location over that period. The effect of occupation on wages declines from 18.0 percent to 14.8 percent. With China added to the sample, projecting backwards wages in Shanghai and Beijing, the percent of wage variance explained by city of residence drops from 80.5 percent in 1982 to 71.8 percent in 2009, and, inversely, the percent of variance explained by occupation increases from 14.2 percent to 21.8 percent. In short, including China in the sample increases the wage variance between rich and poor cities in 1982 (Beijing and Shanghai would have been the poorest cities sampled in 1982), and wage growth in China more than compensates for the tendency towards polarization elsewhere by 2009. But the general pattern remains, that the city in which one lives is several times more important in determining wages than the type of labor that person performs.
China has not (yet) had a larger effect on the global distribution of wages because, until recently, wages in China were so low that their high rates of growth barely registered on a global scale, but high rates of economic growth in China (and India) have the potential for changing the face of global inequality and stratification in the not-to-distant future. But the relative convergence of income and wages for other geographical locations in the world is much less pronounced, and the relative importance of location in shaping income and wage differentials (even when including China in the analysis) remains very high to this day.
Modeling global wages
Table 2 offers insights into the covariates of wage growth between 1982 and 2009. But instead of focusing exclusively on national characteristics to estimate national wage trends or individuals to estimate individual wage trends, models in Table 2 combine individual-level COC characteristics with a measure of national development. Each model is analyzed with China (B) and without China (A) (we should note that because wages in 1982 for Beijing and Shanghai are estimated using more aggregate data than the UBS data, these models tend to underestimate the wage impact of disaggregated characteristics such as sex or experience).
Estimating (ln)wage change, 1982 to 2009
p <.1; ** p < .05.
A: no China, B: with China.
Observations are weighted by the number of large cities in the relevant country.
Model 1 offers a concise estimate of broad wage trends. The dependent variable is the change in the log wage from 1982 to 2009. The first independent variable, the average wage for each city in 1982 (‘urban average’), tests for (beta) convergence between cities. A negative coefficient, for example, suggests that wage growth is greater in lower wage cities. Second, the urban gap, the wage gap between each COC and their urban average, tests for convergence within cities. Like the urban average, a negative coefficient signifies greater wage growth for lower wage earners in each city (or falling wage inequality within that city). The final term tests for an interaction of the first two. Convergence between cities cannot vary by a COC’s position within a city (the city average moves the same for all city residents), so the interaction term captures variations in within-city convergence by the average wage level of cities. Because the effect of the term increases with the average wage, a negative coefficient would signal that within-city convergence is stronger where the average wage was higher.
Without China, the only significant effect is the interaction term: there was significantly more convergence (falling inequality) within high-wage cities than low-wage cities. With Chinese workers added to the sample, we find significant convergence between cities (a product of rapid wage growth in China). Combining the urban gap and interaction terms, we find that low-wage cities experienced rising inequality, again driven by trends in Chinese cities, while wage distributions in high-wage cities moved in the opposite direction.
Models 2 and 3 estimate the impact of various individual-level COC characteristics on wage growth between 1982 and 2009. Skill is an index ranging from 0 to 1 primarily based on educational requirements, but also reflecting other ‘skilled’ demands (e.g. knowledge of multiple languages). Experience is measured in years. Sex (female = 1) and trade (involved in the production of tradable goods = 1) are dichotomous variables. Focusing on Model 2, we find that workers in tradable goods, more experienced workers and female workers experienced significantly faster wage growth. The results are fairly consistent with and without China.
Moving to Model 3, we add a measure of national development (national wage = GDP per capita/2000) in 1982 to allow us to test for any relationship between the effects tested in Model 2 and national productivity levels in 1982. Again, the impact of skill on wage growth was not significant between 1982 and 2009. But wages for workers in tradable goods grew faster than for other workers, particularly for workers in low-income countries. This final caveat vanishes when we add China to the sample – wage growth for tradable goods workers is consistently higher across high- and low-income cities. Other results are generally consistent with Model 2. As in Model 1, cities with lower productivity in 1982 experience faster overall wage growth to 2009 as represented by the negative coefficient for national wage.
