Stern, D., Finkelstein, N., Stone, J. R., III, Latting, J., & Dornsife, C. (1994). Research on School-to-Work Programs in the United States (MDS-771). Berkeley: National Center for Research in Vocational Education, University of California.

Two-Year Colleges and Proprietary Schools

An increasing proportion of vocational and technical education is taking place in less-than-four-year postsecondary schools, both public and proprietary. Although federally supported vocational education is still limited to occupations not ordinarily requiring a bachelor's or advanced degree, the postsecondary schools prepare for a higher-paying segment of these occupations than secondary schools do. Nurses and various kinds of technicians are among the large occupational groups in which large proportions of individuals report that they received their qualifying training in two-year colleges or proprietary schools (U.S. Department of Labor, Bureau of Labor Statistics, 1992). Unlike those who qualified by means of secondary vocational programs, these postsecondary qualifiers obtain significantly higher earnings than individuals who report that their jobs required no special training at all (Bowers & Swaim, 1992).

Analytic Issues

To evaluate the effects of postsecondary education or training, it is necessary to take account of the fact that, unlike high school which is more or less compulsory, the decision to enroll in postsecondary education is voluntary. To inform policy, it is not enough to measure the difference in well-being between those who do and do not enroll. The question is whether individuals who do not (or who do) enroll would be better off if they did (or did not). For certain kinds of training programs, this can be answered by conducting an experiment in which individuals are randomly assigned to treatment and control groups. However, it is more difficult to conduct a random-assignment experiment to test the effects of attending a two-year college. Perhaps the nearest thing to a true experimental study on the effects of education is Angrist and Krueger's (1992) analysis of labor market outcomes for Vietnam-era young men whose participation in higher education was affected by the draft lottery. While their finding that ordinary non-experimental analysis apparently has not overstated the returns to schooling is of great interest, the pre-1992 Current Population Survey (CPS) data they were using does not permit specific analysis of nonbaccalaureate, postsecondary education. Consequently, inferences about the effects of two-year colleges or other postsecondary education must rely on naturally occurring variation in behavior.

Individual choices divide the population at several junctures. Some enter postsecondary education, some do not. Those who enter postsecondary education must choose among several kinds of institutions; and in most of these institutions, they are given a choice of fields to study. Some hold paid jobs while in school, others do not. Some finish their programs, some do not. Some join the labor force after leaving school, others do not. Each of these choices may be affected by unmeasured variables that also affect subsequent economic success. If so, simply comparing economic outcomes for the subpopulations created at each choice point will not give an accurate measure of the true effect of each choice, even when observed differences among the subpopulations are taken into account.

Logically, self-selection can reflect either absolute or relative advantage (see Maddala, 1983, pp. 257-260). In comparing individuals who attend community college with others who do not participate in any postsecondary education, an absolute advantage for the community college group means that their economic performance would exceed that of the other group even if both groups had the same amount of schooling, be it community college or no postsecondary education. Relative advantage means that the high school group's performance would be better than the community college group's if neither group participated in postsecondary education, but the community college group would do better than the high school group if both groups attended community college. Absolute advantage entails a hierarchical ordering; relative advantage implies a more pluralistic pattern of rational self-selection (Willis & Rosen, 1979).

If community college students have an absolute advantage over individuals who possess only high school diplomas, then ordinary statistical procedures that do not take account of this advantage will overestimate the benefit of attending community college by attributing to community college attendance differences that should properly be attributed to the prior advantage of those who attend. However, if the advantage of community college students is relative and not absolute, then ordinary statistical procedures which fail to take account of it will actually underestimate the benefit of community college attendance. The reason is that standard procedures would assume that individuals who did not attend would have obtained the same benefit if they did attend and would estimate that benefit as the difference in earnings (or other outcome) not attributable to other measured differences between attenders and non-attenders--but this contradicts the idea of relative advantage, which implies that non-attenders would not in fact obtain the same benefit if they did attend community college.

