In February 2000, when the current American economic expansion passed its 107th month, President Clinton proudly announced that this expansion was the most sustained in US history, surpassing even the “golden age” of the 1960s. Praise for American economic performance extends well beyond our borders. Throughout the developed world, the American economy now serves as a model of dynamic growth and job creation. European politicians envy the low rate of unemployment here–half the official levels in many European countries–and marvel at all the new jobs.

To be sure, American economic performance has its critics. Some analysts argue that American and European unemployment rates are not so different, once one takes into consideration the vastly higher rate of imprisonment here.1 Incarceration rates in the United States rose dramatically in the 1990s, and prisoners are drawn from a segment of the population with high rates of unemployment. Adjusting the unemployment rate in the United States for incarceration rates could add as much as two percentage points to the unemployment rate, which would push the US rate above that of a number of European countries.

A more common criticism points to the costs of the American strategy of employment growth. American success is founded on “flexible” labor markets, which allow employers–especially in the context of globalization, NAFTA, and the WTO–to hire and fire employees relatively easily, reorganize employment structures in response to market conditions, and adjust wages as needed, especially in a downward direction. As a result, job growth is accompanied by persistent poverty, continuing high levels of inequality, and the growth of poorly paid, dead-end service sector jobs. While these critics acknowledge the recent American success at creating jobs, they also argue that the jobs are “lousy”–with low pay and little chance for improvement.

Are the critics right? Is the American “jobs miracle” based on the expansion of lousy jobs? More precisely, what is the distribution of job quality–the balance of good and bad–in the current expansion, and how does this distribution compare to earlier job expansions? Before celebrating the American model, and urging its emulation elsewhere, we need answers to these questions.

An advertisement in the Economist by the Pfizer company in 1999 reports one line of response:

But what about the quality of the new jobs created? The figures about the American labour market tell us a quite different story from the “trash-job-and-working-poor” litany that we so often hear. Since 1983 about 50 percent of the new net jobs created in the US economy–about 15 million–were in the managerial and professional sector, and adding the medium-skilled occupations, the figure rises to over 80 percent. Furthermore, around 70 percent of the new net jobs were in occupations remunerated above the median income for all full-time employees.2

The assertions in Pfizer’s ad are based on a widely cited report on the topic prepared by Joseph Stiglitz when he was chairman of the President’s Council of Economic Advisors.3 The report is based on a study of the 1994-1996 expansion. Using Current Population Survey data provided by the US Census Bureau, Stiglitz constructed an “occupation-by-sector matrix”: in effect, a table listing 22 different sectors of the economy and 45 occupations in each sector. Typical examples of sectors are “durable goods manufacturing,” “educational services,” and “wholesale trade”; typical examples of occupations include “machine operators,” “teachers (except college and university),” “secretaries,” and “engineers.” This dissection of the labor market yielded a total of 990 potential kinds of “jobs” (cells in the matrix). After eliminating empty and small cells, some 250 or so jobs, accounting for roughly 95 percent of total employment, remained. Stiglitz and his colleagues then calculated the median weekly earnings of full-time employees in each of the 250 cells and defined “job quality” by the distribution of these cell medians. Intuitively, good jobs were the jobs in above-average categories and lousy jobs were in the below-average categories. More precisely, “good jobs” were defined as all cells in this job matrix with median earnings above that of the median cell for the whole matrix, and “bad jobs” were defined as cells with median earnings below the median cell.

The final step in the analysis was to calculate the change in the number of people in each cell between 1994 and 1996. The central finding was striking: roughly 68 percent of all net job growth was among the good jobs, and roughly 50 percent of all net job growth was in jobs in the top three deciles of the earnings distribution.4 The conclusion offered in the report, and summarized in the Pfizer ad, was that the job expansion was strongly weighted towards the creation of good jobs.

The Stiglitz report received a great deal of press at the time of its release. Unfortunately, its political influence is greater than its analytical insight, in two principal ways. First, while it is true that many of the newly created jobs are good, it is also true that lots are lousy and there has been little expansion in between. In short, the true story is one of job polarization, and not simply growth at the top. Second, while job growth for white Americans has been concentrated at the good end of the spectrum, job growth for blacks and Hispanics has been concentrated at the lousy end. The full story then is also one of racial division in the labor market.

