Globalization and Income Polarization in the Developed Countries

Transcription

1 Globalization and Income Polarization in the Developed Countries by Gary Burtless* THE BROOKINGS INSTITUTION March 12, 2007 * John C. and Nancy D. Whitehead Chair in Economic Studies, The Brookings Institution, Washington, D.C. The research described in this paper was performed under a grant from the Japan s Economic and Social Research Institute (ESRI). I am very grateful for this research support. I am also indebted to Sarah Anders and Charles Baschnagel of Brookings for excellent research assistance. The findings and conclusions reported in the paper are solely those of the author and do not represent the views of ESRI or the Brookings Institution.

2 Globalization and Income Polarization in the Developed Countries 1. Introduction In the decade and a half after 1980 the United States experienced a startling increase in inequality. The after-tax incomes of poor Americans shrank, the incomes of middle-class families stagnated, and the incomes of the richest households grew strongly. The rise in U.S. inequality was driven by a sustained growth in the inequality of market incomes, especially of pre-tax wage income (Burtless 1999; Burtless and Jencks 2003). Few other industrialized countries experienced such a dramatic increase in after-tax income inequality, but almost all saw a rise in disparities of household market income after 1980 (Förster and Pearson 2002). In many European countries, growing unemployment, especially among the young and less skilled, contributed to the trend toward greater pre-tax income inequality. One explanation for the rise in inequality is globalization. The economies of industrial countries have become more closely linked with one another and with economies in the developing world. This is the result of deliberate policy and remarkable technical progress. After 1960 the governments of many rich and poor countries agreed to reduce the barriers that impede cross-border movements of goods, services, and capital flows. Faster and bigger airplanes move people and goods more quickly and cheaply than was possible in the past. Lower communication costs, driven by the revolution in information and materials technologies, make it cheaper and easier to integrate production of goods and services across international boundaries. Although most national savings is still invested in home country assets, ever larger pools of capital flow freely across national borders in search of the highest returns. According to a popular view, the declining barriers to cross-border trade and capital flows have hurt low-skill and middle-class workers in the industrial countries. This view has been forcefully advanced by opponents of free trade, who warn that liberalized trade with poor countries in Asia, Africa, and Latin America will undermine rich countries labor protection laws, threaten their social safety nets, eliminate good jobs, and reduce the wages of unskilled and semi-skilled workers. The majority of economists who have studied the direct effects of international trade remain skeptical of the most extravagant of these claims (Burtless 1995; Feenstra 2000). With few exceptions economists find little evidence that rising trade is the main - 1 -

3 source of growing wage or income disparities in the rich countries. Nonetheless, economists who support free trade must face an uncomfortable fact. A classic doctrine of modern trade theory, the factor price equalization theorem, predicts that under plausible assumptions the elimination of trade barriers between a high-wage and a low-wage country can lead to reductions in some real wages in the country with higher wages. Consistent with the theory, the increase in manufactured exports from the developing world to industrialized countries has been accompanied by a loss of manufacturing employment in the industrialized countries and erosion in the relative wages and job prospects of unskilled and medium-skill workers. Even though they seem plausible, of course, the assumptions behind the factor price equalization theorem are not necessarily true. The trends in manufacturing employment and wage disparities could be due to factors that are completely unrelated to trade. Even if the assumptions of the factor price equalization theorem are approximately true and the trends in manufacturing employment and relative wages are partly due to liberalized trade, it is important to establish whether trade has been a major contributor to the trend in inequality or played only a minor role. Many economists think the most persuasive explanation for widening wage disparities is a shift in employers demand for labor linked to the introduction of new kinds of production techniques and management methods. These kinds of change have occurred in industries where trade is not very important as well as those where international trade plays an important role. Technological innovation and the entry of new market participants has put pressure on employers to change their production methods in ways that require a more able and skilled work force. Employers have persisted in this strategy in spite of rising wage premiums to the highly skilled, premiums that now make it more expensive to hire a highly skilled work force than was the case in the 1960s or 1970s. The remainder of the paper is organized as follows. The next section summarizes statistics that show the magnitude of wage inequality changes experienced by OECD countries in recent decades. Section 3 describes the importance of wage changes in the growth in overall inequality. Labor earnings represents only one source of income received by households. Changes in the distribution of other sources of income and in the household structure of families have also played important parts in determining the distribution of personal income. While the paper considers evidence from a number of OECD countries, much of the detailed analysis is - 2 -