Visualizing global wage trends and the H-O model
We now return to one of the key questions posed in the introductory section. For the modernizing paradigm, upward mobility and the transformations observed in world inequality and global stratification as a result of the growth of China (and more recently India) are most likely to be the outcome of changing patterns of trade and shifting wage differentials between skilled and unskilled workers. As outlined by the H-O trade model, within cities or regions, occupations that require greater skills receive higher wages, and the extent of these differentials should be highest in lower-income nations, and lowest in relatively higher-income countries. Our results for recent decades are compatible with these expectations.
Average wage differentials between the skilled and unskilled indeed were lowest in Western Europe and North America/Oceania, with the highest values coming from Latin America, Africa and Asia. For example, in 2009, more skilled workers in Mumbai earned 5.17 times more in wages than less skilled workers, while in the United States the average ratio was 1.99.
From this perspective, we would predict that the relative gap in wages between wealthy and poor countries would decline with globalization. Regional wage gaps should be smaller (as trade increases) for workers that produce tradable goods than non-tradable goods because employers can more easily arbitrage wage differences: bus driving cannot be outsourced but most manufacturing can.
To better assess these propositions, Figure 8 charts wage trends for eight hypothetical worker groups representing COCs from the sample based on regression coefficients estimating the contribution of various characteristics to wages in 1982, 1991, 2000 and 2009. 5 In addition to weighting the contribution of an estimated ‘national wage’ (to assess whether specific COC wages simply reflect employment opportunities within a given nation) and skill, the model assesses the extent to which wages were shaped by the tradability of the productive activity. 6
Estimated (In)wage in constant dollars in high and low income cities, by skill and trade (1982 to 2009).
We also included all possible interaction permutations: for example, do returns to skill vary across high and low wage cities? Hence, Figure 8 allows us to integrate traditionally national wage models (that emphasize on skill and other individual characteristics) and international wage models (with an emphasis on national characteristics and international relations, e.g. trade), by testing not only for independent effects (e.g. the impact of skill on wages), but also interactive effects over time (e.g. trends in wages for low-skill workers in low income countries producing tradable goods versus low-skill workers in low income countries producing non-tradable goods).
We regressed the log wage of 444 COCs on these seven variables – national wage, skill, trade and four interaction variables – in 1982, 1991, 2000, 2009. We then imputed values representing eight hypothetical worker groups using values of 0 and 1 for skill and trade and using the average national wage of the five least and most productive countries represented in the sample.
In Figure 8, our estimated national wage is the single most important variable shaping wage levels (we estimate the variance explained by regressing location [log of national GDPPC] on the residual after controlling for skill and employment in the production of tradable goods). Alone, national location explains between 67.9 percent of the cross-sectional variation in 1982 and 78.7 percent in 2009. These results are generally consistent with the ANOVA results in Figure 6, and there is no evidence in our sample that national location is becoming less important.
Skill also has a positive and significant association with wage. When we interacted skill with national wage the coefficient was negative in 1982 and 1991, meaning that returns to skill were not as great in high-income regions (the effect of location is larger in Model 2, because the larger wage ratios in low-income cities are masking the wage differentials between cities). That this skill*national wage interaction variable is no longer significant in 2000 and 2009 indicates that the skill gaps in low- and high-income cities have converged, and our findings suggest that this is being driven primarily by rising wages among less skilled workers in poor cities.
There is a negative coefficient associated with trade in each of the eight single-year models. But the single year models also show some convergence between COCs in tradables and non-tradables, as manifested by a smaller negative coefficient in 2009 than in 1982 for trade. This result is consistent with Table 2.
