## Abstract

Vocational education is a large part of the high school curriculum, yet we have little understanding of what drives vocational enrollment or whether these courses help or harm early careers. To address this deficiency, we develop a framework for curriculum choice, taking into account ability and preferences for academic and vocational work. We test model predictions using detailed transcript and earnings information from the National Longitudinal Survey of Youth (1997). Our results are twofold. First, students positively sort into vocational courses, suggesting that the belief that low-ability students are funneled into vocational coursework is unlikely true. Second, we find higher earnings among students taking more upper-level vocational courses—a nearly 2 percent wage premium for each additional year, yet we find no gain from introductory vocational courses. These results suggest: (1) policies limiting students' ability to take vocational courses may not be welfare-enhancing, and (2) the benefits of vocational coursework accrue to those who focus on depth over breadth.

## 1.  Introduction

Since the publication of A Nation at Risk in 1983, policy makers and politicians have turned attention to a perceived educational decline of American youth. Stagnant high school graduation rates, declining test scores, and signs that many college entrants are ill prepared for college and the workforce have all contributed to this perception. Many states have responded to this alarm by increasing high school graduation requirements, typically specifying a minimum number of years spent studying academic subjects such as English, mathematics, science, and social studies.

Figure 1.

Course-Taking by American High School Students

Figure 1.

Course-Taking by American High School Students

In our analysis we draw particular attention to factors that predict vocational course-taking (choices) and the ensuing impacts (outcomes) on college-going and wages. While the few existing studies have found largely positive impacts of vocational course-taking on earnings (Mane 1999; Bishop and Mane 2004, 2005; Meer 2007), and mixed evidence on high school graduation and college attendance (Maxwell and Rubin 2002; Agodini and Deke 2004; Neumark and Rothstein 2005, 2006; Cellini 2006), these and others frame vocational course-taking as a “track” rather than as a marginal curriculum choice as is traditionally done for other subjects (for example, mathematics). Our analysis thus departs from previous work in several regards. First, we develop a framework for curriculum choice in which students make curriculum and college decisions in response to information about ability and preferences for vocational or academic (course)work. Second, we treat vocational course-taking as a marginal decision, rather than a choice of “track,” which our empirical evidence suggests is appropriate. Third, we observe students in the labor market to evaluate wage gains from additional vocational courses, allowing for differential returns to general versus specialized vocational coursework. Lastly, we observe how returns vary across vocational fields and by local labor market characteristics.

We use detailed longitudinal transcript and labor market information for respondents in the National Longitudinal Survey of Youth 1997 (NLSY97) to examine these questions, exploiting a rich set of background and ability measures available to us. We ultimately find that more vocational courses are associated with higher wages, on the order of 1.8 to 2.0 percent for each year of specialized vocational coursework, and lower incidence of “idleness” (neither working nor in school) after high school. But, these gains are not uniformly distributed across all vocational course-takers. Separating vocational coursework into “higher” and “lower” levels, corresponding to introductory courses and specialized coursework within a vocational discipline, we find that wage gains are driven entirely by upper-level courses, largely in technical fields, and among non-college graduates. We estimate no wage gain to an additional introductory level vocational course. In addition, we find little evidence that vocational coursework decreases the likelihood of college graduation, though we find suggestive evidence that it may deter the marginal college-goer, implying that those pulled out of college might have been the least likely to graduate. Analogously, we also find that although the labor market value of nonvocational coursework is entirely explained by increased college-going and graduation, the value of vocational coursework is unaffected by accounting for college enrollment and completion.

Our results corroborate a model of positive selection into high school course of study. Supplementary exercises exploiting variation in graduation requirements to instrument for curriculum choice supports this, showing that gains accrue to students who select into vocational coursework but that students induced into additional courses see no advantage. The policy implications from these findings are straightforward and twofold. First, students appear to positively sort into vocational courses, suggesting that the commonly held belief that low-ability or low-effort students are funneled into vocational coursework is unlikely to be true (a negative selection story). Thus, policies that limit students’ ability to take these courses—for example, increasing requirements in other disciplines—may not be welfare-enhancing. Second, the benefits of vocational coursework accrue to students who specialize, rather than those who take multiple vocational courses in several areas. Hence, CTE programs should allow for depth in any topic offered. Recent trends toward more specialized CTE concentrations or “pathways” (away from general, non-specialized coursework) may, therefore, be smart policy.

## 2.  Background and Framing

### Prior Work

Compared with an expansive literature on the returns to additional years of schooling, there is relatively little evidence on the labor market consequences of high school curriculum in the United States, and even less work on effects of vocational course-taking in particular. The value of vocational education has also been subject to debate in both developing and transitional economies, with prior research finding minimal to positive effects on education and labor market outcomes (Malamud and Pop-Eleches 2010; Attanasio, Kugler, and Meghir 2011; Eichhorst et al. 2015). Those studies that evaluate the impact of vocational coursework largely treat vocational curriculum as a binary choice—students are either on the academic or vocational track, as opposed to evaluating the impact of an additional course as one might in, say, mathematics or English. For example, Meer (2007) examines the relationship between “track” (academic, general, business vocational, technical vocational) and earnings later in life. He finds evidence of comparative advantage: Students currently on a technical vocational track would not gain from switching to an academic track, nor would those currently on the academic track benefit from more vocational education.

This is not the only study to find benefits to vocational coursework. Pooling data across several cohorts, Mane (1999) finds positive labor market payoffs to vocational coursework and demonstrates that these were increasing through the 1980s, particularly for noncollege bound students. In newer research, Dougherty (2018) finds large high school graduation effects from students who are just accepted into career academies in Massachusetts compared with those who just miss the cutoff. Analyses using older cohorts have generally supported the same conclusions (Gustman and Steinmeier 1982; Kang and Bishop 1989; Bishop and Mane 2005). Yet, selection is an issue in evaluating vocational programs and only Dougherty uses exogenous variation, though he does not study earnings as an outcome. Using the same data as us, Zietz and Joshi (2005) point out that participation in the vocational track is much more common among disadvantaged students. Recent studies commissioned by the U.S. Department of Education as part of the national assessment of CTE required by Perkins reauthorization used various quasi-experimental methods applied to longitudinal student-level data from two locales (Philadelphia and San Diego), one state (Florida), and one nationally representative high school cohort (Education Longitudinal Study of 2002). These studies found mixed evidence of the effects of CTE on secondary, postsecondary, and labor market outcomes (Furstenberg and Neumark 2005; USDOE 2014). In more recent work, Hanushek et al. (2016) find initial benefits of vocational training, yet they find these come at the cost of adaptability to changing labor market conditions in later life. Taken as a whole, a few patterns emerge. There is more evidence than not of short-term labor market gains from vocational coursework, and maybe long-term losses (Hanushek et al. 2016), in addition to new evidence that the offer of vocational coursework might benefit students in terms of high school graduation (Dougherty 2018), and that these courses are more likely to be populated by less-advantaged students.

A related body of work explores the effect of participation in School-to-Work (STW) programs on education and labor market outcomes using early waves of the NLSY97. This literature has evolved mostly in parallel to the above-mentioned literature on high school curriculum and vocational education, which is surprising because STW programs can be seen as a close substitute, in purpose if not structure, to traditional vocational coursework. The results, again, are mixed. Neumark and Joyce (2001) find few effects of STW participation on behaviors associated with later college attendance but find positive effects on students’ beliefs that they will complete high school and their beliefs about future labor market participation. Neumark and Rothstein (2005, 2006) find some positive benefits on college attendance and the likelihood of later employment, although these accrue largely for boys and in programs with direct links to the labor market (internships and cooperative education); they in fact find some negative effects from tech-prep.2 Cellini (2006) uses a sibling fixed effects model and focuses only on tech-prep, finding that participants are more likely to complete high school and attend a two-year college, resulting in higher overall attainment, but that this comes at the cost of a decline in four-year college attendance.

A separate literature has evolved that estimates the consequences of high school curriculum choice, much of which is summarized in Altonji, Blom, and Meghir (2012). In the United States, much of this work focuses specifically on mathematics course-taking (Altonji 1995; Levine and Zimmerman 1995; Rose and Betts 2004; Joensen and Nielsen 2009; Goodman 2019), generally finding that more math has a positive association with earnings. Importantly, unlike the literature on returns to taking a particular track, this literature treats each course as a decision on a continuum, allowing for marginal as opposed to discrete changes in curriculum choice. We build on this work by explicitly examining both high school curriculum choice and outcomes, with a focus on the number of vocational courses students choose to take.

Blending these two literatures, we treat curriculum as a continuum, a decision informed by figure 2, which depicts the cumulative distribution of the total number of vocational credits taken by respondents in the NLSY97. Figure 2 suggests (1) vocational courses account for a nontrivial share of high school coursework—roughly 50 percent of students take four years’ worth of vocational coursework; (2) an overwhelming majority of students take at least one vocational course in high school; and (3) the distribution is smooth, rather than bimodal, as strong tracking would imply. We frame the decision to take vocational coursework as a result of students learning about their comparative advantages and preferences for vocational or academic (course)work. In response to baseline expectations and received signals about fit, students rationally decide whether to pursue a more vocational or academic focused curriculum as high school progresses. Students consider both human capital and informational consequences of their choices, as curriculum also provides information about labor market prospects. Although several studies of college persistence and major choice have this dynamic flavor (e.g., Arcidiacono 2004; Stinebrickner and Stinebrickner 2011; Stange 2012), this has been mostly absent from studies of high school curricula. Related, Malamud (2011) finds that exposure to a broader curriculum reduces job switches later in life (interpreted as mistakes) by improving match quality at the expense of greater specialization.

Figure 2.

Cumulative Distribution Function of Vocational Course-Taking

Notes: Figure shows cumulative distributions of total vocational courses taken. One credit is equal to one hour per day for a full academic year. N = 4,414.

Figure 2.

Cumulative Distribution Function of Vocational Course-Taking

Notes: Figure shows cumulative distributions of total vocational courses taken. One credit is equal to one hour per day for a full academic year. N = 4,414.

### Theoretical Framework

To frame our empirical analysis, we consider a stylized model (similar to Altonji 1995) of high school curriculum choice, college enrollment, college completion, and labor market entry in which students are forward-looking but face uncertainty about their ability. We assume that students are endowed with ability along two dimensions: academic ability $(αa)$ and vocational ability ($αv).$ These endowed abilities are augmented by investments in human capital through curriculum choices manifested in the number of academic and vocational courses students take—$A$ and $V$, respectively. We define human capital stock $K$ at time $t$ as a function of ability and curriculum choice, including college:
$Kt=ka,tαa,A,kv,tαa,V.$
(1)

We assume that academic ability ($αa$) influences performance in academic subjects in high school and college, and the likelihood of college graduation, whereas vocational ability ($αv$) influences performance in vocational subjects in high school. Allowing for effects of $αj$ on $k∼j,t$ or on performance in opposite-type courses does not change inference, rather, it simply suggests that there are complementarities. Neither component of ability is known at the start of high school, though students have prior beliefs, which depend on fixed characteristics ($X$) we assume are observed by the econometrician. Students learn about $αa$ and $αv$ via performance in academic and vocational subjects, respectively, manifested in grades conditional on covariates—that is, grades relative to what would be expected. For instance, students who perform better in academic courses than their baseline characteristics would predict will revise upward their belief about $αa$. These posterior beliefs then influence students’ subsequent decisions about high school curriculum, graduation, and college enrollment. The likelihood of four-year college graduation, $pr(college)=g(ka,X)$, then depends on academic capital, which is a function of high school curriculum and ability, and on covariates observable to the econometrician. We assume graduation from a two-year institution can depend on $kv$ as well.

We define the labor market as having multiple sectors, delineated by highest degree completed, which define the distribution from which workers draw wages. Both ability components impact productivity in the labor market, though the importance of each may vary by sector; that is, $ka$ may be more valuable in the college-educated labor market whereas $kv$ may be more valuable in the non-four-year college-educated sector. Thus, (log) wage at time $t$ in sector $j$ depends on accumulated skills that are a function of ability and education—$ka,t(αa,A)$ and $kv,t(αv,V)$—plus covariates such as work experience, year, and local labor market conditions.
$wtj=fka,t,kv,t,X$
(2)
for $j={$drop out, high school, two-year college, four-year college}.

During high school, students receive utility that depends on curriculum (the mix of academic and vocational coursework) and expected performance in the chosen curriculum. Utility during college depends on expected performance, covariates, and high school curriculum. Utility in the labor market depends only on log wages here, but might also include a hedonic preference for job type. Students sequentially choose high school curriculum (or dropout) and then whether or not to enroll in two- or four-year college to maximize the present discounted value of lifetime utility. Before each of these decisions, students update their beliefs about own academic and vocational ability.

