Abstract

We use data from the Baccalaureate and Beyond Longitudinal Study and propensity score weighting methods to estimate the effect of a double major on bachelor's degree recipients’ earnings within four years after college graduation. We classify each of a student's two majors in a double major combination as either “higher- or lower-paying,” based on the rank order of the average earnings of each major among single major students. Our analyses yield three main findings. First, within one year after graduation, double major graduates earn significantly less relative to their single major peers with the same higher paying major; however, by four years after graduation, their earnings are similar to those with the single higher paying major and significantly higher relative to those with the single lower paying major. Second, we find that double major graduates are more likely to be employed, work longer hours, and pursue graduate education than their single major peers four years after graduation. Finally, transcript data suggest that double major graduates take fewer classes in the higher paying major, which may explain their initial earnings penalty relative to those with the higher paying single major.

1.  Introduction

The economic premium of a college degree has been growing steadily over the past three decades (Taylor, Fry, and Oates 2014), and wage variations among college graduates have increased as well (Altonji, Arcidiacono, and Maurel 2016). College majors are significantly associated with wage differentials among college graduates (Rumberger and Thomas 1993; Arcidiacono 2004). The economic pecking order among college majors is remarkably consistent in studies based on different datasets and methods. Generally speaking, college graduates with majors in such fields as business, science, math, and engineering enjoy a substantial earnings advantage, whereas those with majors in humanities and the arts, history, and education are ranked at the lower end of the earnings spectrum (Altonji, Blom, and Meghir 2012).

Among college graduates, a substantial proportion pursue a double major1 (Del Rossi and Hersch 2008, 2016; Hemelt 2010). Among the nationally representative sample of participants in the National Survey of College Graduates (NSCG), 26.9 percent of college graduates who earned bachelor's degrees in 2010 had a double major (Del Rossi and Hersch 2016). Although scholars have extensively researched returns on college major choices, very few have investigated the effects of a double major. Those who have done so have found a small overall wage premium (1 to 3 percent) associated with a double major that varies across institutional types and double major combinations (Del Rossi and Hersch 2008, 2016; Hemelt 2010). However, directly comparing students with a double major against those with a single major raises the potential issue of self-selection bias. For instance, students in certain fields of study may be more likely to have a second major than those in other fields of study. The Baccalaureate and Beyond Longitudinal Study 2008--2012 (B&B:08/12) indicates that 17 percent of students in languages and literature have a second major, whereas only 5 percent of students in engineering and computer science do. In addition, major selection decisions could be associated with student characteristics—both observable and unobservable. For example, in the B&B:08/12, respondents with a double major had higher grade point averages (GPAs), on average, than their single major counterparts in the same first major. Neglecting important variables that are correlated with both double major choices and labor market outcomes could result in biased estimates. Although researchers in previous studies attempted to reduce bias by considering students’ demographic variables and basic institutional characteristics (including institutions’ Carnegie Classifications), many other factors, such as student academic performance, motivation, and occupational aspirations, could figure prominently in the decision to pursue a double major.

Our analyses contribute to the extant literature by applying propensity score weighting methods to estimate the effects of a double major on post-college outcomes, including earnings, employment status, and graduate school enrollment. Propensity score weighting methods balance single and double majors with respect to observable characteristics. Because detailed transcript data are available in the B&B:08/12, we are able to estimate propensity scores by using high school and college credits and other academic measures. In addition, our estimates of the effect of a double major are based on a more informative benchmark. In previous studies, researchers typically compared double major students with their single major counterparts in the first or second reported major. These comparisons are not particularly helpful because the order in which double majors are reported could be idiosyncratic. For example, graduates may report majors based on the dates their degrees were awarded, perceived importance, or personal affinity. In this study, we classify each of a student's two majors in a double major combination as either “higher- or lower-paying” based on the rank order of the average earnings of each major among single major students. For example, if a student has a double major in economics and education, regardless of the order in which these two majors are reported, economics is the higher-paying major and education is the lower-paying major. In other words, our estimates represent the earnings premium/penalty of having a double major relative to a single higher-/lower-paying major.

Our results show that one year after graduation, double major graduates earn significantly less than their peers with the single higher-paying major and have a small, insignificant earnings premium relative to peers with the single lower-paying major. The benefits of a double major improve over time. By four years after graduation, the earnings penalty of a double major relative to the single higher-paying major disappears, and the earnings premium over those with the single lower-paying major increases. We further extend the analyses by including employment status and graduate school enrollment. We find that graduates with a double major are more likely to be employed, work longer hours, and pursue graduate education than their single major peers four years after graduation. Finally, double majoring may affect students’ course selection decisions and occupational choices. Transcript data show that double major graduates did not take extra classes in the higher paying major that might have made them more competitive in the labor market. Moreover, when one of the majors has low earnings power, a double major graduate is more likely to have a low-paying occupation.

The rest of the paper is organized as follows. First, we provide an overview of relevant literature about major selection and economic returns associated with double majors. We then describe the data and methods used to estimate the impact of double majors on various post-baccalaureate outcomes, and present findings of our analyses using descriptive statistics and regression models. Finally, we discuss our findings and offer some concluding remarks.