Strategy of Analysis
To evaluate the quality of newly-created jobs in the current economic expansion, we examined the entire period of expansion in the 1990s (recall that Stiglitz only studied 1994-96), compared the recent period with the “Golden Age”of the 1960s, and studied the role of gender and race in each period.5 The two pivotal tasks for exploring the distribution of job quality during these expansions are to categorize jobs, and to figure out how to measure the “quality” of different categories of jobs.

Following Stiglitz, we classified jobs on the basis of occupation-by-sector matrices, using data from the Current Population Survey (CPS)–45 occupations by 22 sectors in the 1990s (the identical categories as those used by Stiglitz) and, because of limitations in the available data, 32 occupations by 21 sectors in the 1960s.6 In principle there would be 990 jobs (45 x 22) within this matrix in the 1990s data and 672 jobs (32 x 21) in the 1960s data, but not all of these cells had any CPS survey respondents in them at both the beginning and end of the expansion.7 Our analysis, therefore, is based on changes in the number of people in those cells in which there were sample cases in both the first year of the expansion and the last: 746 job categories for the 1990s, and 296 in the 1960s.

On the measurement of job quality: ideally we would like to rank job categories from best to worst on the basis of an index of job properties, including wages, job security, working conditions, fringe benefits, and career potentials. In practice, however, the only consistent indicator of job quality available is the earnings of people in these jobs. Moreover, while many other job attributes are obviously important to people, most desirable features of jobs are correlated with earnings. So, again following Stiglitz’s general method, we measured job quality by the median hourly earnings of full-time employees in the job categories derived from the occupation-by-sector matrix.8 In 1992, the best of the 746 jobs was lawyers in wholesale trade with median hourly earnings of $43.23, while the worst was the occupational category “fabricators, assemblers, inspectors, and samplers” in the “social services” sector with median earnings of $4.32. (For jobs with large numbers of people in them, the best job was “health diagnosing” in the “other medical services” sector, with median hourly earnings of $30.75; the worst was “private household service workers” in the “private household services” sector, with median hourly earnings of $5.32).

Our goal, then, is to measure the relative contributions to the job expansion of jobs of differential quality defined by the median hourly earnings of job categories. In doing this, we face a methodological problem: the cells in our job matrix are of vastly different sizes. For example, “engineers” (occupation) in “social services” (sector) is quite small, with roughly 750 people in that job across the United States in 1999, whereas “teachers except college and university” (occupation) in “educational services” (sector) includes 3.6 million people in that year. It makes little sense, therefore, to simply chart the contribution of each cell to the overall expansion of jobs, since the large categories will tend to contribute more to job growth simply because they are bigger. We could correct for these differences in the number of people in different jobs by shifting to an analysis of the rates of growth of different jobs–that is, the change in employment in a cell divided by the size at the beginning of a period. But the cells with the highest growth rates might be small cells that do not contribute much to the overall expansion of jobs. A massive expansion in a tiny, well-paid job cell–tripling the number of engineers in social services, for example–would not imply a large contribution to overall job growth.

Instead, we grouped jobs of similar earnings-defined quality into larger categories with more or less equal numbers of people at the beginning of a job expansion. To do this we first rank-ordered the cells in the matrix from best to worst: that is, from the highest median weekly earnings to the lowest. We then grouped this ranked-ordered set of cells into ten ordered categories, each containing close to 10 percent of the labor force at the beginning of a job expansion.9 We refer to these aggregated categories of jobs as “job-quality deciles.” The bottom decile contains the roughly 10 percent of the labor force in jobs with the lowest median earnings; the highest decile contains the roughly 10 percent of the labor force in jobs with the highest median weekly earnings; and so on. These job-quality deciles are the primary categories we will use in assessing what jobs have contributed the most to the expansion of jobs in the American economy. To convey a sense of what sorts of jobs fall within each of the deciles, the three largest job categories within each decile in the 1990s are given inTable 1.

Not everyone will find this way of classifying jobs and job quality satisfactory. Even though we have divided the job structure into hundreds of job categories, many of these categories remain quite heterogeneous. For example, the category “college and university teachers in the educational services sector” includes community college faculty earning $30,000 per year and professors in elite law schools earning over $250,000 per year. Similarly, “other executives, administrators, and managers” in “finance, insurance, and real estate” comprises CEOs of multinational insurance corporations and executives in local real estate companies. So, while 20 percent of the net job expansion in the 1990s was generated by the highest job-quality decile, this increase might not reflect an expansion of high-paying jobs: a big expansion in “college and university teachers in the educational services sector” might have occurred principally because community colleges were hiring new faculty.