4 based on wage and income developments in the United States. The following section examines trends in inequality, relative wages, and employment across industries and assesses whether these trends are plausibly explained by increased world trade and the reduction in cross-border barriers to overseas goods and capital. The last section describes and evaluates several remedies for the adverse impact of greater inequality. These policy responses are worth considering and adopting, regardless of whether income polarization has been caused by globalization or some other factor. 2. Trends in inequality Between 1980 and 2000 disparities in both wages and disposable incomes rose in the majority of rich countries for which comparable income distribution statistics are available. Figure 1 shows estimated Gini coefficients for 16 countries in the Organization for Economic Cooperation and Development (OECD). The Gini coefficient is a standard statistic for measuring economic inequality. It ranges from 0 (when all families or people have identical incomes) to 1 (when all income is received by a single family or individual). The data come from the Luxembourg Income Study (LIS), which is a cross-national project that assembles and tabulates income distribution statistics using consistent methods for all member countries. The estimates for each country are based on an identical measure of after-tax, after-transfer equivalent household income. Household income includes wages, net self-employment earnings, occupational pensions, interest, dividends, and other capital income, as well as cash and nearcash government transfer payments. Estimated payroll taxes and personal income taxes are subtracted from gross income to form an estimate of a household s disposable cash and near-cash income. Households differ in size, so it is necessary to adjust each household s income to reflect this fact. One popular adjustment, used in figure 1, assumes that a household s spending requirements increase in proportion to the square root of the number of household members. Under this assumption, a family of four needs twice as much income as a single individual living alone to achieve the same standard of living. After calculating the equivalent income per person in every household, analysts then rank every person in the national population from lowest to highest in terms of equivalent income and calculate the nation s Gini coefficient. Changes in the income distribution. The estimates displayed in Figure 1 show the Gini coefficient of after-tax, after-transfer income. The bars in the chart show estimated Gini coefficients for a year between 1979 and (The year differs depending on when countries - 3 -

5 conducted their household surveys.) The bold triangles indicate Gini coefficients for a later year, between 1999 and Countries are ranked from left to right in the order of their Gini coefficients in the earlier period. In twelve of the sixteen countries, disposable income inequality rose between the earlier and later periods. Inequality fell in the Netherlands, France, Switzerland, and Ireland. It increased 18% or more in Sweden, Finland, Belgium, the United Kingdom, and the United States. The exact reasons for the increase differ across countries. Research by OECD economists suggests, however, that much of the rise in inequality was driven by increases in market income inequality (Förster and Pearson 2002). Even countries that did not experience any increase in disposable income inequality often saw an increase in market income inequality. Market income consists of gross labor earnings as well as income from a household s property and capital holdings. Increasing inequality in labor earnings is an important part of the explanation for why disposable incomes have become less equal in many OECD countries. At least in the United States, almost all of the increase in income disparities between poor and middle class families was concentrated in the period between 1979 and This is true in the case of after-tax income as well as market income. Figure 2 shows trends in real equivalent household income at selected points in the U.S. income distribution between 1979 and The top panel shows trends in after-tax, after-transfer income, the same concept of income tabulated for other OECD countries in figure 1. The bottom panel shows trends in market incomes, that is, pre-tax incomes exclusive of government transfer payments. For each income concept the chart shows income trends at three positions in the income distribution at the 20 th, 50 th, and 95 th percentiles. These three points correspond to the relative positions of a low-income, middle-income, and high-income family, respectively. Income growth after 1979 was clearly much faster at the top of the U.S. income distribution than it was either in the middle or at the bottom. Americans at the 95 th percentile saw their market incomes climb 51% between 1979 and 2004, and their disposable incomes rose 54% in the same period. In the middle of the U.S. distribution, market incomes rose 15% while disposable income climbed 23%. At the 20 th percentile, market income rose just 4% while disposable income increased 13%. Changes in government tax and transfer policy had a moderate impact on the final income distribution, but plainly the main factor producing more inequality was the increase in market income disparities

6 The United States experienced three recessions between 1979 and 2004, in , , and The effect of each recession is clearly visible in year-to-year income movements, especially of market incomes. Year-to-year movements in market income are much bigger than movements in disposable income, and the difference is particularly noticeable for Americans near the bottom of the distribution. Between 1979 and 1983, when U.S. unemployment reached its highest rate since the Great Depression, market income at the 20 th percentile fell 18 percent, while income after taxes and transfers fell only 11 percent. 1 Increases in government transfers as well as reduced tax payments partly offset the drop in market incomes caused by the recession. Between 1979 and 1993 the relative income gap between the middle class and the poor widened, but there was very little further increase in this gap after In fact, the gap between the market incomes of the middle class and poor narrowed markedly during the lengthy economic expansion of the 1990s. Although the relative income difference between the middle class and poor stabilized after 1993, the gaps between top income recipients and both the middle class and the poor continued to grow. A notable feature of American distributional trends after 1993 is the remarkable increase of the relative incomes of top income recipients, with faster gains at successively higher points in the distribution. Figure 3 illustrates this development using tabulations based on income estimates of the U.S. Congressional Budget Office (CBO 2006). The CBO estimates are similar to those shown in figure 2, but they are partly based on income reported to the government on income tax forms. (The tabulations reported in figure 3 are based solely on data collected in Census Bureau household surveys.) The CBO estimates reflect a more comprehensive and accurate assessment of the incomes of top income recipients. Figure 3 shows income trends for four groups of households those in the bottom one-fifth of the disposable income distribution, those in the middle one-fifth of the distribution, those with incomes between the 95 th and 99 th income percentiles, and those in the top 1% of the income distribution. The income trends in the lowest three income categories are broadly similar to the trends shown in the top panel of figure 2. The data for the highest income category, however, shows dramatically faster income gains than the gains experienced by any of the lower income 1 The bottom panel ranks individuals according to their market income, so the household at the tenth percentile of the market distribution is not the same as the household at the tenth percentile of the distribution after taxes and transfers