There have been some preliminary attempts to measure whether self-selection occurs in two-year colleges and, if so, whether it reflects absolute or relative advantage. Grubb (1990) compared annual earnings of individuals who entered community colleges with those who completed high school only. He found evidence that men who entered community colleges had unobserved characteristics that correlated positively with earnings. However, the selection coefficients for females who entered community college were statistically insignificant, as were the selection coefficients for males or females who completed high school only. In sum, he found no clear pattern of self-selection.

Hollenbeck (1992) compared labor market outcomes of individuals who had participated in postsecondary technical education with those of others who had enrolled in academic postsecondary education with the intent of completing a baccalaureate. He found significant coefficients indicating an absolute advantage for the technical group. This surprising result may be attributable to the fact that the sample was the 1980 senior cohort from the High School and Beyond (HSB) survey, and the dependent variable was the logarithm of hourly wage in spring 1986--so students who received baccalaureate degrees might not have had sufficient time to demonstrate their advantage in the labor market. Analysis of older cohorts will be necessary to establish whether selection effects are important.

Led by Heckman (1979), econometricians in the 1980s tried various procedures to correct for selection bias. However, when applied to training programs, these procedures yielded results that were very sensitive to assumptions about unobserved variables and that differed from results obtained by actual controlled experimentation (LaLonde, 1986). This led many to conclude that experiments with random assignment were necessary.

While random assignment of individuals to treatment and control groups does eliminate the influence of unmeasured variables in large samples, it also can create other problems. Unlike medical research, evaluations of education and training programs cannot administer a placebo. Therefore, individuals who are turned away from a program may feel some disappointment, after they have decided they wanted to participate and have made the commitment of applying. This may have a discouraging effect on their subsequent behavior. The result would be worse performance by individuals in the control group than if they had never heard about the program. This may explain why losers in the lottery for admission to New York City career magnet high schools were less likely to show up for ninth grade (see Tables 16-18). The existence of such a "disappointment effect" would tend to bias the evaluation in favor of finding positive results for the treatment group.

Another analytic issue has to do with how educational attainment is specified. In the varied world of postsecondary education, individuals may acquire certificates and degrees in various combinations. Perhaps it matters whether someone who obtains a baccalaureate, for example, has also picked up an associate's degree along the way. Unfortunately, most of the empirical work on returns to postsecondary education has simply classified individuals according to their highest educational attainment. Hollenbeck (1992) is an exception: He includes an indicator of whether individuals have received a vocational certificate or degree prior to obtaining a baccalaureate. The association of this indicator with wages and earnings is often statistically significant, but its sign is sometimes positive and sometimes negative.

A conventional way to measure the payoff from education or training is to compute the internal rate of return, defined as the value of the discount rate that equates the present value of costs and benefits. The private economic benefits from investing in education or training include increased earnings and consumption benefits. Social benefits include these plus any externalities. Private costs include tuition and other out-of-pocket expenses plus earnings foregone while attending school. For postsecondary students, foregone earnings are a major component of cost. Social costs also include any public subsidies.

Mincer (1974) introduced a technique for estimating the private rate of return to schooling directly from a regression in which the logarithm of earnings is a linear function of years of schooling and other predictors. The coefficient on years of schooling is an estimate of the rate of return. This technique has become widely used, but its derivation is sometimes forgotten. In Mincer's derivation, it is assumed that each year of schooling is a year of foregone earnings. (In this tradition, out-of-pocket expenses are sometimes assumed to be paid from earnings during the summer, when school is not in session.) That is, work life begins when schooling ends.

This assumption no longer fits the facts for postsecondary students, however. CPS data reveals that 45% of college students were employed in 1959, and 56% in 1986. Among 21-year-old college students in the National Longitudinal Survey of Youth Labor Market Experience (NLSY), employment rose from 52% in the spring of 1979 to 63% in spring 1986. The NLSY data shows employment rates are slightly higher among two-year than among four-year college students (Stern & Nakata, 1991). Many two-year colleges have accommodated working students by scheduling more classes at night.