An alternative strategy, therefore, would be to ignore occupation and sector entirely and simply treat jobs as earnings-generating employment contracts. We could then study how this job-earnings distribution changes during job expansions. Much recent research on growth of earnings inequality does precisely this. But while direct analyses of the earnings distribution are certainly important, it is also important to have a clear idea of how jobs are changing. Employers do not simply make employment offers at a specified earnings level; they make job offers to do particular kinds of things (occupations) within particular kinds of firms (sectors) at particular levels of earnings. Jobs within the occupation-by-sector matrix are a rough proxy for types of jobs created by firms. We want to know whether or not the occupations and sectors with the highest paying jobs–defined by median earnings–are the ones that are growing the most rapidly.

Patterns of Job Growth
Consider, then, the distribution of job quality in the net job expansion during the long, sustained employment boom of the 1990s. Figure 1[Acrobat Reader required for all figures] presents the contribution of each of the job-quality deciles (defined at the beginning of the employment expansion in 1992) to the growth of jobs between 1992 and 1999. The results are striking. The biggest contributor to the job expansion was the highest decile: over 20 percent of the net expansion of jobs during the 1990s came from these jobs. So far, the results seem consistent with the Stiglitz study.

The second biggest contributor, however, was the worst decile, which contributed about 17 percent of the net job expansion. Nearly 40 percent of the total net job expansion occurred among the very best and the very worst kinds of jobs in the American economy. The jobs that contributed least to the job expansion occur in the second through fifth deciles of job quality. In 1992, these accounted for almost 40 percent of the labor force, but only 14 percent of the net expansion of jobs came from these categories.

Overall, then, the 1990s job expansion was dominated by the net expansion of employment among relatively good jobs–the top three deciles accounted for almost 50 percent of job expansion. But it was also marked by strong polarization in the pattern of employment growth: strong expansion at the tails of the job-quality distribution, combined with weak growth among average to just-below-average jobs. So the skeptics who emphasize lousy jobs also have a point.

To understand the polarization of job growth in the 1990s, consider the contrast (see Figure 2) between the pattern for the 1990s and the pattern for the 1960s. In the 1960s, lowest decile contributed the least to job creation: less than 2 percent of the net job growth came from this category of jobs. The bottom four deciles, in fact, collectively generated less than 20 percent of job expansion. (In the 1990s, the bottom decile alone generated 17 percent of net job expansion.) In addition, while job expansion at the top of the job structure was strong in the 1960s, the top deciles did not generate quite as high a proportion of total job expansion as in the 1990s. In the earlier decade, just over 40 percent of the job expansion came from the top three job-quality deciles, whereas in the 1990s these categories generated nearly 50 percent of the job expansion. Finally, in the 1960s the two middle deciles of the job structure generated 30 percent of the job expansion, compared to only 12 percent in the 1990s. Overall, then, the 1960s were a period of strong employment growth in the middle and upper segments of the job distribution, whereas the 1990s were characterized by much more polarized job expansion with particularly robust expansion among high-end job categories.10

These claims about job polarization in the 1990s might be challenged in either of two ways: by reference to problems of youth employment or numbers of part-time workers.

As has frequently been noted, the past several decades have seen a fairly steady decline of employment in manufacturing and growth of employment in services, including substantial growth in low-end service jobs. One of the iconic images of deindustrialization is the replacement of well-paid, high-skill industrial jobs with hamburger-flipping jobs. Because these low-end service jobs, especially in food services, are disproportionately filled by teenagers and young adults, it is possible that the rapid expansion of the lowest-paid job categories in the 1990s was largely the result of more jobs for young workers. The polarization shown in Figure 1 might not matter much if the rapid growth of jobs in the bottom decile was mainly an expansion of opportunity for early employment experience for teenagers.

To assess this objection, consider what happens when we confine our attention to full-time employees between the ages of thirty and 55.111As Figure 3 shows, the pattern is virtually identical to the pattern for all full-time workers in both the 1990s and the 1960s. The polarization of job expansion in the 1990s cannot be attributed to the expansion of employment opportunities for young workers.

