文章基本信息

标题：Is the market for college graduates headed for a bust? Demand and supply responses to rising college wage premiums - includes related articles - Special Issue: Earnings Inequality
作者：John H. Bishop
期刊名称：New England Economic Review
印刷版ISSN：0028-4726
出版年度：1996
卷号：May-June 1996
出版社：Federal Reserve Bank of Boston

Is the market for college graduates headed for a bust? Demand and supply responses to rising college wage premiums - includes related articles - Special Issue: Earnings Inequality

John H. Bishop

Demand for college-graduate workers was strong during the 1980s (Blackburn, Bloom, and Freeman 1989; Katz and Murphy 1990; Kosters 1989; Freeman 1991). Their relative wage rose, and the share of 18- to 24-year-olds attending college rose in response. Have the demand and technology shocks that produced this result now run their course? Is the supply response large enough to stop or reverse the 1980s escalation of the relative wages of college graduates?

Read superficially, U.S. Bureau of Labor Statistics (BLS) projections appear to suggest that the answer to these questions is "Yes." In the most recent BLS report, the growing supply of college graduates was projected to outstrip growth in demand by 330,000 annually (Shelley 1994). An even larger gap between supply and demand had been projected in 1992 (Shelley 1992). Looking at earlier projections, some in the press even reported that the college graduate labor market was about to go bust. New York Times reporter Louis Uchitelle, for example, led off an article titled "Surplus of College Graduates Dims Job Outlook for Others" with the following:

Hundreds of thousands of jobs, once performed creditably without a college degree, are today going to college graduates as employers take advantage of an oversupply.... At roughly 25 percent of the work force - higher than in any other industrial nation - college graduates outstrip the demand for their skills, the Labor Department reports (June 18, 1990, p. 1).

Most economists, however, do not believe a surplus of college graduates and other skilled workers exists now or is likely to develop anytime soon. The Secretary of Labor and the Chief Economist's office within the U.S. Department of Labor apparently give little credence to the BLS projections of a college graduate surplus. Skills shortages are a common theme of Secretary Robert Reich's speeches and of policy initiatives of the department.

Who is right? Is a bust of the college graduate labor market on the horizon? In fact, a closer reading of the latest BLS projections suggests that it is not. The future is predicted to be much like the past. Since the recent past has been characterized by low unemployment and rising relative wages for college graduates, the BLS data can also be interpreted as predicting more of the same.

Past BLS projections have not been particularly successful in predicting changes in the market for college graduates. The BLS projected a strong market for college graduates in 1970, just prior to the bust of the 1970s. The BLS projected a weak market in 1980, just prior to the 1980s boom. Moreover, the task of projecting the number of jobs "requiring a college degree" into the future is essentially impossible. The methods used to make the projections are not well adapted to the task. BLS projections published in 1981, 1983, and 1985 underestimated the growth of managerial and professional jobs and overestimated the growth of lower-skill jobs. The methods used to project occupational employment inevitably miss an important portion of the rise in skills that is under way in our economy.

An alternate methodology for projecting occupational employment is available. Regressions predicting occupational employment shares with a trend, unemployment, the trade deficit, and the share of workers using personal computers did a better job projecting the early 1990s than the BLS. Both methods, however, missed predicting the slowdowns in the growth of clerical, technical, and craft jobs.

The updated model projects that professional, technical, and managerial jobs will account for 60 percent of job growth between 1990 and 2005. Skill differentials between college-level jobs and other jobs continue to expand, even in the most recent data. And because the college-age population cohort is small, the increase in the share of the cohort attending school has not produced a substantial increase in the ratio of new BAs to total employment. Rates of college completion are not high enough to flood the college graduate market, and U.S. youth are not overeducated relative to those of other nations, as has been claimed. In fact, young Europeans and East Asians spend more years in school than young Americans.

These findings have important policy implications. They suggest that we should raise high school standards, increase student financial aid, make tuition tax-deductible, and stop increasing tuition at public colleges.

What Do the 1994 BLS Projections of Supply and Demand for Graduates Really Imply?

Let us begin by examining the projections of supply and demand made by the U.S. Bureau of Labor Statistics in 1992 and 1994. The 1992 report states that "estimates of available entrants to the college graduate labor force (supply) will average ... 406,000 more than demand" (Shelley 1992, p. 16). The 1994 report predicts that "nearly 25 percent of new entrants are expected to settle for jobs that do not require a college degree" (Shelley 1994, p. 9). Both sound quite negative about the future demand for college graduates. Table 1 presents the numbers. In 1992, BLS projected a significant deterioration of the supply-demand balance during the 1990s. The annual increase in the number of "underemployed" graduates was projected to be equal to 31 percent of the annual flow of BAs into the labor force (both immigrants and new graduates) during the period. The share of underemployed college graduates was projected to increase from 19.9 percent in 1990 to 25.9 percent in 2000.

However, when later projections were made, predictions of employment growth in professional jobs were revised upward, from the 340,500 per year of the 1991 projections to 477,000 per year in the 1993 projections and to 461,000 in the 1995 projections (Silvestri and Lukasiewicz 1991; Silvestri 1993; BLS 1996). This has improved BLS's projected outlook for college graduates. Consequently, in the most recent report, the gap between supply and demand is expected to grow only slightly, from 22 percent of the gross increase in the supply of BAs to 24 percent after 1992. The BLS also projects that the growth of the "underemployment" share will accelerate somewhat after 1992.

[TABULAR DATA FOR TABLE 1 OMITTED]

These projections can be interpreted in two very different ways. Many reporters have interpreted them as implying that the market for college graduates is about to deteriorate. This interpretation comes from focusing on projected gaps between demand and supply in the future. Focusing instead on how the future is expected to differ from the recent past, one arrives at a different conclusion. The 1994 report projects that the supply-demand balance for college graduate workers for the 1992-2005 period will be rather similar to the conditions that prevailed during the previous eight years. Since unemployment rates of college graduates remained low and relative wage rates grew substantially from 1984 to 1992, the BLS projections really predict a continuation of the strong labor market for college graduates that characterized the 1980s. They also point out, quite correctly, that a strong market for college graduates does not imply that all college graduates will have professional, technical, managerial, or sales representative jobs.

Are the BLS Estimates of Jobs "Requiring" a College Degree and of "Underemployed" College Graduates Credible and Reliable?

The BLS assesses the current demand-supply balance for college graduates by defining a set of jobs that "require" a college degree and then counting up the number of college graduates who do not have one of these jobs. The workers being categorized are not asked whether they believe a college degree is required or useful in their job. The classification is based on the match between reported education and reported occupation. Workers with fewer than 16 years of schooling are automatically counted as having jobs that do not require a college degree. Workers with 16 or more years of schooling are classified as "underemployed" when the reported occupation appears not to "require" a college degree.

This classification of occupations is inherently arbitrary and idiosyncratic to the analyst. When Sargent and Pfleeger (1990) did the analysis, the BLS concluded there were 18.1 million college-level jobs in 1988. When Hecker (1992) reestimated the number two years later, he found 21.8 million college-level jobs (a 20 percent increase). Yet the validity of the whole effort to measure "underemployment" depends on this classification being done correctly in every detail, not only for the present but also for up to 15 years in the future. This is essentially impossible.