Since a year of postsecondary schooling can no longer be considered a year of foregone earnings, the coefficient on years of schooling in a regression for the logarithm of earnings can no longer be interpreted as a rate of return. It simply measures the percentage increase in earnings associated with an additional year of schooling. Computing the rate of return to schooling requires comparing this benefit with the cost of schooling, which must be measured directly.

Unfortunately, there do not seem to be any studies on the actual amount of foregone earnings while in college. Existing longitudinal surveys contain enough information to estimate foregone earnings with reasonable accuracy, but no one appears to have done this analysis yet. The findings from such an analysis could have decisive implications for the rate of return. For instance, Stern and Nakata (1991) found that the rate of return to two years of college for a male whose annual college expenses were $3,000 would be 14.1% if he earned $1,500 a year while in college; however, if he earned $7,500 a year as a student, the rate of return would be 29.6%.

Freeman (1974) recognized the importance of this issue in his analysis of the economic payoff from training in proprietary business colleges and technical institutes. He observed that these schools made special efforts to reduce the opportunity cost of students' time by scheduling classes in the evening or in concentrated blocks during the morning or afternoon. As a result, a year of instruction in proprietary schools cost less than a year of college in foregone work hours. Using data on average work hours for students in proprietary schools and colleges, Freeman converted the actual time and money cost of schooling into year-equivalents. He used this to adjust the rate of return to schooling, estimated from a regression equation of the logarithm of earnings on reported years of schooling. The adjusted estimate represents a rate of return per true year-equivalent invested in schooling.

In addition to complicating the estimation of the rate of return to schooling, the fact that many students are employed while in school also raises other issues. One is, which comes first? Standard procedures for estimating the economic outcomes of schooling are still formulated as if students go first to school then to work. In fact, however, many young people spend several years after high school drifting from one job to another, as we have already pointed out, while at the same time dropping into and out of postsecondary education. At some point, a definable career interest eventually starts to gel. An actual work experience may be the precipitating factor in forming this stable identification with a particular line of work or a particular employer. Having identified desirable work, it may then become necessary to return to school in order to fill in missing qualifications or become eligible for advancement. Work experience may therefore precede schooling in preparing for eventual long-term employment. If so, work experience prior to and during the course of schooling may affect postschool success differently than work experience after schooling has been completed.

Some indication that work experience and schooling operate jointly in determining subsequent career success has come from studies of employed students reviewed in an earlier section of this paper. Stephenson (1981, 1982) analyzed data from the National Longitudinal Survey of Young Men and found that employment during college was positively related to wages a few years afterward. Stephenson treated work experience as exogenous, ignoring the possibility that employment during college and subsequent success in the labor market might both result, at least in part, from other variables such as ability or ambition. However, San (1986), using the same dataset, also found that work during college was positively associated with earnings a few years later, even allowing for work during college to be endogenous. As mentioned above, this is consistent with findings for high school students: namely, a positive correlation between time spent working while in school and subsequent employment or earnings (Bishop et al., 1985; D'Amico, 1984; Meyer & Wise, 1982; Mortimer & Finch, 1986; Ruhm, 1993; Stern & Nakata, 1989).

While these results point to the possibility that work and school may interact positively to influence subsequent career success, there is also evidence that working while in college increases the probability of dropping out (Kohen et al., 1978; Ehrenberg & Sherman, 1987). Consistent with Astin (1975), Ehrenberg and Sherman find this effect is smaller if students are working on campus, perhaps because their work is related to what they are studying or because on-campus jobs are more likely to accommodate students' schedules. The relatedness of school and work may be important not only in reducing the degree to which work interferes with school, but also in producing positive effects later.