7 groups. Surprisingly, much of the gain at the very top is attributable to increases in labor compensation rather than to extraordinary windfalls from capital gains (see also Piketty and Saez 2006). Wage developments. Trends in U.S. wage inequality have been extensively documented in recent papers (see Autor and Katz 1999; Card and DiNardo 2002; Autor, Katz, and Kearney 2006; and Lemieux 2006). It should be stressed, however, that more fragmentary time trend data from other rich countries also show increased wage inequality in most of them. Summarizing this evidence, the OECD reports that the proportional gap between the earnings of full-time working men at the 10 th and 90 th percentiles of the wage distribution increased an average of 14.5% between the 1970s and the 1990s. The relative wages of younger workers fell 9.4% over the same period, while minimum wage workers saw their relative earnings decline 13.6% (OECD 2004a, Chart 3.3). Figure 4 shows OECD estimates of the change in gross earnings inequality among full-time, year-round workers in 17 rich countries between 1994 and In 14 of the 17 countries, earnings disparities among full-time workers increased. In 5 of the countries, the relative disparity in pay increased 10% or more. Figure 5 shows relative changes in the pure price of U.S. labor, which is most plausibly measured using the hourly gross wage received by wage and salary employees. The chart displays relative pay trends separately for men, in the top panel, and women, in the bottom panel. It covers the period from 1973, when this kind of wage data was first collected, up through Both panels show three indicators of relative pay. The top line in each panel shows the ratio of hourly pay earned by a worker at the 95 th wage percentile and a worker earning the median wage. This ratio is usually referred to as the wage ratio. The lower two lines show the and the wage ratios. The figure provides clear evidence of widening U.S. pay disparities among both men and women over almost the entire period since There was a distinctive break in one of the trends, however. Whereas the relative earnings differences between middlerank and top U.S. earners has continued to grow over the entire period, the gap between low- and middle-rank earners widened until about but then shrank or stabilized after the late 1980s. Evidently, the factors that pushed down the relative wages of poorly paid workers in the 1980s disappeared or became less important after By 2005 the relative wage difference between poorly paid and average-wage U.S. workers was approximately the same as it had been in the early 1970s

8 It is not easy to reconcile this long-term pattern with the hypothesis that intense competition with unskilled workers in newly industrializing countries is the main factor driving wage disparities in the United States. If workers in China and southeast and south Asia were the major new competitors influencing wage developments, classic trade theory would predict that the most important adverse effect would be on U.S. workers with the lowest skills and wages. The standard theory explaining the link between trade and wages is the Heckscher-Ohlin model of international trade. This model explains the pattern of international trade by reference to the relative abundance of factors of production among trading partners. The model predicts that between two countries, say, North and South, which share the same technology, North will export commodities that are produced with relatively more of the factor of production that is relatively abundant in North and will import commodities produced with relatively more of the factor of production that is relatively abundant in South. Two theorems derived from the Heckscher-Ohlin model, the factor price equalization theorem and the Stolper-Samuelson theorem, deal explicitly with the effect of trade on wages, land rents, and other factor prices. The factor price equalization theorem, proved by Paul Samuelson in a pair of well-known articles, shows that under the assumptions of the Heckscher-Ohlin model and a regime of unrestricted free trade, the prices of the factors of production will be equalized among trading partners (Samuelson 1948 and 1949). That is, if the assumptions of the model are true, free trade between the United States and China will equalize U.S. and Chinese wages for equivalent labor and will equalize rents for a standard unit of land, even though the factors of production cannot move freely across the Pacific Ocean. If one accepts the assumptions of the theorem, liberalized trade between the United States and China represents good news for most Chinese workers and bad news for U.S. workers whose skills are equivalent to those of a typical worker in China. Because most Chinese workers skills are probably equivalent to those of a less-skilled American, factor price equalization will imply that less-skilled U.S. workers must accept a reduction in their real wage when U.S.-China trade is fully liberalized. The Stolper-Samuelson theorem implies that an increase in the domestic price of a commodity, brought about by a higher tariff or additional protection, will raise the real price of the factor of production that is used relatively intensively in producing that commodity (Stolper and Samuelson 1941). If shoes are produced using labor intensively and land sparingly, whereas wheat is produced using land intensively and labor sparingly, then an increase in tariff protection - 7 -