First, the occupational coding system used by the Current Population Survey (CPS) and the Census is not reliable and comprehensive enough to allow accurate measurement of a concept like "underemployment." Census Bureau studies have found that between 18.3 and 27.3 percent of the individuals recorded as professionals, technicians, or managers in one interview are recorded in a less skilled occupation in a subsequent interview four to seven months later (U.S. Bureau of the Census 1972).

Substantial errors also occur in measuring educational attainment. Between 5.5 and 9 percent of respondents recorded as having more than 16 years of schooling in one interview are recorded as having fewer than 16 years of schooling in a later interview. If errors in reporting occupation and schooling are uncorrelated with each other, measurement error raises the estimated "underemployment" share by as much as 12 to 18 percentage points.(1)

Consequently, reporting and coding errors are responsible for many of the apparent mismatches between an individual's occupation and his or her education. How else can one explain the 9.6 percent of college teachers and the 5.4 to 6.5 percent of lawyers, physicians, and secondary school teachers who claim not to have completed 16 years of schooling (BLS 1990, Table F-3)? The unreliability of individual measures of occupation and education means that counts of mismatches between schooling and occupation derived from micro CPS data have little validity. True mismatches between education and occupation are a lot less common than these statistics suggest.

The second problem is the lack of symmetry in the handling of possible mismatches between educational qualifications and occupation. Large numbers of workers without college degrees say they occupy jobs that most people would agree "require" at least a four-year degree. In 1988, 44,000 lawyers, 42,000 social scientists, 46,000 natural scientists, 33,000 physicians, 61,000 college teachers, 143,000 elementary and secondary school teachers, and 363,000 engineers said they had not completed four or more years of college (BLS 1990, Table F-3). The BLS does not classify these individuals as "undereducated." By ruling out the possibility of undereducation, the conceptual framework makes inevitable a conclusion that there are too many college graduates.

The third problem is the great heterogeneity of the college graduate category. Ten percent of college graduates cannot write a brief letter explaining an error made on a credit card bill or determine the discount from an oil bill for early payment (National [TABULAR DATA FOR TABLE 2 OMITTED! Center for Education Statistics (NCES) 1995a, pp. 38, 40, 66). As one can see in Table 2, these graduates are less likely to have professional, technical, or managerial jobs and much more likely to have service or laborer jobs (NCES 1995a, p. 95).

More than 40 percent of young adults with associate or bachelor's degrees cannot calculate change from a menu. Seventeen percent of young college graduates read at a level below the typical eleventh-grader (Kirsch and Jungeblut 1986). How can someone with high-school or lower reading and math levels be considered "underemployed" or "overeducated" in a secretarial, carpentry, or retail sales job? For such individuals, the problem is "undereducation," not "underemployment."(2)

The fourth problem is that for most occupations, the question "Does it require a college degree?" does not have a "yes" or "no" answer. It is a matter of degree. Some employers structure their management jobs in ways that make the skills normally developed in college absolutely essential; at other employers such skills are very helpful, and at still others the skills are of little advantage. The magnitude of the college graduate productivity advantage also depends on the quality of the alternative labor supply. If the competence of those who ended their schooling with high school deteriorates, as it did during the 1970s (Bishop 1991), the demand for college graduates will increase. The correct answer to the question of whether a college degree is required is "It depends." It depends on circumstances that analysts and researchers have little knowledge of and no ability to forecast a decade ahead.

The BLS projection exercise apparently assumes that occupations have specific educational or basic skill "requirements." This job requirements view of the world is illustrated in Figure 1. Job A requires a basic skill level of at least [M.sub.1] while job B requires basic skills at or above [M.sub.2]. Exceeding these skill thresholds, however, very quickly yields no further increases in productivity. Once the threshold is reached, diminishing returns set in with a vengeance. People working in job A who have an [M.sub.2] skill level are classified as "underemployed." Is this how basic skills influence job performance? Let us test the job requirements hypothesis.

The job requirements view predicts that everyone in job A will have at least the [M.sub.1] skill level, and that as a result of diminishing returns, the steep productivity increase near the [M.sub.1] skill level quickly becomes a very gentle slope as the individual's skill increases. In other words, the relationship between test scores and job performance should have a convex shape (a large negative second derivative). The job requirements view also implies that the impact of basic skills tests on job performance should diminish as schooling increases.

Data collected by the U.S. Department of Labor's Employment Service to validate the General Aptitude Test Battery were used to test these hypotheses. This data set contains data on job performance, the nine GATB "aptitudes," and background data on 36,614 individuals in 159 different occupations. Professional, managerial, and high-level sales occupations were not studied but the sample is quite representative of the 71.1 million workers in the rest of the occupational distribution. It ranges from drafters and laboratory testers to hotel clerks and knitting-machine operators. A total of 3,052 employers participated. (See Box 1, "Description of the Job Performance Study," below.)

The results are presented in Figure 2 and Table 3.(3) [TABULAR DATA FOR TABLE 3 OMITTED! For general academic ability, the hypothesis of diminishing returns was rejected in seven of the eight occupation groups. The exception was sales clerks, where the squared term on general academic ability was significantly negative and general academic ability had a positive effect on performance only when the test scores were no more than one-half a standard deviation above the mean.

For psychomotor skills, however, the hypothesis of diminishing returns was accepted at the 10 percent level for operatives, service workers and sales workers, and for all workers combined. The second derivatives are not so large, however, that the sign of the relationship reverses within the range of actual data. For all workers and for operatives, the derivative of performance with respect to the psychomotor test scores at one standard deviation above the mean of the test is 52 to 57 percent of the derivative when test scores are one standard deviation below the mean. For service workers, the derivative one standard deviation above the mean is 37 percent of the derivative at one standard deviation below the mean.

These results suggest that the job requirements model has some validity for psychomotor skills but not for the basic academic skills that are the primary objective of schooling. Both this analysis and studies conducted by others have found that the underlying relationship between basic academic skills and performance in a specific job is smooth, continuous, and close to linear (Hunter 1983).

The Record of BLS Projections of the Demand and Supply for College Graduates

Despite the difficulties, since 1970 the BLS has published biennial projections of the supply-demand balance. The starting point of its projections are its forecasts of occupational employment growth. It then projects changes in the proportion of particular occupations that "require a college degree," the number of bachelor's degrees to be awarded per year, and the [TABULAR DATA FOR TABLE 4 OMITTED] annual rates of flow into and out of jobs by workers with a college degree. Comparisons are then made between the projected number of job openings "requiring a college degree" and the projected flow of college graduates seeking work, producing estimates of the number of "underemployed" college graduates. Column 3 of Table 4 presents BLS's projection of the annual increase in the number of "underemployed" college graduates during the projection period. Column 4 presents the projected 10-year change in the share of college graduates who are "underemployed."