2.  Literature Review

Although students may have myriad reasons for pursuing a double major, preparing for work is by far the most common reason. Pitt and Tepper (2012) provided the most recent and thorough description of double majoring through a Web-based survey of nine selective colleges. Their results show that 68 percent of students choose to pursue a double major to be competitive in the labor market. Other important reasons include expressing self-identity, pursuing complementary subjects, and gaining a breadth of knowledge. Zafar (2012) also confirmed that students choose a double major to improve their prospects in the labor market.

Although students may have high expectations regarding the economic benefits of a double major, research in this area has revealed only a small earnings premium. Del Rossi and Hersch (2008) provided the first estimate of the return on double majoring using data from the 2003 NSCG. They found that having a double major increases earnings by 1.4 percent relative to having a single major in the first major field for all bachelor's degree recipients. Also using data from the 2003 NSCG, Hemelt (2010) found that college graduates with a double major earn, on average, 3.2 percent more than their counterparts with a single major.

Despite many insights from existing studies on the effects of college majors on post-baccalaureate outcomes, important gaps remain with respect to the effects of a double major. First, any estimation of the effects of a double major depends critically on the reference group. Prior studies suggest that the combination of majors matters when it comes to the economic return of a double major. Del Rossi and Hersch (2008) found that combining a high paying major (i.e., business, engineering, or science/math) with a less lucrative major (i.e., arts/social science or education) results in a 7 to 50 percent wage premium relative to a single low-paying major, but no significant gain compared with a single high-paying major. Del Rossi and Hersch (2016) reported similar results after applying the same strategy to recent data from the 2003 NSCG. Hemelt (2010) examined earnings differentials across major combinations and found that students with a first major in physical science, biology/life science, computer science, and mathematics can expect significant wage premiums from adding a second major. Furthermore, a second major in business administration, computer science, or engineering is most profitable, regardless of the first major. The heterogeneous effects of different major combinations suggest that double majoring may be beneficial compared with single majoring in one of the majors, and may be useless or even disadvantageous compared with single majoring in the other major. It is therefore essential to specify the benchmark when estimating the return associated with a double major.

Second, the effect of a double major varies across students’ academic characteristics and institutional characteristics. Del Rossi and Hersch (2008) found that the premium of a double major is greater (2.3 percent) for graduates whose highest degree is a bachelor's degree, but nonsignificant for those with a post-bachelor's degree. Hemelt (2010) revealed that the effect varies across types of institutions, ranging from 4 percent at research and comprehensive universities to no effect at liberal arts colleges. These findings suggest that students’ academic characteristics and institutional characteristics could be associated with choosing and completing double majors. The availability of detailed information on individual academic characteristics and institutional characteristics in the B&B:08/12 provides an excellent opportunity to control for these differences and reduce potential bias. Moreover, in previous studies scholars have used ordinary least squares (OLS) regression, assuming the selection of double majors can be effectively accounted for when controlling for a set of observed individual and institutional characteristics. In this analysis, we use a propensity score weighting approach to more adequately account for confounding factors. Benefitting from a rich set of individual and institutional characteristics in the B&B:08/12, we attempt to address or at least alleviate self-selection bias.

Last but not least, it is unclear to what extent having a double major increases employment prospects and graduate school enrollment rates shortly after college graduation. Del Rossi and Hersch (2008, 2016) found little evidence that graduates with double majors have greater job match quality, higher job satisfaction, or higher job mobility than those with single majors in one of the same fields. To our knowledge, researchers have not yet examined the effect of having a double major on graduates’ employment status and graduate school enrollment. Carnevale, Cheah, and Strohl (2012) reported a 9 percent unemployment rate among recent college graduates based on data from a pooled sample of participants in the American Community Survey 2009–10, indicating that post-college unemployment remains an important issue and the likelihood of unemployment differs across college majors. Altonji, Arcidiacono, and Maurel (2016) also reported differential graduate degree attainment rates by field of undergraduate study. Thus, this study includes employment status and graduate school enrollment as outcomes of interest to understand how a double major may affect graduates’ trajectories after graduation.

3.  Data and Methods

Data, Sample, and Variables

Our main data source is the B&B:08/12, which is a nationally representative sample of individuals who received bachelor's degrees during the 2007–08 academic year. It provides information about respondents’ education and work experience one year and four years after completing their bachelor's degrees. To identify double major choices, we merged data from the B&B:08/12 with detailed course, term, and degree records from the Postsecondary Education Transcript Collections. Degree records in the transcript data report the first major, second major, and minors for each degree received by students. Data on double majors from the B&B:08/12 study are superior to data from other sources (e.g., the NCGS) because the dataset: (1) includes both double degrees and double majors associated with a single degree; (2) provides official transcript data rather than self-reported data on double majors, and (3) clearly differentiates double majors from minors.

Double majors do not necessarily lead to a second degree. Institutional policies vary as to whether a double major can be counted as a double degree. Usually, a student pursuing two separate major programs within a college/school would earn a single degree, whereas a student pursuing two distinct degree types from two separate colleges/schools within a university (e.g., from the College of Arts and Sciences and the Wharton School at the University of Pennsylvania, or the Wheelock College of Education and Human Development and the School of Social Work at Boston University)2 would earn a double degree. In this study, our definition of double majors includes both double majors associated with a single degree and double degrees received at a single institution within a four-year period. In the B&B:08/12 data, 220 respondents had earned double degrees and 1,240 respondents reported double majors associated with a single degree.3 One may be concerned that double degree recipients are systematically different from those who pursue a double major for a single degree. We found that in our sample, students with double degrees earned more credits (146) on average than those with a double major for a single degree (129). For these reasons, we included indicators for both double degree holders and total credits in our empirical models. In addition, we restricted the sample to double majors pursued at a single institution within a four-year period because pursuing double degrees and majors at separate institutions usually requires much more coursework than doing so at the same institution.

