Double_Major

Education Economics Vol. 18, No. 2, June 2010, 167–189 The college double major and subsequent earnings Steven W. Hemelt* University of Maryland, Baltimore County, Baltimore, Maryland, USA shemel1@umbc.edu StevenHemelt 0000002008 00 2008 & Francis Original Article 0964-5292 (print)/1469-5782 EducationFrancis 10.1080/09645290802469931(online) CEDE_A_347161.sgm Taylor andEconomics In this study I examine the relationship between graduating from college with two majors rather than one and labor market earnings using the 2003 National Survey of College Graduates. Because institutions are heterogeneous both in terms of overall quality and in the availability of opportunities to double major, I attempt to control for such overarching institutional differences and explore their effects on premiums to completing a double major. On average, I find a double major to earn 3.2% more than his/her single major counterpart. I also find evidence that premiums to double majoring differ across types of institutions: ranging from a near 4% premium at Research and Comprehensive universities to no effect at Liberal Arts colleges. Finally, I investigate the degree to which choices of first and second major academic disciplines affect earnings premiums. Keywords: double major; earnings return Introduction Over the past few decades, the earnings gap between college and high school educated workers has grown (Lemieux 2006). One response to this has been a marked rise in the proportion of high school graduates enrolling in college (Kane 1999). In times of growing enrollments, students necessarily seek ways in which to set themselves apart. One possible mechanism through which this distinction can occur is the structure of undergraduate education. High school graduates not only make decisions about whether to enroll in college, they also make decisions about what to study while there. One potential structure that has been little studied is the decision to undertake dual academic majors. Although recent studies have investigated the returns to different majors, little is known about the value of the decision to ‘double major,’ or complete two sets of degree requirements. It is important to understand the definition of a double major. A double major is required to complete two sets of requirements – one for each academic discipline that comprises the degree. Double majors ultimately receive one Bachelor’s degree and are therefore distinct from students who earn two separate Bachelor’s degrees. Since double majors at a given institution complete approximately the same number of credits as single majors, the difference of interest lies in the composition of undergraduate education rather than in the amount of education (classes, credits, etc.). Officially declaring a second major is not costless to students, and implies an expectation of some return. Students who double major may anticipate they will be better prepared for the labor market, with a more diverse set of acquired skills. Double major degrees may also act as signals of motivation or adaptability to potential employers. *Email: shemel1@umbc.edu ISSN 0964-5292 print/ISSN 1469-5782 online © 2010 Taylor & Francis DOI: 10.1080/09645290802469931 http://www.informaworld.com 168 S.W. Hemelt There is some anecdotal evidence that the choice to double major has been growing in popularity over the past decade. As early as 1990, the New York Times noted that many Liberal Arts colleges were limiting the number of credits required for a major so that students could take a broader variety of courses (Fowler 1990). Over a decade later, Lewin (2002) cites marked increases in the percentage of double majors at universities across the county. Only 14% of Georgetown’s graduating class of 1996 were double majors, yet by 2002 that figure had climbed to 23%. Washington University in St Louis saw a 14 percentage point increase in the proportion of students declaring double majors between its graduating classes of 1997 and 2002 (Lewin 2002). Along with the increased prevalence of double majors at institutions across the country rages a debate concerning the usefulness of such combinations after college. Some university administrators doubt double majors create any real labor market advantage. Others fear that meeting multiple sets of departmental requirements stops students from pursuing a broader undergraduate education (Lewin 2002). Administrators favorable to the concept of double majors believe that the ability to think clearly and critically across disciplines is of more value in the current labor market than overspecialization (Fowler 1990). In this paper, I raise the question ‘Do double majors generate tangible labor market benefits'’ I focus not only on average earnings effects of double majoring, but explore the degree to which such effects vary by institutional type and estimate earnings effects conditional on choices of primary and secondary academic fields. Estimating the value of a double major over a single major is obviously important to current and future college students. It is also important for university administrators. There may be some level of anxiety among university administrators concerning possible costs double majors impose on universities. Double majors that take more than four years to graduate necessarily lower four-year graduation rates. An important objective for universities is to matriculate, educate, and graduate students. Some institutions may question the value of having students pursue two majors, possibly slowing their graduation. Yet, Del Rossi and Hersch (2008) find no significant difference in time to Bachelor’s degree completion for those with and without double majors. Also, to the extent that double majors ultimately generate higher earnings and better employment for students, such benefits may trickle back to universities in terms of alumni-giving, boosterism, or more non-pecuniary benefits – such as a bolstered academic reputation. In order to estimate returns to double majors, I use data from the 2003 National Survey of College Graduates (NSCG). I make use of the extensive information available on single major choices compared with double major choices, along with other personal background characteristics. Due to the fact that earning a double major at one university can be very different in terms of both program availability and difficulty from another, I also use information on the ‘Carnegie Classification’ of schools provided in the 2003 NSCG to control for institutional differences that might affect decisions about double majoring, as well as economic prospects. In the following section, I provide some background relevant to understanding patterns and outcomes of double majoring. I next describe my empirical approach. I then present my results, and finally consider the implications of the main findings. Background Extant literature has addressed questions surrounding the impact of college major choice on earnings. Finnie and Frenette (2003) examine earnings differences by Education Economics 169 major field of study for three different cohorts (1982, 1986, and 1990) of Canadian Bachelor’s-level university graduates. They find Engineering, Health, Commerce, Computer Science, and Mathematics/Physics to yield the highest returns in terms of earnings; Economics and Education to be in the middle; and other Social Sciences and Agricultural/Biological Sciences to generate the lowest returns. Their model