Quite clearly the BLS effort to project the supply-demand balance for college graduates has been a failure. Compare the predicted changes in the share of college graduates underemployed (Column 4) to the actual changes (Column 5). At the beginning of the 1970s, BLS projected a decline in the share of college graduates who were underemployed during the subsequent decade. Instead, the share underemployed grew substantially. Changes in the relative wage of young college graduates provide an additional ex post criterion for evaluating the accuracy of BLS's projections of supply-demand balance (Columns 7 and 8). If the projection had been correct, the relative wage of college graduates should have also risen during the period. Instead, the college premium fell 6.7 to 7.6 percentage points by 1980 (Column 8).

At the end of the 1970s, BLS was projecting large surpluses of college graduates during the 1980s. According to the projection made in 1978, the surplus of college graduates was going to grow at a rate equal to 30 percent of the annual flow of bachelor's degrees awarded and the "underemployment" share was going to rise 5.5 percentage points by the end of the decade. The rise in the "underemployment" share was instead only 1.7 percent. If the projections had been correct, relative wages of college graduates should have fallen; instead they rose by 23 to 26 percentage points.

In reality, demand responds to supply and supply responds, with a lag, to demand. An increase in the supply of college graduates with computer science degrees, for example, lowers wages for the group, and this allows some companies to undertake projects not feasible before and it induces other companies to keep development work in the United States rather than moving it abroad. Wage rates and job-finding difficulties influence enrollment decisions and choice of major, so supply responds to demand. Getting "college-level" jobs also depends on personal qualities - initiative, work habits, and the like - of the graduates. The BLS projection model has oscillated between predicting large decreases and large increases in the share of college graduates who are "underemployed" because it omits feedback loops and other key determinants of employment patterns. If one wants to project shares of college graduates in "non-college" jobs, a better approach is to estimate a historical model (or system of equations) using variables that can be forecasted into the future.

The evidence suggests that the BLS's methods of translating occupational projections into projections of the demand-supply balance for college graduates are seriously flawed. But the problems are not limited to the way in which occupational employment distributions are translated into numbers of college-level jobs. The BLS's occupational projections are also seriously flawed. The BLS systematically under-projects the growth of skilled jobs and over-projects the growth of unskilled jobs.

Biases in BLS Projections of the Growth of Managerial and Professional Jobs

A myth is abroad in the land that job growth is coming (or will come) primarily from low-skill jobs ("McJobs"). In 1987, Levin and Rumberger, for example, stated:

In summary, the evidence suggests that new technologies are unlikely to have a profound effect in upgrading the education and skill requirements of jobs, and that most new jobs or job openings will be in occupations that require relatively low skills and education (1987, p. 344).

In 1990, Mishel and Teixeira predicted:

Growth in skill levels from occupational upgrading will actually slow down in the 1990s. In fact, future growth rates in skill levels are likely to be only one-fourth to one-third as fast as those in the recent past (1990, p. 1).

In 1995, Basil Whiting, a former deputy assistant secretary in the Department of Labor, wrote:

Labor Department projections show that most new jobs in the economy at the turn of the century will not be those of technicians but rather in the more prosaic and lower-paid fields of hospitality, retail sales, clerical work, janitoring and other service occupations (Whiting and Sayer 1995, p. 11).

All of these writers based their forecasts on BLS occupational projections. Levin and Rumberger justified their reliance on BLS projections, as follows:

On the basis of their past record they are still likely to provide a better indication of how the overall job market will look in the future than generalizations from a few casual observations, guesswork, or simple extrapolations of past trends. The point is that none of the latter devices has come close to the accuracy of the BLS forecasts in a world where - by their nature - no forecasts will be perfect (1987, p. 338).

How good is the past record of BLS projections of job growth? BLS projected in 1981 that professional, technical, and managerial jobs, which were 24.9 percent of the nation's jobs in 1978, would account for 28 percent of employment growth between 1978 and 1990 (see Table 5, line 3). Operatives, laborers, farm laborers, and service workers, 37 percent of employment in 1978, had been projected to account for 35.4 percent of employment growth during the period. Columns 4 and 6 of Table 5 tell us what actually happened: Professional, technical, and managerial jobs accounted for 53.6 percent of 1978-90 job growth and operative, laborer, and service jobs accounted for only 8.7 percent of the growth.

The passage of time has produced two additional opportunities to compare projected growth to actual growth: BLS's 1983 and 1985 projections of occupational employment growth through 1995. BLS projected in 1983 that professional, technical, and managerial jobs (PT&M) would account for 30.7 percent of job growth from 1982 to 1995. In fact, PT&M jobs accounted for 53 percent of employment growth. Operative, laborer, and service jobs (OL&S) were projected to account for 30.8 percent of job growth, but in reality accounted for only 15.7 percent of job growth.

For the 1984 to 1995 period, BLS projected that PT&M employment would account for 38.8 percent of employment growth and that OL&S would account for 28.7 percent of growth. Here again, they were far off the mark. For the 1984 to 1995 period, PT&M accounted for 58.3 percent of job growth, and OL&S accounted for 15.9 percent (row 5 of Table 5). Figure 3 presents a comparison of projected and actual growth by broad occupational category (rather than occupational shares of total growth, as in Table 5) for the 1984-95 period. As before, modest growth (22 percent over 11 years) was projected for professional and managerial jobs. Actual growth rates were much larger: 35 percent for professional jobs and 51 percent for managerial jobs. The BLS projection for technical jobs [TABULAR DATA FOR TABLE 5 OMITTED! matched the actual growth of 29 percent, however. For the operative, laborer, and service categories, each component was overestimated.

Why Have BLS Projections Been So Far Off the Mark?

BLS occupational projections contain many possible sources of error. Projections of final demand shares may be wrong. The input-output matrix is often quite old, and this contributes to errors in projecting value-added shares. For example, unanticipated changes in the federal deficit and exchange rates made export and import shares of industry output particularly difficult to predict in the 1980s. Industry-specific productivity growth may also be in error, resulting in incorrect predictions of industry employment.

Substantial changes also have occurred in the occupational composition of specific industries, and this has often been a major source of error in occupational projections. BLS derives occupational employment demand by multiplying projected industry employment totals by an assumed, industry-specific, occupational share vector. Adjustments are made to these vectors when BLS studies of the introduction of new technology indicate that changes can be anticipated by the end of the forecast period.(4) Since studies cannot be funded for every industry and for every technological innovation, and the effects of these changes are difficult to foresee 10 years in advance, many of the changes that will occur in the composition of occupational demand within industries are missed by BLS projections.

The BLS obtains its estimates of the occupational composition of employment in specific industries from a survey of establishments, the Occupational Employment Statistics (OES). When the BLS made the projections of 1990 occupational employment in 1981, only one wave of OES survey data was available for most states and industries. The projections for the 1984-95 period were made more difficult by a 1982-83 change in the occupational classification system. Comparability over time is also threatened by the periodic changes in the industry-specific list of occupations that respondents receive on their questionnaire. The Handbook of Methods describes what is done when data are thought to be of doubtful comparability: "When an occupation is added, deleted or changed in definition from one OES survey to the next, extrapolated trends are not developed: the current-year ratios for these occupations are held constant in the preliminary projected matrix" (BLS 1982, p. 143).

