Archived Information

Direct Evidence on the Impact of Academic Courses

With these considerations in mind, we are prepared to assess the evidence. Seven data sets have been used in a dozen separate studies to provide information that is directly relevant to the question at hand. The data sets are:

All seven data sets are nationally representative of their target populations. Most followed students for no more than six years after high school; however, the ETS data and the NLS-72 followed students for considerably longer. The twelve studies in which the data sets were analyzed differed in their model specifications, even among studies that used the same data set. The most important differences include whether they used controls for cognitive ability, and if so, from what point in time; whether educational attainment was taken into account, and if so, through a sampling restriction, or by statistical controls in the regression model; whether they used a method other than OLS regression to address selection issues; and whether the main predictor variable reflects academic courses or an overall academic program. The twelve studies examined many types of labor market outcomes, but only four outcomes were considered in more than one study, and I will concentrate on them. The four outcomes are: (1) hourly wages; (2) weeks or months employed in a given time frame; (3) annual earnings; and (4) attainment of occupational status.

Effects on Hourly Wages

Eight studies examined effects of academic coursework on hourly wages. Their findings are summarized in table 1.

Studies with controls for test scores prior to coursework. The studies that best match our theoretical specification are those that control for cognitive skills prior to high school (to take into account differential propensities to enroll in academic courses), but not after high school (to avoid masking the effects of course work that occur through gains in skills). Four studies followed this design, and these indicate a small, inconsistent relation between coursework and wages for men, and a modest tendency towards positive effects for women, with some contrary findings. None of the estimates for men were statistically significant in the original studies, and as many were negative as were positive. Hotchkiss and Dorsten (1987) did not distinguish between effects for men and effects for women, and they found effects close to zero.

Both Grasso and Shea (1979), using NLS-LME, and Daymont and Rumberger (1982), using NLS-Y from a decade later, found that women who engaged in more academic study obtained higher wages in the first few years after high school. Grasso and Shea reported a substantial gain of nearly 10 percent for women who had enrolled in a college-preparatory program, compared to those in the general track. Daymont and Rumberger observed a gain of just over 1 percent for each academic credit. At first glance, this appears a much smaller impact than that observed by Grasso and Shea, since college-preparatory students take only two to three more academic courses during their four years of high school than general-track students. However, the effect of college-track membership reflects differences in levels as well as varied numbers of courses. Hence, these two studies suggest that an academic program of study is beneficial for the wages of young women who do not go on to college.

However, Gardner (1984), using the same data set as Daymont and Rumberger (1982) and following respondents for two years longer, obtained non-significant negative coefficients for white and minority females. This result is especially surprising since Gardner's estimates reflect participation in an academic program of study consisting of at least four years of English, three years of math, and two of science and social studies. Still, a number of other design differences could account for the varying results. First, while Daymont and Rumberger used ninth grade GPA to control for cognitive skills, Gardner used an ability test. On the one hand, the test is undoubtedly a more reliable control, and Daymont and Rumberger's finding could thus be overstated. On the other hand, Daymont and Rumberger explained that the reason they did not use this variable is that it was missing for about half the cases. Moreover, survey respondents who were out of high school at the time of the survey (half the sample or more) would have taken the test after completing their high school course work, and this would tend to obscure the impact of academic courses if the impact occurs by adding to cognitive skills. Finally, Gardner included only respondents who were employed full time; this could have eliminated an important source of variation in wages. These problems raise doubt about Gardner's findings, and suggest we should not use them to overturn Grasso and Shea's and Daymont and Rumberger's conclusions.

Other studies. Altonji (1992), analyzing NLS-72, recognized the problem of controlling for ability at the end of high school, but the survey began with high school seniors and did not obtain earlier test scores. Consequently, Altonji reported results with and without the test score controls. Without these controls, Altonji found coefficients of .018, .003, -.001, -.002, and .012, for math, science, English, social studies, and foreign language courses, respectively. These average to .006, or an increase of 0.6 percent for each additional course (see Jencks et al., 1979, p. 27-28, for a clear exposition of transforming logarithmic coefficients to percentages).

To adjust for selection bias in this model, Altonji used between-school variation in course taking as an instrumental variable to identify the effects of individual students' course work. This approach assumes that some schools induce students to take more academic courses, increasing students' enrollment propensities apart from individual background and test score differences. Using this method, Altonji found even smaller effects of course work on wages, although the effect for foreign language remained positive and statistically significant. Noting the negative coefficients for English and social studies, Altonji estimated the impact of taking an additional year each of math, science, and foreign language. This analysis indicated benefits of over 3 percent using OLS, but less than 2 percent using the instrumental variables approach. In another model, Altonji included a control for curricular track; this analysis yielded similar estimates for course work effects, and a non-significant impact of 2.6 percent for college-track membership.