predicted the actual labor market earnings most accurately for Engineering, Computer Science, and other Health graduates. Other studies have investigated how students actually go about choosing between college majors (Berger 1988; Montmarquette, Cannings, and Mahseredjian 2002). Berger (1988) proposes a lifecycle model concerning major choice, and finds some evidence that an individual’s probability of choosing a certain major rises as the present value of the predicted future earnings stream for that major increases. A related question that comprises much of the interest in the value of different college majors is whether or not such distinctions in educational content explain the gender gap in earnings. Several studies have investigated the degree to which college major choice along with course-taking behavior can explain this gap (Daymont and Andrisani 1984; Brown and Corcoran 1997; Eide 1994). The majority of these studies find that college major choice does account for a nontrivial portion of the gender earnings gap. While questions surrounding the underlying mechanisms that cause men and women to pursue different types of majors that are then rewarded differently in the labor market are very interesting, the main focus of this paper is to explore a specific compositional choice open to all college students – the double major. The choice to pursue a double major is certainly informed by personal preferences and ability, but it is probably also affected by the type of university a student attends – in terms of the course and major offerings available, the overall quality of the institution, as well as more intangible ‘cultural’ factors that operate within institution types. A vast literature has examined the effects of college quality on earnings.1 In the face of some countervailing evidence, most of these studies have found statistically significant, modestly sized earnings effects of attending a ‘high-quality’ institution (Zhang 2005). One difficulty in all of these studies is how to measure college quality. Zhang (2005) confronts this issue head on and analyzes the effect of college quality on earnings using four different measures of institutional quality: mean SAT scores of the entering class, Barron’s selectivity rankings, Carnegie Classifications, and in-state undergraduate tuition and fees. No matter the metric used, Zhang (2005) finds the effect of college quality on earnings to be positive and significant. While this may be true, an additional and fairly consistent finding amidst this literature – reaching as far back as James et al. (1989) – states that ‘what matters most is not which college you attend but what you do while you are there …’ (James et al. 1989, 252). This concern over the composition of students’ education while in college is a key motivation for investigating double major effects. In the context of both direct earnings and wage gaps, college major matters. Further, the type of institution a student chooses to attend can hold earnings implications as well. As graduates now enter the workforce with double major degrees, these types of questions must be extended to such dual majors. As with single majors, a second major may provide additional training that imparts specific skills to students that enhance labor force productivity. Double major degrees may also act as signals to employers for flexible and dedicated candidates capable of inter-disciplinary thinking. Del Rossi and Hersch (2008) begin to step beyond examining returns to only single major choices – also using the 2003 NSCG – and estimate returns to different 170 S.W. Hemelt combinations of double majors for both men and women. They find double majoring to increase earnings by 2.3% relative to having a single major. These first estimates are interesting and suggestive. There currently exist no other empirical estimates addressing the question of potential earnings benefits to double majors. The present study seeks to provide additional estimates relevant to the topic and add to the understanding of double major benefits provided by Del Rossi and Hersch (2008). Further, I add to this small existing literature by controlling for university attributes that might affect double major prevalence as well as associated earnings. Given considerable variability across institutions in terms of general quality as well as the availability of opportunities to double major, I develop a model that controls for these overarching differences and estimates average premiums to double majoring. I also define ‘field of study’ groups a bit more specifically. This increased specificity becomes helpful when interpreting results. Finally, I explore the degree to which the academic fields of both first and second majors affect associated double major earnings premiums. Data and methods In order to examine the returns to double majors, I use data from the 2003 NSCG. This was a survey of individuals who received a Bachelor’s or advanced degree from a US institution (prior to April 2000), were living in the United States, and were under the age of 76 years. The National Science Foundation was responsible for the administration of the survey and the compilation of all included data. Given the research question of interest, I work with two main samples. I limit the first sample to only those individuals who possess one Bachelor’s degree (no advanced degrees, no two-year degrees, no multiple Bachelor’s degrees, etc.) in order to examine potential direct earnings benefits to double majoring. I also utilize a second sample in which I re-include all Master’s and other advanced degree holders (PhD, MD, JD, etc.) in order to examine the possibility of more indirect benefits. I limit both main samples to a working-age population (ages 65 years and under) of individuals participating full-time in the labor market at the time of the study. Finally, I restrict both samples to include only four-year institutions characterized as: Research, Doctorate Granting, Comprehensive, or Liberal Arts. I am not interested in two-year schools or in specialized schools of law, art, or music. In total, the working samples contain approximately 26,900 and 49,100 observations, respectively. The outcome variable of interest is gross total earnings from the prior year (2002). The 2003 NSCG also contains a number of important background characteristics on individuals, which include race, ethnicity, parents’ education levels, and the number of children under the age of 18 at home. Main models For the 2003 NSCG samples outlined above, I estimate models of the following basic form: ln( EARN i ) = β1EXPi + β2 Pi + β 3 PEi + β 4 Mi + β5 DMi + α c + ε i (1) where the dependent variable is a measure of all pre-deduction earnings of individual i in 2002. The key independent variables include a vector of 15 college major dummy Education Economics 171 variables (Mi), each representing a single group of majors, and a dummy variable indicating whether or not individual i earned a double (second) major of any type (DMi). To capture the effect of experience on earnings, I include a measure of each individual’s years of general job experience as a quadratic (EXPi). Personal and demographic characteristics of each individual also influence earnings. Pi is a vector of such personal characteristics, including gender, race and ethnicity,2 the number of children under 18 years old living at home, and marital status. Additionally, given differences in labor force