As far as the order of double major is concerned, the first and second majors are clearly differentiated in the B&B:08/12 transcript data for those who have two majors associated with a single degree.4 However, for those with double degrees, no indicators in the B&B:08/12 data show the order of degrees; thus, we arranged the degrees in chronological order. If double degrees were conferred on the same date (90 students), we assigned a random order for degrees.

We also linked college selectivity data from the Barron's Admissions Competitiveness Index 2008. The Barron's index categorizes institutions into seven selectivity groups based on entering students’ class ranks, high school GPAs, and average SAT scores, and the percentage of applicants admitted. Furthermore, we retrieved institutional characteristics from the Integrated Postsecondary Education Data System.

In this analysis, we focused on a variety of post-college outcomes: earnings, employment status, working hours, and graduate school enrollment. We calculated earnings as the annual salary associated with the respondent's primary job.5 We measured employment status based on a series of binominal outcomes: in/out of labor force, unemployed/employed, full-time/part-time job, and the number of hours worked per week.6 We measured graduate school enrollment based on whether the graduate was enrolled in a post-bachelor's degree program or had received a post-bachelor's degree.7

The initial sample included 15,720 respondents who received baccalaureate degrees from a four-year institution between July 2007 and June 2008 for whom major information and transcript data are available. Among this initial sample, 9.2 percent reported having a double major. Table A.1 in the appendix reports sample size by employment and enrollment status. After excluding respondents without information on employment and enrollment status in 2009 and 2012 follow-ups, the analytical sample was composed of approximately 13,280 to 13,700 respondents. When using graduate school enrollment as the dependent variable, we used the entire analytical sample. When examining labor force participation, we restricted the sample to those who were not enrolled in any postsecondary institutions at the time of the interviews, yielding a sample of around 10,160 to 10,660 respondents. When examining employment, we further limited the sample to those who were in the labor force. When using earnings, full-time jobs, or working hours as outcomes, we limited the sample to those who were employed (either full-time or part-time) and not enrolled at any higher education institutions at the time of the interviews (about 8,800). We used multiple imputation to address missing values in covariates.8 Although no values were missing for the earnings variables, zero earnings were reported by ten employed respondents in 2009 and 140 employed respondents in 2012. We excluded these respondents when examining earnings. All analyses reported in this study were weighted by panel and transcript weights since variables from multiple waves of interviews and transcript data are included. Because the B&B:08/12 is a complex survey with clustering and stratification, we used the Taylor linearized method (Demnati and Rao 2004) while considering primary sampling units (i.e., institutions) and strata for variance estimation, if not specified otherwise.

Methods

Our estimation begins with a baseline OLS model:
ln(Yi)=α0+α1Di+α2Xi+Majorj+ui,
(1)
where ln(Yi) is the logarithm of individual i’s annual salary for his or her primary job; Di is an indicator of whether individual i graduated with a double major; and Xi is a set of variables related to demographic information, family background, student academic characteristics, institutional characteristics, and labor market variables that have been theoretically motivated and empirically tested in previous studies (Thomas and Zhang 2005; Robst 2007). Demographic variables include gender and race/ethnicity (i.e., black, Hispanic, Asian, and other races). Family background variables include family income and first-generation college graduate. Academic variables include undergraduate GPA, total credits earned, additional bachelor's degree, and post-bachelor's degree. Because double major choice is likely to be affected by students’ high school preparation, we also included variables related to high school curriculum (i.e., SAT, college credits earned in high school, number of Advanced Placement credits) to alleviate omitted variable bias. Institutional characteristics indicate the level of institutional control and selectivity of the schools where students earned their bachelor's degrees. Finally, labor market variables include age, tenure, tenure squared, and job-relatedness of major. We converted all financial variables into 2012 constant dollars. The descriptive statistics of main variables by double major status are shown in tables A.2 and A.3. The comparision between double major and single major students reveals patterns similar to those found in previous resarch (Del Rossi and Hersch 2008, 2016; Hemelt 2010; Pitt and Tepper 2012)—that is, graduates with a double major exhibited stronger academic achievement and family socioeconomic status.

Majorj is a set of dummy variables representing the major field of study. Note that the specification of college majors is important because it provides a benchmark against which earnings differentials can be evaluated. Conventionally, the first major field is used as the single major benchmark to determine whether pursuing a second major can increase one's earnings potential (Del Rossi and Hersch 2008, 2016; Hemelt 2010). However, the wage premium associated with a double major may depend on the benchmark against which earnings are compared. For example, a student with a double major in economics and education may earn less than students with a single major in economics but more than students with a single major in education, suggesting the importance of examining the returns associated with different major combinations. However, analyzing returns on double major combinations requires a large sample size. For example, in our B&B:08/12 sample, after we collapsed major categories to twenty based on the two-digit Classification of Instructional Programs code, the approximately 1,500 students with a double major had 250 major combinations. Alternatively, one could use broad areas of study. For example, Del Rossi and Hersch (2008) categorized majors into five broad areas (art/social science, business, education, engineering, and science/math) and thereby yielded fifteen major combinations, ignoring the order of the two majors. This approach, however, masks variations in earnings premiums across majors within a broad area of study.