Given these data problems and BLS's focus on projecting over 500 different occupations, it is easy to see why BLS chose not to systematically extrapolate past trends in occupational staffing ratios derived from OES or other data into the future, but rather to rely on the judgment of analysts who can take data quality problems into account. Systems that rely on the judgment of analysts are inherently conservative, however. Unfortunately, occupational staffing ratios are seldom stable over periods of 10 years or more, and it is better to start with an assumption that trends are stable than that the ratios themselves are stable.

It does not appear that problems in extrapolating changes in occupational staffing ratios have become less serious as experience with the OES survey has accumulated. The PT&M share of projected job growth has not increased much in later projections, and the 1987 and 1989 projections seem as wide of the mark as the 1985 projection (Table 5). With 64 percent of the projection period already completed, the actual PT&M share of job growth is running 25.8 percentage points above the share projected in 1987. With 58 percent of the projection period already completed, the actual PT&M share is running 29.3 percentage points above the share projected in 1989.

Can Other Projection Methods Succeed Where the BLS Method Has Failed?

Ron Kutscher, the Associate Commissioner at BLS responsible for projections, has said "One could never hope that a projection of the future is entirely accurate" (1991, p. 253). A natural response to the criticisms of BLS methods is to ask, "Can one do better?" Shani Carter and I attempted to do better in a 1991 paper. We showed that both a simple linear extrapolation of occupational share trends for the 1972-80 period and a regression-based projection did a better job of predicting 1990 shares than the BLS (Bishop and Carter 1991b).

Hindsight is always better than foresight, however. A projection constructed by someone with knowledge of the actual outcome will generally be better than projections developed without such knowledge. Probably the only really fair test of the validity of alternative forecasting methodologies is to look back at projections published in the past and compare results.

This section of the paper offers such a test. The 1991 regression-based projections of Bishop and Carter are compared to the BLS projections published in the same year.(5) Summary results on the high-skill and low-skill shares of projected and actual job growth are presented in rows 8 and 11 of Table 5. Projected shares of job growth through 2000 will be compared to shares of actual job growth through 1995. Professional, technical, and managerial jobs actually accounted for 74 percent of job growth between 1990 and 1995. Operatives, laborers, and service workers accounted for 10.5 percent. Bishop and Carter's (B&C) projections clearly come closer to the 1990-95 reality than the BLS projections. B&C projected a 70 percent PT&M share of job growth and a OL&S share of 2 percent. BLS by contrast projected a PT&M share of 41 percent (46 percent when translated into CPS data) and an OL&S share of 27 percent.

In a paper published in early 1992, Bishop presented projections based on a variety of linear specifications, rather than a logit specification (Bishop 1992b). Summary results for these projections are shown in rows 12 through 14 of Table 5. These projections also outperform the BLS projections. The best of the projections for the professional, technical, and managerial category have the same four variables on the right-hand side as the preferred model of Bishop and Carter (1991a). In this preferred model, occupational shares are a function of a time trend, the unemployment rate, the merchandise trade balance as a share of GDP, and the share of workers with computers on their desks.

Let us take a more disaggregated look at how the projections are doing, one-third of the way through the 15-year projection period. Which occupations were accurately projected by both methodologies? Which occupations surprised both B&C and BLS? Figure 4 compares actual growth during the 1990-95 period to projected growth. (The 15-year projections were sized to a five-year period by the simple expedient of dividing percentage growth projections by 3.) Both projections missed three important developments: sharp slowdowns in the growth of craft jobs, clerical jobs, and technical jobs. In areas of disagreement about growth rates - for managers, professionals, operatives and laborers, and service workers - the B&C methodology produced more accurate predictions. For a description of the B&C model and its stability, see Box 2.

Are the Wage Premiums Paid for Skill Continuing to Rise?

The wage differential between college graduates and high school graduates grew more slowly after the mid to late 1980s than it did in the preceding decade. It still appears to be growing, however. In Katz and Murphy's (1990) data, the weekly wage differential for workers with one to five years of work experience stabilized at a high level beginning around 1985, but the differential for workers of all other experience levels continued to grow through 1990, the latest year of their data series (Murphy and Welch 1993). In Mishel and Bernstein's (1992) data, the differential for workers of all experience levels rose 7.8 percent between 1987 and 1990 and then fell 2.9 percent in 1991. Figure 5 presents the most recent data on the wage premium received by those with four or more years of college. For males, the wage premium rose from 57 percent in 1989 to 64 percent in 1992 and 71 percent in 1993. For females, the premium rose from 63 percent in 1989 to 68 percent in both 1992 and 1993.

The 1991-92 recession caused many companies to cut back their hiring of college graduates. The unemployment rate among managers and professionals, 2.0 percent in the first quarter of 1989, rose to 3.5 percent in September 1992 and then fell to 2.6 percent during 1994. This setback was not the beginning of a crash in the market demand for college graduates; it was a temporary effect of the recession. Blue-collar workers were hurt much more by the recession than managers and professionals. The unemployment rate of operatives and laborers, which was 7.7 percent in the first quarter of 1989, rose to 11.4 percent in July 1992 and was still 9.0 percent during 1994. BLS data on median weekly earnings can also be brought to bear on the issue of recent trends in wage premiums for skill. Between the second quarter of 1991 and the second quarter of 1995, the annual rate of increase of nominal wages was 1.4 percent for operatives and laborers, 1.6 percent for service workers, 1.7 percent for craft workers, 1.2 percent for clerical workers, 3.3 percent for technicians, 2.9 percent for managers and 3.7 percent for professionals. In summary, the very latest data on trends in occupational wage differentials suggest that skill differentials continue to widen.

The latest data on employment growth have similar implications. Between October 1994 and October 1995, professional jobs grew by 753,000 and managerial jobs by 837,000, while service jobs rose by only 31,000 and all other jobs declined by 346,000.

Is the Supply Response Large Enough to Flood the Market with New BAs?

The one event that could invalidate my prediction of continued high wage premiums for college graduates is a massive increase in the number of college graduates trained in well-paid fields like science, engineering, and business. How likely is such a flood? The high economic payoffs to college during the late 1980s and the 1990s have increased enrollment in college and the proportion of high school graduates who complete at least one year of college [ILLUSTRATION FOR FIGURE 6 OMITTED]. Non-completion rates have remained high, however, so enrollment increases during the 1980s have had only a modest effect on the share of 25- to 29-year-old high school graduates who have completed a four-year degree or more. Many adults have gone back to school and completed their degree, however, and this has resulted in a substantial increase in the ratio of BAs awarded to the number of 22-year-olds - from 22.5 percent in 1980 to 31.0 percent in 1992 [ILLUSTRATION FOR FIGURE 7 OMITTED]. This ratio is projected to increase further to 35.3 percent in the year 2000, a 57 percent increase over 1980 (NCES January 1995b).