The model delineates several tradeoffs in the choice of curriculum. First, high school vocational courses could have lower psychic costs than academic ones for students with high $αv$ (low $αa$). Second, they could be productivity-enhancing for students who enter the non-(four-year) college job sector. Third, they could provide students with the opportunity to acquire more information about $αv$, which should enable better decisions about college enrollment. Productivity in the non-college sector is an important factor in college enrollment decisions, yet many students may lack individual-specific information about this parameter. Exposure to vocational curriculum may be one way to acquire it.

However, vocational coursework could have drawbacks. If the vocational track is less rigorous, it may leave students ill-prepared for college and reduce the likelihood of graduating college among those who enroll. Additionally, vocational concentrators may experience reduced productivity in the college job sector if academic coursework is more productivity-enhancing than vocational among college graduates. Similarly, reduced collegiate performance may also mean that vocational students have a greater psychic cost of going to college than those on the academic track. Lastly, students on the vocational track will necessarily have less information about $αa$ than those who pursue the academic track. This may cause some who would benefit from enrolling in college to forego higher education and the subsequent returns.

One implication of the model is that providing the option of studying a vocational curriculum has ambiguous effects on many outcomes overall because the effects are likely to be quite heterogeneous. For students with diffuse priors about their abilities, exposure to vocational coursework could either increase or decrease college enrollment depending on whether estimates of $αa$ and $αv$ are revised upward or downward. Regardless, the additional option should improve welfare even among those who reduce college enrollment rates because it reduces dropout, interpreted as bad matches between students and college.

### Predictions

We use this framework to derive predictions for our empirical analysis. Above we describe a story about selection, akin to a Roy-type model. Those students who expect higher returns to vocational coursework will be more likely to enroll in those courses. Although this may seem obvious, it clarifies an important yet subtle point: Studying curriculum choice does not lend itself well to the randomization framework often used in educational research. Why? Imagine that we randomly assigned some students to vocational coursework and others not and did not allow for defiers—what should we then expect? We should expect all compliers to be worse off if they had good information ex ante. Those who would have chosen a vocational track in the absence of the experiment (the always-takers) will see positive returns whereas those who would have otherwise enrolled in college-prep coursework (the compliers) should see negative returns compared with their counterfactual. In this framework, only those who would have made ex ante poor decisions would be made better off by having someone else (the experimenter) choose differently for them. The problem with this hypothetical experiment is that the econometrician does not know what students would have taken in the absence of random assignment. This would be akin to randomly assigning college majors, which is an awful idea.

This thought experiment helps frame our predictions. These predictions result from two elements of our theoretical framework. First, that the value of college-prep coursework accrues though an increased likelihood of college attendance and graduation. Thus, we should observe that earnings gains from nonvocational coursework should be largely if not entirely explained by college-going. Second, that students have reasonably good information and self-select. Thus, we should observe that experimental variation in vocational course-taking, which we induce through graduation requirements as a supplemental test, should have zero effect on earnings. In our empirical exercises that follow, we find that each of these predictions bears out.

## 3.  Data Sources and Sample

### The NLSY97 and Transcripts

We examine schooling and labor market activity for respondents in the NLSY97 cohort. These data include 8,984 individuals who were ages 12 to 18 years when they were first interviewed in 1997. The survey is representative of all American youth at that time period and respondents have been followed annually with information on educational attainment, labor market experience, and family formation. High school transcripts were collected from respondents’ high schools in two waves by the National Opinion Research Center (NORC), who administers the NLSY. The first was conducted in 1999–2000 for 1,391 respondents who were no longer enrolled in high school at that time. Transcripts for 4,618 additional respondents were collected in 2004. In total, transcript data were collected for 6,009 respondents.

NORC used course catalogs to categorize all transcript course titles to a common scheme according to the Secondary School Taxonomy (SST-R) formulated by the Department of Education. To create uniformity in credit-taking, courses are converted into “Carnegie credits” where one Carnegie credit is defined as a course that meets every day for one period for an entire school year. Similarly, grades for each course were converted by NORC into a common (0–4) metric.3 The SST-R describes three exhaustive categories of Career/Technical Education coursework, which we refer to generally as vocational (including family and consumer sciences education), general labor market preparation, and specific labor market preparation (Bradby and Hudson 2007). Here we omit the very small number of courses under family and consumer sciences in vocational and place them under “other” coursework.

Importantly, this careful course coding allows us to disaggregate courses into “low” and “high” level courses. For vocational courses, lower-level courses are those classified as “1st course” on transcripts. Upper-level courses include courses beyond the introductory level, including: “2nd or later courses,” “Specialty course,” or “Co-op/Work Experience” in the transcripts. This disaggregation allows us to analyze the benefits of breadth—taking many courses in different fields, and depth—specializing in a particular vocational field. We conduct an analogous disaggregation for core courses (English, math, science, and social studies) with the purpose of separating more and less difficult versions of a particular course within a subject. In this case, transcripts identify courses as either “basic,” “regular,” “Advanced and Honors,” “Specialized,” and “AP/IB.” In this case, lower courses are “basic” or “regular,” and higher courses are the remainder.

### Sample Definition

We restrict our sample to exclude those lacking transcripts in all years of high school, those who never completed ninth grade (for whom high school curriculum is not relevant), and respondents with extreme or illogical totals in their transcripts. Specifically, we exclude high school graduates who either (1) had fewer than four years of high school transcripts; (2) had taken fewer than two courses in each of math, English, science, and social studies; or (3) had more than thirty-six or fewer than twenty total credits. For high school dropouts, the limiting conditions are necessarily more relaxed as any small number of credits can be consistent with dropout. Thus, for dropouts we restrict to those who (1) completed at least ninth grade; (2) had taken at least one-half course in math, English, science, and social studies; and (3) had taken more than four total credits ever. The resulting “Analysis sample” used to study curriculum choice and postsecondary outcomes includes 4,414 respondents. Lastly, we restrict wage analysis to a “Wage sample” of 3,708 students whom we determined entered the labor market and had a valid wage record. To determine labor market entry we use monthly enrollment arrays from the NLSY97 and determine enrollment in each semester (fall, spring). We then define labor market entry as the first of four consecutive non-enrolled semesters that are followed by no more than one enrolled semester thereafter.4 In specification checks following our main analyses we show that relaxing these assumptions does not alter our results.

Table 1 compares the Analysis and Wage samples to the full NLSY among a broad set of demographic characteristics. Comparing columns 1 and 2 shows that the Analysis sample is slightly more advantaged than the full NLSY, but on average is representative of the NLSY in whole. Importantly, differences between the two are largely determined by whether NORC could retrieve transcripts, rather than any behavior of the respondent herself. A comparison of columns 2 and 3 shows the Wage and Analysis samples are nearly identical, with few respondents lost between the two. Our empirical exercises focusing on wages pool multiple observations for each individual over time, resulting in 19,029 person-year observations. In our evaluations of idleness, we use the same 3,708 respondents but include all person-year observations, not only those with wages, after age 18 years, resulting in a sample of 44,774. Approximately one-third of our Analysis sample obtain a four-year college degree, one third attend college without earning a degree, 8 percent earn a two-year degree, 19 percent obtain a high school diploma but no postsecondary education, and 3 percent do not graduate from high school.

Table 1.
Sample Comparison
Full NLSYAnalysis SampleWage Sample
Mean(sd)Mean(sd)Mean(sd)
Demographics
Male 0.51 (0.50) 0.48 (0.50) 0.49 (0.50)
Black 0.26 (0.44) 0.22 (0.42) 0.23 (0.42)
Hispanic 0.21 (0.41) 0.17 (0.37) 0.17 (0.38)
Other race 0.04 (0.19) 0.04 (0.20) 0.04 (0.19)
Mother < high schoola 0.39 (0.49) 0.36 (0.48) 0.36 (0.48)
Mother > high schoola 0.14 (0.35) 0.17 (0.38) 0.17 (0.37)
Ever aida 0.44 (0.50) 0.37 (0.48) 0.39 (0.49)
South, age 12a 0.32 (0.47) 0.31 (0.46) 0.32 (0.47)
Poverty ratioa 2.83 (2.70) 3.36 (2.89) 3.23 (2.75)
Bilinguala 0.02 (0.15) 0.03 (0.17) 0.03 (0.17)
Public high school 0.89 (0.32) 0.92 (0.28) 0.92 (0.27)
AFQT (Z) 0.00 (1.00) 0.00 (1.00) −0.04 (1.01)
AFQT missing 0.21 (0.41) 0.16 (0.37) 0.16 (0.37)
Course-taking
Vocational low 2.43 (2.24) 2.34 (1.96) 2.38 (1.95)
Vocational high 0.99 (1.42) 0.95 (1.32) 0.96 (1.34)
Core low 9.75 (2.73) 9.88 (2.65) 9.94 (2.62)
Core high 4.71 (3.19) 4.87 (3.20) 4.73 (3.13)
Electives 7.44 (2.81) 7.50 (2.69) 7.44 (2.70)
Any vocational attainment 0.95 (0.22) 0.95 (0.22) 0.96 (0.19)
<High school 0.13 (0.33) 0.03 (0.17) 0.02 (0.15)
High school 0.26 (0.44) 0.19 (0.39) 0.21 (0.41)
Some college 0.31 (0.46) 0.35 (0.48) 0.34 (0.47)
2-year degree 0.07 (0.25) 0.08 (0.28) 0.08 (0.27)
4-year degree 0.24 (0.43) 0.35 (0.48) 0.34 (0.47)
Observations 8,984 4,414 3,708
Full NLSYAnalysis SampleWage Sample
Mean(sd)Mean(sd)Mean(sd)
Demographics
Male 0.51 (0.50) 0.48 (0.50) 0.49 (0.50)
Black 0.26 (0.44) 0.22 (0.42) 0.23 (0.42)
Hispanic 0.21 (0.41) 0.17 (0.37) 0.17 (0.38)
Other race 0.04 (0.19) 0.04 (0.20) 0.04 (0.19)
Mother < high schoola 0.39 (0.49) 0.36 (0.48) 0.36 (0.48)
Mother > high schoola 0.14 (0.35) 0.17 (0.38) 0.17 (0.37)
Ever aida 0.44 (0.50) 0.37 (0.48) 0.39 (0.49)
South, age 12a 0.32 (0.47) 0.31 (0.46) 0.32 (0.47)
Poverty ratioa 2.83 (2.70) 3.36 (2.89) 3.23 (2.75)
Bilinguala 0.02 (0.15) 0.03 (0.17) 0.03 (0.17)
Public high school 0.89 (0.32) 0.92 (0.28) 0.92 (0.27)
AFQT (Z) 0.00 (1.00) 0.00 (1.00) −0.04 (1.01)
AFQT missing 0.21 (0.41) 0.16 (0.37) 0.16 (0.37)
Course-taking
Vocational low 2.43 (2.24) 2.34 (1.96) 2.38 (1.95)
Vocational high 0.99 (1.42) 0.95 (1.32) 0.96 (1.34)
Core low 9.75 (2.73) 9.88 (2.65) 9.94 (2.62)
Core high 4.71 (3.19) 4.87 (3.20) 4.73 (3.13)
Electives 7.44 (2.81) 7.50 (2.69) 7.44 (2.70)
Any vocational attainment 0.95 (0.22) 0.95 (0.22) 0.96 (0.19)
<High school 0.13 (0.33) 0.03 (0.17) 0.02 (0.15)
High school 0.26 (0.44) 0.19 (0.39) 0.21 (0.41)
Some college 0.31 (0.46) 0.35 (0.48) 0.34 (0.47)
2-year degree 0.07 (0.25) 0.08 (0.28) 0.08 (0.27)
4-year degree 0.24 (0.43) 0.35 (0.48) 0.34 (0.47)
Observations 8,984 4,414 3,708