In this analysis, we used higher paying majors and lower paying majors as the benchmarks. The classification of higher/lower paying major is individual-specific and based on average earnings of single major graduates after controlling for important individual and college characteristics. Specifically, we regressed earnings on major dummies and important covariates based on the sample of single major graduates.9 The coefficient for each major from this model can be interpreted as the major's earnings power, which we used to establish an economic ranking of majors (see table A.4 for a list of major categories ranked by estimated earnings power). Between the two majors of any double major combination, the one with a larger coefficient was the higher paying major, and the other one was the lower paying major. Our estimates represent the economic return of a double major relative to either the lower paying major or the higher paying major. For other outcome variables, including employment status and graduate school enrollment, we used probit models.

It is important to recognize that the relationships presented in equation 1 should not be interpreted as causal. Although regression models control for observable differences between graduates with a double major and those with a single major, the decision to pursue a double major is probably subject to preferences, financial constraints, beliefs, and expected benefits from the respective majors. Because probit and OLS approaches could yield biased estimates if ui were correlated with the decision to pursue a double major, we used propensity score weighting to address the self-selection problem. As a supplement, in order to attribute all estimated effects of double majoring to selection bias, we also used techniques developed by Altonji, Elder, and Taber (2005) to quantify how strong the relationship between double majoring and unobservable determinants of outcomes would have to be relative to the strength of the relationship between double majoring and observable determinants of outcomes.

The propensity score weighting method (Rosenbaum and Rubin 1983) is used to make the control group's outcomes represent the counterfactual outcomes of the treatment group by rendering the weighted groups similar with respect to observable characteristics. Because propensity score weighting offers many advantages by being easy to implement, accommodating most types of outcome analysis and retaining most cases from the original sample, it is an attractive technique that is becoming increasingly popular in multiple disciplines (Guo and Fraser 2015). In this study, because the number of students with a double major is limited, the weighting method is especially attractive since it retains as many observations as possible. Specifically, we used an inverse probability weighted regression adjustment estimator to estimate the average treatment effect on the treated (ATT). The inverse probability weighted regression adjustment estimator has a double-robust property, meaning that the estimator is consistent as long as the specification for predicting either the propensity score or the outcome is correct (Hirano and Imbens 2001).

We first performed a logistic regression for choosing a double major conditioned on a set of covariates, Z, and predicted the propensity score for each observation. Z is composed of gender, age, age squared, race/ethnicity, family income, parental education, SAT scores, Advanced Placement credits, college credits earned in high school, institutional control, institutional selectivity, and geographic regions. Including high school Advanced Placement credits and college credits earned in high school is important because these credits can reduce the credit requirement for a double major. Robustness checks were conducted by including different sets of covariates in estimating propensity scores. Based on the predicted propensity scores, we created weights for estimating ATT following suggestions made by Nichols (2008) and Guo and Fraser (2015):
ω(W,z)=W+(1-W)e^(z)1-e^(z),
(2)
where W corresponds to the value of treatment (1, 0) and e^(z) corresponds to the propensity score (Guo and Fraser 2015). Identifying the ATT requires a weak overlap condition. Figure 1 shows the kernel density of propensity scores for single and double majors, indicating a relatively large region of overlap between these two groups.
Figure 1.

Propensity Score of Treatment and Control Groups

Figure 1.

Propensity Score of Treatment and Control Groups

Table 1 presents standardized differences of covariates before and after the weighting. Researchers have suggested that a standardized difference greater than 0.1 indicates an imbalance (Austin 2009). Results in table 1 indicate that no standardized difference is greater than 0.1 after student academic variables (i.e., SAT, college credits earned in high schools, Advanced Placement credits, institutional control, and institutional selectivity) are included in the propensity score prediction. Column 4 further added geographic regions as covariates because they may also affect labor market outcomes. The propensity score weighting (PSW) estimations in the following analysis will rely on propensity scores obtained from column 4 as the model passed the balanced test and included the most comprehensive list of covariates.