9 for shoes will boost the real wage received by laborers. By implication, a reduction in tariff protection for shoes will reduce the real wage. Accepting for a moment the assumptions of the theorem, a reduction in protection for apparel and footwear, which use unskilled labor relatively intensively, will tend to reduce the real wage received by unskilled American workers. The earnings trends displayed in figure 5 show this pattern of relative wage change during the period from the mid- or late 1970s up to the late 1980s, but after 1988 the relative wages at the bottom of the distribution stopped falling and started to rise. The surge in imports from the developing world into the United States did not slow in In fact, its momentum increased. The one trend that continued after 1988 was the growth in the wage differential between workers at the top of the distribution and workers in the middle and at the bottom. In 1988 wage earners at the 90 th percentile earned exactly 2.0 times the wage of workers earning the median wage. By 2005 this ratio increased to 2.25 for women and to 2.31 for men. The earnings gap between median wage earners and earners at the 95 th percentile increased even faster, driven mainly by a much faster rate of increase in the real wages of earners at the top. This pattern is also reflected in annual wage earnings reported to the U.S. Social Security Administration. According to income reports in household Census surveys, the annual earnings of the median U.S. full-time, year-round earner increased a little less than 5% between 1990 and The Social Security Administration s annual wage data show that real annual earnings at the 95 th percentile increased 26% between 1990 and 2005, while at the 99 th percentile earnings increased 37%, and at the th percentile they rose 82%. 2 Since the late 1980s the main factor driving up U.S. wage inequality has been the exceptional growth of wages among workers in the top one-tenth of the distribution. For workers further down the wage ladder, earnings inequality stopped increasing and in fact probably declined. Any explanation of recent inequality trends must account for this distinctive pattern of relative wage movement. 2 The Social Security Administration publishes annual tabulations of wage earnings reported on W-2 tax forms. For example, its tabulations of 2005 earnings can be found here: The tabulations can be used to estimate the exact distribution of total wage and salary earnings for virtually all wage and salary workers in the United States, including earners with annual wages over $10 million. Using these published tabulations, I estimated the real earnings gains reported in the text. Note that my estimate of the median change in real earnings is based on Census survey data on wage earners with full-time, year-round jobs. In contrast, the Social Security Administration tabulations cover all workers who had taxable wage and salary earnings in a year

10 One trend that pushed up U.S. earnings inequality during the 1980s was the growth in the earnings premium paid to workers with advanced skills. A simple indicator of general skill is a worker s school attainment. Since the 1960s the most common educational qualification in the United States is attainment of a secondary school diploma. Consequently, high school graduates earnings are often used as a benchmark for assessing the payoff to higher levels of education. Figure 6 shows estimates of the log earnings difference between high school graduates and two groups of workers with higher levels of school attainment, college graduates and workers with at least one post-college degree. To measure the premium for college and post-college degrees, I regressed the logarithm of workers annual labor earnings on age and educational attainment for years between 1968 and 2005 using the Census Bureau s Current Population Survey files. In order to reduce the sampling variability of the displayed results, estimates shown in figure 6 reflect the centered average of regression coefficients for five successive calendar years. The estimation sample includes full-time, year-round workers between 25 and 64 years old who have a valid report of their annual labor income, including both wages and net self-employment earnings. 3 The top and bottom panels of figure 6 show sizeable increases in the earnings premium enjoyed by degree holders during much of the period after Female degree holders saw their educational pay premium rise substantially between 1980 and 1992, but their gains since 1992 have been very small (see lower panel of figure 6). The increase in the educational pay premium persisted over more years for male degree holders, but their gains appear to have slowed or stopped in the late 1990s (top panel). Since college completion has become more common among working-age Americans, the rise in the payoff to advanced schooling has occurred against a backdrop of an increasing relative supply of welleducated workers. This leads many labor economists to infer that the rising earnings premium for higher education must have signaled a rise in the relative demand for highly educated workers. The fact that the earnings premium for college and post-college degrees stopped 3 In order to offset the effects of year-to-year changes in the Census Bureau s top-coding procedures, I top-coded earnings in every year using a simple and uniform procedure. Reported earnings that exceeded the 97 th percentile of male earnings in a given year were recoded to the 97 th percentile value for male earners, and earnings reports that exceeded the 99 th percentile of female earnings in a given year were recoded to the 99 th percentile value for female earners. This procedure means that the estimated education premiums do not capture the full earnings advantage enjoyed by well educated earners in the top 3% of the male earnings distribution and in the top 1% of the female distribution. Thus, the estimates almost certainly understate the increase in the education premium, especially for men