Overall, most segments experienced at least minor wage growth, but finished the period in 2009 with wages (adjusted for inflation) that were similar to when they started in 1982. The one big exception involves the wages for unskilled workers engaged in the production of tradables in low-income cities: wages for these workers converged rapidly with those prevailing in non-tradable activities (and these results did not change when China was added to the sample and observations were weighted by city population). While this result is consistent with the H-O model, the lack of divergence between unskilled-trade good producers and unskilled-nontrade good producers in high income cities is not. Since 1991, wages for skilled-trade good producers in high income cities have fallen relative to those of their nontrade good producing compatriots and skilled-trade good producers in low income cities. ‘Skill’ is being redefined in high-income cities as some formerly more privileged workers are forced to compete more intensively with now similarly ‘skilled’ workers in low income cities.
Summary of findings
The data exercise we conduct in this article provides compelling evidence that an ascribed characteristic, the nation in which workers are located, continues to be the major force shaping the distribution of wages worldwide. For our sample as a whole, the past 30 years show considerable stability in the global distribution of wages, and transcendent mobility was rare across the COCs in our data. But the data also indicate that more recently in China, as well as in some wealthier countries, the COCs included in our data did experience some significant relative movement within the global distribution of wages. Producers of tradable goods in poorer cities gained ground relative to both workers in non-tradable activities within those same cities, and tradable good producers in high-wage cities.
These conclusions illustrate the ways in which more disaggregated wage data can help us better specify patterns and trends in world inequality. The specific findings discussed in this section also help us understand why the interpretation of patterns and trends in global inequality remains a field riddled by contention. Critical perspectives, interpret the data to support the notion that polarization and inequality persist in the distribution of wealth and income among nations. As we discussed earlier in this article, even today, after years of rapid growth in China and more recently India, many advancing a critical paradigm would continue to argue that the extent of convergence is limited (either because many poor nations continue experiencing low rates of growth, or because growth in China and India is itself forecast to be short-lived and limited).
But looking at the same trends, the modernizing paradigm sees a different pattern. While recognizing the existence of a persistent gap between poor and wealthy nations through most of the 20th century, this gap is attributed primarily to internal characteristics of these nations, and the recent convergence experienced in world inequality is argued to demonstrate that modernization and greater integration into world markets provide opportunities for poor nations to catch up with wealthy ones.
We would contend that the findings reported in this article require that we critically reevaluate each of these paradigms, as they are both hindered by blind spots that limit their ability to productively understand the character of current transformations in world inequality and global stratification. This is the topic to which we turn below.
Discussion
Often, the critical paradigm on world inequality and stratification underestimates the extent to which poor or ‘peripheral’ areas, once having such a status, might undergo transformations that generate upward mobility. Because many studies using such a perspective understand polarization and inequality to be at the very core of the operation of the markets and the world-economy, it is hard from such a standpoint to even consider how greater participation in the latter can at any time benefit underprivileged areas. But the findings reported in this article suggest that the 1990s and 2000s indeed enhanced opportunities for lower skilled workers involved in tradables in poorer countries to increase their wages relative to all other COC clusters. Critical perspectives should make a greater effort to theorize the world-historical changes that allowed underprivileged populations to gain significant ground relative to more privileged populations around the world in recent decades.
Of course, some authors working within the critical paradigm always have been open to exploring ways in which current transformations in world inequality might be challenging the equilibrium of high inequality and polarization that characterized previous centuries. Wallerstein’s work (e.g. 1996) consistently insists that the capitalist world-economy has a life cycle, and that its inherent polarization might give way to very different distributional outcomes in the future. On a somewhat different track, the important work of Giovanni Arrighi (e.g. Arrighi, 1994; Arrighi and Drangel, 1986) challenged the notion that world-economic processes altogether preclude the upward mobility of poor nations: drawing on Schumpeter (1942), Arrighi emphasized instead the multiple and varied ways in which enterprises and rulers seek to enhance their status, in ways that lead to constant innovation and what Joseph Schumpeter labeled ‘creative destruction’. Such approaches certainly avoid the blind spots that often prevail within the critical paradigm, and are conducive to a more productive understanding of current transformations in world inequality and global stratification.