Table 1.
Standardized Difference of Covariates after Propensity Score Weighting
Sex, age, race (1)(1) + Family background (2)(2) + Academic performance (3)(3) + Regions (4)
Sex 0.002 0.002 0.009 0.011 
Age 0.001 0.001 0.006 0.007 
White 0.001 0.001 0.005 0.005 
Black or African American 0.001 0.001 0.004 0.003 
Hispanic 0.000 0.000 0.003 0.001 
Asian 0.002 0.002 0.002 0.002 
Other 0.001 0.001 0.005 0.006 
Family income 0.001 0.000 0.004 0.002 
First generation 0.060 0.002 0.011 0.010 
SAT 0.055 0.001 0.004 0.003 
College credits in high school 0.229 0.221 0.009 0.012 
Advanced Placement credits 0.113 0.109 0.007 0.008 
Public 0.136 0.131 0.021 0.029 
Private not-for-profit 0.124 0.118 0.006 0.009 
Private for-profit 0.151 0.143 0.007 0.010 
Most competitive 0.070 0.068 0.003 0.005 
Highly competitive 0.289 0.278 0.007 0.009 
Very competitive 0.027 0.034 0.003 0.002 
Competitive 0.002 0.006 0.002 0.002 
Less competitive 0.109 0.100 0.004 0.004 
Noncompetitive 0.081 0.077 0.001 0.001 
Special focus 0.002 0.002 0.009 0.011 
Sex, age, race (1)(1) + Family background (2)(2) + Academic performance (3)(3) + Regions (4)
Sex 0.002 0.002 0.009 0.011 
Age 0.001 0.001 0.006 0.007 
White 0.001 0.001 0.005 0.005 
Black or African American 0.001 0.001 0.004 0.003 
Hispanic 0.000 0.000 0.003 0.001 
Asian 0.002 0.002 0.002 0.002 
Other 0.001 0.001 0.005 0.006 
Family income 0.001 0.000 0.004 0.002 
First generation 0.060 0.002 0.011 0.010 
SAT 0.055 0.001 0.004 0.003 
College credits in high school 0.229 0.221 0.009 0.012 
Advanced Placement credits 0.113 0.109 0.007 0.008 
Public 0.136 0.131 0.021 0.029 
Private not-for-profit 0.124 0.118 0.006 0.009 
Private for-profit 0.151 0.143 0.007 0.010 
Most competitive 0.070 0.068 0.003 0.005 
Highly competitive 0.289 0.278 0.007 0.009 
Very competitive 0.027 0.034 0.003 0.002 
Competitive 0.002 0.006 0.002 0.002 
Less competitive 0.109 0.100 0.004 0.004 
Noncompetitive 0.081 0.077 0.001 0.001 
Special focus 0.002 0.002 0.009 0.011 

Notes: Propensity scores used in column 1 are predicted by sex, age, and race; column 2 added family income and parental education; column 3 further added high school academic performance variables; column 4 further added geographic regions. Values in bold indicate the risk of imbalance (>0.1).

Finally, we ran a weighted least squares regression using the propensity score weights, controlling the same set of variables, X, as in the OLS model. We also incorporated sampling weights into the PSW analysis by multiplying propensity score weights and sampling weights (Guo and Fraser 2015). DuGoff, Schuler, and Stuart (2014) showed that combining PSW with survey weighting is necessary to achieve unbiased treatment effect estimates that are generalizable to the original survey target population. Because we used a two-step approach, standard errors in the weighted least squares were not consistent; therefore, we used bootstrapping to obtain correct standard errors (Hirano and Imbens 2001). Note that students sampled in the B&B:08/12 are nested within colleges. To preserve the correlations within colleges during the bootstrapping process, we used the college as the unit of observation instead of individuals, yielding the bootstrapped version of clustered standard errors.

4.  Results

Double Majors and Earnings

In table 2, we present OLS estimates of the impacts of double majoring on earnings in 2009 (one year after graduation) and 2012 (four years after graduation). The values in the left half of table 2 are relative to respondents’ higher paying majors—in other words, the wage premium of having a second major relative to only having the single higher paying major in the combination—and the values in the other half are relative to respondents’ lower paying majors. We incrementally added control variables to the models. In 2009, results show that graduates with a double major earned about 6.9 percent less (−0.071 log points in column 3)10 than those with the single higher-paying major when controlling for student demographic and family background, institutional control and selectivity, and student academic performance. Additionally, controlling for graduates’ postgraduation experience (i.e., work experience, working hours, job-relatedness of the major, and post-bachelor's degree) did not change the magnitude of the coefficient but substantially reduced the standard errors. The right half of table 2 indicates that those with a double major had a significant earnings premium relative to those with the single lower paying major. Yet, after controlling for postgraduation experience, the premium is no longer significant.

Table 2.
Effects of a Double Major on Earnings (Ordinary Least Squares)
Compared with Higher Paying MajorsCompared with Lower Paying Majors
(1)(2)(3)(4)(5)(6)(7)(8)
Dependent: 2009 earnings (1 year post-BA) 
Double major −0.050 −0.063 −0.071* −0.071** 0.108** 0.100** 0.095** 0.025 
 (0.040) (0.039) (0.039) (0.029) (0.045) (0.044) (0.043) (0.030) 
Second degree 0.117* 0.124* 0.125* 0.072 0.151* 0.160* 0.156* 0.080 
 (0.070) (0.069) (0.072) (0.053) (0.072) (0.070) (0.072) (0.054) 
N 8,580 8,580 8,580 8,540 8,580 8,580 8,580 8,540 
Dependent: 2012 earnings (4 years post-BA) 
Double major 0.060 0.045 0.044 0.013 0.183*** 0.168*** 0.169** 0.114*** 
 (0.045) (0.043) (0.042) (0.036) (0.047) (0.044) (0.044) (0.035) 
Second degree 0.054 0.079 0.092 0.024 0.093 0.118 0.128 0.042 
 (0.091) (0.090) (0.095) (0.062) (0.090) (0.089) (0.094) (0.062) 
N 7,800 7,800 7,800 7,800 7,800 7,800 7,800 7,800 
Important controls 
Demographic & family background 
Institutional control & selectivity   
Academic performance     
Postgraduate experience       
Compared with Higher Paying MajorsCompared with Lower Paying Majors
(1)(2)(3)(4)(5)(6)(7)(8)
Dependent: 2009 earnings (1 year post-BA) 
Double major −0.050 −0.063 −0.071* −0.071** 0.108** 0.100** 0.095** 0.025 
 (0.040) (0.039) (0.039) (0.029) (0.045) (0.044) (0.043) (0.030) 
Second degree 0.117* 0.124* 0.125* 0.072 0.151* 0.160* 0.156* 0.080 
 (0.070) (0.069) (0.072) (0.053) (0.072) (0.070) (0.072) (0.054) 
N 8,580 8,580 8,580 8,540 8,580 8,580 8,580 8,540 
Dependent: 2012 earnings (4 years post-BA) 
Double major 0.060 0.045 0.044 0.013 0.183*** 0.168*** 0.169** 0.114*** 
 (0.045) (0.043) (0.042) (0.036) (0.047) (0.044) (0.044) (0.035) 
Second degree 0.054 0.079 0.092 0.024 0.093 0.118 0.128 0.042 
 (0.091) (0.090) (0.095) (0.062) (0.090) (0.089) (0.094) (0.062) 
N 7,800 7,800 7,800 7,800 7,800 7,800 7,800 7,800 
Important controls 
Demographic & family background 
Institutional control & selectivity   
Academic performance     
Postgraduate experience       