The proportionate increase in the total number of BAs awarded, however, is much smaller because the low birth rates of the 1960s and '70s mean that fewer individuals are in the 20- to 30-year-old age cohort that typically receives most of the BAs. As a result, the number of BAs awarded as a percentage of total employment fell from 1.09 percent in 1974 to 0.95 percent in 1980 and 0.96 percent in 1992. It is projected to fall even further, to 0.88 percent in 2000 and 0.86 percent in 2005. Relative to the stock of college graduates, the number of new BAs has declined even more precipitously. Thus, despite the technology-driven shift in employer demand in favor of college-educated workers, the ratio of new graduates to total employment has not risen.

To make matters worse, the number of college graduates retiring from the labor force is increasing every year, as the veterans who went to college under the GI bill retire from the work force. As a result, the ratio of workers with a college degree to those with a high school degree or less is projected to grow no more rapidly in the 1990s than it did during the 1980s (Bishop 1992b, Table 4).

Is the Rapid Growth of College-Level Jobs since 1970 an Aberration?

Still another way to test the reasonableness of our projections of continued strong growth of demand for college graduates is to look at trends in managerial and professional jobs abroad and in the supply of well-educated workers in other industrialized nations. Let us define the rate of increase in skills as the difference between the growth rate of professional, technical, and managerial (PT&M) jobs and the growth rate of manual jobs (service, craft, operative, laborer, and farm occupations). For the United States, the rate of skill increase was 1.6 percent per year during the first half of the twentieth century, 1.9 percent per year between 1950 and 1970, 2.8 percent per year between 1970 and 1981 and 2.5 percent per year during the 1980s.

The rise in skills is proceeding even more rapidly in Europe and East Asia (Table 6). The Japanese rate was 4.3 percent per year in the 1970s and 3.3 percent per year in the 1980s, the German rate was 3.7 percent per year in the 1970s and 2.5 percent in the 1980s. Finland's rate was 6.4 percent per year in the 1970s and 5.1 percent per year in the 1980s, Korea's 4.2 percent per year in the 1970s and 6.0 percent per year in the 1980s. Three countries - Canada, Norway, and the United Kingdom - now have proportionately more professional, technical, and managerial workers than the United States, and other countries are close behind.

The supply of college-educated workers has been increasing rapidly all over the world. During the 1970s and 1980s, the share of the population of working age who were university graduates grew at an annual rate of 3.3 percent in the United States, 3.6 percent in Japan, 2.8 percent in Germany, 5.6 to 5.8 percent in Sweden and Norway, 3.1 percent in Belgium and 4.0 percent in Canada (OECD 1989).

The share of the work force that has graduated from university is higher in the United States than in any other country (columns 4 and 5 of Table 6). But many Europeans would argue that the bachelor's degrees awarded at the second- and third-rank American colleges and universities that educate the vast bulk of students reflect a lower standard than the French licence or the Dutch Doctoraal examen. High school graduation standards are also higher in Europe and Asia. Completing secondary school requires 14 years rather than 13 years of attendance in some countries. In others, high standards result in many students having to repeat grades. In many cases, the material Americans study in the freshman year of college is taught to Asians and Europeans in secondary school.

Any U.S. lead in the share of the work force with college education is also a legacy of policy initiatives that are 30 to 50 years old. The share of 25- to 29-year-olds who have graduated from college is no higher now than it was 17 years ago. In terms of flows - that is, numbers currently in school - the nations of Northern Europe and East Asia 'have either caught up or surged ahead. Using age-specific school enrollment rates, the OECD has calculated the expected number of full-time equivalent years of schooling (K to 12 or post-secondary) received between ages five and 29. These estimates are presented in the last column of Table 6. For the United States, the figure is 14.8 years. The comparable figure is 15.9 in West Germany and France, 15.8 in Belgium, 15.4 in Canada and Finland, 15.6 in Norway and Denmark, and 16 in the Netherlands.

In sum, the rapid growth of college-trained workers and of college-level jobs is not uniquely American. American youth are neither more nor better educated than their counterparts in Japan and Northern Europe. Thus, the "America is so well educated, much of it must be unnecessary" argument offered by Louis Uchitelle and others is based on a premise that is no longer valid, if it ever was.

Policy Implications of These Findings

Professional, technical, managerial, and high-level sales workers currently (October 1995) account for 38.4 percent of United States employment, 42.2 percent of hours worked, and about 59 percent of the earnings received by all workers.(10) By comparison, craft workers, operatives, and laborers outside of [TABULAR DATA FOR TABLE 6 OMITTED] construction receive only about 19 percent of the compensation paid in the economy. The competitiveness of the American work force is in reality more a function of the cost and quality of managerial, professional, and technical workers than it is of the blue-collar factory workers who are normally the focus of discussions of competitiveness.

The short-run consequence of a shortage of highly qualified workers is higher wage premiums for the skilled. The long-run consequence may be loss of comparative advantage in industries that make heavy use of managerial, professional, and technical workers. The high cost of hiring managerial and professional workers in the United States is already inducing firms to look elsewhere for these skills. For example, many software companies now economize on expensive American programmers and systems analysts by contracting with subsidiaries in Bulgaria, Russia, and India to develop code for new programs.

In 1991, Hewlett-Packard picked a Frenchman to head its troubled PC division and moved the division's headquarters to his home town of Grenoble, France. Since then, HP has staged a dramatic comeback in the PC market. Manufacturing time was cut, pricing became more aggressive and, as a result, HP moved from fourteenth to sixth largest PC producer in the world. Production, which had been spread across 12 plants, was concentrated into just two, one of which is in Grenoble (The Economist 1993). Hewlett-Packard is not alone. In 1991 and 1992, Du Pont moved the headquarters of its electronics division to Tokyo and its agricultural products division and part of its fiber and polymer business to Switzerland. IBM moved its networking systems division headquarters to the United Kingdom (Lublin 1992). Is this the start of a trend?

During the 1980s, 18- to 64-year-old college graduates with a business major earned nearly three times what high school graduates of the same gender earned (Kominski and Sutterlin 1992). Since social rates of return to college are now at postwar highs, substantial increases in supply are desirable. This would simultaneously reduce the supply of unskilled workers, so skill premiums should fall and unskilled wages should rise. It would not be a tragedy if a major increase in college completion rates lowered the wage premium paid business B.A.s over high school graduates to only 100 percent rather than 200 percent. Indeed, competitiveness would improve and income inequality would decline.

Substantial increases have already occurred in the proportion of young people getting college degrees, but these have not been sufficient to stop the seemingly inexorable rise in the college wage premium. Still bigger increases in college enrollments have been prevented by a rapid escalation of tuition charges at public colleges and the limited availability of need-based financial aid (Bishop 1992a). During the 1980s, tuition charges rose 48 percent more than student ability to pay out of current earnings.