Notes: Analysis sample restricts to respondents with transcripts in all years they attended high school and who attended at least ninth grade. Course-taking restrictions are: took at least 2 of each core course for high school graduates; took 1 of each core course if tenth grade+ dropout; took between 20 and 36 credits for graduates; took at least 4 credits for dropouts. Wage sample consists of respondents who have a reported wage >$2.00/hour after labor market entry, which is defined as the first interview after four consecutive non-enrolled semesters. Ever aid indicates if respondent's family ever received government aid. Public high school indicates if primary high school was public. Gifted indicates if respondent ever took a “gifted” class. Core credits include English, Math, Science, Social Studies. Electives include Language, Art/Music, Physical Education, and Other. Low/High core indicates upper- or lower-level course within subject-grade. Low vocational indicates entry level course. High vocational indicates second course or higher, or an internship/experiential learning. NLSY: National Longitudinal Survey of Youth; sd: standard deviation; AFQT: Armed Forces Qualification Test. aRespondents with missing values were recoded to 0 and an indicator equal to 1 is included in regression models indicating those with missing values. ### The Distribution of Course-Taking We begin here by highlighting the fact that nearly all students take some vocational coursework—approximately 95 percent, as shown in table 1. We also show that on average students take 3.3 years’ worth (the equivalent of 3.3 year-long, one hour per day courses) of vocational coursework, equal to about 12 percent of their 25.5 total credits. On average, students take just shy of one of these 3.3 vocational credits in an upper-level or work experience course. How does this compare with nonvocational course-taking? Figure 3 shows the distribution of credits taken by field, demonstrating not only considerable variation in vocational courses compared with other core and non-core coursework, but also visually demonstrating that the modal value is greater than zero. The extent of vocational coursework taken by U.S. high school students relative to traditional academic subjects, even among the most academically oriented, has not been previously appreciated. Importantly, we interpret the smoothness of the distribution of vocational course-taking in figures 2 and 3 as confirmatory evidence of our treatment of vocational education as a marginal decision as opposed to a discrete choice of track. Figure 3. Credits Taken in Each Subject Notes: Figures show histograms of credits taken, defined as courses completed, even if it is an F or course that does not accumulate credit hours. One credit is equal to one hour per day for a full academic year. N = 4,414. Figure 3. Credits Taken in Each Subject Notes: Figures show histograms of credits taken, defined as courses completed, even if it is an F or course that does not accumulate credit hours. One credit is equal to one hour per day for a full academic year. N = 4,414. Figure 4 depicts the average number of courses taken by our sample in each vocational category. The most common types of vocational courses taken are transportation and industry, business and management, computer technology and keyboarding, and general labor market preparation.5 Advanced vocational courses are offered and taken in each of these categories, except for general labor market preparation and keyboarding. Given this heterogeneity in high school curriculum, it is natural to investigate the sources and consequences of this variation. Figure 4. Number of Carnegie Credits by Low/High Experience, by Field Notes: Figure shows number of Carnegie credits earned in each type of vocational course. Low courses are entry level. High/experience are upper-level or internship/work-based experience courses. One credit is equal to one hour per day for a full academic year. N = 4,414. CTE: career and technical education; GMLP: General Labor Market Preparation. Figure 4. Number of Carnegie Credits by Low/High Experience, by Field Notes: Figure shows number of Carnegie credits earned in each type of vocational course. Low courses are entry level. High/experience are upper-level or internship/work-based experience courses. One credit is equal to one hour per day for a full academic year. N = 4,414. CTE: career and technical education; GMLP: General Labor Market Preparation. ## 4. Choices: Explaining Course-taking Behavior ### Descriptive Evidence: Course-Taking We now turn our attention to the predictors and temporal pattern of vocational course-taking in high school. Figure 5 presents the average number of courses taken by subject, separately by grade. Though English and foreign language are taken with the same frequency at all grades—students take about one English credit, equal to one full year, and 0.5 language credits per year, on average—others exhibit strong time trends. For instance, vocational courses are taken with much greater frequency in eleventh and twelfth grades, as students approach graduation and ostensibly have more control over their schedules. Social studies courses exhibit a similar, though weaker, trend. On the other hand, math and science course-taking drops considerably in twelfth grade. This suggests that there exist opportunities for students’ experiences early in high school to inform later curriculum choices, and that students’ specialization in vocational coursework has a distinct temporal aspect. This temporal pattern is an important feature of our conceptual model, where early course outcomes inform beliefs about ability and subsequent decisions. Figure 5. Number of Credits Taken by Subject and Grade Notes: Students who dropout are only counted above in grades they attempted. One credit is equal to one hour per day for a full academic year. N = 4,414. Figure 5. Number of Credits Taken by Subject and Grade Notes: Students who dropout are only counted above in grades they attempted. One credit is equal to one hour per day for a full academic year. N = 4,414. Figure 6 plots the number of credits taken by subject area separately by the Armed Forces Qualification Test (AFQT) quartile. Though total credits are comparable across AFQT scores, there exists substantial variation in subject mix. Vocational courses and foreign language courses exhibit the most noticeable relationship with AFQT. High AFQT students take about two thirds as many vocational courses as low AFQT students, and about three times as many foreign language courses. Higher AFQT students also tend to take more science courses. It is important to note that for most students, the AFQT is administered during their high school careers—the mean and median respondent took the AFQT approximately at age 14.5 years and nearly all had taken the test by age 16 years—thus we cannot rule out the impact of course-taking on the AFQT itself. On the other hand, the AFQT is likely not informed by the most salient tradeoff with vocational coursework: language courses. That is, the AFQT tests only mathematical and English language ability, not foreign language skill or science. Thus, if vocational courses pull students out of foreign language, science, or social studies courses, we should expect to see no effect on AFQT, allaying concerns about timing. Figure 6. Number of Credits Taken by Subject and Armed Forces Qualification Test (AFQT) Quartile Notes: Shapes mark average total number of credits taken in each subject, by AFQT quartile. One credit is equal to one hour per day for a full academic year. N = 3,703 with non-missing AFQT. Figure 6. Number of Credits Taken by Subject and Armed Forces Qualification Test (AFQT) Quartile Notes: Shapes mark average total number of credits taken in each subject, by AFQT quartile. One credit is equal to one hour per day for a full academic year. N = 3,703 with non-missing AFQT. ### Empirical Specification: Course-Taking To examine predictors of vocational course-taking more rigorously, we estimate reduced-form linear regression models using the number of credits taken in each subject, $j$, as the dependent variable. In each specification we focus on vocational courses taken in tenth through twelfth grades, as there are greater opportunities for students to choose vocational courses after freshman year and this allows us to model choice in grades 10 through 12 as a function of revealed ability (grades) in ninth grade.6 Guided by the theoretical framework outlined earlier, we are interested in three classes of explanatory variables, which we sequentially enter into equation 3. First, we include a vector of observable student characteristics, such as gender, race, and family background as described in the summary statistics tables. Second, we include AFQT as a composite measure of ability. Lastly, as we expect curriculum decisions to be influenced by new information about students’ “fit” with academic or vocational curricula, we also include measures of academic performance in math and English in ninth grade, $GPAicore9th$, and a control for whether the math or English courses were high (difficult) or low level, $High/Lowicore9th$.7 Though students have prior beliefs about their fit with different curricula, performance early in high school provides additional information with which expectations can be revised. Students who perform well in early core academic subjects might then pursue a more academic, or less vocational, curriculum. Lastly, we include $τsc$, a full set ninth-grade cohort ($c$) by state of residence in twelfth-grade ($s$) fixed effects. That is, comparing within students who entered ninth grade in the same year and who lived in the same state in twelfth grade, when the bulk of vocational coursework is taken. Our empirical specification is: $VocCreditsi=γ0+γ1Xi+γ2AFQTi+γ3GPAicore9th+γ4High/Lowicore9th+τsc+ɛi.$ (3) The estimated coefficients combine several of the structural parameters described in the dynamic model. In particular, the relationship between AFQT and number of vocational credits taken ($γ2$) combines the effect of ability on expected performance in high school and college, which influences the flow utility from curriculum, on the likelihood of college graduation, and on labor market outcomes. Thus, the fact that high AFQT students are less likely to take vocational courses may be explained by any of these mechanisms. Similarly, we interpret $γ3$ as the importance of new information revealed about “fit” with an academic curriculum in the form of ninth grade GPA. ### Results: Course-Taking Table 2 presents results from equation 3. As prior work has documented, we find that male students and those from disadvantaged backgrounds (mother has high school degree or less, low income, received public assistance) are more likely to take vocational courses. We also find that rural and southern students take more vocationally oriented coursework. Vocational course-taking also decreases sharply with measured ability (AFQT). This is true even when high school dropouts are excluded in specification 4. Table 2. Vocational Courses Taken, Grades 10—12 All Vocational CoursesVocational LowVocational High (1)(2)(3)(4)(5)(6)(7) Male 0.598*** 0.596*** 0.558*** 0.570*** 0.555*** 0.333*** 0.264*** (0.065) (0.064) (0.065) (0.066) (0.067) (0.051) (0.038) Black −0.181* −0.388*** −0.422*** −0.423*** −0.208** −0.158* −0.004 (0.093) (0.095) (0.096) (0.099) (0.105) (0.081) (0.061) Hispanic −0.434*** −0.573*** −0.586*** −0.577*** −0.398*** −0.265*** −0.105* (0.095) (0.098) (0.098) (0.102) (0.116) (0.090) (0.062) Other −0.690*** −0.734*** −0.719*** −0.712*** −0.476*** −0.230** −0.246*** (0.123) (0.124) (0.125) (0.129) (0.138) (0.106) (0.071) Mother < HS 0.112 0.092 0.089 0.080 0.075 0.029 0.042 (0.083) (0.082) (0.082) (0.085) (0.087) (0.067) (0.048) Mother > HS −0.454*** −0.395*** −0.393*** −0.414*** −0.381*** −0.252*** −0.123** (0.085) (0.084) (0.083) (0.085) (0.088) (0.066) (0.049) Ever aid 0.341*** 0.270*** 0.248*** 0.310*** 0.234*** 0.204*** 0.013 (0.078) (0.078) (0.078) (0.081) (0.082) (0.063) (0.046) South, age 12 0.342*** 0.310*** 0.328*** 0.367*** 0.037 0.073 −0.058 (0.078) (0.077) (0.078) (0.080) (0.156) (0.121) (0.082) Rural, age 12 0.283*** 0.263** 0.257** 0.230** 0.303** 0.275*** 0.066 (0.109) (0.108) (0.108) (0.110) (0.125) (0.095) (0.068) Poverty ratio −0.001*** −0.000*** −0.000*** −0.001*** −0.000*** −0.000*** 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Main HS public 1.120*** 1.072*** 1.064*** 1.087*** 1.100*** 0.734*** 0.315*** (0.086) (0.086) (0.086) (0.087) (0.096) (0.072) (0.053) Gifted −0.765*** −0.518*** −0.395*** −0.403*** −0.423*** −0.302*** −0.110** (0.070) (0.073) (0.081) (0.082) (0.086) (0.063) (0.049) Bilingual −0.049 −0.095 −0.079 −0.090 −0.032 0.199 −0.242*** (0.176) (0.172) (0.172) (0.177) (0.188) (0.154) (0.076) AFQT-Z −0.387*** −0.332*** −0.354*** −0.365*** −0.254*** −0.082*** (0.042) (0.045) (0.046) (0.047) (0.037) (0.027) Core GPA 9th −0.135*** −0.149*** −0.167*** −0.141*** −0.007 (0.038) (0.039) (0.040) (0.032) (0.022) Math 9th, high −0.086 −0.078 −0.060 −0.097 0.025 (0.116) (0.118) (0.120) (0.091) (0.072) English 9th, high −0.171* −0.204** −0.155* −0.168** 0.021 (0.087) (0.089) (0.092) (0.066) (0.052) State-cohort FE Yes Yes Yes HS grads only Yes Yes Yes Yes Year 9th Yes Yes Yes Yes R2 0.105 0.122 0.126 0.14 0.231 0.215 0.131 Observations (N4,414 4,414 4,414 4,165 4,165 4,165 4,165 All Vocational CoursesVocational LowVocational High (1)(2)(3)(4)(5)(6)(7) Male 0.598*** 0.596*** 0.558*** 0.570*** 0.555*** 0.333*** 0.264*** (0.065) (0.064) (0.065) (0.066) (0.067) (0.051) (0.038) Black −0.181* −0.388*** −0.422*** −0.423*** −0.208** −0.158* −0.004 (0.093) (0.095) (0.096) (0.099) (0.105) (0.081) (0.061) Hispanic −0.434*** −0.573*** −0.586*** −0.577*** −0.398*** −0.265*** −0.105* (0.095) (0.098) (0.098) (0.102) (0.116) (0.090) (0.062) Other −0.690*** −0.734*** −0.719*** −0.712*** −0.476*** −0.230** −0.246*** (0.123) (0.124) (0.125) (0.129) (0.138) (0.106) (0.071) Mother < HS 0.112 0.092 0.089 0.080 0.075 0.029 0.042 (0.083) (0.082) (0.082) (0.085) (0.087) (0.067) (0.048) Mother > HS −0.454*** −0.395*** −0.393*** −0.414*** −0.381*** −0.252*** −0.123** (0.085) (0.084) (0.083) (0.085) (0.088) (0.066) (0.049) Ever aid 0.341*** 0.270*** 0.248*** 0.310*** 0.234*** 0.204*** 0.013 (0.078) (0.078) (0.078) (0.081) (0.082) (0.063) (0.046) South, age 12 0.342*** 0.310*** 0.328*** 0.367*** 0.037 0.073 −0.058 (0.078) (0.077) (0.078) (0.080) (0.156) (0.121) (0.082) Rural, age 12 0.283*** 0.263** 0.257** 0.230** 0.303** 0.275*** 0.066 (0.109) (0.108) (0.108) (0.110) (0.125) (0.095) (0.068) Poverty ratio −0.001*** −0.000*** −0.000*** −0.001*** −0.000*** −0.000*** 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Main HS public 1.120*** 1.072*** 1.064*** 1.087*** 1.100*** 0.734*** 0.315*** (0.086) (0.086) (0.086) (0.087) (0.096) (0.072) (0.053) Gifted −0.765*** −0.518*** −0.395*** −0.403*** −0.423*** −0.302*** −0.110** (0.070) (0.073) (0.081) (0.082) (0.086) (0.063) (0.049) Bilingual −0.049 −0.095 −0.079 −0.090 −0.032 0.199 −0.242*** (0.176) (0.172) (0.172) (0.177) (0.188) (0.154) (0.076) AFQT-Z −0.387*** −0.332*** −0.354*** −0.365*** −0.254*** −0.082*** (0.042) (0.045) (0.046) (0.047) (0.037) (0.027) Core GPA 9th −0.135*** −0.149*** −0.167*** −0.141*** −0.007 (0.038) (0.039) (0.040) (0.032) (0.022) Math 9th, high −0.086 −0.078 −0.060 −0.097 0.025 (0.116) (0.118) (0.120) (0.091) (0.072) English 9th, high −0.171* −0.204** −0.155* −0.168** 0.021 (0.087) (0.089) (0.092) (0.066) (0.052) State-cohort FE Yes Yes Yes HS grads only Yes Yes Yes Yes Year 9th Yes Yes Yes Yes R2 0.105 0.122 0.126 0.14 0.231 0.215 0.131 Observations (N4,414 4,414 4,414 4,165 4,165 4,165 4,165 Notes: One course (dependent variable) is one full year. Math/English high are indicators = 1 if respondent was in upper-level English/Math courses in ninth grade. Columns 4—6 are restricted to high school (HS) graduates only. Core credits include English, Math, Science, Social Studies. Electives include Language, Art/Music, Physical Education, and Other. Low/High core indicates upper or lower-level course within subject-grade. Low vocational indicates entry level course. High vocational indicates second course or higher, or an internship/experiential learning. Robust standard errors in parentheses. FE: fixed effects; GPA: grade point average; AFQT-Z: Armed Forces Qualification Test normalized to a z-score. *p < 0.10; **p < 0.05; ***p < 0.01. As noted earlier, vocational course-taking has a strong temporal dimension: It is taken largely in tenth and twelfth grades. Thus, information revealed during high school could potentially influence course-taking decisions over time. We test this in specification 3, which includes ninth-grade GPA in math and English as a covariate, plus controls for whether these courses were upper or lower level, and specification 4 which restricts to only high school graduates. We find that students performing poorly in these academic subjects, conditional on our rich set of covariates, are more likely to take vocational courses later in high school, even controlling for AFQT. That is, a one-point increase in ninth-grade math and English GPA decreases total vocational course-taking by 0.15 of a one-year course. This represents a roughly 5 percent decrease in vocational course-taking off from a base of approximately three total vocational courses, on average. This result is consistent with a key aspect of our theoretical framework, where students use early course performance to update their beliefs about their fit with the academic/college sector or vocational/noncollege tracks, and these beliefs in turn influence course-taking. In column 5, we include a full vector of cohort–state fixed effects to observe whether state-level variation in either vocational course offerings, graduation requirements, or state-level labor market conditions might affect the relationships we observe. Adding these fixed effects reduces the minority and income gradients, but has little impact on the relationship between ninth-grade GPA (or AFQT) and course-taking, suggesting that this type of variation is not the primary driver of course-taking behavior. The final two columns present estimates separately for low- and high-vocational coursework, respectively. Results in these final two specifications reveal a subtle yet important distinction—ability and freshman year GPA, in addition to many of the fixed demographic characteristics we observe concerning childhood circumstances, are far more predictive of taking low-level vocational courses than they are of upper-level or specialized vocational courses. In particular, we find that a negative signal of ability manifested in low grades in core courses in ninth grade, conditional on course level and AFQT, increases the number of low-level vocational courses in grades 10 through 12, but has no impact on taking upper-level, or specialized, vocational courses. Put differently, low ability, or learning about low ability, leads students to take a broader array of vocational courses rather than to specialize in a particular type of vocational training. Similarly, one could also point out that the AFQT explains much more variation in low-level vocational coursework than it does upper-level coursework. The importance of this result is underscored in subsequent analyses demonstrating that (1) whereas lower-level vocational courses deter students from college, upper-level courses at worst do no harm, and (2) whereas upper-level vocational courses benefit students in the labor market, low-level vocational courses have no impact. ## 5. Outcomes: College and the Labor Market ### Empirical Specification: College Attendance and Completion We begin our analysis of the consequences of high school vocational coursework by observing the relationship between high school curriculum choice and both college attendance and completion. To do so, we first estimate three linear reduced-form equations where the dependent variable, $College$, is a binary indicator of college attendance in a two-year college, a four-year college, and then any college. We then repeat this exercise for college completion conditional on college enrollment. In both cases, we condition on having completed high school, though results are similar if not. Our main explanatory variables, $VocHighi$ and $VocLowi$, measure the number of high and low vocational credits earned during high school. Because the number of vocational credits is likely correlated with credits earned in other subjects, we control for a vector of credits earned in core (high and low) and elective subjects (language, art/music, physical education, and “other” courses) as defined below: $Collegei=β0+β1VocLowi+β2VocHighi+β3CoreLowi+β4CoreHighi+β5Electi+β6Xi+β7AFQTi+τsc+εi.