experience by gender, I interact the gender dummy with marital status, the number of children under 18 years old at home, and both job experience variables. I include measures of parental education levels for each individual as a measure of some aspect of ability and the number of children under 18 years old living at home as a control for resource demand. The vector of parental education dummies (PEi) is divided into two sets of variables, one for the respondent’s mother and one for the respondent’s father, each structured as follows: high school dropout, high school graduate, some college, college graduate, Master’s degree, advanced/professional degree – with the reference category for each parent as high school dropout. In the main models, I also include institution-type dummies (αc) based on the Carnegie Classifications to control for university attributes that might affect double major prevalence as well as associated earnings. In all models, heteroskedasticityrobust standard errors are calculated as observations are weighted by the inverse of the probability that the individual is included in the sample.3 Single and double major choices The 2003 NSCG collects information on first and second majors for each individual’s first Bachelor’s degree, most recent degree, and highest degree. For each, respondents are asked to list their ‘primary field of study’ and then ‘second major,’ if applicable. As the primary interest here is on the effects of undergraduate double majoring, I derive all single and double major variables from information on respondents’ first Bachelor’s degree. In order to meaningfully categorize majors within groups, I sort the over 150 specific major codes provided in the 2003 NSCG into 15 major categories. This grouping scheme is based upon the actual survey’s grouped list of academic fields provided to respondents and the common institutional setup of ‘schools’ within universities (i.e. Arts, Sciences, Education, Engineering, etc.). Table 1 presents a list of these categories along with the specific academic majors they represent. I model double major choices in three ways. First, I include a general measure of double major directly, along with the single major choices. This double major dummy is not mutually exclusive to the set of first/single major academic categories. Allowing information on observations to vary in this manner generates the ‘average premium’ attributable to any double major (as the coefficient on the double major dummy variable). I then interact the general double major dummy variable (DMi) with all of the first major categories (Mi). The coefficients on these interaction terms show the average premium of adding a second major to a known first major. Finally, I break up the general double major dummy into specific categories of second majors (labeled in the same fashion as the first majors) in order to investigate the average premiums associated with adding a specific type (field) of second major to any first major. Ideally, one would want to control for selection into the double major track in all of the above models. Unfortunately, given the data in the 2003 NSCG, there is no way 172 Table 1. S.W. Hemelt List of academic major categories. (8) Languages and Linguistics English language, literature, and letters Linguistics Other Foreign languages and literature (9) Health Sciences Audiology and speech pathology Health services and administration Health/medical assistants Health/medical technologies Medical preparatory programs (e.g. pre-dentistry, pre-medical, pre-veterinary) Nursing (four years or longer program) Pharmacy Physical therapy and other rehabilitation or therapeutic services Public health (including environmental health and epidemiology) Other health/medial sciences (1) Agricultural Sciences Animal sciences Food sciences and technology Plant sciences Agricultural economics Other agricultural business and sciences (2) Architecture Architecture Environmental design (3) Biological/Life Sciences Biochemistry and biophysics Biology, general Botany Cell and molecular biology Ecology Genetics, animal and plant Conservation and natural resources Microbiological sciences and immunology Nutritional sciences Pharmacology, human and animal Zoology, general Other biological sciences (10) Social Sciences Psychology (clinical, counseling, educational, experimental, general, industrial/organizational, social, and other) Public affairs (4) Business and Administration Public policy Accounting Anthropology and archeology Actuarial science Criminology, criminal justice services Business administration and management Home economics Business, general Pre-law studies Business and managerial economics Economics Business marketing/marketing management Geography Financial management History, history of science Marketing research International relations Operations research Political science and government Communications, general Area and ethnic studies Journalism Social work Other business, administration, communications Other social sciences (5) Computer and Information Sciences Computer and information sciences, general Computer programming Computer science Computer systems analysis (11) Physical Sciences Astronomy and astrophysics Atmospheric sciences and meteorology Chemistry (except biochemistry) Earth sciences Education Economics Table 1. (Continued). Geology, geological studies Oceanography Physics Other physical sciences 173 Data processing Information services and systems Other computer and information sciences (6) Education Education administration (12) Liberal Arts Computer teacher education Liberal arts and general studies Counselor education and guidance Library science Elementary teacher education Other liberal arts Mathematics teacher education Physical education and coaching (13) Math and Statistics Applied mathematics Pre-school/kindergarten/early childhood teacher education Mathematics, general Science teacher education Statistics Special education Other mathematics Social science teacher education Other education (14) Visual and Performing Arts (7) Engineering Dramatic arts Aerospace, aeronautical and astronautical Fine arts, all fields engineering Music, all fields Agricultural engineering Other visual and performing arts Architectural engineering Bioengineering and biomedical engineering (15) Other Fields Chemical engineering Fields not listed Civil engineering Parks, recreation, leisure, and fitness Computer and systems engineering studies Electrical, electronics, and communications engineering Engineering sciences and mechanics Environmental engineering Engineering, general Geophysical and geological engineering Industrial and manufacturing engineering Materials engineering, ceramics and textiles Mechanical engineering Metallurgical engineering Mining and minerals engineering Naval architecture and marine engineering Nuclear engineering Petroleum engineering Engineering-related technologies Other engineering 174 S.W. Hemelt to completely sort out the effects of ‘motivation’ on earnings from the strict effects of obtaining a double major degree. The survey does not include exogenous information on double major choice that could be exploited to control for such selection issues. Even theoretically it is difficult to conceive of a potential instrumental variable that would be related to whether or not a student earns a double major but not related to his/her future earnings. In the absence of such information, I estimate models that control for some key institutional differences expected to affect both earnings and double major prevalence. While the causality of these results should be interpreted with some caution, they provide very interesting and suggestive descriptive evidence surrounding earnings benefits to double majors. Carnegie classification institutional model A quick glance through any four-year university’s course offerings and degree requirements will attest to the sometimes complicated and very diverse group of requirements specifying how one can ‘double major.’ Some universities encourage double majoring, while others have little well-documented information on the topic. Additionally, some double major programs may apply credits to dual-degree seekers more generously than others. This wide range of requirements, informational access, incentives, and difficulty levels surrounding double majors makes general inferences regarding the topic quite challenging. There are several reasons why prevalence rates and returns to double majoring may be expected to vary across institutional types. For one, institutional barriers to double majoring across colleges within large, multi-college universities are likely to be much greater than at smaller, single-college Liberal Arts institutions. Second, in schools where double majoring is therefore more prevalent, the additional return to a double major, above and beyond a strong single major, may be insignificant. Third, at larger institutions with greater barriers to double majoring, successfully navigating the requirements to do so may be a mechanism through which students choose to ‘set themselves apart’ from their peers. Considering this, at types of institutions where double majoring is relatively more cumbersome, the premiums attributable to an individual’s choice to complete a double major may be larger. Finally, different universities are likely to have distinct policies on how one can double major. These institutional policy differences may also affect the premiums attributable to double majoring. To the extent that a double major both imparts additional substantive skills to students and signals to employers a candidate of high personal motivation and discipline, the type of university where the double major was earned matters. Controlling for these across-university differences is therefore crucial in assessing overall as well as field-specific returns to double-majoring. A best-case scenario for assessing such returns would include data where institutions are uniquely labeled. Within each institution would also be a significant number of individual observations. This would allow for the estimation of a true university fixed-effects model that exploits only the variation in earnings of single and double majors within a given university. The 2003 NSCG does not include university-specific identifiers. Yet, even in the absence of such identifiers, the need to address important university differences persists. The question then becomes a matter of grouping – how to group like universities. The 2003 NSCG includes data that classify each university using the ‘Carnegie Classification’ system of universities. This basic classification system was developed Education Economics 175 by the Carnegie Commission on Higher Education in 1970.4 One of the intents in creating such categories was to meaningfully label institutional differences. The classification system itself takes into account the types degree programs offered and research activity, both of which measure some aspect of quality. The main groups include Research, Liberal Arts, Comprehensive, Doctorate Granting, two-year institutions, specialized schools of medicine, art, music, and more. The 2003 NSCG classifies universities into a total of 17 groups. Of these 17 groups, eight represent traditional four-year institutions that educate undergraduates. It is these eight that are of primary interest.5 In order to compare double majors with single majors at similar types of universities, I estimate the main models including controls for each classification category. By exploiting the variation in double major choice only within each of these categories, more accurate return estimates are obtained. Results In order to get a feel for the prevalence of double majoring, Figure 1 graphs the mean percentage of students earning double majors over time based on the cross-sectional ‘year of graduation’ information available in the 2003 NSCG, for both main samples. Contrary to the anecdotal evidence provided earlier that the choice to double major has been rising in popularity, these data show a downward trend in double major degree prevalence among graduates.6 Yet, this trend could be caused by a variety of factors not directly related to dual major ‘popularity’ among students. For one, institutional policies concerning double majors may have undergone significant changes throughout these decades – making it relatively more or less difficult for students to complete the necessary requirements. Also, an individual’s conception of a ‘double major’ is likely to have changed over time. Therefore, a graduate from an earlier decade who now works in an area relatively unrelated to his/her college major may in fact record that they were a double major in college. These types of definitional Figure 1. Double major prevalence. Mean (weighted) % of double majors 50.00 40.00 30.00 20.00 10.00 0.00 Grad BA Only Before 1970s 36.63 39.24 1970Ð1980 29.36 28.79 1980Ð1990 23.00 21.80 1990Ð2000 20.23 18.77 Cohorts (by year of graduation) Figure 1. Double major prevalence. 176 S.W. Hemelt changes are sure to affect survey reporting and prevalence rates of double majoring. Nevertheless, from this trend, it is clear that double majors have made up and continue to make up a nontrivial portion of graduating collegiate degree-holders. General double major premium In the sample containing only Bachelor’s degree holders, 23.41% are double majors. When graduate degree holders are included, this percentage increases to 24.91 – indicating that those with graduate degrees are a bit more likely to have an undergraduate double major. Some of the most popular first majors in both samples are Business and Administration, Social Sciences, Education, and Engineering. Further, individuals with Business and Administration, Social Sciences, or Education first majors are also most likely to have pursued a double major. I present additional descriptive statistics that further characterize both samples in Table 2. From the development of the institutional controls model in columns (1), (2), and (3) of Table 3, it is clear that double majors command a greater market return than their single major counterparts – with an average premium of 3.2%. This premium is not significantly affected by including measures of parental education levels, suggesting that the double major effect is not due to differences among graduates in family educational background or values. The general double major premium is also not substantially affected by including the controls for institutional differences described earlier. This further suggests that the earnings effects of double majoring are not due to general differences across academic institutions. Even for a college graduate with a relatively lower-earning major, this premium implies a nontrivial potential increase in earnings. Evaluated at the most common sample characteristics, a female Liberal Arts single