11 increasing in the middle or late 1990s suggests that some of the factors pushing up relative demand for highly educated workers slowed or the availability of college-educated workers increased. Globalization might offer a partial explanation for this development, at least up to the time the college pay premium stopped growing. The integration of large, poor countries into a world economy that also includes rich industrialized nations made highly educated and skilled workers in the rich countries relatively more scarce. If China, India, and southeast Asia become fully integrated with OECD economies, highly skilled workers in the bigger world economy will be relatively much less abundant than they were in an integrated global economy that only included rich OECD countries. It is a little hard to believe, however, that this kind of integration by itself could erode the relative position of middle-skill workers in rich countries about as fast as it reduced the relative position of rich countries unskilled workers. Figure 7 shows estimates of the earnings penalties suffered by U.S. workers if they have failed to complete secondary school. The estimates were obtained using the same method and with the same sample described in the previous paragraph. They show that the pay differential between high school dropouts and graduates widened between 1980 and the late 1990s, but the differential has not widened much in recent years. 3. Polarization of household incomes Whatever the ultimate cause of the increase in U.S. pay disparities, its impact on overall income inequality has been noticeable and direct. The great majority of American families with a working-age adult rely mainly on labor earnings to support themselves. When hourly and annual earnings grow more unequal, household incomes will grow more unequal as well. Figure 8 shows the trend in annual earnings disparities among year-old Americans who are heads of household or who are married to the head of a household. Each earner included in the sample is a full-time, year-round worker. In order to make comparisons easier, I have tabulated the same earnings ratios displayed in figure 5 above, which shows long-term trends in hourly pay disparities among all U.S. adults who hold a wage and salary job. The three indicators shown in the figures are the 95-50, 90-50, and earnings ratios. Both figure 5 and figure 8 show broadly similar trends. The disparity between earners at the 50 th and 90 th earnings percentiles increased almost continuously between 1979 and 2000, and the proportional difference between earners at the 50 th and 95 th percentiles increased even faster. In the lower ranks of the earnings

12 distribution, however, the gap between low earners and median earners increased through the middle or late 1980s, but then stopped growing. Impact of rising wage inequality. One way to assess the contribution of wider pay disparities to the overall growth in U.S. income inequality is to ask how much income inequality would have changed if wage disparities had not increased. Using U.S. Census Bureau survey data, I have performed this calculation for years between 1979 and My analysis procedure is simple to explain in the case of earnings inequality among workers of the same sex, say, men. Annual earnings inequality among men rose between 1979 and 2004, as we have seen (top panel of figure 8). To preserve the same amount of earnings inequality in the two years, it is necessary to assign males who had a low rank in the 2004 wage distribution more earnings than was reflected in their responses to the 2004 income survey. It is also necessary to assign earners who had a high rank in the 2004 wage distribution less earnings than was reflected in their responses to the 2004 income survey. The simplest way to accomplish this is to assign to 2004 workers the earnings level to which their rank in the male earnings distribution would have entitled them if they had been workers in This preserves the exact 1979 male earnings distribution, but it ignores the change in average male earnings between the two years. My procedure is to assign to each 2004 male worker the income to which his rank in the 2004 distribution would have entitled him in 1979, multiplied by the ratio of total male earnings in 2004 to total male earnings in This method ensures that the exact shape of the 1979 distribution of relative male earnings is preserved in the simulated 2004 male earnings distribution, but the sum of 2004 male earnings is the same in both the original and the simulated 2004 distributions. An identical procedure can also be used to assign the earnings distribution observed among men in 2004 to men in This alternative procedure provides estimates of how the income distribution would have changed between 1979 and 2004 if the 2004 distribution of relative male earnings had been observed in both years. These same methods can be used to assess how much the overall income distribution was affected by changes in earnings inequality among women. In the calculations described below, I analyze the effects of changes in earnings inequality among men and women who are heads of household or who are married to the head of a household. (Both kinds of earners will be referred to as heads of household in the remainder 4 The analytical method for estimating the impact of pay disparities on income inequality follows the method developed in Burtless (1999)