$ (4) While it is important to note that in most cases students taking upper-level vocational courses do so after taking an introductory vocational course, as demonstrated above, most students in fact take multiple vocational courses. Thus, we are comparing students who choose to take several different introductory vocational courses with those who take their courses in a linear sequence. For example, on average, students take just over two low-level vocational courses (that is, two one-year courses) and one upper-level vocational course. We verify this further in our robustness specifications. ### Empirical Specification: Wage To examine labor market consequences of vocational course taking we stack wage observations and estimate reduced-form linear regression models that exploit cross-sectional variation in high school course-taking. Our primary labor market outcome is the log of real hourly wage, which is observed at multiple ages and thus pooled across all observations for individuals in our wage sample. We begin by including only $VocHighi$ and $VocLowi$ and then sequentially include additional covariates to observe how estimates change when nonvocational coursework ($Corei,Electi$), ability ($AFQT$), and college attendance and completion ($Postseci$) are accounted for. The inclusion of either of these types of controls (nonvocational courses and postsecondary outcomes) alters the interpretation of the estimated parameters and thus is an important modeling choice, a subject we discuss more thoroughly in the next subsection. Our preferred specification is then: $lnwageit=β0+β1VocLowi+β2VocHighi+β3CoreLowi+β4CoreHighi+β5Electi+β6Xi+β7AFQTi+ψPostseci+τsc+φt+εit.$ (5) Our controls ($Xi)$ include all the demographic and high school characteristics listed in panels A and B of table 1, as well as cohort by state of residence in high school fixed effects $(τsc$), secular wage-year indicators ($θt$), indicators for region of residence and Metropolitan Statistical Area status at interview, and an indicator for enrollment. All specifications account for the within-individual correlation across observations by clustering standard errors on the individual. ### An Aside: Course Substitution, Postsecondary Outcomes, and the Interpretation of Parameter Estimates Because students’ time is fixed, one substantive modeling choice is whether and how to control for the number of credits earned in other subjects. Without controls for other courses, that is core courses and electives, parameters $β1$ and $β2$ in equation 5 should be interpreted as the ceteris paribus change in log wage associated with an additional (high or low) vocational credit. Because each additional vocational course could be taken at the expense of another course (math, language, science, etc.), or what would otherwise be non-course time, the nature of the treatment likely varies across the population. The three panels in figure 7 depict the implicit nature of the tradeoff between vocational and other types of courses. Vocational courses are most clearly traded off with electives (art/music, foreign language, physical education, and “other” courses). The correlation between vocational and core academic courses (English, math, science, and social studies) is also negative, but much weaker. Thus, the “average treatment” in the model without controls for other courses can be thought of as taking one vocational course but fewer electives. Figure 7. Pairwise Joint Distributions of Core, Elective, and Vocational Credits Accumulated for High School Graduates Notes: Figures show scatterplots of total credits accumulated for high school graduates. Core courses include English, Math, Science and Social Studies. Electives include Art/Music, Foreign Languages, Physical Education and “Other” courses. One credit is equal to a one hour per day full year course. N = 4,414. Figure 7. Pairwise Joint Distributions of Core, Elective, and Vocational Credits Accumulated for High School Graduates Notes: Figures show scatterplots of total credits accumulated for high school graduates. Core courses include English, Math, Science and Social Studies. Electives include Art/Music, Foreign Languages, Physical Education and “Other” courses. One credit is equal to a one hour per day full year course. N = 4,414. Controlling for the number of credits taken in other subjects alters this interpretation. As a thought experiment, imagine that we divided courses into three types: core, vocational, and electives, and then for each respondent determined what share of total coursework each constituted. Including all three measures in our empirical specification would violate the full rank assumption (perfect collinearity) and one term would have to be omitted in the regression equation. Thus, holding course-taking in all other subjects constant, parameters $β1$ and $β2$ should be interpreted as the outcome change associated with taking one more high or low vocational credit at the expense of what would otherwise be non-course time, controlling for selection on observables. That is, the full rank assumption relies empirically on variation in the total number of courses taken across students. In this case, the opportunity cost of an additional course depends on how students potentially use this marginal hour among available activities, such as study hall, paid or volunteer work, homework, or leisure. Furthermore, controls for course taking in other subjects may also account for additional forms of selection if course-taking in other subjects is a marker for traits influencing outcomes that also correlate with vocational course-taking. Surprisingly, prior literature on the effects of course taking is mostly silent on this important conceptual issue, with the exception of Altonji (1995). His models include number of credits taken in eight subjects (English, social studies, math, science, foreign language, fine arts, industrial arts, and commercial), omitting time devoted to study halls, certain vocational courses, physical education, and home economics. He concludes that “one would expect all the elements of [the coefficient vector] to be greater than or equal to 0” since the specification controls for any displacement effects on other course-taking. In examining the labor market effects of math courses, Rose and Betts (2004) do not mention non-math course taking as the primary opportunity cost of taking an additional math course. Their main empirical model includes the number of math courses in six different levels, thus each coefficient should be interpreted as the effect of taking more courses in one level of rigor holding constant the number of math courses in the other levels. In these models, students are explicitly trading off math course-taking with time allocated to other subjects or non-school time (which of these, though, remains unclear). Subsequent specifications include number of courses in English, science, and foreign language, also using detailed curriculum categories. Interpreting the coefficients in these models is difficult because an additional year spent in upper-level English, for instance, may have a different opportunity cost against which the effect is measured than one spent in basic biology. Goodman (2019) shows that curricular reforms have a moderate and significant impact on math course-taking, but also a modest insignificant impact on other course-taking. His two-sample instrumental variables estimates of the wage effects of math courses implicitly assumes that reforms cannot impact wages through these non-math courses, either because reforms do not affect non-math course-taking or because such courses have no labor market impact. In our preferred model we follow in the spirit of Altonji (1995) and include a vector containing the number of credits earned in all other subjects, but show similar specifications with only vocational credits included. In an ancillary set of regressions we model the share of courses taken in each subject, leaving electives as the omitted category—and including a continuous measure of the total number of courses taken. Although directionality in each of these specifications is similar, leading to similar policy conclusions, for reasons mentioned above interpretation differs. ### Results: College Attendance and Completion Before examining labor market outcomes, panel A of table 3 shows conditional mean differences in postsecondary attendance and completion across high school course-taking as described in equation 4. Note that here we restrict only to high school completers to avoid confounding the number of credits taken with graduation. Results are nonetheless robust to their inclusion. We find little relationship between vocational coursework and college attendance on aggregate, resulting from opposite-signed coefficients on two-year college (positive) and four-year schools (negative) for upper-level vocational courses, and point estimates near zero for low-level vocational courses. As expected, we find that students taking more core and elective credits are less likely to attend two-year colleges and more likely to attend four-year institutions. Table 3. Dependent Variables Indicate Two- or Four-Year College Attendance or Completion, Conditional on Completing High School Panel A: Number of Credits Attend Two Year (1)Attend Four Year (2)Attend any (3)Earn Two Year (4)Earn Four Year (5)Earn any (6) Vocational credits, low −0.001 −0.003 −0.004 −0.005 0.005 0.003 (0.004) (0.004) (0.004) (0.008) (0.007) (0.005) Vocational credits, high 0.009 −0.007 0.002 0.021* 0.016* 0.010 (0.006) (0.006) (0.005) (0.011) (0.009) (0.007) Core credits, low −0.015*** 0.034*** 0.019*** −0.001 0.014** 0.018*** (0.004) (0.004) (0.004) (0.009) (0.006) (0.005) Core credits, high −0.030*** 0.053*** 0.024*** 0.018* 0.030*** 0.040*** (0.003) (0.004) (0.003) (0.010) (0.006) (0.005) Elective credits −0.007** 0.023*** 0.017*** −0.001 0.017*** 0.016*** (0.003) (0.003) (0.003) (0.008) (0.004) (0.004) R2 0.166 0.344 0.218 0.301 0.229 0.227 Panel B: Share of Credits Share vocational low*100 0.001 −0.007*** −0.006*** −0.001 −0.003* −0.003** (0.001) (0.001) (0.001) (0.002) (0.002) (0.001) Share vocational high*100 0.004*** −0.008*** −0.004*** 0.006** −0.001 −0.001 (0.002) (0.001) (0.001) (0.003) (0.003) (0.002) Share Core low*100 −0.002* 0.002** 0.000 0.000 −0.001 0.001 (0.001) (0.001) (0.001) (0.003) (0.002) (0.002) Share Core high*100 −0.006*** 0.008*** 0.002** 0.005 0.004** 0.007*** (0.001) (0.001) (0.001) (0.003) (0.002) (0.002) Total credits −0.013*** 0.029*** 0.016*** 0.002 0.018*** 0.020*** (0.003) (0.003) (0.002) (0.006) (0.004) (0.003) R2 0.166 0.346 0.218 0.301 0.230 0.228 Controls and observations for both panels If attended college Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes AFQT Yes Yes Yes Yes Yes Yes State-Cohort FE Yes Yes Yes Yes Yes Yes Observations (N4,165 4,165 4,165 919 2,495 3,414 Panel A: Number of Credits Attend Two Year (1)Attend Four Year (2)Attend any (3)Earn Two Year (4)Earn Four Year (5)Earn any (6) Vocational credits, low −0.001 −0.003 −0.004 −0.005 0.005 0.003 (0.004) (0.004) (0.004) (0.008) (0.007) (0.005) Vocational credits, high 0.009 −0.007 0.002 0.021* 0.016* 0.010 (0.006) (0.006) (0.005) (0.011) (0.009) (0.007) Core credits, low −0.015*** 0.034*** 0.019*** −0.001 0.014** 0.018*** (0.004) (0.004) (0.004) (0.009) (0.006) (0.005) Core credits, high −0.030*** 0.053*** 0.024*** 0.018* 0.030*** 0.040*** (0.003) (0.004) (0.003) (0.010) (0.006) (0.005) Elective credits −0.007** 0.023*** 0.017*** −0.001 0.017*** 0.016*** (0.003) (0.003) (0.003) (0.008) (0.004) (0.004) R2 0.166 0.344 0.218 0.301 0.229 0.227 Panel B: Share of Credits Share vocational low*100 0.001 −0.007*** −0.006*** −0.001 −0.003* −0.003** (0.001) (0.001) (0.001) (0.002) (0.002) (0.001) Share vocational high*100 0.004*** −0.008*** −0.004*** 0.006** −0.001 −0.001 (0.002) (0.001) (0.001) (0.003) (0.003) (0.002) Share Core low*100 −0.002* 0.002** 0.000 0.000 −0.001 0.001 (0.001) (0.001) (0.001) (0.003) (0.002) (0.002) Share Core high*100 −0.006*** 0.008*** 0.002** 0.005 0.004** 0.007*** (0.001) (0.001) (0.001) (0.003) (0.002) (0.002) Total credits −0.013*** 0.029*** 0.016*** 0.002 0.018*** 0.020*** (0.003) (0.003) (0.002) (0.006) (0.004) (0.003) R2 0.166 0.346 0.218 0.301 0.230 0.228 Controls and observations for both panels If attended college Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes AFQT Yes Yes Yes Yes Yes Yes State-Cohort FE Yes Yes Yes Yes Yes Yes Observations (N4,165 4,165 4,165 919 2,495 3,414 Notes: Sample is limited to high school graduates. Core credits include English, Math, Science, Social Studies. Electives include Language, Art/Music, Physical Education, and Other. Low/High core indicates upper or lower-level course within subject-grade. Low vocational indicates entry level course; high vocational indicates second course or higher, or an internship/experiential learning. Controls include: gender, race, mother's education, rural or South at age 12, family poverty ratio in 1997, public primary high school, any gifted courses, any bilingual education, and fixed effects (FE) for ninth-grade cohort and twelfth-grade state of residence. Robust standard errors in parenthesis. AFQT: Armed Forces Qualification Test. *p < 0.10; **p < 0.05; ***p < 0.01. Repeating the same exercise in columns 4–6 for respondents who attended two- or four-year college as their highest grade suggests that there is little observational evidence that vocational courses decrease graduation rates for those enrolled. In fact, we find evidence that, conditional on enrollment, students earning more upper-level vocational credits are more likely to graduate by nearly 2 percentage points. In panel B of table 3 we replicate our model using the share of credits accumulated in each field and include an additional control for the total number of credits accumulated (estimating the same model without total credits yields empirically similar results). In this case, as an actualization of the thought experiment described above, we omit the share of courses that were electives to satisfy the full rank requirement. Results are largely similar to panel A. We find that whereas increasing the share of low vocational courses decreases college attendance for four-year schools, increasing the share of courses dedicated to specialized, or upper-level, vocational course-work increases enrollment to two-year schools and decreases four-year enrollment. Conditional on college enrollment, we find that those with higher shares of coursework dedicated to introductory vocational courses are marginally less likely to earn a degree on an order of about a 3-percentage-point decrease for each 10 percent increase in the share of courses that are introductory vocational. These conflict with results in panel A, which showed no relationship, highlighting the importance of credits versus shares. Nonetheless, these coefficients in panel B are small and marginally different from zero. We also find that those with larger shares of upper-level vocational coursework are marginally more likely to graduate from two-year schools by about 6 percentage points for each 10-percentage-point increase in upper vocational share. Taken together, these results suggest that vocational courses may deter the marginal student from college, at least as measured by shares, but that there is little aggregate impact on graduation rates conditional on enrollment. Put differently, vocational courses may pull those students out of college who may be least likely to graduate. Though noisy, this is suggestive of a learning process where vocational coursework provides students with valuable labor market skills and additional information about their comparative advantage in the vocational or college labor market. An alternative explanation is that the opportunity to take vocational coursework compels some students to graduate from high school, but these students are unlikely to attend college. This would result in lower college enrollment rates, but college graduation conditional on enrollment will be unaffected as the marginal high school graduate is unlikely the marginal college attendee. Finally, we draw attention to the coefficient estimates on core and elective course-taking. In particular, taking additional courses, or increasing shares in these courses, increases the likelihood of both college enrollment and completion. Although this is not surprising, we highlight this result to tie our earlier theoretical prediction with the wage regressions we present below—in particular, that the labor market value of nonvocational coursework operates through increased college attendance. ### Results: Wages and Idleness In table 4, we begin our wage analysis by pooling observations and estimating unconditional correlations between vocational (low and high) course-taking and log hourly wage of employed NLSY respondents. We find, as expected, that those earning more low-level vocational credits earn less—about 2 percent per year of low vocational coursework. This is not surprising because high school graduates who take more of these courses tend to be lower-achieving and more disadvantaged, attributes that have independent effects on labor market outcomes. Conversely, we find a positive relationship between upper-level vocational coursework and earnings on the order of a 1 percent increase with each additional year of coursework. We next include our full set of covariates (excluding AFQT), finding that indeed accounting for childhood circumstances, school type, and a host of other controls both explains away the entirety of the negative relationship between low vocational coursework and wages, and increases the observed wage gains associated with upper-level vocational courses. Table 4. Dependent Variable Is Log (Real) Hourly Wage (1)(2)(3)(4)(5)(6)(7) Vocational credits, low −0.022*** −0.004 −0.001 −0.000 0.001 0.002 0.001 (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) Vocational credits, high 0.011** 0.016*** 0.017*** 0.017*** 0.017*** 0.021*** 0.018*** (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) Core credits, low 0.001 0.001 0.001 −0.004 (0.003) (0.003) (0.003) (0.003) Core credits, high 0.021*** 0.017*** 0.020*** 0.008** (0.003) (0.003) (0.003) (0.003) Elective credits 0.001 0.000 0.001 −0.004 (0.003) (0.003) (0.003) (0.003) Degree Yes State-cohort fixed effects Yes Yes AFQT Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes R2 0.009 0.208 0.219 0.220 0.226 0.269 0.291 Observations (N3,708 3,708 3,708 3,708 3,708 3,708 3,708 Observations*Years (N19,029 19,029 19,029 19,029 19,029 19,029 19,029 (1)(2)(3)(4)(5)(6)(7) Vocational credits, low −0.022*** −0.004 −0.001 −0.000 0.001 0.002 0.001 (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) Vocational credits, high 0.011** 0.016*** 0.017*** 0.017*** 0.017*** 0.021*** 0.018*** (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) Core credits, low 0.001 0.001 0.001 −0.004 (0.003) (0.003) (0.003) (0.003) Core credits, high 0.021*** 0.017*** 0.020*** 0.008** (0.003) (0.003) (0.003) (0.003) Elective credits 0.001 0.000 0.001 −0.004 (0.003) (0.003) (0.003) (0.003) Degree Yes State-cohort fixed effects Yes Yes AFQT Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes R2 0.009 0.208 0.219 0.220 0.226 0.269 0.291 Observations (N3,708 3,708 3,708 3,708 3,708 3,708 3,708 Observations*Years (N19,029 19,029 19,029 19,029 19,029 19,029 19,029 Notes: Sample is wage sample. Wages are in real 2010 dollars. Wage sample consists of respondents who have a reported wage >$2.00/hour after labor market entry—defined as the first interview after four consecutive non-enrolled semesters. Controls include: gender, race, mother's education, rural or South at age 12, family poverty ratio in 1997, public primary high school, any gifted courses, any bilingual education, year entered ninth grade, year of wage record, and Metropolitan Statistical Area status when wage was recorded. A one-unit change in credits is equal to a one-hour, one-year course. Core credits include English, Math, Science, Social Studies. Electives include Language, Art/Music, Physical Education, and Other. Low/High core indicates upper or lower-level course within subject-grade. Low vocational indicates entry level course; high vocational indicates second course or higher, or an internship/experiential learning. Standard errors clustered on individuals in parentheses. AFQT: Armed Forces Qualification Test.