A second objection is that the restriction of our analysis to full-time workers could also affect the results. A central theme in contemporary discussions of changes in labor markets and employment relations is the growth of various forms of nonstandard and “flexible” work. While Figure 1 shows very strong expansion of employment in the top three deciles of the employment structure, these results could be misleading if there was a massive growth in part-time employment in jobs in the lowest deciles.

Figure 4 indicates that this, too, was not the case in the 1990s.12Broadly, we see the same kind of polarization in the net job expansion among part-time employees as among full-time employees. If anything, the contribution to net job expansion of the top deciles for part-time work is even greater than for full-time work: nearly two-thirds of the net job expansion in part-time work was generated by the jobs in the top three job-quality deciles.13 The pattern in Figure 1, therefore, cannot be attributed to the restriction to full-time jobs.

Race and Gender
So, in both the “Golden Age” of the 1960s and the “new economy” of the 1990s, the job categories at the high end of the job distribution contributed disproportionately to job expansion. In the 1990s, but not the 1960s, the very bottom of the job structure also contributed substantially to job expansion, and the middle of the distribution made only a marginal contribution. The net result is a polarized job expansion.

Taking race and gender into account sharpens the contrasts between the two expansions and clarifies the nature of the polarization. Consider the contributions of each job-quality decile to the net expansion of employment in the 1960s and the 1990s within each of the race-gender categories. Because of data limitations, in the 1960s we can only distinguish four categories: white males, white females, black males, and black females. In the 1990s we can divide the white racial group into Hispanic and non-Hispanic whites. The results, presented in Figure 5 and Figure 6, reveal quite dramatic variations, both over time and across various categories.14

In the 1960s, gender differences in the patterns of job expansion were much sharper than racial differences. Virtually none of the job expansion for white men or black men in the 1960s occurred in the bottom four job-quality deciles: these four deciles accounted for -2.6 percent of the net job expansion for white men and -23.1 percent of the net job expansion for black men. In contrast, for women of both races job expansion was heavily concentrated in the bottom deciles of the job-quality distribution: fully 48 percent of the net job expansion for white women and 52 percent for black women occurred in the bottom four deciles. At the other end of the job-quality distribution, 67 percent of the net job expansion for white men and 44 percent of the job expansion for black men occurred in the top three deciles, compared to 12 percent of the net expansion for black women and 13 percent for white women. Although racial differences in patterns of job expansion among men were not negligible in the 1960s, they were relatively muted compared to the dramatic difference between men and women.

In contrast, in the 1990s (see Figure 6), the racial differences in patterns of net job expansion are bigger than the gender differences. Among non-Hispanic whites–men and women–job expansion is heavily concentrated in the top deciles: in the 1990s, 62 percent of the net job expansion for white men and nearly 90 percent for white women occurred in the top three job-quality deciles. The polarization we observed for the labor force as a whole is completely absent among white men (only 6 percent of the net job expansion for white men occurred in the lowest decile of the job structure), and present only in a muted way among white women, for whom 13.4 percent of the net job expansion occurred in the bottom decile of jobs.

The patterns for non-white men and women in the 1990s differ sharply from the pattern for whites. For all four non-white race-gender categories, job expansion was concentrated in the lower deciles of the job-quality distribution. The bottom two deciles accounted for 50 percent of the net job expansion for Hispanic women, 35 percent for Hispanic men, 25 percent for black women and 28 percent for black men (compared to 15 percent of the net job expansion for white women and only 4 percent for white men). Polarization of job expansion is also present among blacks and Hispanic men, although not among Hispanic women.

The pattern of polarized job expansion in the 1990s we observed inFigure 2, therefore, has a very strong racial character. This is illustrated in Figure 7, where the contributions of each job-quality decile to net employment expansion in the 1990s is broken down into the racial categories.15 In 1992, at the outset of the 1990s job expansion, 30 percent of those in jobs below the median were black or Hispanic. If no process of racial polarization existed in the employment expansion then roughly 30 percent of the net expansion of jobs in these categories would also have been filled by blacks and Hispanics. The actual figure is over 95 percent. In 1992, 70 percent of employees in jobs below the median were white, yet less than 5 percent of the net expansion of these job categories in the 1990s was filled by whites. At the other end of the job-quality spectrum, 77.9 percent of the job expansion in the top two deciles was filled by whites.