13 of the paper.) These male and female workers earned 87% of all labor income reported in the CPS file in The other 13% of wage and self-employment income was earned by secondary earners in households, that is, household members who were neither the head of household nor married to the head of a household. Table 1 shows the results of the calculations just described. It displays trends in the inequality of U.S. household income between 1979 and 2004 under various scenarios. The income definition used is equivalent (or household-size-adjusted ) gross personal income in other words, equivalent income measured on a pre-tax basis. 5 Nearly all estimates of inequality are affected by top-coding, a procedure that restricts the range of income reports contained in the microdata file. The Census Bureau changed its top-coding procedures from time to time between 1979 and To eliminate the effects of these changes, I applied a uniform procedure to top-code all labor earnings in the estimation file at the 98.5 th percentile for each year included in the analysis. As a result, the estimates of inequality understate true income disparities and the estimated inequality trends understate the true rise in inequality between 1979 and later years. The top line in Table 1 shows the trend in pre-tax income inequality in four years, 1979, 1989, 2000, and The first three years are the final years of lengthy economic expansions; 2004 is the last year of Census data available when the analysis was performed. These estimates show that overall inequality increased 10% between 1979 and 1989 and an additional 5% between 1989 and The total increase in pre-tax inequality between 1979 and 2004 was thus 15.7%. The second line shows what the trend in U.S. pre-tax income inequality would have been if earnings inequality among male household heads had remained unchanged after Instead of rising 15.7% between 1979 and 2004, overall inequality would only have increased 9.2%, that is, a little less than two-thirds of the increase actually observed. This estimate implies that about 38% of the rise in overall inequality can be explained by increased earnings disparities among male heads of household. Another way of calculating the effect of rising male earnings 5 It is more complicated to perform the simulations using income measured on an after-tax basis. If the simulation requires that a household s pre-tax income be raised or lowered compared with the amount reported on the survey, a post-tax analysis requires us to adjust the income and payroll taxes the household pays. The U.S. tax system is mildly progressive and nonlinear, so the required tax adjustment is not straightforward. It turns out, however, that the percentage change in the Gini coefficient of income inequality was approximately the same between 1979 and 2004 using either a pre-tax or after-tax income measure. On a pre-tax basis, the Gini coefficient increased 15.7% between 1979 and On an aftertax basis, it increased 16.2%. It therefore seems unlikely that an after-tax analysis would produce results that differ substantially from those shown in Table

14 inequality is to consider how much overall inequality would have changed if the shape of the male earnings distribution had remained fixed at its shape in The third row shows the result of this calculation. The Gini coefficient of overall inequality would have increased from to 0.389, and this change is about 48% of the change in overall inequality between 1979 and Using this estimation procedure, it appears a little over half of the growth in overall inequality was due to the increase in male earnings inequality. Assuming the two procedures give us a lower and upper bound on the direct effect of male earnings inequality, they suggest that between 38% and 52% of the growth in pretax income inequality is explained by increased labor income inequality among men who are heads of households. The next two lines show simulation estimates of the effect of holding constant the shape of the female earnings distribution, either using the 1979 shape as a baseline or using the 2004 distribution as a baseline. Using either estimate, it appears that increased female earnings inequality had little effect on the overall income distribution. As we have seen, this is not because female earnings inequality remained constant between 1979 and On the contrary, it increased almost as much as earnings inequality among men (see figures 5 and 8). There are two main reasons that increased inequality in women s earnings had a much smaller net impact than the growth in male earnings inequality. One reason is that men s earnings represent a bigger percentage of a typical household s income, so the level of a male head s earnings will usually be more important in determining the household s rank in the income distribution. In 1979, for example, three-quarters of all labor income earned by heads of households and their spouses was earned by men. Only about a quarter was earned by women. In 2004 a little less than two-thirds of household heads income was earned by men, with the remainder earned by women. A second reason that higher female earnings inequality did not contribute more to higher overall inequality is that increasing pay disparities among women often lifted the incomes of households in the bottom half of the U.S. income distribution. Some women with aboveaverage earnings are the only breadwinners in a below-average-income family. Increased inequality in female earnings will bring these households incomes closer to the middle, reducing overall inequality. The combined effects of higher male and female earnings inequality are displayed in the sixth and seventh rows of Table 1. Not surprisingly, these estimates are very similar to those showing the impact of increased male inequality by itself. Between 39% and

15 54% of the rise in overall income inequality is traceable to the rise in earned income inequality among men and women who are heads of American families. Impact of other trends. Results in the bottom four lines of Table 1 show the effects of two other changes in the U.S. economy and demographic structure. One notable change has been the shift in the correlation structure of husband and wife earnings. Before the early 1960s, men with high earnings were likely to be married to a spouse who earned below-average wages. This was because the husband s high earnings allowed his spouse to devote her time and energy mainly to work in the home. A remarkable transformation in women s position in the U.S. labor market has contributed to a change in the correlation of husband and wife earnings. The great majority of working-age married women are now employed, and the husband-wife earnings correlation is now positive. Furthermore, the correlation has grown more positive over time, contributing to the growth in overall income inequality (Karoly and Burtless 1995). The Spearman rank correlation of earnings of working husbands who head families and their working wives increased from to between 1979 and Using the same methodology developed in Burtless (1999), I have examined the role of the husband-wife earnings correlation on the overall income distribution by holding constant this correlation and then calculating the change in the Gini coefficient. To hold the husband-wife correlation constant at the 1979 level, for example, I adjusted earnings amounts in the 2004 CPS file to duplicate the correlation pattern observed in There are several ways to make this adjustment. My procedure was to calculate the earnings ranks of men and women separately for both 1979 and Suppose the man with rank r in 1979 was married to a woman with rank s in the 1979 female earnings distribution. To preserve the 1979 husband-wife correlation in 2004, I assigned the 2004 man with rank r (if he was married) to the woman with rank s in the 2004 distribution. To implement this procedure I assumed that the earnings of the woman with rank s was available to the family headed by the married male with rank r. After the correlation of husband and wife earnings is held constant in this way, the Gini coefficient is recalculated. As the estimates in rows 8 and 9 of Table 1 show, the rising correlation of husband-wife earnings had a noticeable impact on the trend in overall inequality between 1979 and Row 8 shows that 84% of the observed change in overall inequality would have occurred, even if the husband-wife earnings correlation remained unchanged. This implies that 16% of the rise in overall inequality occurred because of an increase in the correlation of husband and wife earned