*p < 0.10; **p < 0.05; ***p < 0.01.

We next include the AFQT, a commonly used proxy of “ability” in the model. We know from summary statistics that students with higher AFQT scores take fewer vocational courses, suggesting that its inclusion might increase the observed relationship between vocational coursework and wages. Yet, specification 3, which includes AFQT, does not affect results.

We also might think that more vocational coursework crowds out academic subjects or electives, which might affect wages as well. Column 4 then controls for the number of credits earned in other subjects, separating elective and core academic subjects (math, English, science, and social studies) and omits AFQT for the time being. This specification also distinguishes high- and low-level academic subjects (e.g., advanced algebra versus basic math). Consistent with results in Rose and Betts (2004), advanced coursework is much more advantageous than basic coursework. Comparing columns 2 and 4, the coefficient on advanced vocational work does not change, suggesting that vocational course-taking does not crowd out other courses that also have impacts on wages (not conditioning on college or AFQT). Alternatively, credits earned in other subjects may be a marker for academic ability, which further controls for the negative selection bias. Because these factors may offset, we are not able to distinguish these two distinct channels. Controlling for all factors in column 5 reconciles this, yet has little impact on results, other than reducing the gain to upper-level core courses by one-third. In other words, selection on ability, at least as measured by AFQT, does not appear to affect the observed relationship between vocational coursework and wages, conditional on our full set of controls.