Notes: Standard errors clustered at institutional level are in parentheses. Panel and transcript weights are applied, and variances are estimated by Taylor linearized method. Estimations are based on the respondents who were employed and not enrolled in 2009 and 2012, respectively. Variables for demographics, institutional control and selectivity, and academic performance are listed in Appendix table A.2. Postgraduation experience includes work experience and its squared term, working hours per week, relatedness of colleges majors with occupations, and whether the student has a post-bachelor's (BA) degree.

*p < 0.1; **p < 0.05; ***p < 0.01.

By four years after graduation, graduates with a double major had improved their positions relative to their peers with a single major. Unlike in 2009, we found no significant earnings penalty for those with a double major relative to those with the single higher paying major. Compared with those with the single lower paying major, graduates with a double major earned 12.1 percent more (0.114 log points in column 8), which is much larger than the estimate for 2009. After controlling for academic performance and postgraduation experience, the coefficient of having a second degree is not significant, suggesting that there is no additional effect of having a double degree beyond having a double major.

In table 3, we report estimates based on the PSW approach. The results from PSW, which estimate the ATT of a double major, confirm the results from the OLS estimation. In 2009, the earnings of those with a double major were about 7.1 percent lower (−0.074 log points) than what they would have earned if they had graduated with the single higher paying major, and this earnings penalty is significant at the 0.05 level. In contrast, those with a double major earned 2.5 percent more than what they would have earned if they had graduated with the single lower paying major, although this small benefit is not statistically significant. In 2012, the earnings for those with a double major did not differ significantly from what they would have earned if they had graduated with the single higher paying major; however, they earned about 10.2 percent (0.097 log points) more than what they would have earned if they had graduated with the single lower paying major, and this earnings premium is significant at the 0.05 level. It is noteworthy that having a graduate degree (coefficient not reported in table 3 due to space limit) yields a 3 percent earning premium in 2012.

Table 3.
Effects of a Double Major on Earnings (Propensity Score Weighting)
Compared With Higher-Paying MajorCompared With Lower-Paying Major
Dependent: 2009 earnings (1 year post-BA) 
Double Major −0.074** 0.025 
 (0.032) (0.034) 
N 8,540 8,540 
Dependent: 2012 earnings (4 years post-BA) 
Double major 0.041 0.097** 
 (0.040) (0.039) 
N 7,800 7,800 
Compared With Higher-Paying MajorCompared With Lower-Paying Major
Dependent: 2009 earnings (1 year post-BA) 
Double Major −0.074** 0.025 
 (0.032) (0.034) 
N 8,540 8,540 
Dependent: 2012 earnings (4 years post-BA) 
Double major 0.041 0.097** 
 (0.040) (0.039) 
N 7,800 7,800 

Notes: Bootstrapped standard errors clustered at institutional level are in parentheses. Estimations are based on the respondents who were employed and not enrolled at any postsecondary institutions in 2009 and 2012, respectively. Control variables in the outcome function are the same as in the full model (columns 4 and 8 in table 2). Propensity scores are estimated from the preferred logit model (column 4 in table 1). BA = bachelor's degree.

**p < 0.05.

It is important to recognize that PSW methods rely on the assumption of selection on observables. Nonobservable factors might drive students’ decisions to pursue a double major, such as motivation and personal interests. As it is difficult to find a valid instrumental variable for the decision to double major, we applied techniques described by Altonji, Elder, and Taber (2005) to assess the bias arising from selection on unobservables using information about selection on observables. Specifically, we derived the ratio of selection on unobservables to selection on observables that would be required to “explain away” the effect of double majors estimated from OLS models. This ratio is 1.5 in 2009 and 2.6 in 2012 for the earnings model that controls for a comprehensive list of student academic and college characteristics. In other words, selection on unobservables would need to be at least 1.5 times stronger than selection on observables to attribute the entire estimated effect of a double major to selection bias. Thus, it is unlikely that the estimated effects of a double major revealed by the OLS and PSW models are dominated by selection bias.