16 incomes. If we instead hold constant the correlation structure of husband-wife earnings at the 2004 level, the Gini coefficient of overall inequality would have increased from to 0.389, an increase that is 90% of the change in inequality actually observed. This estimate implies that 10% of the rise in overall inequality was due to a higher correlation of husband and wife earnings. Marriage patterns and household living arrangements have also changed. More American families contain only a single adult member, and fewer contain a married couple. This represents an advantage for some single adults, because their incomes are spread among fewer family members. It is a disadvantage for others, because they are deprived of the potential earnings contributions of a spouse or adult partner. Equivalent income is more equally distributed among people who live in married-couple families than it is among people who live in single-adult-head households. The proportion of Americans who live in married-couple families is shrinking. In 1979, 74 percent of adults and children lived in married-couple households. By 2004, the fraction had shrunk to 64 percent. To obtain a simple estimate of the impact of this development on overall inequality, I adjusted the sampling weights on the 2004 CPS file to preserve the distribution of families at their observed proportions in I classified families in three categories: married-couple families, single-male-adult families, and single-female-adult families. The bottom two rows in Table 1 show how overall inequality would have changed assuming that the proportion of persons living in each family type had remained unchanged. These estimates suggest that between 14% and 21% of the jump in inequality can be traced to the changing pattern of marriage and household composition. In sum, these calculations suggest that between 39% and 54% of the change in overall U.S. income inequality was directly caused by the increase in earned income inequality. In addition, between 10% and 16% was caused by the growing correlation of earned incomes received by husbands and wives, and between 14% and 21% was caused by a shift in Americans household living arrangements. The remainder of the increase was caused by a variety of miscellaneous factors, including changes in the distribution of capital income and government transfer payments and changes in the relative importance of different income sources. 6 Note that 6 Capital income is more unequally distributed than labor income, so an increase in the relative importance of capital income will produce an increase in overall inequality. Karoly and Burtless (1995) perform a detailed and comprehensive decomposition of the sources of change in the U.S. income distribution between 1959 and 1989, but their analysis is restricted to families containing a working-age

17 some of the individual factors probably reduced inequality, though on balance these factors caused inequality to rise. For many of the factors just mentioned, it is hard to see any direct connection with globalization. Few experts would claim, for example, that the trends in family composition and toward higher correlation of husband and wife earnings are attributable to the influence of trade or global outsourcing. The trends in marriage and household composition were already well underway even before wages began to grow less equal after the mid-1970s. These trends are mainly due to changes in social attitudes toward the importance of marriage and the role of women in the family and in the labor market. Similarly, the trend toward a higher correlation of husbands and wives earnings began long before imports from developing countries (aside from natural resources) played an important role in U.S. markets. Contrary to a widespread impression about changing work patterns among women, a disproportionate share of women s gains in employment, annual hours of work, and earnings since the late 1970s have been concentrated among women in relatively affluent households. 7 This is hardly a pattern we would expect to see if rising female employment rates were mainly the result of growing inequality among men and the shrinking wages of trade-dislocated husbands. Some critics of globalization may claim that intense competition from overseas producers is the ultimate source of the family composition trends and earnings correlation trends described above. These claims are not logically well supported or believable, however. In sum, increased earnings inequality among working men and women who head households is an important reason for increasing income polarization in the United States, but it probably accounts for one-half or less of the growth in overall income inequality. Changes in family composition, a sharp rise in the correlation of married partners earnings, and other adult. The tabulations in Table 1 cover a more recent period and cover the entire noninstitutionalized U.S. population, including children, working-age adults, and the population older than For example, among year-old men who have wages in the middle one-fifth of the male earnings distribution, the real earnings contribution of a spouse increased by only about one-third (34%) between 1979 and Among men in the same age group whose earnings were in the top one-fifth of the male earnings distribution, the contribution of spouses earnings increased by 143% during the same period. The enormous difference is explained by three main factors: There was a smaller decline in the percentage of men in the top income group who were married; there was a bigger proportional increase in the fraction of wives in the top income group who entered employment; and there was a bigger percentage increase in the wages earned by the working wives of men in the top income group. This last development reflects the growing correlation between husband and wife earnings. See Burtless (1999)