We next address concerns that the local education environment, for example, graduation requirements, labor markets, state-level vocational funding or state/cohort-level attitudes toward vocational education, might bias our results. To account for this, we include a full set of cohort by state of twelfth-grade attendance fixed effects. We find that their inclusion has little impact on results, suggesting that concerns about the local environment biasing our results are not immediate. Lastly, in column 7, we add controls for postsecondary attendance and attainment. As expected, we find that the majority of wage gains associated with core coursework are explained by controlling for higher education and an ability proxy. In fact, the 2 percent upper core course advantage is more than halved and signs on lower core and electives are now negative, though indistinguishable from zero. Yet, we find that the positive wage gains associated with additional upper-level vocational courses are unaffected by the inclusion of these factors, suggesting a 1.8-log-point wage premium for each additional year of upper-level vocational coursework. We take this as prima facie evidence that the labor market value of vocational coursework accrues largely to non-college graduates, and the labor market value of nonvocational coursework is largely, if not entirely, explained by its contribution to postsecondary attainment.

In sum, our estimates suggest a near 2 percent wage gain associated with each additional year of upper-level vocational coursework, yet no gain is associated with increasing breadth in introductory vocational courses. We take this as evidence of gains to specialization in high school labor market training. Moreover, these results persist, conditional on a measure of nontechnical labor market skill in the AFQT, a host of student-level controls including parental income, state-cohort fixed effects, parent's education, high school type, and urban/rural designation in adolescence.

As a specification check, in table 5 we repeat our preferred specification from table 4 using an alternative specification modeling the share of courses taken rather than the count, as we did in panel B of table 3. Again, results largely confirm our preferred specification but with a different interpretation. In table 5 we estimate that a 1-percentage-point increase in the share of upper-level vocational courses increases wages by just over one-half of one percent, even after accounting for our full set of controls. To put these results in comparable terms, we estimate the change in share of courses associated with a one-course increase by regressing the share of upper-level vocational courses on the number of vocational courses taken, finding that one extra course is equivalent to a nearly 4 percent increase in the share of courses taken. Because students in the wage sample take roughly twenty-five courses on average, where a one-course increase is equivalent to a 4 percentage point change, this confirms our intuition of a tradeoff with elective courses. Multiplying this 4-percentage-point increase by the 0.006 coefficient estimate suggests a 2.4 percent increase in wages—larger but similar in magnitude to the 1.8 percent increase estimated in the credit count model in table 4—suggesting these results do not rely heavily on functional form assumptions.

Table 5.
Dependent Variable Is Log Hourly Wage, Course-Taking Defined in Shares of Total
(1)(2)(3)(4)(5)(6)
Share Vocational low*100 −0.005*** −0.001 0.000 0.000 0.000 0.001
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Share Vocational high*100 0.003* 0.004*** 0.005*** 0.005*** 0.005*** 0.006***
0.001 (0.001) (0.001) (0.001) (0.001) (0.001)
Share Core low*100 −0.004*** −0.001 −0.000 0.000 0.000 −0.000
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Share Core high*100 0.006*** 0.005*** 0.004*** 0.005*** 0.005*** 0.003***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Total credits    0.003* 0.004** −0.001
(0.002) (0.002) (0.002)
Degree      Yes
State-cohort fixed effects     Yes Yes
AFQT   Yes Yes Yes Yes
Controls  Yes Yes Yes Yes Yes
R2 0.052 0.220 0.226 0.226 0.269 0.291
Observations (N3,708 3,708 3,708 3,708 3,708 3,708
Observations*Years (N19,029 19,029 19,029 19,029 19,029 19,029
(1)(2)(3)(4)(5)(6)
Share Vocational low*100 −0.005*** −0.001 0.000 0.000 0.000 0.001
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Share Vocational high*100 0.003* 0.004*** 0.005*** 0.005*** 0.005*** 0.006***
0.001 (0.001) (0.001) (0.001) (0.001) (0.001)
Share Core low*100 −0.004*** −0.001 −0.000 0.000 0.000 −0.000
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Share Core high*100 0.006*** 0.005*** 0.004*** 0.005*** 0.005*** 0.003***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Total credits    0.003* 0.004** −0.001
(0.002) (0.002) (0.002)
Degree      Yes
State-cohort fixed effects     Yes Yes
AFQT   Yes Yes Yes Yes
Controls  Yes Yes Yes Yes Yes
R2 0.052 0.220 0.226 0.226 0.269 0.291
Observations (N3,708 3,708 3,708 3,708 3,708 3,708
Observations*Years (N19,029 19,029 19,029 19,029 19,029 19,029

Notes: Shares are the share of total courses taken in each subject group multiplied by 100. Share of courses that are electives is the omitted category. Sample is wage sample. Wages are in real 2010 dollars. Wage sample consists of respondents who have a reported wage >$2.00/hour after labor market entry—defined as the first interview after four consecutive non-enrolled semesters. Controls include: gender, race, mother's education, rural or South at age 12, family poverty ratio in 1997, public primary high school, any gifted courses, any bilingual education, year entered ninth grade, year of wage record, and Metropolitan Statistical Area status when wage was recorded. A one-unit change in credits is equal to a one-hour, one-year course. Core credits include English, Math, Science, Social Studies. Electives include Language, Art/Music, Physical Education, and Other. Low/High core indicates upper or lower-level course within subject-grade. Low vocational indicates entry level course; high vocational indicates second course or higher, or an internship/experiential learning. Standard errors clustered on individuals in parentheses. AFQT: Armed Forces Qualification Test. *p < 0.10; **p < 0.05; ***p < 0.01. Above we explicitly consider wage gains as opposed to earnings differentials, where the latter also account for nonemployment. To round out our analysis of labor market outcomes, we estimate a modified version of table 4 with a binary indicator for idleness as the dependent variable, defined as neither working nor in school.8 With no need to limit the sample to those entering the labor market or to those employed, we use our entire analysis sample of 4,414 individuals and limit to observations on or after age 19 years, when ostensibly all are out of high school. Moving across columns 1–3 of table 6 shows the effect of adding sequential sets of control variables on the relationship between course-taking and idleness, as in tables 4 and 5, with similar results. With a full set of controls in column 4, we find little relationship between idleness and vocational coursework of any type. Yet, disambiguating the sample between those who ever attended college and those who never attended shows that students who never attended college and took more upper-level vocational coursework were less likely to be idle, on the order of a 1.3-percentage-point decrease in the incidence of idleness for each year of upper-level vocational coursework, and 0.08 percentage point for each low-level course. There is little relationship between core or elective credits and idleness for non-college students. When we restrict the sample to students who attended any college, we find no relationship between idleness and high school vocational coursework, suggesting that decreased idleness is attributable to increased employment and exclusively for students who do not attend postsecondary schooling. Table 6. Dependent Variable Is Idleness (Not Working for Wage and Not Enrolled) (1)(2)(3)(4)(5)(6) Vocational credits, low 0.001 −0.002 −0.002 −0.002 −0.008* 0.002 (0.002) (0.002) (0.002) (0.002) (0.004) (0.002) Vocational credits, high −0.005* −0.005** −0.005* −0.004 0.013** 0.001 (0.003) (0.003) (0.003) (0.003) (0.005) (0.003) Core credits, low −0.004*** −0.004*** −0.004*** −0.001 0.005 −0.003 (0.002) (0.002) (0.002) (0.002) (0.004) (0.002) Core credits, high −0.014*** −0.011*** −0.011*** −0.005*** −0.003 −0.006*** (0.001) (0.002) (0.002) (0.002) (0.005) (0.002) Elective credits −0.007*** −0.004*** −0.005*** −0.002 −0.010** −0.001 (0.001) (0.001) (0.001) (0.001) (0.004) (0.001) Attended college No Yes Degree Yes Yes Yes State-cohort fixed effects Yes Yes Yes Yes AFQT Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes R2 0.015 0.036 0.057 0.072 0.121 0.061 Observations (N4,414 4,414 4,414 4,414 961 3,453 Observations*Years (N44,774 44,774 44,774 44,774 9,819 34,955 (1)(2)(3)(4)(5)(6) Vocational credits, low 0.001 −0.002 −0.002 −0.002 −0.008* 0.002 (0.002) (0.002) (0.002) (0.002) (0.004) (0.002) Vocational credits, high −0.005* −0.005** −0.005* −0.004 0.013** 0.001 (0.003) (0.003) (0.003) (0.003) (0.005) (0.003) Core credits, low −0.004*** −0.004*** −0.004*** −0.001 0.005 −0.003 (0.002) (0.002) (0.002) (0.002) (0.004) (0.002) Core credits, high −0.014*** −0.011*** −0.011*** −0.005*** −0.003 −0.006*** (0.001) (0.002) (0.002) (0.002) (0.005) (0.002) Elective credits −0.007*** −0.004*** −0.005*** −0.002 −0.010** −0.001 (0.001) (0.001) (0.001) (0.001) (0.004) (0.001) Attended college No Yes Degree Yes Yes Yes State-cohort fixed effects Yes Yes Yes Yes AFQT Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes R2 0.015 0.036 0.057 0.072 0.121 0.061 Observations (N4,414 4,414 4,414 4,414 961 3,453 Observations*Years (N44,774 44,774 44,774 44,774 9,819 34,955 Notes: Sample is analysis sample, ages 19 and older. Core credits include English, Math, Science, Social Studies. Electives include Language, Art/Music, Physical Education, and Other. Low/High core indicates upper- or lower-level course within subject-grade. Low vocational indicates entry level course; high vocational indicates second course or higher, or an internship/experiential learning. Standard errors clustered on individuals in parentheses. AFQT: Armed Forces Qualification Test. *p < 0.10; **p < 0.05; ***p < 0.01. ### Heterogeneity, Sensitivity, and Alternative Hypotheses We next turn our attention to heterogeneity in wage gains and sensitivity to alternative hypotheses, model specifications, and sample selection. We begin by decomposing observed wage gains by vocational field of study, though sample size constraints prohibit us from a full set of interactions between occupational field and high/low denomination. Results are shown in table 7, where coefficients are ordered from the most to least frequented vocational courses. We find that gains are driven by Transportation & Industry (including construction trades, mechanics and repair, transportation, and production), Business & Management (including business management, services, and marketing), and Health Care. Notably, Transportation & Industry and Business are the most frequented vocational fields. We find large and negative coefficients on Industrial Arts as classified as General Labor Market Preparation, meaning introductory coursework. The NLSY does not provide details on which courses were included but only 2 percent of respondents earned any credits in this field. No other fields have negative coefficients that are statistically different from zero. We also find no gain from either Keyboarding or Computer Technology, a nontrivial finding considering that computer access was far from universal in the late 1990s and early 2000s when NLSY respondents were in high school, yet coefficients are positive and we cannot statistically rule out they are similar to those driving results. These results highlight heterogeneity in gains associated with vocational coursework both across specialized versus introductory coursework and across fields of specialization. Table 7. Dependent Variable Is Log (Real) Hourly Wage (1)(2)Mean EarnedAny Earned Core credits, low 0.001 −0.004 9.9 1.00 (0.003) (0.003) Core credits, high 0.020*** 0.008*** 4.7 0.98 (0.003) (0.003) Elective credits 0.001 −0.004 7.4 1.00 (0.003) (0.003) Transport & industry 0.014*** 0.013*** 1.02 0.61 (0.004) (0.004) Business/management 0.015** 0.011* 0.64 0.43 (0.006) (0.006) GLMP keyboarding 0.007 0.002 0.35 0.47 (0.014) (0.013) GLMP general −0.004 −0.004 0.30 0.24 (0.007) (0.007) Computer tech. 0.009 0.005 0.29 0.27 (0.010) (0.010) Agriculture 0.017* 0.015 0.19 0.10 (0.009) (0.009) Service 0.000 −0.001 0.17 0.11 (0.006) (0.006) Education & child care −0.006 −0.008 0.12 0.11 (0.009) (0.009) GLMP technical education −0.020 −0.016 0.09 0.10 (0.016) (0.016) Health care 0.022** 0.022** 0.09 0.06 (0.010) (0.010) Public & protective services −0.018 −0.017 0.04 0.03 (0.013) (0.012) GLMP industrial arts −0.057** −0.051** 0.02 0.02 (0.026) (0.025) GLMP other −0.016 −0.012 0.01 0.01 (0.021) (0.019) Degree Yes Controls Yes Yes AFQT Yes Yes State-cohort fixed effects Yes Yes R2 0.27 0.292 Observations (N3,708 3,708 Observations*Years (N19,029 19,029 (1)(2)Mean EarnedAny Earned Core credits, low 0.001 −0.004 9.9 1.00 (0.003) (0.003) Core credits, high 0.020*** 0.008*** 4.7 0.98 (0.003) (0.003) Elective credits 0.001 −0.004 7.4 1.00 (0.003) (0.003) Transport & industry 0.014*** 0.013*** 1.02 0.61 (0.004) (0.004) Business/management 0.015** 0.011* 0.64 0.43 (0.006) (0.006) GLMP keyboarding 0.007 0.002 0.35 0.47 (0.014) (0.013) GLMP general −0.004 −0.004 0.30 0.24 (0.007) (0.007) Computer tech. 0.009 0.005 0.29 0.27 (0.010) (0.010) Agriculture 0.017* 0.015 0.19 0.10 (0.009) (0.009) Service 0.000 −0.001 0.17 0.11 (0.006) (0.006) Education & child care −0.006 −0.008 0.12 0.11 (0.009) (0.009) GLMP technical education −0.020 −0.016 0.09 0.10 (0.016) (0.016) Health care 0.022** 0.022** 0.09 0.06 (0.010) (0.010) Public & protective services −0.018 −0.017 0.04 0.03 (0.013) (0.012) GLMP industrial arts −0.057** −0.051** 0.02 0.02 (0.026) (0.025) GLMP other −0.016 −0.012 0.01 0.01 (0.021) (0.019) Degree Yes Controls Yes Yes AFQT Yes Yes State-cohort fixed effects Yes Yes R2 0.27 0.292 Observations (N3,708 3,708 Observations*Years (N19,029 19,029 Notes: Wages are in real 2010 dollars. Wage sample consists of respondents who have a reported wage >$2.00/hour after labor market entry—defined as the first interview after four consecutive non-enrolled semesters. Controls include: gender, race, mother's education, rural or South at age 12, family poverty ratio in 1997, public primary high school, any gifted courses, any bilingual education, year entered ninth grade, year of wage record, and Metropolitan Statistical Area status when wage was recorded. A one-unit change in credits is equal to a one-hour, one-year course. Standard errors clustered on individuals in parentheses. GLMP: General Labor Market Preparation; AFQT: Armed Forces Qualification Test.