Legislators and college presidents often justify the escalation of college tuition as only fair, given the high wages graduates receive as adults. Setting tuition high is claimed to be a way of helping those who cannot afford college at the expense of rich college graduates. This is a myth. The promised increases in financial aid are never sufficient to hold college students from low-income families harmless. The primary outcomes are fewer students, fewer graduates, and higher wages for those who complete college. College enrollment and graduation rates are highly responsive to tuition levels. Regression models (Bishop 1992a) imply that raising public college tuition by 50 percent ($893 per year) would lower the enrollment of 18- to 19-year-old women by 16 percent, lower the enrollment of 20- to 24-year-old women by 21 percent, and lower the number of B.A.s awarded to women by 11.8 percent.(11)

Elasticities of demand for and supply of college graduates are such that a 12 percent reduction in the supply of college graduates increases their wage relative to that of high school graduates by about 5.8 percent, or $1886 per year in 1992 dollars.(12) In the new long-run equilibrium that results, the present discounted value of after-tax earnings over the course of the graduate's career goes up $23,100, much more than the $3574 of additional tuition payments.(13) Those who graduate from college gain from a high-tuition policy. Two groups lose: those who are prevented from attending and graduating from college and those who never planned to go to college in the first place. They suffer a decline in their real wage because the number of high school graduates competing for the limited number of low- and medium-skilled jobs has gone up.

The implication of this discussion is that low tuition levels in public colleges (and tax deductions for college tuition) are both an effective and a fair way of increasing the supply of college graduates. Other ways of increasing the supply of college graduates are expanded financial aid; higher academic standards in high school, to reduce college drop-out rates; expansion of advanced placement programs, so as to shorten the time required to earn a degree; and preference for immigrants with high-level scientific and technical training over immigrants with little education and few skills.

1 Let us make the standard assumption that measurement error is random (that is, uncorrelated over time and uncorrelated across questions). Then, 18.3 to 27.3 percent of respondents reporting a PTM occupation in one interview reporting a non-PTM occupation in another interview implies that 0.8526 to 0.9039 of the individuals who are truly in a PTM occupation report themselves in a PTM occupation {[(1 - .273).sup..5] = .8526}. The estimated proportion of true college graduates who report having less than a college degree is 0.9539 to 0.9721 {[(1 - .09).sup..5] = .9539}. Thus, the estimated proportion of true college graduates with PTM jobs who underreport either their occupation or schooling ranges between 0.1213 and 0.1867 {.1867 = (1 - .8526 * .9539)}.

2 Robst's (1995) analysis of PSID data indicates that the prestige ranking of the college one attends also has large effects on the probability of being "underemployed." Those who attended colleges in the bottom fifth of the prestige ranking had twice the likelihood of being counted as "underemployed" of those who attended colleges in the top quartile.

3 Selection effects generate a negative bias in coefficients on years of schooling. In a selected sample like accepted job applicants or job incumbents, one cannot argue that these omitted unobservable variables are uncorrelated with the included variables that were used to make initial hiring decisions and, therefore, that coefficients on included variables are unbiased. When someone with 10 years of formal schooling is hired for a job that normally requires 12 years of schooling, there is probably a reason for that decision. The employer saw something positive in that job applicant (maybe the applicant received particularly strong recommendations from previous employers) that led to the decision to make an exception to the rule that new hires should have 12 years of schooling. The analyst is unaware of the positive recommendations, does not include them in the job performance model and, as a result, the coefficient on schooling is biased toward zero. This phenomenon also causes the estimated effects of other worker traits used to select workers for the job such as previous relevant work experience to be biased toward zero. Consequently, the results presented should not be viewed as estimates of the structural effect of schooling on worker productivity. The test score results are not similarly biased, however, because firms using aptitude tests similar to the GATB for selecting new hires were excluded from the regression.

4 This characterization of how occupational staffing patterns were projected is based on U.S. Bureau of Labor Statistics, Handbook of Methods, Bulletin 2134-1, 1982, p. 143, and conversations with Ron Kutscher, Associate Commissioner responsible for projections.

5 The regression equations used for this exercise are found in Table B2 of Bishop and Carter (1991a).

6 The estimates of the number of PCs in use in business were made by Future Computing/Datapro Inc. and can be found in Table 1340 of The Statistical Abstract, p. 179. They are derived by cumulating the numbers of machines sold. A very low scrap rate of 3.4 to 6 percent, depending on the year, was assumed. Where possible, vendor reports were used to allocate sales of computers between categories of end user - business, education, and home. Quite often, however, rules of thumb were used to make these allocations. Future Computing is no longer in business so more detailed information on how the series was constructed and data for 1989 are not available. CPS surveys in 1989 and 1993 provide data on the proportion of workers who use computers at work (NCES 1993c, p. 434, and 1994, p. 439). The proportionate growth rate produced by comparing the two surveys was applied to the 1989 value of the PC use variable from Future Computing. Projections of PC use for succeeding years were made by extrapolation. Projections were based on an assumption that the unemployment rate in 2005 would be 5.5 percent and the merchandise trade deficit would be 1.4 percent of GDP.

7 B&C estimated a number of alternative models in order to test the sensitivity of results to changes in functional form and specification and in the scenario projected for the year 2000. Such tests were needed because only 18 years of data were available on which to estimate the forecasting model, and theory did not yield only one plausible specification. The results of some of these tests are detailed in Bishop and Carter (1990). While specification and scenario did affect projected occupational shares, all of the specifications yielded substantially larger increases in skilled jobs than the BLS projections. Other findings were robust with respect to specification and scenario as well.

8 The estimates are available from the author upon request.

9 The addition of 1990-95 data significantly changed the projections of employment in 2005 for three key occupations. Managerial Workers: In the 1991 paper the growing use of PCs appeared to be the primary explanation of the accelerating growth of managerial jobs during the 1980s. Despite the continued rise in PC use, the recession of the early 1990s caused a larger than expected slowdown in the growth of managerial jobs. B&C forecasts overpredicted managerial jobs by 2 to 4 percent between 1990 and 1993 but then got back on track when unemployment fell to 5.5 percent in 1994 and 1995. The updated estimate of the model gives unemployment a more important role and PC use a less important role in the determination of managerial jobs. Clerical Workers: The 1991 B&C model overpredicted the growth of clerical jobs in the early 1990s, underestimating the negative effect of the PC revolution. The updated regression reflects this by giving PC use a bigger role in the determination of clerical employment. Factory Operatives: In the 1991 B&C model PC use had a large, statistically significant, negative effect on operative employment. Operative jobs did not decline nearly as much as the model predicted, so the updated regression model assigns less weight to PC use and the result is an increase in projected employment of factory operatives.

10 Employment and Earnings (November 1995, pp. 27, 34). High-skill sales workers include sales representatives outside of retail and services and proprietors and supervisors in the retail and service industry. The median weekly wage for full-time managerial and professional workers is 46 percent above the overall median. High-skill sales workers' wages are 30 percent above and technicians' wages 14 percent above the overall average. These ratios are multiplied by the occupation's share of hours worked to calculate the share of earnings going to these high-skill occupations.