Putting these various observations together, we can safely say that:

1. The pattern of job expansion among white men is virtually the same in both the 1960s and 1990s: it is heavily weighted at the high end of the job structure with little polarization of employment growth.

2. The pattern of job expansion for white women changes dramatically across the two periods. In the 1960s, this expansion was concentrated in the middle and bottom of the employment structure; in the 1990s, it has been heavily concentrated at the top. In the 1960s the profile of job expansion among white men differed sharply from that among white women; in the 1990s, the two patterns have substantially converged.

3. Racial differences in patterns of employment expansion have increased to the point that the 1990s can be characterized as a period of racially polarized job expansion. The net expansion of jobs at the bottom of the employment distribution is overwhelmingly dominated by minorities whereas the expansion of employment at the top is strongly dominated by whites.

Does it Matter?
Does the sustained expansion of employment in the United States in the 1990s conform, then, to the rosy picture of the “jobs miracle” touted at home and abroad? It is certainly true that many new jobs have been created in the United States during this period. And if one dichotomizes jobs into two simple categories–good jobs (above the median) and bad jobs (below it)–then most of the American jobs expansion in the 1990s occurred among “good jobs”: about two-thirds of the net expansion was among jobs in the sixth to tenth deciles of the job-quality distribution. So the evidence does not support a simple summary judgment that crummy jobs dominate the job expansion.

But it should now be clear that this summary judgment tells only part of the story. In the 1960s, the sustained job expansion was unequivocally a process of upgrading the employment structure. The 1990s expansion was characterized by a polarization of employment opportunities, albeit a polarization weighted toward the high end of the job structure. Moreover, this polarized pattern of job expansion was highly racialized. Employment for whites–both men and women–expanded sharply among the better jobs in the employment structure, whereas new jobs for blacks and Hispanics concentrated at the bottom of the employment structure.

This pattern of employment expansion has deep implications for the nature of social inequality in the United States. First, it suggests that the problem of poverty in the United States increasingly concerns the working poor rather than primarily people marginalized from the system of employment altogether. This is not to say that the link between poverty and unemployment and exclusion from the labor force has disappeared, but rather that an increasing proportion of poor people are working full-time in jobs that pay below poverty-level wages. To seriously tackle poverty in America today requires more than just getting poor people into jobs; it requires changing the quality of jobs available to them.

Second, the very slow rate of growth of jobs in the lower-middle range of job quality suggests that it will become increasingly difficult for people working in the very worst jobs to move up in the employment structure. Most upward job mobility is to jobs that are only modestly better than the job one holds. People in the bottom decile of employment are unlikely to make a jump directly to jobs in the sixth to tenth deciles. Since jobs in the second through fifth deciles of the employment structure have been growing at about a third of the rate of the labor force as a whole, people employed in the rapidly expanding bottom job decile face limited opportunities for improvements in employment.

Third, the pattern of job expansion in the 1990s suggests significant transformations in the structure of racial stratification. Since the 1960s, there has been a considerable expansion of employment of African Americans and other racial minorities in what are loosely described as middle-class jobs. The proportion of doctors, lawyers, professors, managers, and executives who are African American has increased significantly. Among these higher-level jobs, therefore, there has been a gradual deracialization. Among jobs at the bottom of the employment structure, on the other hand, the 1990s witnessed a process of deepening racialization. Only 2 percent of the expansion of jobs among non-Hispanic whites occurred in the jobs below the median job category, compared to nearly 60 percent of net job expansion among blacks and Hispanics combined.

So, what is to be done? What sorts of public policies are suggested in light of these trends in the American employment structure? Of course, any thorough analysis of policy alternatives would have to consider many more issues in current labor market trends than simply the macro-patterns of job expansion studied here:the patterns of inequalities within these jobs categories; the problem of contingent and part-time work; the patterns of mobility across these job categories for different demographic groups; the relationship between patterns of job creation and things like firm size, technical change, and linkages to the global economy; and so on. Still, we think our analysis has something to say about current debates about government policy. What we offer here, therefore, is a short discussion of a range of policy directions that bear on the specific problem of the polarized pattern of job expansion rather than a comprehensive discussion of government policies, labor markets, and employment structures.