*p < 0.10; **p < 0.05; ***p < 0.01.

Having decomposed aggregate gains both by field and by specialized versus introductory coursework, we now turn to a series of specification checks. Although we have shown various iterations of our main wage specification in the tables above, we are also interested in testing sensitivity to different definitions and samples. To do so we begin in table 8 by reestimating our preferred wage specification from column 6 of table 4. In column 2, we address concerns that accounting for state-by-cohort level variation is insufficient as course-taking might be more largely influenced by local (school district) conditions. While the data do not allow us to compare within schools, in specification 2 we replace our state-by-cohort fixed effects with county-by-cohort effects, allowing us to compare students in the same ninth-grade cohort who attended twelfth grade in the same county. This addition has little impact on estimates and no impact on interpretation. Although one might still worry that there exists considerable variation in school environments and attitudes toward vocational coursework within counties, we note that we are also accounting for measures of childhood wealth, maternal education, gifted and bilingual schooling, whether students went to a private high school, urban/rural status at age 12, and measures of academic ability in addition to coursework in core and elective fields. To further verify this, in column 3 we add our full set of graduation requirements (English, math, social studies, science, and total credits) to the model to ensure that school-level variation in graduation requirements is not driving results. Again, we find no change to our estimates, likely because effects are absorbed by our state- or county-cohort fixed effects.

Table 8.
Robustness Checks (Dependent Variable Is Log Wage)
(1)(2)(3)(4)(5)(6)(7)(8)(9)
Vocational credits, low 0.001 0.001 0.002 0.001 0.003 0.003 0.011  −0.002
(0.003) (0.003) (0.004) (0.003) (0.004) (0.004) (0.007)  (0.004)
Vocational credits, high 0.018*** 0.020*** 0.019*** 0.017*** 0.019*** 0.020*** 0.021**  0.015*
(0.005) (0.005) (0.005) (0.004) (0.006) (0.005) (0.010)  (0.007)
Core credits, low −0.004 −0.003 −0.006 −0.002 −0.005 −0.003 −0.002 −0.004 −0.004
(0.003) (0.003) (0.003) (0.002) (0.003) (0.003) (0.003) (0.003) (0.003)
Core credits, high 0.008** 0.007** 0.007** 0.010*** 0.009*** 0.009*** 0.009*** 0.008*** 0.008**
(0.003) (0.003) (0.003) (0.002) (0.003) (0.003) (0.004) (0.003) (0.003)
Elective credits −0.004 −0.002 −0.002 −0.003 −0.005* −0.005* −0.006** −0.004 −0.004
(0.003) (0.003) (0.003) (0.002) (0.003) (0.003) (0.003) (0.002) (0.003)
Concentrator        0.032** 0.027
(0.015) (0.018)
Specialist        0.062*** 0.020
(0.018) (0.029)
Sample limitation None None Non-missing requirements Relax sample restrictions Age >23 Hours ≥35 Vocational credits ≤4 None None
Requirements   Yes
County-cohort fixed effects  Yes
State-cohort fixed effects Yes  Yes Yes Yes Yes Yes Yes Yes
Degree Yes Yes Yes Yes Yes Yes Yes Yes Yes
Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes
AFQT Yes Yes Yes Yes Yes Yes Yes Yes Yes
R2 0.291 0.336 0.296 0.288 0.264 0.329 0.302 0.291 0.291
Observations (N3,708 3,708 3,285 4,338 3,575 3,403 2,660 3,708 3,708
Observations*Years (N19029 19,029 16,935 22,674 14,827 14,536 13,170 19,029 19,029
(1)(2)(3)(4)(5)(6)(7)(8)(9)
Vocational credits, low 0.001 0.001 0.002 0.001 0.003 0.003 0.011  −0.002
(0.003) (0.003) (0.004) (0.003) (0.004) (0.004) (0.007)  (0.004)
Vocational credits, high 0.018*** 0.020*** 0.019*** 0.017*** 0.019*** 0.020*** 0.021**  0.015*
(0.005) (0.005) (0.005) (0.004) (0.006) (0.005) (0.010)  (0.007)
Core credits, low −0.004 −0.003 −0.006 −0.002 −0.005 −0.003 −0.002 −0.004 −0.004
(0.003) (0.003) (0.003) (0.002) (0.003) (0.003) (0.003) (0.003) (0.003)
Core credits, high 0.008** 0.007** 0.007** 0.010*** 0.009*** 0.009*** 0.009*** 0.008*** 0.008**
(0.003) (0.003) (0.003) (0.002) (0.003) (0.003) (0.004) (0.003) (0.003)
Elective credits −0.004 −0.002 −0.002 −0.003 −0.005* −0.005* −0.006** −0.004 −0.004
(0.003) (0.003) (0.003) (0.002) (0.003) (0.003) (0.003) (0.002) (0.003)
Concentrator        0.032** 0.027
(0.015) (0.018)
Specialist        0.062*** 0.020
(0.018) (0.029)
Sample limitation None None Non-missing requirements Relax sample restrictions Age >23 Hours ≥35 Vocational credits ≤4 None None
Requirements   Yes
County-cohort fixed effects  Yes
State-cohort fixed effects Yes  Yes Yes Yes Yes Yes Yes Yes
Degree Yes Yes Yes Yes Yes Yes Yes Yes Yes
Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes
AFQT Yes Yes Yes Yes Yes Yes Yes Yes Yes
R2 0.291 0.336 0.296 0.288 0.264 0.329 0.302 0.291 0.291
Observations (N3,708 3,708 3,285 4,338 3,575 3,403 2,660 3,708 3,708
Observations*Years (N19029 19,029 16,935 22,674 14,827 14,536 13,170 19,029 19,029

Notes: Sample is wage sample. Core credits include English, Math, Science, Social Studies. Electives include Language, Art/Music, Physical Education, and Other. Low/High core indicates upper- or lower-level course within subject-grade. Low vocational indicates entry level course; high vocational indicates second course or higher, or an internship/experiential learning. Standard errors clustered on individuals in parentheses. AFQT: Armed Forces Qualification Test.

*p < 0.10; **p < 0.05; ***p < 0.01.

In column 4 we then relax our sample restriction that excluded students with illogical or unusually large/small total credit counts. Results are unchanged. In column 5 we then restrict our wage observations to only those over age 23 years, when the vast majority of students have completed formal schooling. Again, results are unaffected. In column 6 we then restrict to those working thirty-five hours per week or more, again finding no change in results. In column 7 we address concerns about the way in which vocational credits are counted by high schools. In particular, the NLSY's documentation notes that the weighting scheme for certain vocational and enrichment courses can differ from those of academic coursework in a manner that would inflate the total number of Carnegie credits earned.9 Although we address this in our original sample selection criteria, omitting respondents with unusually high credit counts, we further test for sensitivity here by restricting to respondents who earned no more than four Carnegie credits in vocational coursework, with little change to results. Note also that the systematic inflation of the count of vocational coursework would serve to underestimate the true relationship. Any case where this relationship is caused by statewide reporting rules would be accounted for by state-level fixed effects. The only concern would be cases where differential weighting is correlated with unobserved factors correlated with both the number of actual courses taken and earnings, which we find unlikely.

Finally, in columns 8 and 9 we turn our attention to alternatives to our “upper/lower” vocational coursework definition. To do so, we replace our course-level measures with indicators for Vocational Concentrator and Vocational Specialist. The NLSY defines a concentrator as someone who has earned at least three Carnegie credits in a single labor market field. Specialists are those earning at least four credits in a specific field, with at least two of these coming from upper-level courses. In column 8 we replace our continuous low and high measures with these dichotomous and mutually exclusive measures that specifically measure vertical credit accumulation (depth) as opposed to horizontal vocational credit accrual (or breadth) in vocational coursework. Results are consistent with our claim of the value of depth over breadth—we find that concentrators see a small wage premium of 3.2 percent and specialists experience a 6.2 percent wage premium. In column 9 we include both designations for Concentrator/Specialist and credit counts, finding that we still see a statistically significant relationship between specialized vocational coursework and wages.

To further demonstrate that those students taking a large number of low-level, or introductory, vocational courses are indeed experiencing breadth as opposed to racking up multiple low-level credits in a single field, we plot in figure 8 the relationship between the number of low-level vocational courses taken and the number of different vocational fields in which students took vocational coursework. Were the majority of students simply accruing multiple low-level courses in a single field (a different measure of depth) we would expect a flat relationship between credits and the number of fields experienced. We find this is not the case. For example, students who take three Carnegie credits of low-level vocational coursework experience on average nearly three different fields, suggesting that indeed our interpretation of breadth versus depth is warranted.

Figure 8.

Relationship between Number of Different Vocational Fields and Total Low Vocational Courses

Notes: Sample are all respondents who ever took a vocational course. Upper y-axis indicates the number of different fields in which each respondent took a low level vocational by the total number of low-level courses she took (x-axis). The second (lower) y-axis indicates the share of respondents taking each amount of low-level vocational courses. Line is a linear fit.

Figure 8.

Relationship between Number of Different Vocational Fields and Total Low Vocational Courses

Notes: Sample are all respondents who ever took a vocational course. Upper y-axis indicates the number of different fields in which each respondent took a low level vocational by the total number of low-level courses she took (x-axis). The second (lower) y-axis indicates the share of respondents taking each amount of low-level vocational courses. Line is a linear fit.

We have demonstrated that those taking more specialized vocational coursework see higher wages and lower incidence of idleness in early careers. We attribute this to students sorting into coursework that aligns with their talents and preferences, as self-selection models predict. Further, we have taken several steps to identify possible alternative hypotheses. Primary among these is the negative selection story, where less-able students are pushed into vocational coursework. The persistently positive results we observe here are likely inconsistent with this story. As an additional test, in the online Appendix we use a set of instruments that affects enrollment in vocational courses, specifically nonvocational course requirements. Although our results are noisy as the instruments are relatively weak, in these exercises we find little positive effect of additional vocational courses for those induced into taking them, suggesting that students who choose to enroll in vocational coursework are the ones realizing gains, and those induced to enroll due to graduation requirements see little benefit. This is consistent with a positive sorting story, but also one where course requirements induce students into entry level vocational classes, where gains are low. Nonetheless, we take this as further evidence, combined with results in table 4, that policies limiting (upper-level) vocational coursework for students who would have enrolled otherwise would negatively impact wages.