National Data Sources

Several national datasets include information on economic returns to non baccalaureate education and training. The decennial U.S. Census of Population and Housing has the most complete coverage of the population. For the first time, in 1990, the Census question on educational attainment began to distinguish between completion of a two-year associate's degree and completion of some college but no degree. Previously, Census data did not distinguish between these two levels of schooling. Furthermore, holders of two-year degrees are asked if they are vocational or academic. Analysis of 1990 Census data will therefore permit the most detailed analysis of earnings and other payoffs to a two-year degree within various demographic groups. However, the decennial Census has two major drawbacks: it occurs only every ten years, and it is not longitudinal.

Two other surveys, both also conducted by the Census Bureau, partially compensate for these shortcomings of the decennial Census. The Current Population Survey (CPS) draws a random sample of civilian, non-institutional households each month. About 60,000 households are currently targeted for the monthly survey, but about 2,600 of these are not available for interviews (U.S. Bureau of the Census, 1992, p. D-1). The March CPS survey includes detailed questions about income, which permit comparisons of earnings by educational attainment. The Survey of Income and Program Participation (SIPP) is another sample survey of households, focusing on labor force status, earnings, and participation in government transfer programs. The SIPP and CPS samples are structured as short-term panels, with data collected on each individual for about two years. SIPP panels were started in 1984, 1986, and 1987. The 1987 panel included 33,100 eligible individuals of whom 24,400 were interviewed for the life of the panel (U.S. Bureau of the Census, 1992, p. D-2). Because they include more detailed questions on education, training, and employment, the CPS and SIPP permit more thorough cross-sectional studies. The CPS is also useful for measuring aggregate year-to-year trends. However, none of these surveys provides long-term longitudinal data.

The first survey that provides data for estimating the long-term consequences of nonbaccalaureate education, including specifically participation in two-year colleges, is the National Longitudinal Study of the High School Class of 1972 (NLS72), sponsored by the U.S. Department of Education. The original sample consisted of 22,652 randomly selected seniors within a stratified probability sample of U.S. high schools. Following the baseline survey in spring 1972, the sample has been resurveyed by mail in 1973, 1974, 1976, 1979, and 1986. The NLS72 have been used to analyze economic returns to nonbaccalaureate education and training by Grubb (1991, 1992), Hollenbeck (1992), and Kane and Rouse (1992).

Another longitudinal survey sponsored by the U.S. Department of Education is High School and Beyond (HSB). Designed to study both progress through high school and subsequent behavior, HSB started in spring 1980 with a sample of 30,030 high school sophomores and 28,240 seniors, randomly selected within a stratified probability sample of U.S. high schools. Follow-ups have occurred in 1982, 1984, and 1986. Economic returns to nonbaccalaureate education and training in the HSB data have been studied by Hollenbeck (1992); Horn (1989); and Lyke, Gabe, and Aleman (1991).

The National Longitudinal Survey of Youth Labor Market Experience (NLSY) is the latest of five cohorts comprising the National Longitudinal Surveys. The earlier cohorts included older and younger men and women. The surveys have all been conducted by the Center for Human Resources Research at Ohio State University under contract with the U.S. Department of Labor. NLSY began in 1979 with 14- to 21-year-olds and has followed them through 1988. The base year national probability sample included 11,406 civilian respondents and 1,280 youth in the military. Information on labor market experience and further education have been collected annually. Transcript data for about 77% of the civilian respondents have also been collected. A separate file contains week-by-week work histories. Monk-Turner (1986) analyzed returns to two-year college attendance among the NLS young men and young women. The NLSY cohort has now matured to the point where it would be worth analyzing the effects of nonbaccalaureate, postsecondary education and training; several investigators are now working on this, but no results have yet been published or presented at this time.