To shed light on the heterogeneity in the effect of double majoring, we examined whether the earnings power gap between two majors of a double major combination can moderate the economic return associated with a double major. We measured the earnings power gap as the difference in the earnings power between two majors of a double major graduate, which ranges between 0 and 0.5 in our data.11 We then introduced an interaction term between the double major indicator and the earnings power gap in equation 1. Figure 2 depicts the predicted earnings differentials of double major graduates relative to single major graduates. In 2009, the disadvantage of double major graduates relative to those with the single higher paying major increased as the earnings gap grew. By 2012, the earnings penalty of a double major had been reduced. Double major graduates even had higher earnings than those with the single higher paying major if their two majors had similar average earnings. However, when the earnings power gap became large, double major graduates still suffered an earnings penalty compared with their peers with the single higher paying major. Meanwhile, the earnings premium of double major graduates relative to those with the single lower paying major increased in both years as the gap became larger.

Figure 2.

Predicted Earnings Differentials Between Double Major and Single Major Graduates by Earnings Power Gap

Notes: The figure plots the differences between predicted earnings of double major graduates and predicted earnings of single major graduates in the higher (lower) paying major across the earnings power gap ranging from 0 to 0.5, with other covariates fixed at means. The predicted values are derived from the coefficients of propensity score weighting model that regressed earnings on the double major indicator, the interaction term between double major indicator and earnings power gap, and a full list of covariates. BA = bachelor's degree.

Figure 2.

Predicted Earnings Differentials Between Double Major and Single Major Graduates by Earnings Power Gap

Notes: The figure plots the differences between predicted earnings of double major graduates and predicted earnings of single major graduates in the higher (lower) paying major across the earnings power gap ranging from 0 to 0.5, with other covariates fixed at means. The predicted values are derived from the coefficients of propensity score weighting model that regressed earnings on the double major indicator, the interaction term between double major indicator and earnings power gap, and a full list of covariates. BA = bachelor's degree.

In addition to pursuing a second major—a popular choice among college students who wish to acquire knowledge in multiple academic fields—is to pursue a minor, which typically has lower credit requirements. To test whether a minor is associated with economic returns similar to those of a double major, we restricted the definition of “single major” to those with one major and no minors, and incorporated a “minor” indicator into our baseline models for those with one major and a minor in another field. Results in table 4 show that compared with their single major (no minor) peers in the higher paying major, on avarage, a minor generated lower economic returns than a second major in 2009 and 2012.

Table 4.
Effects of a Double Major and a Minor on Earnings
Compared with Higher Paying MajorCompared with Lower Paying Major
OLSPSWOLSPSW
Dependent: 2009 earnings (1 year post-BA) 
Double Major −0.066*** −0.073*** −0.003 −0.012 
 (0.025) (0.026) (0.025) (0.025) 
Minor −0.096*** −0.079*** 0.001 0.010 
 (0.022) (0.024) (0.021) (0.023) 
N 8,760 8,760 8,760 8,760 
Dependent: 2012 earnings (4 years post-BA) 
Double Major 0.009 0.011 0.062** 0.034 
 (0.028) (0.029) (0.028) (0.027) 
Minor −0.072*** −0.068*** 0.000 −0.017 
 (0.022) (0.024) (0.022) (0.027) 
N 7,990 7,990 7,990 7,990 
Compared with Higher Paying MajorCompared with Lower Paying Major
OLSPSWOLSPSW
Dependent: 2009 earnings (1 year post-BA) 
Double Major −0.066*** −0.073*** −0.003 −0.012 
 (0.025) (0.026) (0.025) (0.025) 
Minor −0.096*** −0.079*** 0.001 0.010 
 (0.022) (0.024) (0.021) (0.023) 
N 8,760 8,760 8,760 8,760 
Dependent: 2012 earnings (4 years post-BA) 
Double Major 0.009 0.011 0.062** 0.034 
 (0.028) (0.029) (0.028) (0.027) 
Minor −0.072*** −0.068*** 0.000 −0.017 
 (0.022) (0.024) (0.022) (0.027) 
N 7,990 7,990 7,990 7,990 

Notes: Estimations are based on the respondents who were employed and not enrolled at any postsecondary institutions in 2009 and 2012, respectively. Control variables in the outcome function are the same as in the full model. Propensity scores are estimated from the preferred logit model. OLS = ordinary least squares; PSW = propensity score weighting; BA = bachelor's degree.

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

Mechanisms

In this section, we explored various mechanisms through which double majoring may affect graduates’ earnings. First, we estimated the effects of a double major on student employment outcomes and graduate school enrollment rates (see table 5). Both OLS and PSW results suggest that one year after graduation, having a double major had no significant impact on the likelihood of participating in the labor force, being employed, and enrolling in graduate school, regardless of the comparison group. However, double major graduates were more likely to be employed full-time and reported working more hours per week. Benefits of double majors become salient four years after graduation. Having a double major yielded a significant advantage relative to having the single lower paying major. Double major graduates were more likely to participate in the labor force, be employed, and work longer hours. The positive effects are especially strong for graduate school enrollment, with all models yielding positive and statistically significant coefficients for double majors, regardless of the reference group. This may partially explain the reason why the relative standing of double major graduates improved between 2009 and 2012. Recall that having a graduate degree yields a 3 percent earnings premium in 2012—the positive effect of double majors on graduate enrollment suggests that earnings estimates based on the sample of college graduates not enrolled in graduate school would likely understate the total earnings’ effect of double majors.