major is expected to earn a yearly salary of approximately $32,300 as a fresh college graduate while a male Liberal Arts single major is predicted to early a starting yearly salary of $35,200. Even for these lower-earning majors, the 3.2% general premium of adding a second (double) major translates into an extra $1033 or $1126 in annual earnings, respectively. Simply using the mean salary for the ‘Bachelor’s Only’ sample, a 3.2% premium implies a $2087 earnings increase. Premiums conditional on academic fields of majors While this general premium of adding any second major to an existing first major is an important starting point, it is reasonable to assert that the academic field to which a double major is added, as well at the academic field of the second major itself, could have important impacts on the size and direction of this premium. The general premium attributable to double majoring is smaller than the difference in earnings returns between most single majors. Yet, to the extent that a second major affects labor market productivity differently than simply switching the academic area of a single major, these dependencies must be investigated. Students often have a primary major in mind when considering the choice of whether or not to add a second (or double) major. The first interacted model I develop addresses the following question: Conditional on a specific academic first major field, what is the premium attributable to adding any second/double major' Column (4) of Table 3 presents the results of this model. I find evidence that five groups of first majors can expect significant premiums from adding a second major. These premiums Table 2. n Mean Standard error Variable n Mean Descriptive statistics. Standard error Variable Education Economics Double major interactions Double major Agricultural Sciences + DM Architecture + DM Biological/Life Sciences + DM Business and Administration + DM Computer Sciences + DM Education + DM Engineering + DM Languages and Linguistics + DM Health Sciences + DM Social Sciences + DM Physical Sciences + DM Liberal Arts + DM Math and Statistics + DM Visual and Performing Arts + DM Other Field + DM 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 0.234 0.005 0.008 0.009 0.075 0.007 0.035 0.011 0.010 0.005 0.047 0.005 0.007 0.005 0.008 0.005 0.004 0.001 0.000 0.001 0.002 0.001 0.002 0.001 0.001 0.001 0.002 0.000 0.001 0.000 0.001 0.001 177 Bachelors only sample Double major Agricultural Sciences major Architecture major Biological/Life Sciences major Business and Administration major Computer Sciences major Education major Engineering major Languages and Linguistics major Health Sciences major Social Sciences major Physical Sciences major Liberal Arts major Math and Statistics major Visual and Performing Arts major Other Field major Male African-American, non-Hispanic White, non-Hispanic Asian, non-Hispanic Hispanic Admerican Indian, non-Hispanic Multiple race, non-Hispanic Married Children under 18 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 26914 0.234 0.025 0.009 0.042 0.316 0.033 0.121 0.085 0.028 0.056 0.167 0.018 0.025 0.015 0.035 0.026 0.576 0.071 0.831 0.030 0.049 0.005 0.012 0.707 0.900 0.004 0.001 0.001 0.001 0.004 0.001 0.003 0.002 0.001 0.002 0.003 0.001 0.001 0.001 0.002 0.002 0.004 0.002 0.003 0.001 0.001 0.000 0.008 0.004 0.010 178 Table 2. n 26914 26914 20.418 65223.11 0.084 540.20 Mean Standard error Variable n Mean (Continued). Standard error Variable S.W. Hemelt Experience Wages Graduate degrees sample Double major Agricultural Sciences major Architecture major Biological/Life Sciences major Business and Administration major Computer Sciences major Education major Engineering major Languages and Linguistics major Health Sciences major Social Sciences major Physical Sciences major Liberal Arts major Math and Statistics major Visual and Performing Arts major Other Field major Master’s Advanced Degree (PhD, MD, etc.) Male African-American, non- Hispanic White, non-Hispanic 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 0.249 0.021 0.008 0.058 0.250 0.026 0.140 0.082 0.038 0.057 0.194 0.026 0.028 0.017 0.032 0.022 0.277 0.100 0.563 0.068 0.837 0.003 0.001 0.001 0.001 0.003 0.001 0.002 0.001 0.001 0.001 0.003 0.001 0.001 0.001 0.001 0.001 0.003 0.002 0.003 0.001 0.002 Double major interactions Double major Agricultural Sciences + DM Architecture + DM Biological/Life Sciences + DM Business and Administration + DM Computer Sciences + DM Education + DM Engineering + DM Languages and Linguistics + DM Health Sciences + DM Social Sciences + DM Physical Sciences + DM Liberal Arts + DM Math and Statistics + DM Visual and Performing Arts + DM Other Field + DM 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 49139 0.249 0.004 0.001 0.015 0.069 0.006 0.041 0.010 0.015 0.006 0.058 0.007 0.008 0.006 0.006 0.004 0.003 0.000 0.000 0.000 0.002 0.000 0.001 0.001 0.001 0.000 0.001 0.000 0.001 0.000 0.000 0.000 Table 2. n 49139 49139 49139 49139 49139 49139 49139 49139 0.031 0.046 0.004 0.012 0.719 0.885 20.302 72601.47 0.001 0.001 0.000 0.001 0.003 0.007 0.063 454.33 Mean Standard error Variable n Mean (Continued). Standard error Variable Asian, non-Hispanic Hispanic American Indian, non-Hispanic Multiple race, non-Hispanic Married Children Under 18 Experience Wages Education Economics Note: All descriptives are weighted by the final probability weight provided by and specific to the 2003 NSCG. DM, double major. 179 180 Table 3. S.W. Hemelt Premiums of a second major: conditional on choices of academic disciplines. Dependent variable: ln(total 2002 earnings) Variable Column (1) Column (2) Column (3) Column (4) Column (5) Yes Yes Yes Yes Added controls Include parental education levels' No Yes Yes Include controls for institutional No No Yes differences' First/single majors Agricultural Sciences −0.022 −0.02 −0.036 (0.047) (0.046) (0.046) Architecture 0.201 0.197 0.166 (0.069)*** (0.068)*** (0.067)** Biological/Life Sciences 0.022 0.022 0.017 (0.047) (0.041) (0.040) Business and Administration 0.198 0.201 0.212 (0.034)*** (0.034)*** (0.034)*** Computer Sciences 0.322 0.325 0.334 (0.039)*** (0.039)*** (0.039)*** Education −0.125 −0.117 −0.098 (0.036)*** (0.036)*** (0.036)*** Engineering 0.317 0.317 0.305 (0.037)*** (0.036)*** (0.036)*** Languages and Linguistics 0.064 0.056 0.040 (0.046) (0.046) (0.045) Health Sciences 0.194 0.203 0.212 (0.037)*** (0.037)*** (0.037)*** Social Sciences 0.050 0.051 0.050 (0.036) (0.036) (0.035) Physical Sciences 0.197 0.198 0.196 (0.048)*** (0.048)*** (0.047)*** Liberal Arts 0.020 0.015 0.016 (0.056) (0.056) (0.056) Math and Statistics 0.216 0.218 0.223 (0.052)*** (0.051)*** (0.051)*** Visual and Performing Arts −0.093 −0.095 −0.094 (0.047)** (0.047)** (0.046)** General double major dummy 0.026 0.026 0.032 (0.013)** (0.013)** (0.013)** Conditional on first major discipline Biological/Life Sciences + DM Computer Sciences + DM Health Sciences + DM −0.044 −0.024 (0.049) (0.046) 0.157 0.164 (0.070)** (0.068)** −0.005 0.014 (0.045) (0.041) 0.210 0.202 (0.036)*** (0.033)*** 0.315 0.323 (0.043)*** (0.039)*** −0.098 −0.091 (0.038)*** (0.035)** 0.295 0.295 (0.038)*** (0.036)*** 0.014 0.044 (0.052) (0.045) 0.221 0.210 (0.039)*** (0.037)*** 0.043 0.053 (0.038) (0.035) 0.150 0.189 (0.054)*** (0.048)*** −0.008 0.017 (0.062) (0.056) 0.189 0.211 (0.061)*** (0.051)*** −0.107 −0.091 (0.052)** (0.047)* 0.100 (0.048)** 0.089 (0.043)** −0.130 (0.054)** Education Economics Table 3. (Continued). Dependent variable: ln(total 2002 earnings) Variable Physical Sciences + DM Mathematics and Statistics + DM Conditional on second major discipline First major + Business and Administration First major + Computer Sciences First major + Engineering Observations R-squared 26914 0.156 26914 0.160 26914 0.169 26914 0.169 181 Column (1) Column (2) Column (3) Column (4) Column (5) 0.185 (0.069)*** 0.123 (0.070)* 0.084 (0.023)*** 0.176 (0.047)*** 0.143 (0.047)*** 26914 0.170 Note: Robust standard errors in parentheses. *Significant at 10%, **significant at 5%, ***significant at 1%. Models also include personal covariates and measures of general job experience. Reference category is a single major in ‘Other Fields.’ While models in columns (4) and (5) control for all possible interactions, only the statistically significant pairs are presented. range from −13% for students with a first major in Health Sciences to 18.5% for students with a first major in Physical Sciences. From an earnings standpoint, it appears relatively more beneficial to those majoring in Health Sciences to remain single majors. The second interacted model I develop addresses a related question: What is the premium of adding a specific academic second major field to any first major' The results of this model are presented in column (5) of Table 3. This model will also allow us to examine more closely the academic components of the most profitable double majors. Regardless of one’s first major, there are three types of second majors that appear to be the most profitable potential additions: Business and Administration, Computer Science, and Engineering. Of those double majors who chose a second major in Business and Administration, 39% had first majors in Accounting, Business Management, or Financial Management. Other first majors likely to add a Business and Administration field as a second major were Marketing, Communications, Computer Sciences, Information Systems, Economics, Political Science, General Psychology, and Sociology. The Computer Science second majors declared first majors most frequently in the following areas: Mathematics, Electrical/Electronic Engineering, Accounting, General Business, and Financial Management. Finally, Engineering second majors frequently were individuals who ‘doubled up’ on engineering disciplines. Over 70% of these double majors had first majors in another engineering discipline. Patterns of double majoring among less technical majors, such as Education and Languages/Linguistics second majors, are also quite interesting. For example, outside those Education double majors who studied two related areas of education, the remaining double majors studied generally very content-based first majors, including Mathematics (3.3%), Biology (3%), English Language (5.1%), Foreign Languages (2.3%), and History (5%). Given that education majors generally earn less, it therefore Table 4. Dependent variable: ln(total 2002 earnings) Column (1) 0.037 (0.010)*** 0.037 (0.010)*** 0.043 (0.010)*** Column (2) Column (3) Column (4) Average premium to double major: allowing for graduate degrees. 182 Variable Double major Master’s degree 0.032 (0.010)*** 0.159 (0.010)*** Advanced degree (PhD, professional) Agricultural Sciences Architecture Biological/Life Sciences S.W. Hemelt Business and Administration Computer Sciences Education Engineering Languages and Linguistics Health Sciences Social Sciences −0.022 (0.040) 0.183 (0.057)*** 0.248 (0.032)*** 0.180 (0.029)*** 0.295 (0.034)*** −0.061 (0.030)** 0.322 (0.030)*** 0.126 (0.034)*** 0.238 (0.033)*** 0.111 (0.030)*** −0.013 (0.039) 0.179 (0.056)*** 0.242 (0.032)*** 0.189 (0.029)*** 0.301 (0.033)*** −0.047 (0.029) 0.326 (0.030)*** 0.115 (0.034)*** 0.248 (0.033)*** 0.111 (0.029)*** −0.034 (0.039) 0.140 (0.056)** 0.232 (0.032)*** 0.206 (0.029)*** 0.318 (0.033)*** −0.019 (0.029) 0.306 (0.030)*** 0.102 (0.034)*** 0.257 (0.033)*** 0.110 (0.029)*** 0.537 (0.015)*** −0.030 (0.039) 0.177 (0.055)*** 0.100 (0.031)*** 0.223 (0.028)*** 0.347 (0.033)*** −0.055 (0.029)* 0.310 (0.030)*** 0.029 (0.033) 0.205 (0.032)*** 0.049 (0.029)* Table 4. Dependent variable: ln(total 2002 earnings) Column (1) Column (2) Column (3) Column (4) (Continued). Variable Physical Sciences Liberal Arts Mathematics and Statistics Visual and Performing Arts Education Economics Include master’s and other advanced degree holders in sample' Include controls for master’s and other advanced degrees' Include controls for parental education levels' Include controls for institutional differences' Observations R-squared 0.302 (0.035)*** 0.085 (0.041)** 0.252 (0.040)*** −0.131 (0.038)*** Yes No No No 49139 0.149 0.296 (0.035)*** 0.074 (0.041)* 0.253 (0.039)*** −0.136 (0.039)*** Yes No Yes No 49139 0.154 0.286 (0.035)*** 0.077 (0.041)* 0.261 (0.038)*** −0.131 (0.038)*** Yes No Yes Yes 49139 0.167 0.168 (0.034)*** 0.014 (0.040) 0.213 (0.037)*** −0.129 (0.038)*** Yes Yes Yes Yes 49139 0.211 Note: Robust standard errors in parentheses. *Significant at 10%, **significant at 5%, ***significant at 1%. Reference category is a single major in ‘Other Fields.’ All models also include personal covariates and measures of general job experience. 183 184 S.W. Hemelt makes sense to see an insignificant or even negative premium attributed to adding a second major in Education, especially for students who added an Education field to a more technical or content-based first major field.7 Not surprisingly, many of the second majors in Languages/Linguistics had first majors from the Education group, namely Elementary Education (7%) and Secondary Education (9.3%). Generally speaking, the addition of more technical field as a second major tends to increase earnings premiums relative to only a single major. This is often true even when the first major itself is quite technical. This was true of those individuals who added an Engineering field as a second major, and is also true of those who added a Mathematics or Statistics field as a second major – over 45% of these individuals had first majors in either Computer Science, Mathematics, Chemistry, or Physics. Institutional differences While the academic fields of both the first and second major that comprise a double major affect premiums, differences across institutions in terms of quality may also give rise to different double major earnings effects. As previously mentioned, one factor affecting the Carnegie Classifications used here to control for general institutional differences is the type and quantity of graduate degree programs offered (Master’s and Doctorate/Professional degrees). Therefore, in earnings analyses by type of institution, it will be important to both re-incorporate and control for graduate degree holders trained at different institution types. As a first step, I re-estimate the same progression of models outlined in Table 3 (columns (1)–(3)) on a sample that includes individuals who possess advanced degrees beyond their first Bachelor’s degree. An important question related to the potential earnings returns of double majors is whether or not they also generate other benefits besides those directly tied to earnings. For one, earning a double major may increase the likelihood of pursuing a graduate degree. Subsequently, double majors may indirectly affect future labor market earnings through this ‘graduate school effect.’ Table 4 presents the results of this first step. Examining the point estimates for the average premium generated by a double major reveals possible additional benefits to double majoring. The coefficient on the double major dummy increases by about 1% when graduate degree holders are included, without explicitly controlling for their respective graduate degrees. This implies that the double major effect may now be picking up some of the earnings benefits to holding a graduate degree. Once control variables for these degrees are included in the model, the coefficient on the double major dummy drops back to 3.2%. This movement in the average premium suggests that double majors produce both direct earnings benefits as well as indirect earnings benefits through the increased likelihood of obtaining a graduate degree. Continuing with the sample containing graduate degree holders, I next estimate four separate regressions by institution type (Research, Doctorate Granting, Liberal Arts, and Comprehensive) in order to investigate potential differences across institutions in double major returns. In all four models, Master’s and other advanced degrees (PhD, MD, etc.) command highly significant and consistent earnings premiums – on the order of 15–18% for a Master’s degree, and between 52% and 55% for a doctorate or other advanced professional degree. Further examining the results in Table 5, the premium attributable to a double major appears to vary across types of institutions – from an insignificant effect at Liberal Arts colleges to a significant premium of close Education Economics Table 5. Average premium to double major: conditional on type of institution. Dependent variable: ln(total 2002 earnings) Variable Double major Research Doctorate Granting 0.040 (0.029) 0.180 (0.026)*** 0.555 (0.035)*** −0.120 (0.155) 0.240 (0.124)* 0.140 (0.085)* 0.261 (0.081)*** 0.391 (0.085)*** −0.051 (0.083) 0.263 (0.086)*** −0.013 (0.089) 0.174 (0.096)* 0.007 (0.081) 0.188 (0.094)** −0.003 (0.101) 0.163 (0.107) −0.180 (0.097)* 23.79 7128 0.191 185 Liberal Arts Comprehensive −0.005 (0.023) 0.180 (0.024)*** 0.533 (0.047)*** −0.001 (0.109) 0.604 (0.134)*** 0.131 (0.068)* 0.135 (0.061)** 0.315 (0.075)*** −0.167 (0.062)*** 0.209 (0.068)*** −0.033 (0.071) 0.073 (0.102) 0.022 (0.062) 0.122 (0.075) −0.032 (0.076) 0.210 (0.076)*** −0.209 (0.091)** 29.63 6398 0.235 0.038 (0.015)** 0.157 (0.015)*** 0.523 (0.027)*** 0.028 (0.059) 0.252 (0.104)** 0.065 (0.050) 0.198 (0.045)*** 0.345 (0.051)*** −0.038 (0.044) 0.302 (0.053)*** 0.095 (0.055)* 0.212 (0.047)*** 0.028 (0.046) 0.205 (0.064)*** 0.037 (0.071) 0.208 (0.065)*** −0.074 (0.062) 26.53 15734 0.172 0.039 (0.018)** Master’s 0.150 (0.017)*** Advanced Degree (PhD, MD, etc.) 0.549 (0.021)*** Agricultural Sciences −0.039 (0.060) Architecture 0.153 (0.078)** Biological/Life Sciences 0.102 (0.053)* Business and Administration 0.248 (0.049)*** Computer Sciences 0.329 (0.059)*** Education −0.082 (0.052) Engineering 0.344 (0.048)*** Languages and Linguistics 0.046 (0.057) Health Sciences 0.235 (0.052)*** Social Sciences 0.091 (0.050)* Physical Sciences 0.167 (0.055)*** Liberal Arts 0.022 (0.075) Mathematics and Statistics 0.218 (0.063)*** Visual and Performing Arts −0.138 (0.064)** Mean (weighted) % double majors 21.95 Observations 19879 R-squared 0.208 Note: Robust standard errors in parentheses. *Significant at 10%, **significant at 5%, ***significant at 1%. Reference category is a single major in ‘Other Fields.’. All models also include personal covariates, measures of general job experience, and controls for parenteral education levels. 186 S.W. Hemelt to 4% for both Research and Comprehensive universities.8 In order to verify the real presence of premium differences, I estimate a model including dummies for these four institution types as well as interacted terms between these dummies and the general double major dummy. The interacted terms (double major*university type) are jointly significant at the 5% level, substantiating the presence of real institutional differences in double major earnings effects.9 As hypothesized earlier, one might expect the general return to double majoring to be affected by how ‘set apart’ choosing to double major makes a given student compared with the rest of the study body. The estimates presented in Table 5 at least anecdotally support this idea. At Liberal Arts institutions, where nearly 30% of students are double majors, the earnings premium to a double major is found to be negative, extremely small, and statistically insignificant. At Research institutions, where only 22% of students earn double majors, the premium to doing so is statistically significant with a magnitude of about 3.9%.10 Women versus men One final way in which premiums to double majoring may differ is by gender. Given the fact that labor market experiences can differ quite substantially for men and women, it is important to further consider how these differences may affect premiums to double majoring for each group. To more formally examine whether or not real differences between men and women exist in the context of this earnings question, I perform a Chow test. This test essentially asks whether or not significant enough differences exist between the intercepts and slope coefficients for men and women to justify estimating completely separate models for these two groups. I conclude that such differences do exist.11 Therefore, Table 6 presents results for men, women, and again for the full sample.12 Upon examining the double major dummy coefficients separately estimated for men and women, at first glance it looks as though women may stand to benefit more from choosing to double major than men. Yet, this is not the case. These two genderspecific coefficients are not statistically different from one other,13 implying that both men and women stand to gain from double majoring. While both men and women may glean earnings benefits from double majoring, women do double major at a higher rate than men. I find 25.1% of women to be double majors, compared with only 22.2% of men. This difference in mean proportions is statistically significant at the 1% level, and contextualizes well the insignificance of the difference between the genderspecific double major dummy coefficients. Therefore, overall, we can say that while women appear more likely to double major than men, both stand to equally reap the earnings premium attributable to double majoring. Conclusions The present analysis provides evidence that obtaining a double major leads to nontrivial increases in labor market earnings, above and beyond the returns to specific single majors. On average, I find a double major to earn 3.2% more than his/ her single major counterpart, controlling for a variety individual background characteristics as well as institutional differences. In addition to this direct earnings effect, I find some evidence that double majors produce indirect earnings benefits through a ‘graduate school’ effect. Education Economics Table 6. Average premium to double major: men versus women. 