The Panel Study of Income Dynamics (PSID) is the longest-lived continuous survey of a nationally representative sample. Maintained by the Institute for Social Research at the University of Michigan, it was originally called the Survey of Economic Opportunity when it began in 1966 to assess the impact of anti-poverty programs. Subsequently, it was expanded into a longitudinal study of economic well-being. The 1968 study was based on a nationally representative sample of 4,802 households in forty states. The study has followed the original households as well as the "split-offs" who left home to establish new households. Detailed employment and educational histories have been collected for new household heads and wives over the years and for all heads and wives in 1976 and 1985. There are both household-level and individual-level longitudinal files available for public use.

There is continued interest in using individual earning records that are collected for administrative purposes to analyze the effects of participation in educational programs that cannot be evaluated experimentally. Stevens (1991) has described the possible use of unemployment insurance (UI) earnings data for this purpose. Since 1988, employers in all states have been required to make quarterly wage reports on individuals covered by UI. Social Security numbers can be used to link individuals' reported earnings to their school records in states where the school systems keep Social Security numbers in students' files. At least ten states have used UI data to evaluate vocational education programs. One major drawback of the UI data is that it excludes some workers: self-employed and federal employees, among others. Another major problem is that the records are kept by each state's employment security agency, so an individual who goes to school in one state then goes to work in another state would not have a complete school-and-work record in either state. This latter problem could conceivably be rectified by assembling state files at the national level. Since UI records already exist, it is worth the effort to try to solve these and other problems or to try to work around them. Stevens (1992) reports some early findings that illustrate what might be learned from UI data.

The National Center on the Educational Quality of the Workforce (1992) has produced a crosswalk that compares questions on education and training for the CPS, SIPP, NLS72, NLSY, and HSB surveys. In addition to showing the specific wording of each question, the crosswalk also provides information about skip patterns, demographic questions, and sample characteristics.

Results for Two-Year Colleges

Cross-sectional differences in earnings between graduates of two-year colleges and graduates of high school only are available for all age groups from the Survey of Income and Program Participation (SIPP). Average monthly earnings over a four-month period were reported in spring 1987 (Kominski, 1990, Table 2). Multiplying these by 12 gives the following estimated increments in annual earnings for holders of associate's degrees compared to high school graduates:

More recent estimates are available from the Current Population Survey (CPS). In 1992, the CPS began using the new Census question on educational attainment. The March 1992 survey (U.S. Bureau of the Census, 1992) found the following differences in mean 1991 earnings between individuals with associate's degrees and individuals with only high school diplomas (or equivalent):

The first thing to note about these differences is that they are substantial, ranging from $3,589 among 25-34 year old African-American women to $9,590 for 35-44 year old African-American men. The extra earnings for men with an associate's degree first increase then decrease with age. This suggests that the degree opens additional opportunities for advancement through mid-career. The later decline may reflect either obsolescence of knowledge represented by the degree or perhaps a vintage effect if men who were 55-64 years old in 1992 obtained their associate's degrees at an earlier time and if the quality of degree programs used to be lower. However, the earnings advantage for 55-64 year-old women with associate's degrees is greater than for any other age group. Explaining these divergent patterns will be an interesting challenge for future research. Part of the explanation may have to do with differences between vocational and academic associate's degrees: the CPS question distinguishes between the two; but the published tables, from which the numbers above were taken, do not.

Of course these differences in earnings may reflect prior differences between associate's degree recipients and high school graduates rather than the effect of the associate's degree itself. For instance, individuals who obtain associate's degrees tend to come from more educated or affluent families than those who complete high school only (Lyke et al., 1991, Table B.3). They also possess more of certain abilities, measured by grades and test scores (Tables B.4 and B.5), which lead to both higher educational attainment and higher earnings. Also, there are differences in gender, ethnic composition, and amount of work experience. To obtain a more precise measure of the economic payoff from an associate's degree, it is necessary to control statistically for these variables.

Such statistical control is possible with the detailed data from longitudinal datasets such as the High School and Beyond (HSB) survey and the National Longitudinal Study of the High School Class of 1972 (NLS72). The HSB cohort of seniors in 1980 were re-surveyed in 1986, when most of them were around age 24. Most of those who went to two-year colleges would have finished or dropped out by then. The 1986 HSB senior follow-up therefore provides good information for assessing the short-term payoff from two-year college attendance. The NLS72 cohort was also re-surveyed in 1986, when most of them were around age 32. This provides good information on respondents a little further on in their careers.