11 Average public college tuition charges were $1787 in 1992-93 (NCES 1993c, p. 309). A regression predicting the ratio of BAs awarded to women divided by the mean number of high school diplomas received by women 4 to 10 years previously was used to predict the impact of a 50 percent increase in public college tuition from its 1988 level (Bishop 1991). The ratio of tuition to the forgone earnings of female college students (the wage of female high school graduates with 1 to 5 years of work experience times .75) was assumed to be .1355, its actual level in 1988. The higher tuition policy is assumed to increase this permanently to .20325. The proportionate change in BA awards was calculated by multiplying (.20325 -.1355) x (the coefficient on the tuition variable) x (1 minus the ratio of BAs to high school diplomas in 1989) = (.06775) x (-2.72) x (1 - .362) = 11.75 percent.

12 The relative supply of college graduates is defined as ln[(BAs/HSG)/(1 - BAs/HSG)]: ln[BAs/(HSG - BAs)] = [E.sub.s]. The effect of the 50 percent increase in tuition on [E.sub.s], TP, is -.1843 = .06775 x (-2.72). Using a logarithmic approximation of the model predicting [E.sub.s] in Bishop (1991), we have a formula for the relative supply curve: [E.sub.s] = .89 x ln([W.sub.CG]/[W.sub.HSG])/.75 + TP + S, where S captures the effect on supply of other exogenous variables. Following Blackburn, Bloom, and Freeman (1989), I assume an elasticity of relative demand of -2, so the relative demand curve is [E.sub.d] = -2.0 x ln([W.sub.CG]/[W.sub.HSG]) + D, where D is a variable reflecting other influences on relative demand for college graduates. Setting [E.sub.d] = [E.sub.s], and reorganizing terms to get an expression for the relative wage, we have ln([W.sub.CG]/[W.sub.HSG]) = (TP + S - D)/(-2.0 - .89/.75) = (TP + S - D)/3.19. The long-run impact of the tuition increase on relative wages is (-.18428)/(3.19) = .0578 or 6 percent. If the elasticity of relative demand had been assumed to be -4, the equilibrium increase in relative wages would have been 3.55 percent. Even with this very high elasticity of substitution, the long-run effect of a high tuition policy is to help college graduates and hurt those who do not go to college. Since a 12 percent change in the flow of new BAs takes many years to have comparable effects on the stock of BAs, short-run effects on relative wages would be small, so when the policy is introduced the first few cohorts of college graduates lose out initially, because the wage increase starts out being small. After eight years or so, however, college graduates benefit from the policy change.

13 I seek to calculate the long-run impact of the policy of raising public college tuition charges by 50 percent and keeping them high. Mean earnings of college graduates were $32,629 in 1992, so a 5.8 percent increase is $1886 per year. The marginal tax rate (netting out deductions) on these earnings is assumed to be 35 percent, so at a 5 percent real rate of discount, the PDV at age 21 is $1886. x .65 x 17.89 = $21,931 where 17.89 = (1/.05)(1 - [e.sup.45r]) because the individual is assumed to work continuously until age 66.

References

Adkins, Douglas L. 1975. The Great American Degree Machine. The Carnegie Foundation for the Advancement of Teaching.

Alsalam, Nabeel. 1993. "Interpreting conditions in the job market for college graduates." Monthly Labor Review, February, pp. 51-53.

Bishop, John H. 1977. "The Effect of Public Policies on the Demand for Higher Education." Journal of Human Resources, vol. XII, no. 3, Summer.

-----. 1991. "Achievement, Test Scores and Relative Wages." Workers and Their Wages, edited by Marvin Kosters, pp. 146-81. Washington, D.C.: American Enterprise Institute.

-----. 1992a. "The Growth of Female College Attendance: Causes and Prospects." Center for Advanced Human Resources Studies, Working Paper.

-----. 1992b. "Is a Skills Shortage Coming? An Evaluation of the Bureau of Labor Statistics Projections for 2005." Workforce, Spring.

Bishop, John H. and Shani Carter. 1990. "The Deskilling vs. Upskilling Debate: The Role of BLS Projections." Cornell University, Center for Advanced Human Resource Studies, Discussion Paper 90-14.

-----. 1991a. "The Worsening Shortage of College Graduate Workers." Educational Evaluation and Policy Analysis, Fall.

-----. 1991b. "How Accurate Are Recent BLS Occupational Projections?" Monthly Labor Review, October.

Bishop, John H. and Jane Van Dyk. 1977. "Can Adults Be Hooked on College? Some Determinants of Adult College Attendance." Journal of Higher Education, vol. 48, no. 1, January/February, pp. 39-62.

Blackburn, McKinley, David Bloom, and Richard Freeman. 1989. "The Declining Economic Position of Less Skilled American Men." In A Future of Lousy Jobs?, Gary Burtless, ed., pp. 31-67. Washington, D.C.: The Brookings Institution.

Carey, Max L. 1980. "Evaluating the 1975 Projections of Occupational Employment." Monthly Labor Review, vol. 103, no. 6, June, pp. 10-21.

-----. 1981. "Occupational Employment Growth through 1990." Monthly Labor Review, vol. 104, no. 8, August, pp. 42-55.

Carey, Max L. and Kevin Kasunic. 1982. "Evaluating the 1980 Projections of Occupational Employment." Monthly Labor Review, vol. 105, no. 7, July, pp. 22-30.

College Placement Council. 1985. "Inflation and the College Graduate: 1962-1985." Bethlehem, PA.

-----. 1991. "CPC salary survey." September. Bethlehem, PA.

The Economist. 1993. "The Metamorphosis of Hewlett-Packard." June 19, pp. 77-81.

Fiorito, J. and Robert C. Dauffenbach. 1982. "Market and Nonmarket Influences on Curriculum Choice by College Students." Industrial and Labor Relations Review, vol. 36, no. 1 (October), pp. 88-101.

Freeman, Richard. 1971. The Labor Market for College-Trained Manpower. Cambridge, MA: Harvard University Press.

-----. 1976. The Overeducated American. New York: Academic Press.

-----. 1991. "Labour Market Tightness and the Mismatch between Demand and Supply of Less-Educated Young Men in the United States in the 1980s." Mismatch and Labor Mobility, Schioppa Fiorel, ed., pp. 360-85. New York: Cambridge University Press.

Hecker, Daniel. 1992. "Reconciling conflicting data on jobs for college graduates." Monthly Labor Review, July, pp. 3-21.

-----. 1995. "Further analysis of the labor market for college graduates." Monthly Labor Review, February, pp. 39-41.

Hunter, John. 1983. Test Validation for 12,000 Jobs: An Application of Job Classification and Validity Generalization Analysis to the General Aptitude Test Battery. Washington, D.C.: U.S. Employment Service, Department of Labor.

Katz, Lawrence and Kevin Murphy. 1990. "Changes in Relative Wages, 1963-1987: Supply and Demand Factors." Quarterly Journal of Economics, vol. CVII, pp. 35-78.

Katz, Lawrence, Gary W. Loveman, and David Blanchflower. 1992. "A Comparison of Changes in the Structure of Wages in Four OECD Countries." Cambridge MA: National Bureau of Economic Research, July, pp. 1-47.