187 Dependent variable: ln(total 2002 earnings) Variable Double major Agricultural Sciences Architecture Biological/Life Sciences Business and Administration Computer Sciences Education Engineering Languages and Linguistics Health Sciences Social Sciences Physical Sciences Liberal Arts Mathematics and Statistics Visual and Performing Arts Include parental education levels' Include controls for institutional differences' Mean (weighted) % double majors Observations R-squared Full sample 0.032 (0.013)** −0.036 (0.046) 0.166 (0.067)** 0.017 (0.040) 0.212 (0.034)*** 0.334 (0.039)*** −0.098 (0.036)*** 0.305 (0.036)*** 0.040 (0.045) 0.212 (0.037)*** 0.050 (0.035) 0.196 (0.047)*** 0.016 (0.056) 0.223 (0.051)*** −0.094 (0.046)** Yes Yes 23.41 26914 0.169 Men 0.027 (0.019) −0.091 (0.057) 0.117 (0.084) −0.016 (0.049) 0.201 (0.044)*** 0.324 (0.048)*** −0.121 (0.049)** 0.281 (0.045)*** 0.011 (0.072) 0.220 (0.054)*** 0.083 (0.048)* 0.187 (0.057)*** −0.043 (0.076) 0.207 (0.063)*** −0.109 (0.068) Yes Yes 22.18 16884 0.115 Women 0.039 (0.017)** 0.089 (0.075) 0.314 (0.088)*** 0.059 (0.067) 0.215 (0.051)*** 0.337 (0.068)*** −0.085 (0.052) 0.467 (0.092)*** 0.065 (0.060) 0.208 (0.053)*** 0.004 (0.052) 0.192 (0.092)** 0.097 (0.078) 0.243 (0.085)*** −0.077 (0.063) Yes Yes 25.08 10030 0.089 Note: Robust standard errors in parentheses. *Significant at 10%, **significant at 5%, ***significant at 1%. Reference category is a single major in ‘Other Fields.’ All models also include personal covariates as well as measures of general job experience. I also find evidence that premiums to double majoring vary by the first major to which second majors are added, by the discipline of the second major itself, and by university type. Some of the most profitable combinations involve the addition of technical or business-related second majors to a fairly wide range of first majors. 188 S.W. Hemelt These results also speak to the ongoing policy debate in universities nationwide over the usefulness of such degrees in the post-college labor market. The evidence presented here is quite suggestive of a positive earnings benefit to double majoring. Due to the fact that selection into a double major retains some non-random elements by way of individual student choice, it is not entirely clear how much of this benefit is due to the dual major degree itself and how much is due an intangible personal factor like ‘motivation.’ While these estimates may represent an upper bound on the positive earnings premium associated with a double major degree, it is encouraging that this effect is not diminished by models that control for relevant personal background characteristics as well as important institutional differences likely to affect who gets a double major as well as one’s future labor market prospects. Additionally, I investigate the sensitivity and robustness of these double major premium results to changes in both first and second major academic areas and to changes in institution type. To the extent that double majoring produces tangible labor market earnings benefits to students, these results are of potential help to administrators making decisions regarding the restructuring of double major programs as well as decisions about the level of support such programs and students receive. Acknowledgements Thanks to Dave Marcotte, Lisa Dickson, and two anonymous referees for very helpful comments and suggestions. Of course, any errors and all opinions are the author’s own. Notes 1. For a good exposition of the development of this particular literature, see Zhang (2005). 2. The vector of race and ethnicity dummies is derived from a variable that treats the ‘Hispanic’ ethnicity category as mutually exclusive to the other race categories. 3. These survey-specific probability weights are provided in the 2003 NSCG. 4. Carnegie Foundation for the Advancement of Teaching’s (2006) ‘Basic Classification 5. 6. 7. 8. 9. 10. Technical Details’ – see http://www.carnegiefoundation.org/classifications/index.asp' key=798. The eight classifications are: Research I, Research II, Doctorate Granting I, Doctorate Granting II, Comprehensive I, Comprehensive II, Liberal Arts I, and Liberal Arts II. These differences in weighted sample mean proportions are each statistically significant at the 1% level: comparisons of the future decade’s mean proportion of double majors with the preceding decade’s mean proportion of double majors yield z-scores of −9.18, −12.87, and −6.14, respectively, for the sample including graduate degree holders, and z-scores of −8.03, −10.23, and −5.32, respectively, for the sample containing only Bachelor’s degree holders. The relevant premium point estimates and standard errors are as follows: an Education first major who adds any second major (Table 3, column (4) model) can expect an insignificant premium of 0.012 (0.029), and an individual adding an Education field as a second major to a given first major (Table 3, column (5) model) can expect an insignificant premium of −0.007 (0.027) . It should be noted that three of the four double major dummy point estimates are nearly the same (Research, Doctorate Granting, and Comprehensive), yet in the case of Doctorate Granting institutions, the associated standard error is over 50% larger, accounting for its insignificance. Clearly there is a different premium to double majoring at a Liberal Arts school, compared with the rest. The estimates obtained using the ‘Bachelor’s Only’ sample in place of the ‘Graduate Degree Holders’ sample are similar in all respects. All reported mean proportions of double majors are weighted using the final probability weight provided by and specific to the 2003 NSCG. Education Economics significant at the 1% level (F = 19.64). 189 11. In the context of an unrestricted model, the gender dummy and all interactions are jointly 12. As I am mainly interested in direct earnings benefits to double majors, I choose to use the ‘Bachelor’s Only’ sample here. Yet, all results are qualitatively the same if I switch to the ‘Graduate Degree Holders’ sample. 13. The t-statistic from a two sample t-test equals 0.47. References Berger, M.C. 1988. Predicted future earnings and choice of college major. Industrial and Labor Relations Review 41, no. 3: 418–29. Brown, C., and M. Corcoran. 1997. Sex-Based differences in school content and the male– female wage gap. Journal of Labor Economics 15, no. 3: 431–65. Daymont, T., and P. Andrisani. 1984. Job preferences, college major, and the gender earnings gap. The Journal of Human Resources 19, no. 3: 408–28. Del Rossi, A.F., and J. Hersch. 2008. Double your major, double your return' Economics of Education Review 27, no. 4: 375–86. Eide, E. 1994. College major choice and changes in the gender wage gap. Contemporary Economic Policy 12: 55–63. Finnie, R., and M. Frenette. 2003. Earning differences by major field of study: Evidence from three cohorts of recent Canadian graduates. Economics of Education Review 22, no. 2: 179–92. Fowler, E.M. 1990. Dual majors are gaining popularity. New York Times, 4 September, p. D10. James, E., N. Alsalam, J.C. Conaty, and D. To. 1989. College quality and future earnings: Where should you send your child to college' American Economic Review 79, no. 2: 247–52. Kane, T.J. 1999. The price of admission: Rethinking how Americans pay for college. Washington, DC: Brookings Institution Press. Lemieux, T. 2006. Post-secondary education and increasing wage inequality. NBER Working Paper Series No. 12077. Cambridge, MA. Lewin, T. 2002. For students seeking edge, one major just isn’t enough. New York Times, 17 November, p. 1. Montmarquette, C., K. Cannings, and S. Mahseredjian. 2002. How do young people choose college majors' Economics of Education Review 21, no. 6: 543–56. Zhang, L. 2005. Do measures of college quality matter' The effect of college quality on graduates’ earnings. The Review of Higher Education 28, no. 4: 571–96. Copyright of Education Economics is the property of Routledge and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.