Results from several studies of these and other longitudinal datasets are summarized in Tables 22.1-22.6, which report the incremental yearly earnings associated with two-year college attendance. The differences shown are either in reported earnings for a given year or in hourly wages multiplied by 2000 to convert them into an annual estimate. All the differences shown in Tables 22.1-22.6 are estimated from regression equations that include controls for demographic variables, socioeconomic background, and measured ability. Where indicated, the regressions also control for amount of work experience.

The HSB data was analyzed by Horn (1989), Hollenbeck (1992), and Lyke et al. (1991). Horn's results imply a yearly earnings advantage of roughly $3,000 for those who complete a two-year vocational program, compared to those who complete high school only. Horn's calculation assumes each year's coursework includes 30 credits that are in the same field as respondents' subsequent employment. Hollenbeck's estimate for a vocational completer who attended for 24 months is about $1,800. On the other hand, Lyke et al. find an insignificant difference between community college and high school graduates, where the type of associate's degree (vocational versus academic) is not taken into account. Differences in specification among the three analyses make it difficult to see exactly why these findings vary, but it appears that the early payoff from a vocational associate's degree is greater than from an academic degree.

The NLS72 data have been studied by Grubb (1992, 1993a), Hollenbeck (1992), and by Kane and Rouse (1992). Grubb found a vocational associate's degree is worth in the vicinity of $3,000 a year for females but does not contribute much to additional earnings for males. These estimates are in the same ballpark as Hollenbeck's finding of a $2,000 advantage for those who attended community college and received vocational certificates or degrees; Hollenbeck did not report separate analyses for males and females. Kane and Rouse found additional earnings of almost $5,000 for females who obtained vocational associate's degrees. The increment for males with vocational associate's degrees was only one-tenth as big.

Medrich and Vergun (1994) found that women, but not men, who were employed full-time received an additional boost in annual pay if they were working in a job that was related to their vocational field of study. This analysis looked at earnings from 1990 to 1992 for 18-34 year-old SIPP respondents who attained a postsecondary degree in a vocational field no later than Fall 1990. The result for females is similar to results for high school vocational graduates described earlier.

In the NLS72 data, Grubb (1992, 1993a) and Kane and Rouse (1992) also estimated the financial return to associate's degrees in nonvocational fields. They found little or no gain in earnings for men or women, except for those with associate's degrees in math or science.

As expected, all of these estimates of earnings differences for two-year college graduates, controlling for ability and demographic characteristics, are less than the simple, uncontrolled differences observed in the SIPP and CPS data.

Since most students who enter two-year colleges do not complete associate's degrees, it is important to estimate the payoff from taking courses and not finishing the degree. As shown in Tables 22.1, 21.2, and 22.4, estimates vary. Grubb (1993a) finds little effect of credits for noncompleters. Hollenbeck finds little effect on wages but a substantial effect on hourly earnings in the NLS72 data. He also finds large payoffs for noncompleters in the HSB data. Kane and Rouse find a positive wage payoff for noncompleters in the NLS72 data. It is hard to know what to make of these divergent results. Grubb warns that the variety of programs and local circumstances makes it hazardous to generalize about effects.

Results for Proprietary Vocational Schools

SIPP data shows the following differences between individuals who hold a postsecondary vocational certificate (including proprietary schools) and those who graduated from high school only (Kominski, 1990, Table 2):

These differences increase substantially with age. This suggests some kind of complementarity between proprietary school training and work experience. It may be that individuals with vocational certificates are able to obtain more on-the-job training or to make better use of it. It may also be that individuals tend to enroll in proprietary vocational schools after they have had enough successful work experience to see how they can benefit from this training. Another conceivable explanation is that older workers attended proprietary schools in an earlier time when these schools may have offered programs of higher quality than in more recent years.