Overall, Altonji's work suggests the impact of academic courses on wages is small, but (most clearly in the case of foreign language) probably non-zero. This finding covers the period from 5 to 14 years after high school, and it is averaged across persons with different years of schooling, controlling for educational attainment. In results not displayed in tables, Altonji noted that the positive effects of foreign language courses were stronger for those who did not attend college than for those who did. Unfortunately, Altonji did not distinguish between effects for men and effects for women. Altonji's study, like others that examine courses or credits, lacks information on the type or level of courses within subject areas. A likely reason for stronger effects of math, science, and especially foreign language is that they represent an academic focus involving higher level courses; as noted earlier, students vary little in the number of courses they take in English and social studies, even though they often differ in course levels. Altonji attempted to take course levels into account with additional, separate analyses of students in academic and non-academic tracks, with similarly weak results. This sampling partition confounds differences among students with differences among courses—there is much overlap in self-reported track membership among students in different courses—so it is not surprising that the results did not change much.

Of analyses that controlled for ability using tests administered at the end of high school, most observed negative coefficients for the link between academic courses and wages. Both studies by Kang and Bishop (1986, 1989) found negative effects on wages two years after high school. Hollenbeck found a substantial negative effect for men 14 years after high school but an even larger positive effect for women, comparing those enrolled in the college track with those in the general track during high school.

It is difficult to interpret the coefficients for academic courses and programs in these analyses. If academic course work contributes to wages, it must do so by affecting cognitive skills; at least, no other mechanism has been suggested. Hence, one would expect zero effects in these models. Negative effects could come from selection patterns described earlier; for example, students who took many academic courses, but who did not go to college may be those who benefit least from such courses. Another consideration for the Kang and Bishop studies is that outcomes were measured only two years after high school; perhaps not enough time had passed for the benefits of courses to be reflected in wages.

Time-varying effects, non-linearities, and tradeoffs. Two studies attempted to model changes in effects for individuals over time. Grasso and Shea (1979) observed that for men, the benefits of a college-preparatory curriculum increased with work experience among those who did not attend college. However, a separate analysis of growth in wages indicated no significant curriculum effects, so the authors interpreted the first finding as reflecting a period effect; that is, the college track may have had increasing benefits across the cohorts included in the NLS-LME, who ranged from 14 to 24 years of age in 1966. Analyzing a single cohort, Kang (1984) also found no significant differences in the growth of hourly wages among persons with varied academic training in high school, during the first 21 months after high school completion. However, Altonji (1992) reported steeper growth in wages for students who completed more course work in science and foreign language; results for math, English, and social studies were small and statistically insignificant.

Kang and Bishop (1989) allowed for a non-linear effect of course work by including a quadratic term in their regression equations. They found no accelerating or decelerating effects on hourly wages. However, they observed a positive interaction of academic and vocational courses for males, which they interpreted as an indication that academic courses enhance the effects of vocational courses. These models are provocative, and would be worth replicating in analyses that control for test scores prior to course work.

Summary of effects on wages. The best available evidence suggests that academic courses in high school provide small benefits to wages in the first few years after high school. This conclusion holds more securely for women than men, because separate analyses yielded non-significant and sometimes negative results for men. These effects may have increased during the 1960s and early 1970s, and they may increase for an individual over time, although the evidence supporting that prediction is inconsistent. Analyses of separate academic subjects show the strongest effects for foreign language; assuming this is not merely selection bias, it may mean that such courses promote communication skills and general cognitive abilities that pay off at work (Altonji, 1992), or it may reflect enrollment in high-level academic courses, with which foreign-language enrollment is correlated (Rosenbaum, 1976; Alexander and Cook, 1982).

Effects on Employment

Three studies with controls for prior achievement examined the impact of exposure to academic curricula on employment shortly after high school. Daymont and Rumberger (1982) found that the more academic credits students had accumulated, the less likely they were to be unemployed after high school. (To simplify presentation, I have reversed the sign of the effect on unemployment for display in table 2, which lists effects on weeks employed.) Accumulating vocational credits also aided employment opportunities in Daymont and Rumberger's analyses, so the main contrast is with studies in art, music, physical education, etc., and more importantly, with obtaining fewer credits overall during high school.