If one agrees that the patterns we have documented of polarized job expansion in the 1990s–particularly the racialized form of that polarization–present a problem, then there are two broad categories of policy response. Option one: Don’t worry too much about job polarization per se. Let the market determine the character of the jobs that are created, but in various ways ameliorate the impact of such polarization on the standards of living underwritten by these jobs. If the expansion of jobs generated by the market is polarized, this is what the “New Economy” needs. The task of government is to insure that people in these jobs–the “working poor”–nevertheless live decent lives. Option two: Use public policy to directly affect the patterns of job growth, encouraging in various ways job growth in the middle of the employment structure and discouraging it at the bottom. The pattern of job expansion is not some “natural” result of the operation of efficient markets, but the inevitable result of all sorts of public policies: the nature of the tax code, the institutions of skill formation, the regulation of the employment contract and working conditions, the minimum wage, and laws regulating unions. The task of government is to design such policies in such a way as to rebuild social mobility and expand job opportunities in the middle of the employment structure.

Political responses to the economic polarization in the United States have tended to take the first of these policy directions, and then only in halting ways. The most notable example is the Earned Income Tax Credit (EITC), a provision in the tax code specifically designed to raise standards of living of the working poor above what they can get through the labor market.16 The EITC is the one redistributive program that was expanded in the 1990s. The failed attempt at creating universal health insurance in the early 1990s can also be interpreted as an attempt to partially decouple the standards of living of working people from the pattern of earnings and benefits generated by the labor market. If the EITC were further expanded and comprehensive universal health insurance enacted, it would matter much less whether jobs with low pay were rapidly expanding or if mobility bridges between those jobs and better paying jobs had collapsed.17

Policy reforms like the EITC leave the distribution of job quality generated by the labor market largely unaffected. Indeed, for many policy makers this is one if the virtues of the EITC: the tax code provides income subsidies to the working poor without mucking about with the internal operation of labor markets. This would be fine if there were no negative consequences to an economically polarized employment structure beyond the standard of living of those at the bottom. But employment polarization also undermines opportunities for individual mobility, reduces social solidarity, and reinforces (when polarization follows racial lines) stereotypes and other forms of racial division. Moreover, the amount of income-raising that is politically feasible through the EITC is likely to remain small compared to potential impact of other programs.

The question, then, is whether public policy can effectively shift the pattern of job creation and help generate more middle-range jobs. Here there are two broad kinds of strategies: first, direct intervention by the state to create certain types of jobs through publicly funded employment; second, the creation of incentives and institutional infrastructure to encourage private employers to create such jobs.

Since the triumph of neoliberalism and the demise of Keynesian views of state intervention, the public-sector employment option has been completely off the political agenda in the United States. In the immediate future, there is no real political prospect of launching a major expansion of public employment, except perhaps of public school teachers. Nevertheless, we may soon enter a period in which a serious expansion of public works is once again politically feasible. The oft-noted neglect over the past quarter century of bridges, public transportation, public school buildings, and other state-financed infrastructure creates a need for a considerable expansion of public works. And the large government budget surpluses generated by the economic expansion of the 1990s make such an expansion fiscally feasible as well. An expansion of public works could help shift employment expansion away from personal services, retail, and other low-end jobs, and toward the middle range of the job structure. Particularly if one wants to counteract racial polarization, a significant expansion of public works could be important, since it is easier to direct public works towards specific labor markets and populations than it is to direct private-sector employment.

In a variety of ways, public policy can also have a significant impact on the extent to which private employers create low-end jobs or relatively well-paid, skilled jobs. This is at the heart of the discussion of “high-road” versus “low-road” capitalism: high-road capitalism is characterized by the expansion of relatively well-paid, skilled jobs in the middle of the employment structure; low-road capitalism by the expansion of low-skill, low-wage jobs. So, the question is: what can be done, as Joel Rogers puts it, “to close off the low road, help to pave the high road, and enable workers and firms stuck on one to walk the other.”18

Two policies are especially relevant to closing off the low road: raising the minimum wage and strengthening the labor movement. While modest increases in the minimum wage probably have little effect on the pattern of job creation, a substantial rise–a rise sufficient to give people in low-end jobs a “living wage”–would almost certainly dampen the expansion of jobs at the bottom of the job-quality distribution. Changing labor laws in ways that would facilitate the growth of unions and strengthen their role in regulating labor markets and working conditions could also contribute to dampening low-road job creation. While much would depend upon the specific strategies and visions adopted by unions, a strong union movement has the potential to reduce wage differentials by raising wages at the bottom, making subcontracting to low-wage firms more difficult, and in other ways making “low-road” strategies more costly for employers.