Kirsch, I. S. and A. Jungblut. 1986. Literacy: Profiles of American Adults. Princeton, NJ: Educational Testing Service.

Klein, Deborah. 1984. "Occupational Employment Statistics for 1972-82." Employment and Earnings, January, pp. 13-16.

Kominski, Robert. 1990. What's It Worth? Educational Background and Economic Status: Spring 1987. U.S. Bureau of the Census, Current Population Reports, no. 21, p. 70.

Kominski, Robert and Rebecca Sutterlin. 1992. What's It Worth? Educational Background and Economic Status: Spring 1990. U.S. Bureau of the Census, Current Population Reports, no. 32, p. 70.

Kosters, Marvin. 1989. "Wages and Demographics." Presented at "Wages in the 1980s" Conference, November 3, 1989, American Enterprise Institute, Washington, D.C.

Kutscher, Ron. 1991. "Reply to Bishop and Carter on the Worsening Shortage of College Graduates." Educational Evaluation and Policy Analysis, Fall, pp. 247-52.

Leslie, Larry and Paul Brinkman. 1985. "Student Price Response in Higher Education: The Student Demand Studies." University of Arizona, Tucson, AZ.

Levin, Henry and Russell Rumberger. 1987. "Educational Requirements for New Technologies: Visions, Possibilities, and Current Realities." Educational Policy, vol. 1, no. 3.

Lublin, Joann. 1992. "Firms Ship Unit Headquarters Abroad." Wall Street Journal, December 9, pp. B1 & B9.

Mishel, Lawrence and Jared Bernstein. 1992. Declining Wages for High School and College Graduates. Washington, D.C.: Economic Policy Institute.

Mishel, Lawrence and Ruy Teixeira. 1990. The Myth of the Coming Labor Shortage: Jobs Skills and Incomes of America's Workforce 2000. Washington, DC: Economic Policy Institute.

Murphy, Kevin and Finis Welch. 1993. Inequality and Relative Wages, pp. 1-14. Santa Monica, Calif.: Unicon.

National Center for Education Statistics. 1992a. The Condition of Education: 1992. Washington, D.C.: U.S. Department of Education.

-----. 1992b. Digest of Education Statistics: 1992, Table 352. Washington, D.C.: U.S. Department of Education.

-----. 1993a. The Condition of Education: 1993. Washington, D.C.: Department of Education.

-----. 1993b. Occupational and Educational Outcomes of Recent College Graduates 1 Year after Graduation: 1991. Washington, D.C.: U.S. Department of Education, April.

-----. 1993c. Digest of Educational Statistics: 1993. Washington, D.C.: U.S. Department of Education.

-----. 1994. Digest of Educational Statistics: 1994. Washington, D.C.: U.S. Department of Education.

-----. 1995a. Literacy in the Labor Force, pp. 1-288. Washington, D.C.: U.S. Department of Education, January.

-----. 1995b. Projections of Education Statistics to 2004. Table 28. Washington, D.C.: U.S. Department of Education.

Organisation for Economic Co-operation and Development. 1989. Employment Outlook. Paris: OECD.

Robst, John. 1995. "College quality and overeducation." Economics of Education Review, vol. 14, no. 3 (September), pp. 221-28.

Rumberger, Russell. 1981. "The Changing Skill Requirements of Jobs in the U.S. Economy." Industrial and Labor Relations Review, vol. 34, July, pp. 578-90.

Sargent, Jon and Janet Pfleeger. 1990. "1988-2000 - Outlook for College Graduates." Excerpts from an unpublished BLS Technical Memorandum.

Saunders, Norman. 1993. "Employment Effects of the Rise and Fall of Defense Spending." Monthly Labor Review, April, pp. 3-10.

Shelley, Kristina. 1992. "The Future of Jobs for College Graduates." Monthly Labor Review, July, pp. 13-21.

-----. 1994. "More Job Openings - Even More Entrants: The Outlook for College Graduates: 1992-2005." Occupational Outlook Quarterly, vol. 38, no. 2 (Summer), pp. 4-9.

Silvestri, George. 1993. "Occupational Employment: Wide Variations in Growth." Monthly Labor Review, November, pp. 58-86.

Silvestri, George, and John Lukasiewicz. 1985. "Occupational Employment Projections: the 1984-95 Outlook." Monthly Labor Review, November, pp. 42-59.

-----. 1987. "A Look at Occupational Employment Trends to the Year 2000." Monthly Labor Review, September, p. 46.

-----. 1989. "Projections of Occupational Employment, 1988-2000." Monthly Labor Review, November, pp. 42-65.

-----. 1991. "Occupational Employment Projections." Monthly Labor Review, November, pp. 64-94.

Silvestri, George, John M. Lukasiewicz, and Marcus E. Einstein. 1983. "Occupational Employment Projections through 1995." Monthly Labor Review, vol. 106, no. 11, November, pp. 37-49.

Tsang, Mun, Russell Rumberger, and Henry Levin. 1991. "The Impact of Surplus Schooling on Worker Productivity." Industrial Relations, vol. 30, no. 2, pp. 209-28.

U.S. Bureau of the Census. 1970. "Characteristics of Men with College Degrees: 1967." Current Population Reports, Population Characteristics, Series P20, No. 201, p. 23. Washington, D.C.: U.S. Government Printing Office.

-----. 1972. "Effects of Different Reinterview Techniques on Estimates of Simple Response Variance." Evaluation and Research Program of the U.S. Census of Population and Housing 1960, Series ER 60, No. 11, Tables 30 and 46, pp. 1-100.

-----. 1982. "Money Income of Households, Families and Persons in the United States: 1980." Current Population Reports, Series P-60, No. 132, Table 52. Washington, D.C.: U.S. Government Printing Office.

-----. 1987. "What's It Worth? Educational Background and Economic Status: Spring 1984." Current Population Reports, Series P-70, No. 11, U.S. Government Printing Office, Washington, D.C.

-----. 1991. "Money Income of Households, Families and Persons: 1988 and 1989." Current Population Reports, Series P-60, No. 172. Washington, D.C.: U.S. Government Printing Office.

U.S. Bureau of Labor Statistics. 1982. Handbook of Methods, Bulletin 2134-1.

-----. 1989 and previous years. "Nearly Three-fifths of the High School Graduates of 1988 Enrolled in College." USDL 89-308, June.

-----. 1990. Unpublished Table F-3, "1983-1988 trends in the employment of persons who have 4 or more years of college education and the percent they comprise of total occupation employment, by detailed occupation." pp. 26-38.

-----. 1992a. "Employment Cost Index - Compensation." Monthly Labor Review, March, p. 68.

-----. 1992b. Occupational Projections and Training Data. Bulletin 2401, U.S. Dept of Labor, May 1992.

-----. 1996. Press release, January.

-----. Various years. Employment and Earnings, Table 56 and Table 60, January and April issues.

Whiting, Basil J. and Wade Sayer. 1995. School-to-Work or School-to-What? Philadelphia, PA: Public Private Ventures, October, pp. 1-41.