One might also worry that states or localities where returns to vocational training are (expected to be) higher will offer more, or have better funded, vocational programs. Yet, we find that comparing within state-cohort, or even county-cohort, has little impact on results. Lastly, one might worry that specialization might also include multiple introductory courses within a field. Yet, we show that alternatively defining vocational course-taking by concentrators and specialists yields even larger results, and we also demonstrate that increasing introductory courses corresponds with an increasing number of different vocational fields. In addition, we note that this is all true conditional on a very broad set of observable characteristics, courses taken in other fields, a measure of ability, and accounting for postsecondary attendance and attainment. As a result, we interpret our findings as strongly suggestive evidence that those students who choose to take vocational coursework see benefits in the labor market to specialized courses, in particular in technical fields and among those who do not earn a postsecondary degree. This does not imply that nonvocational students would benefit from more (specialized) vocational coursework, but it does strongly suggest that were vocational coursework removed from the curriculum, a sizable share of students, in particular those not planning to attend college, would be worse off.

There are caveats to these conclusions, in addition to those mentioned above. Foremost among these is that we only observe the early careers of NLSY respondents—the mean age in the most recent year of the wage sample is 28.5 years. Recent work by Hanushek et al. (2016) suggests that increasingly specialized training in high school limits workers’ ability to adapt to changing labor market conditions later in life. This is a question we and others would be wise to address in future research with a direct focus on the U.S. context.

## 6.  Conclusion and Policy Implications

We weigh in on a longstanding debate over labor market training in U.S. high schools. Our analysis focuses on factors leading students to take vocational courses and how these courses affect transitions to college and the workforce. We couch our analysis within a sorting framework where students learn about preferences for and skill in academic or vocationally oriented (course)work.

Using detailed course-taking and labor market information from the NLSY97, we find that each additional year of advanced vocational coursework during high school is associated with a near 2 percent increase in wages. This result is robust to an extensive set of individual controls for student background, ability, location, and cohort effects, for any displacement effects on the completion of other courses and effects on postsecondary attainment. Thus we conclude that each additional course of introductory vocational coursework has no benefit (or harm) in the labor market. We interpret this finding as the value of depth over breadth.

Our analysis reveals several subtle yet sharp distinctions that lead to direct policy implications. First, we find that vocational coursework may have informational value, enabling students to make more informed choices about their likely fit with college. Because taking more advanced vocational coursework is associated with lower four-year-college enrollment rates but no reduction in college completion, this implies that students induced out of four-year college by a vocational secondary curriculum may have been least likely to earn a degree. Early exposure to vocational curricula may thus facilitate better college enrollment decisions and fewer ex post “mistakes.” Concerns about high student debt among college dropouts has made improving college enrollment decisions an important policy priority. Second, we demonstrate that whereas wage gains associated with nonvocational courses (core and electives) are entirely explained by college enrollment, wage gains from upper-level vocational courses are unaffected by controlling for college enrollment and completion, suggesting that these courses do in fact have value in the labor market. Lastly, we demonstrate that gains accrue to those students who select into vocational coursework. Those induced into additional courses due to graduation requirements see no benefit. These last two points speak directly to criticisms that vocational education is “preparing students for jobs that don't exist,” or that it is a “dumping ground” for low-ability students. The results we uncover do not support either of these conclusions.

How schools should prepare students for post-graduation—college, the workforce, citizenry—remains a timely policy question. This is nowhere more evident than in the recent shift from “college for all” to “college and career,” and in the simultaneous drive toward increased academic standards through the Common Core, along with an increasing drive toward workforce preparedness in Career-Tech pathways. The research presented here makes a strong case for keeping vocational education an accessible part of the high school curriculum, and that students and policy makers should focus on the value of depth over breadth.

## Notes

1.

The field has moved toward the use of the term “career and technical education,” including in the title of the 2006 Perkins Act reauthorization, to differentiate current career-focused education from past vocational education. Throughout we use the terms “vocational education,” “career-tech,” and “CTE” interchangeably.

2.

Tech-prep is a designation of STW programs focused on workforce training in applied fields, such as manufacturing, automotive, or construction.

3.

For the wage analysis, we construct earned credits as those for which students either received a passing (non-F) or satisfactory mark in the course. For the course-taking analysis credits attempted are the credits that a student would have earned had she passed the course. In cases where the number of Carnegie credits is not reported, we impute Carnegie credits as the modal number of credits that a particular student earned in courses in the same field. When a student lacks a comparable course, we then impute to the modal value of all students taking the same exact course title in the same grade in the full sample.

4.

These definitions are taken from Kreisman and Rangel (2015) who also use NLSY data. We do not restrict wage observations by the number of hours worked, but in a robustness check we restrict to jobs of thirty or more hours per week and find nearly identical results.

5.

Table A.1 in the online Appendix, which can be accessed on Education Finance and Policy’s Web site at www.mitpressjournals.org/doi/suppl/10.1162/edfp_a_00266, describes the full list of vocational courses in the NLSY97.

6.

Results are similar if we include ninth-grade course-taking in columns 1 and 2.

7.

Including math and English grade point averages (GPAs) separately yielded empirically equivalent results.

8.

We estimate the model using ordinary least squares as we include a large number of state-cohort fixed effects, which induces bias in nonlinear models. We verify that the linearity assumption does not change results by reestimating the model without the fixed effects, using both a logistic regression and ordinary least squares.

## Acknowledgments

We thank the Institute for Research on Poverty's Emerging Scholars Grants program and the Smith Richardson Foundation for funding and support, and to Daniela Morar and Julian Hsu for excellent research assistance. We also thank seminar participants at the University of Michigan, the University of Tennessee, and Kansas State University for helpful comments.

## REFERENCES

Agodini
,
Roberto
, and
John
Deke
.
2004
. The relationship between high school vocational education and dropping out.
Princeton, NJ
:
Mathematica Policy Research Report No
.
8879
400
.
Altonji
,
Joseph G.
1995
.
The effects of high school curriculum on education and labor market outcomes
.
Journal of Human Resources
30
(
3
):
409
438
.
Altonji
,
Joseph G.
,
Erica
Blom
, and
Costas
Meghir
.
2012
.
Heterogeneity in human capital investments: High school curriculum, college major, and careers
.
Annual Review of Economics
4
(
1
):
185
223
.
Arcidiacono
,
Peter
.
2004
.
Ability sorting and the returns to college major
.
Journal of Econometrics
121
(
1
):
343
375
.
Attanasio
,
Orazio
,
Kugler
, and
Costas
Meghir
.
2011
.
Subsidizing vocational training for disadvantaged youth in Colombia: Evidence from a randomized trial
.
American Economic Journal: Applied Economics
3
(
3
):
188
220
.
Bishop
,
John H.
, and
Ferran
Mane
.
2004
.
The impacts of career-technical education on high school labor market success
.
Economics of Education Review
23
(
4
):
381
402
.
Bishop
,
John H.
, and
Ferran
Mane
.
2005
.
Raising academic standards and vocational concentrators: Are they better off or worse off?
Education Economics
13
(
2
):
171
187
.
,
Denise
, and
Lisa
Hudson
.
2007
.
The 2007 revision of the career/technical education portion of the secondary school taxonomy: Technical/methodological report
(
NCES 2008-030
).
Washington, DC
:
National Center for Education Statistics, Institute of Education Sciences, U.S. Department of Education
.
Cellini
,
Stephanie Riegg
.
2006
.
Smoothing the transition to college? The effect of tech-prep programs on educational attainment
.
Economics of Education Review
25
(
4
):
394
411
.
Dougherty
,
Shaun
.
2018
.
The effect of career and technical education on human capital accumulation: Causal evidence from Massachusetts
.
Education Finance and Policy
13
(
2
):
119
148
.
Eichhorst
,
Werner
,
Nuria
Rodríguez-Planas
,
Ricarda
Schmidl
, and
Klaus
F. Zimmermann
.
2015
.
A road map to vocational education and training in industrialized countries
.
ILR Review
68
(
2
):
314
337
.
Furstenberg
,
Frank
, Jr.
, and
David
Neumark
.
2005
.
School-to-career and post-secondary education: Evidence from the Philadelphia educational longitudinal study
.
NBER Working Paper No. w11260
.
Goodman
,
Joshua S.
2019
.
The labor of division: Returns to compulsory high school math coursework
.
Journal of Labor Economics
37
(
4
):
1141
1182
.
Gustman
,
Alan L.
, and
Thomas L.
Steinmeier
.
1982
.
The relation between vocational training in high school and economic outcomes
.
ILR Review
36
(
1
):
73
87
.
Hanushek
,
Eric A.
,
Guido
Schwerdt
,
Ludger
Woessmann
, and
Lei
Zhang
.
2016
.
General education, vocational education, and labor-market outcomes over the life-cycle
.
Journal of Human Resources
52
(
1
):
48
87
.
Hudson
,
Lisa
.
2013
.
Data Point: Trends in CTE coursetaking
.
NCES 2014-901
.
Washington, DC
:
National Center for Education Statistics
.
Joensen
,
Juanna S.
, and
Helena S.
Nielsen
.
2009
.
Is there a causal effect of high school math on labor market outcomes
?
Journal of Human Resources
44
(
1
):
171
198
.
Kang
,
Suk
, and
John
Bishop
.
1989
.
Vocational and academic education in high school: Complements or substitutes
?
Economics of Education Review
8
(
2
):
133
148
.
Kreisman
,
Daniel
, and
Marcos
Rangel
.
2015
.
On the blurring of the color line: Wages and employment for black males of different skin tones
.
Review of Economics and Statistics
97
(
1
):
1
13
.
Levine
,
Phillip B.
, and
David J.
Zimmerman
.
1995
.
The benefit of additional high-school math and science classes for young men and women
.
Journal of Business & Economic Statistics
13
(
2
):
137
149
.
Malamud
,
Ofer
.
2011
.
Discovering one's talent: learning from academic specialization
.
ILR Review
64
(
2
):
375
405
.
Malamud
,
Ofer
, and
Cristian
Pop-Eleches
.
2010
.
General education versus vocational training: Evidence from an economy in transition
.
Review of Economics and Statistics
92
(
1
):
43
60
.
Mane
,
Ferran
.
1999
.
Trends in the payoff to academic and occupation-specific skills: The short and medium run returns to academic and vocational high school courses for non-college-bound students
.
Economics of Education Review
18
(
4
):
417
437
.
Maxwell
,
Nan L.
, and
Victor
Rubin
.
2002
.
High school career academies and post-secondary outcomes
.
Economics of Education Review
21
(
2
):
137
152
.
Meer
,
Jonathan
.
2007
.
Evidence on the returns to secondary vocational education
.
Economics of Education Review
26
(
5
):
559
573
.
Neumark
,
David
, and
Mary
Joyce
.
2001
.
Evaluating school-to-work programs using the new NLSY
.
Journal of Human Resources
36
(
4
):
666
702
.
Neumark
,
David
, and
Donna
Rothstein
.
2005
.
Do school-to-work programs help the “forgotten half”?
NBER Working Paper No. 11636
.
Neumark
,
David
, and
Donna
Rothstein
.
2006
.
School-to-career programs and transitions to employment and higher education
.
Economics of Education Review
25
(
4
):
374
393
.
Rose
,
Heather
, and
Julian
R. Betts
.
2004
.
The effect of high school courses on earnings
.
Review of Economics and Statistics
86
(
2
):
497
513
.
Rosenbaum
,
James E.
2001
.
Beyond college for all: Career paths for the forgotten half
.
New York
:
Russell Sage Foundation
.
Stange
,
Kevin M.
2012
.
An empirical investigation of the option value of college enrollment
.
American Economic Journal: Applied Economics
4
(
1
):
49
84
.
Stinebrickner
,
Todd R.
, and
Ralph
Stinebrickner
.
2011
.
Math or science? Using longitudinal expectations data to examine the process of choosing a college major.
NBER Working Paper No. 16869
.
Symonds
,
William C.
,
Robert
Schwartz
, and
Ronald F.
Ferguson
.
2011
.
Pathways to prosperity: Meeting the challenge of preparing young Americans for the 21st century
.
Cambridge, MA
:
Pathways to Prosperity Project, Harvard University Graduate School of Education
.
U.S. Department of Education (USDOE)
.
2014
.
National assessment of career and technical education: Final report to Congress
.
Available
https://www2.ed.gov/rschstat/eval/sectech/nacte/career-technical-education/final-report.pdf.
Accessed 16 May 2019
.
Zietz
,
Joachim
, and
Prathibha
Joshi
.
2005
.
Academic choice behavior of high school students: Economic rationale and empirical evidence
.
Economics of Education Review
24
(
3
):
297
308
.
Zimmermann
,
Klaus F.
,
Constanza
Biavaschi
,
Werner
Eichhorst
,
Giulietti
,
Michael J.
Kendzia
,
Alexander
Muravyev
,
Janneke
Pieters
,
Núria
Rodríguez-Planas
, and
Ricarda
Schmidl
.
2013
.
Youth unemployment and vocational training
.
Foundations and Trends in Microeconomics
9
(
1–2
):
1
157
.