18 factors unrelated to globalization account for half or more of the growth in inequality since Even if all of the increase in earnings inequality among working Americans were due to global trade, only about half of increased income inequality could be explained by this development. 4. Impact of globalization Evidence of deeper global integration is plentiful. Reliable evidence of its exact impact on inequality is more elusive (Burtless 1995; Feenstra 2000). Figure 9 shows a simple measure of the role of international trade in the largest industrialized countries, including Japan, the United States, and the four biggest EU economies. The chart shows the sum of exports and imports as a percentage of national GDP. For the EU countries, the chart shows the average of exports plus imports as a percentage of GDP in the four countries. The EU countries and the United States have seen a sizeable increase in the share of international trade in their economies. Between 1970 and 2005 the foreign trade share of GDP increased about 0.7 percentage points a year in the four EU countries and 0.4 percentage points a year in the United States. The growth of the trade share was considerably slower in Japan. Of course, much of the growth in EU trade was among EU member countries. Measured as a percentage of the EU s GDP, Eurostat estimates that the external trade of the EU increased 6.8 percentage points between 1995 and 2005, rising to 23.5% of GDP by This is only a little lower than the share of external trade in the U.S. economy (26.8% of GDP in 2005) and in Japan (27.2% of GDP). Figure 10 shows the shifting sources of U.S. merchandise imports over the past quarter century. The American government divides exporting countries into three broad groups. The group of industrial countries includes Canada, countries in western Europe, and rich countries in the Asia-Pacific region. All these nations have incomes similar to those in the United States. Member countries of the Organization of Petroleum Exporting Countries (OPEC) are principal suppliers of oil to the United States. The remainder of the world s exporters include developing or newly industrialized countries, such as Mexico, China, South Korea, and Malaysia. Measured as a share of U.S. national income, imports from the rich industrialized countries nearly doubled between 1972 and Imports from lower income non-opec countries increased even faster, however. Imports from these countries tripled as a percentage of U.S. GDP between 1972 and 1993, and they more than doubled between 1993 and Merchandise imports from non- OPEC developing countries now account for a larger share of U.S. imports than imports from the rich industrialized countries

19 In recent years Japan has seen a similar shift in the sources of its imports. Since 1993 imports originating from non-opec developing countries have increased from 2.1% to 5.4% of Japan s GDP. For both Japan and the United States, much of the increase is explained by a surge in imports from China. Chinese imports were equal to 0.5% of Japan s GDP in 1993 and to 2.4% of GDP in Imports from China represented 0.5% of U.S. GDP in 1993 and increased to 2.0% of GDP by The rise in imports from China is particularly important because prevailing wages in that country are so far below those in other newly industrializing countries that export heavily to the rich countries. The U.S. Bureau of Labor Statistics measures the hourly compensation costs of manufacturing production workers in the United States and 32 of its major trading partners. Using current exchange rates, the BLS compares compensation costs across countries to determine differences in the hourly price of employing production workers. In 2005, for example, average hourly compensation costs in EU-15 countries were 16% higher than those in the United States. In Japan, compensation was 8% lower than in the U.S. Among developing countries, Mexico and China are the two most important U.S. trading partners. Mexican hourly compensation costs are just 11 percent of those in the United States (U.S. BLS 2006a). Compensation in China, however, is lower still. While the estimates for China are not strictly comparable to those for other countries, a plausible estimate is that Chinese hourly compensation costs are only about 3% of those in the EU, Japan, or the United States. They are less than a third of compensation costs in Mexico (Lett and Bannister 2006; U.S. BLS 2006a). Many critics of globalization see evidence of the adverse effects of trade in employment statistics. Manufacturing, which once offered good job opportunities to unskilled and semiskilled workers in rich countries, accounts for a dwindling percentage of employment. In the G-7 countries, the unweighted average of manufacturing employment fell from 29.4% of total employment in 1970 to just 16.5% of employment in 2005 (U.S. BLS 2006b). In spite of the fact that international trade has become a bigger part of nearly all national economies, a declining percentage of the workforce is employed in industries where trade was traditionally important. If we define the tradable goods industries as manufacturing, mining, and agriculture, then the fraction of the U.S. workforce employed in tradable industries has been shrinking more or less steadily since the late 1960s (see figure 11). These industries employed 32% of working Americans in 1970, but the share of workers in traded goods industries fell to less than 14% in