Closing off the low road, obviously, is not enough. Unless this is also combined with policies that encourage the expansion of middle-range jobs and the acquisition of skills needed to fill those jobs, the result will simply be a decline in job opportunities for people currently at the bottom of the job-quality distribution. Improved education and expanded programs of vocational training, particularly directed towards black and Hispanic communities, would help, since an inadequate supply of skilled labor impedes the creation of skilled jobs. But simply expanding the supply of skilled labor does not automatically call forth the employer demand for such labor. An effective sustained strategy for expanding middle-range employment needs to link such training to an industrial policy that creates real incentives for employers to invest in the right kinds of employment expansion.19

None of these proposals can be implemented in a serious way if public policy is driven by the neoliberal belief that markets should be maximally unconstrained, the state should be minimally interventionist, and taxes should be as low as possible. Instead, these policies would require a dramatic increase in the resources and energy of an affirmative state committed to counteracting the inegalitarian dynamics of markets. This, in turn, would require significant increases in taxation on the beneficiaries of the long expansion of the 1990s. We have the resources to pursue such policies; the question is simply whether an effective political coalition can be forged to press for them. In the absence of political will, we are likely to see more of the same.


1 Bruce Western and Katherine Beckett, “How Unregulated Is the US Labor Market? The Penal System as a Labor Market Institution,”American Journal of Sociology,104 (January 1999): 1030-60.

2 Mauricio Rojas, “The End of Work Fallacy,” The Economist,20-26 February 1999, p. 28.

Annual Report of the Council of Economic Advisors(Washington: United States Government Printing Office, 1997).

4 It is important to note that these results refer to net job expansion rather than job creation per se. There is always a simultaneous process of the creation of new jobs and the destruction of already existing jobs. When we observe that a particular cell in the occupation-by-sector job matrix increased by 10,000 over a period of time, this could mean the creation of 25,000 new jobs and the destruction of 15,000 old jobs. We see only the net effect of these two processes.

5 The technical details of the data we use are presented in the appendix to this paper, available at: Appendix.htm.

6One other note on the 1960s data: because of problems in the CPS surveys in the early 1960s we could not use the data for the first two years—1961 and 1962—of the employment expansion. Our data analysis thus covers the period 1963-1970. For details of the CPS data problems in the 1960s, see the technical appendix to this paper.

7 Stiglitz dropped all cells from this matrix with fewer than ten cases; we have decided to include all cells in which there are cases in both of the years in any time period over which we are assessing job expansion. None of the substantive results we will report would be different if we excluded these small cells.

8 A couple of brief further technical notes: (1) We are using hourly earnings to index job quality rather than weekly earnings, as in the Stiglitz report. The results are not substantively affected by this shift, but we felt that hourly earnings was a better measure of job quality. (2) In order to rank-order the cells of the occupation-by-industry matrix by median earnings as accurately as possible, we combined the CPS samples for the entire period of a job expansion and calculated the median earnings (in constant dollars) of incumbents of these jobs for this expanded sample. This means, in effect, that the quality of jobs is being evaluated by the median earnings over the entire period of a job expansion rather than simply at the beginning. (3) In the 1960s, respondents were not directly asked about their hourly earnings for the current job. Instead they were asked about their earnings for the longest job held the previous year. We therefore had to use this retrospective data to calculate the median earnings of the occupation-by-sector matrix.

9 Since jobs come in lumpy units, it is not possible to aggregate the rank ordered jobs into groups each containing exactly 10 percent of the labor force. Thus, for example, of the ten deciles in 1992, eight contained between 9.6 percent and 10.5 percent of the labor force, one (the lowest decile) contained 11.4 percent of the labor force and one (the third decile) 8.2 percent. None of the patterns we will be examining are significantly affected by these deviations from equal decile categories.