Tables 23.1-23.4 show results of studies using the HSB and NLS72 data to control for individual ability and demographic characteristics in estimating the payoff to proprietary school training. In the short-term data from HSB, Lyke et al. (1991) find proprietary school graduates make about $1,000 a year more than high school graduates. This is more than the earnings advantage they found for individuals with associate's degrees (vocational and academic together). Hollenbeck (1992) estimated a payoff of approximately $2,000 a year from a year in proprietary school--again, somewhat more than his own estimate of the payoff from a community college vocational certificate or degree.

In the NLS72 data, with a somewhat older group, the estimates by Grubb (1991) and by Kane and Rouse (1992) show an earnings gain of $1,000 to $2,000 for proprietary school graduates. For noncompleters, Grubb finds no significant payoff. This is a perceptibly smaller payoff than these investigators found for community college vocational graduates. Comparing the NLS72 and HSB estimates does not replicate the rising payoff among older individuals as found in the SIPP data.

Summary

The results for two-year colleges and proprietary schools can be briefly summarized as follows: Attendance at a two-year college is associated with substantially higher earnings for females who complete associate's degrees, especially in vocational fields. There is not much evidence of a financial return to associate's degrees for males, except possibly in math and science. Evidence on the returns for noncompleters is mixed. CPS data shows the 1991 earnings advantage of associate's degree holders over high school graduates with no postsecondary education was in the range of $1,000 to $9,000 when no adjustment is made for differences in family background or prior achievement in school. The unadjusted difference is highest in the 35-44 age group for men, 55-64 for women. When prior differences are statistically controlled using longitudinal data from the NLS72 and HSB surveys, most studies find a 1985 earnings advantage in the range of $2,000 to $4,000 at age 24 to 32 for associate's degree holders compared to high school graduates. Studies sometimes show bigger returns from a vocational than from an academic associate's degree, but sometimes the reverse.

Attendance at a proprietary vocational or technical school is also associated with gains in earnings. Unadjusted for prior differences, the annual earnings advantage for proprietary school attendees over high school graduates with no postsecondary education is in the range of $1,000 to $6,000, according to 1987 SIPP data, which also shows the greatest difference at age 55-64. When differences in family background and prior school achievement are statistically controlled, the estimated 1985 earnings advantage of proprietary school graduates over high school graduates is $1,000 to $2,000 at age 24 to 32. Studies using HSB data have found a slightly bigger advantage for proprietary graduates than for two-year degree holders around age 24, but studies using NLS72 data have found the opposite result around age 32.

Bigger earning gains do not necessarily imply that one program is more socially profitable than another. Different education and training programs serve different clienteles. They also have different costs, and different divisions of cost between public and private. Definitive judgment of the relative efficiency of different programs would require, first, that the programs be applied to similar populations or, if this is not possible, that differences among the participants in different programs be statistically controlled. Second, costs should be accurately measured. Unfortunately, not enough attention has been given to measuring costs, in particular the opportunity cost of foregone earnings for participants in education or training programs. Further analysis of existing data from longitudinal studies could yield valuable information about this.

Existing longitudinal studies should also keep following their samples as they age. One reason is to resolve the puzzling age pattern of earnings differences associated with two-year colleges and proprietary schools that appears in cross-sectional data. Cross-sectional differences are biggest after age 35 for male graduates of two-year college and after age 55 for proprietary school attendees and female graduates of two-year colleges. But existing longitudinal datasets that include the period when most individuals attended these schools have so far followed them only to age 32 (NLS72), age 24 (HSB), or age 30 (NLSY).

Stern, D., Finkelstein, N., Stone, J. R., III, Latting, J., & Dornsife, C. (1994). Research on School-to-Work Programs in the United States (MDS-771). Berkeley: National Center for Research in Vocational Education, University of California.