Gustman and Steinmeier (1982) obtained a positive coefficient on weeks employed for males, but a negative coefficient for females, and neither was statistically significant. Sample sizes in these analyses of NLS-LME were very small—only 83 males and 52 females—which could account for the imprecision of the estimates. Hotchkiss and Dorsten's (1987) observation of essentially no relation between a concentration in academic course work and employment undoubtedly reflects their sample, which included students currently enrolled in post-secondary education, the destination for most students who focused on academic studies in high school. Hotchkiss and Dorsten also reported that more concentration in academic courses reduced the likelihood of unemployment shortly after high school. Again, however, the contrast was mainly with attending postsecondary schooling rather than with participation in the labor force.

Gustman and Steinmeier had much larger samples in their analyses of NLS-72, but unfortunately these data lack information on cognitive skills prior to high school, and by including subsequent test scores, effects of academic courses may have been obscured. The same must be said of Meyer's (1982) otherwise useful model which examines effects on weeks worked during the first, fourth, and seventh calendar years after high school. Kang and Bishop's (1986, 1989) analyses of HSB Seniors also adjust for skills at the end of high school, and although they find no sign of direct benefits of academic course work, I will argue subsequently (in considering indirect effects of course work) that their results provide evidence that academic studies contribute positively to employment by improving cognitive skills.

In sum, only one study employed a model that meets the theoretical requirements outlined earlier, with a sample of adequate size (Daymont and Rumberger, 1982). This study suggests that students who take more academic courses are less likely to be unemployed after high school.

Effects on Annual Earnings

Analyses of annual earnings yield the most inconsistent and intractable results among the four outcomes we are considering. Four studies considered the impact of an academic program on annual earnings using controls for test scores prior to course enrollment (see table 3). Three obtained positive coefficients of non-trivial sizes for men (Grasso and Shea, 1979; Gustman and Steinmeier, 1982; and Gardner, 1984). However, using an older and earlier sample, Griffen and Alexander (1978) found that earnings for college-track males were substantially lower than those of other students, among those who did not attend college. The authors were puzzled by this finding and, in a footnote, pointed out its consistency with Bowles and Gintis' (1976) view that attitudes learned in the college track are not well suited to jobs of non-college men. The result may also be an artifact of selection, in that college-track students who did not go on to college may have had characteristics that made them poorly suited for work as well as further schooling; this may have been more true in the past when curricular tracks defined students' programs more rigidly than at present (Moore and Davenport, 1988).

In light of problems noted above with analyses by Gustman and Steinmeier (1982) and Gardner (1984), one may have the most confidence in Grasso and Shea's (1979) results. Although they reported positive coefficients, the results were not reliably different from zero for males or females. Without offering specific coefficients, Olneck (1979) also reported that two national surveys from the 1960s indicated that completing a college-preparatory curriculum in high school had no impact on earnings for men who attained the same levels of schooling.

Studies that controlled for test scores after high school show the familiar pattern of zero or negative effects, with the notable exception of Hollenbeck (1993). In his analyses of NLS-72, academic-track males and especially females earned substantially more 14 years after high school, compared with students of similar ability from the general track.

Based on research summarized above suggesting small positive effects on both wages and on employment, one would have expected to see benefits for annual earnings. Yet the evidence does not support this expectation.

Effects on Occupational Status

In contrast to effects on annual earnings, effects on occupational status are more consistent and yield to straightforward interpretation. All three studies listed in Table 4 show some sign of positive effects of participation in academic study on occupational status, and results in two of the three were statistically significant. Both Griffin and Alexander (1978) and Grasso and Shea (1979) conducted separate analyses of high school graduates who did not attend college (among other subsamples). Griffin and Alexander observed a non-significant negative coefficient for college-track membership, but a significant positive effect for enrollment in math and science courses. Each additional course contributed about eight-tenths of a point on a scale of approximately 100 points. Students who took three years of math and science would end up about two and a half points higher than students who took only one year of each [(6 - 2) x .811 = 2.444]. This is a small impact, considering that 13 years had passed since high school graduation. Grasso and Shea found male academic-track students about the same amount ahead of their non-academic peers, but the difference was not statistically significant, and the effect for females was close to zero.

Hotchkiss and Dorsten (1987) did not restrict their sample to persons who stopped their schooling after high school. Hence, their finding of an advantage of a third of a point for students who concentrated on academic courses in high school could reflect differences among students in the propensity to acquire further schooling. In a second analysis, however, the authors introduced an adjustment for the propensity to enter the labor force. This correction takes into account differential likelihood of continuing in school, and it produced a similar result for the impact of academic studies on occupational status.

These studies suggest that students who take more academic courses in high school find jobs with higher status soon after high school.  Although this effect is not reflected in wages or earning at entry level, it may yield greater benefits over time.

[Why Should We Expect a Linkage?]

[Indirect Evidence: Effects of Academic Skills]