10 We have conducted similar decompositions of net job expansion for the expansionary periods in the 1970s (1975-1980) and the 1980s (1983-1989). The results suggest that these two decades were transitional between the job expansion pattern of the 1960s and the 1990s. The 1975-1980 job expansion looks like a muted version of the 1960s pattern—the bottom deciles contribute somewhat more to the job expansion than in the 1960s, while the top deciles contribute a bit less than the 1960s. The polarization pattern in the 1990s begins to appear in the 1980s, although not as sharply.

11 A technical note on these results: the job quality deciles in Figure 3are the same as in Figures 1and 2—that is, these have been calculated on the basis of the entire full-time labor force sample, not simply the thirty- to 55-year-old adult sample. This means that there is no longer 10 percent of the older adult labor force within each of these job quality “deciles,” since younger employees are more concentrated in the lower deciles. In particular, only 8.8 percent of thirty- to 55-year-old employees are in the bottom decile, yet about 13.9 percent of net job expansion among mature adults comes from this category of jobs.

12 It was not possible to do a separate analysis for part-time employees using the 1960s CPS data.

13 In order to facilitate comparisons between Figure 4 for part-time jobs and Figure 1for full-time jobs, we have used the same job quality deciles in Figure 4as were used in Figure 1—that is, they have been calculated on the basis of median hourly earnings of jobs in the occupation-by-sector matrix for the full time sample (rather than recalibrating these deciles just for part-time workers). This makes the results of Figure 4 particularly striking, since part-time employment is more concentrated in the bottom deciles of the overall employment structure than is full-time employment. Thus, for example, only 3 percent of part-time employees were employed in the highest job-quality decile in 1992 (the beginning of the 1990s job expansion), yet over 20 percent of the net job expansion of part-time work occurred in this decile. If we recalibrate the job-quality deciles to reflect the distribution of job-quality strictly within the part-time sample, then this growth at the top appears even stronger: roughly 45 percent of the net expansion of part-time employment in the 1990s was generated within the top job-quality decile when these deciles are calculated among part-time jobs.

14 The job quality decile categories in Figures 5 and 6 are the same as those used in the analysis of the full time labor force as a whole. This means that once we break the analysis down into separate race-gender categories then there is no longer approximately 10 percent of the relevant category within each of these job quality deciles. These figures thus present, for each race-gender group, the distribution of net job expansion for that group across the job quality categories defined for the entire labor force.

15 The aggregate numbers in Figure 7are slightly different since we have not included “other races” (mainly Native Americans and Asians) in the calculations.

16 The EITC is a limited negative income tax for people in the paid labor force. It gives people an income supplement if their annual earnings fall below a certain level and various other criteria are satisfied.

17 A more radical policy reform to partially decouple standards of living from the labor market would be a universal basic income grant that guaranteed all citizens an above-poverty standard of living, regardless of their employment status. Whereas the EITC can be considered a subsidy for low-wage work and thus potentially increases the supply of workers willing to work for low wages, a basic income makes it possible for people to exit the labor market entirely, thus reducing the supply of workers for low-paid jobs. For an extended discussion of basic income, see Philippe Van Parijs, “A Basic Income for All,” Boston Review,October/November 2000. For a comparison of Basic Income with other related proposals, see Erik Olin Wright, “Reducing Income and Wealth Inequality: Real Utopian Proposals,”Contemporary Sociology (January 2000): 143-155.

18 Joel Rogers, “The Folks Who Brought You the Weekend: Labor and Independent Politics,” in Audacious Democracy,eds. Steve Fraser and Joshua Freeman (New York: Houghton Mifflin, 1997), pp. 247-261.

19 An example of such institutional innovation in skill formation and labor markets is the much heralded Wisconsin Regional Training Partnership (WRTP). The WRTP was formed in the mid-1990s in an effort to solve two problems in worker training in the metalworking industry: that publicly provided training often had little to do with the actual skills needed within production; and that employers faced a chronic collective action problem when providing high-level, flexible training to employees, since each employer is worried that if they provide high levels of training to employees, other employers will poach the trained workers. The WRTP solution to these problems involves firms, unions, the state vocational education program and a university-based labor market research institute working together in collective institutions for skill formation and job upgrading. For a description of the Wisconsin Regional Training Partnership, see