Moreover, understanding processes of polarization and inequality to have a strong spatial dimension that can persist over time, but eventually be subject to change, is not a new insight in the social sciences. Adam Smith (1976 [1776]), for example, noted that the relative privilege of towns vis-à-vis the countryside in his time (e.g. the prevalence of higher wages and prices in urban areas) was more a product of capacity of town residents to restrict access to their markets and urban citizenship than of the relative capacities (e.g. skills) of the populations of the countryside. For Smith, the capacity of urban populations to hoard opportunity and to subject rural populations to institutional exclusion generated long-lasting spatial inequalities between town and country. But eventually, he argued, these inequalities became so pronounced, that they would come to generate strong incentives for all to transgress trade restrictions and shift a growing number of production processes to the countryside. While Adam Smith had in mind the changing gap between urban and rural areas, the parallels with changing patterns of inequality between poor and wealthy countries are obvious – and some of these parallels are explored in greater detail in Korzeniewicz and Moran (2009).
The modernizing paradigm has its own blind spot in understanding recent changes in world income distribution as the simple outcome of greater efforts by rulers and people in poor countries to take advantage of market opportunities for their particular skill endowments. Such a paradigm does not engage enough in the critical thinking required to explore the relationship between skill, wages, and world inequality. Instead, they follow the wake left by modern economists in approaching the distinction between the skilled and the unskilled as a categorical difference denoting specific technical capacities. As critical social scientists exploring world inequality and global stratification, we should not be restricted by such an assumption, and in fact have the responsibility to go further and specify the conditions through which ‘skill’ is socially and spatially constructed.
Here again we can draw on Adam Smith, for whom the craft system was primarily a means for regulating the intensity of competition among ‘skilled’ workers, while leaving competition open among ‘unskilled’ workers: ‘The intention [of such] regulations is to restrain the competition to a much smaller number that might otherwise be disposed to enter into the trade. The limitation of the number of apprentices restrains it directly. A long term of apprenticeship restrains it more indirectly, but as effectually, by increasing the expence of education’ (1976 [1776]: Book I, p. 133). Under his analysis, skill emerges not as the specific technical capacities associated with performing certain tasks, but as a socially constructed category (often, around ascribed criteria) that entails a distinction between workers facing lesser or greater competition among their ranks (this is essentially similar to arguments advanced more recently by authors such as Tilly, 1999, and Wright, 2009, who approach skill as socially constructed through processes that include opportunity hoarding and the institutional development of competitive advantages).
In this regard, the results discussed in this article provide evidence of what we might label global opportunity hoarding. The crucial importance of location in accounting for wage inequality on a global scale highlights the persistent role of ascriptive criteria in shaping worldwide stratification: as noted in Korzeniewicz and Moran (2009), this persistence is easily missed in most prevailing studies of wage inequality, as these studies tend to focus almost exclusively on high-income nations. But for most of the 20th century, focusing on high-income nations also made it seem almost natural to emphasize purportedly ‘internal’ characteristics of those nations (e.g. the strength of organized labor, the race between technology and education, sex differences in rates of labor force participation), while ignoring a larger framework of opportunity hoarding and exclusion that had a significant impact in shaping labor markets worldwide (most centrally, restrictive immigration and trade policies that protected some workers from competition but intensified it among others; but also state policies in middle- and low-income countries that for decades sought to highly regulate participation in the world market).
Focusing exclusively on nations as units of analysis limits our ability to see the intersection of national stratification and global inequality in processes of opportunity hoarding and exclusion and the manners in which these processes might be subject to substantive change. Further advancing such a perspective requires a paradigmatic shift not only in theoretical frameworks, but also in the units of analysis chosen to analyze social data (and to conduct their actual collection and aggregation). The project reported in this article is intended as but a small and initial step in that direction.
Footnotes
Acknowledgements
We would like to thank David Smith and the anonymous referees of IJCS for their useful comments on an earlier version of this article.
Funding
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.