Table 5.
Effects of a Double Major on Employment and Graduate School Enrollment
2009 (1 Year Post-BA)2012 (4 Years Post-BA)
Higher Paying MajorLower Paying MajorHigher Paying MajorLower Paying Major
OLSPSWOLSPSWOLSPSWOLSPSW
1 = In labor force; −0.110 −0.049 −0.040 −0.016 0.140 0.176 0.183 0.289** 
0 = Out of labor forcea (0.128) (0.141) (0.133) (0.189) (0.124) (0.132) (0.124) (0.117) 
N 9,920 9,920 9,920 9,920 10,600 9,620 10,600 9,620 
1 = Employed; 0 = Unemployedb −0.108 −0.062 −0.028 0.010 0.109 0.245 0.216 0.297* 
 (0.124) (0.125) (0.131) (0.138) (0.129) (0.171) (0.133) (0.180) 
N 9,520 9,520 9,520 9,520 9,700 8,710 9,700 8,710 
1 = Full-time; 0 = Part-timec −0.093 −0.101 0.087 0.127* 0.069 0.115 0.142 0.223 
 (0.092) (0.071) (0.096) (0.068) (0.127) (0.162) (0.123) (0.148) 
N 8,590 8,590 8,590 8,590 8,900 7,920 8,900 7,920 
Log (Working hours per week)c 0.008 0.015 0.063* 0.084** 0.034 0.055 0.055** 0.063** 
 (0.030) (0.033) (0.033) (0.033) (0.022) (0.030) (0.022) (0.028) 
N 8,590 8,590 8,590 8,590 9,380 7,800 9,380 7,800 
Graduate school enrollmentd 0.054 0.029 −0.012 0.012 0.228*** 0.209** 0.158* 0.145* 
 (0.078) (0.084) (0.082) (0.088) (0.080) (0.089) (0.083) (0.087) 
N 13,380 13,380 13,380 13,380 12,990 12,000 12,990 12,000 
2009 (1 Year Post-BA)2012 (4 Years Post-BA)
Higher Paying MajorLower Paying MajorHigher Paying MajorLower Paying Major
OLSPSWOLSPSWOLSPSWOLSPSW
1 = In labor force; −0.110 −0.049 −0.040 −0.016 0.140 0.176 0.183 0.289** 
0 = Out of labor forcea (0.128) (0.141) (0.133) (0.189) (0.124) (0.132) (0.124) (0.117) 
N 9,920 9,920 9,920 9,920 10,600 9,620 10,600 9,620 
1 = Employed; 0 = Unemployedb −0.108 −0.062 −0.028 0.010 0.109 0.245 0.216 0.297* 
 (0.124) (0.125) (0.131) (0.138) (0.129) (0.171) (0.133) (0.180) 
N 9,520 9,520 9,520 9,520 9,700 8,710 9,700 8,710 
1 = Full-time; 0 = Part-timec −0.093 −0.101 0.087 0.127* 0.069 0.115 0.142 0.223 
 (0.092) (0.071) (0.096) (0.068) (0.127) (0.162) (0.123) (0.148) 
N 8,590 8,590 8,590 8,590 8,900 7,920 8,900 7,920 
Log (Working hours per week)c 0.008 0.015 0.063* 0.084** 0.034 0.055 0.055** 0.063** 
 (0.030) (0.033) (0.033) (0.033) (0.022) (0.030) (0.022) (0.028) 
N 8,590 8,590 8,590 8,590 9,380 7,800 9,380 7,800 
Graduate school enrollmentd 0.054 0.029 −0.012 0.012 0.228*** 0.209** 0.158* 0.145* 
 (0.078) (0.084) (0.082) (0.088) (0.080) (0.089) (0.083) (0.087) 
N 13,380 13,380 13,380 13,380 12,990 12,000 12,990 12,000 

Notes: Control variables include demographic and family background, academic performance, and institutional control and selectivity. Propensity scores are estimated from the preferred logit model. BA = bachelor's degree; OLS = ordinary least squares; PSW = propensity score weighting.

aEstimations are based on the respondents who were not enrolled.

bEstimations are based on the respondents who were in the labor force and not enrolled.

cEstimations are based on the respondents who were employed and not enrolled.

dEstimations are based on all respondents who had participated 2009 and 2012 interviews and had valid transcript data.

*p < 0.1, **p < 0.05, ***p < 0.01.

To understand how changes in the analytical samples between 2009 and 2012 might be related to earnings estimates at these two time points, we limited the sample to those who, as of 2012, had never enrolled in any degree programs after receiving their bachelor's degrees, and then reestimated the impact of a double major on graduates’ earnings. Results are reported in table 6. Estimates suggest that among those who had never enrolled in graduate programs as of 2012, there was almost no penalty for graduates with double majors compared with their peers with the higher paying single major in 2009. This result suggests that many double major graduates who earned significantly less than their single major peers chose to attend graduate school. To further test this, we ran a separate regression (not reported here due to space limitations) and found a significant negative relationship between earnings in 2009 and graduate school enrollment in 2012. In other words, many graduates with double majors whose salaries were relatively low immediately after college graduation decided to attend graduate school. This result is further confirmed by the lower panel of table 6. When we excluded students who had ever enrolled in graduate programs as of 2012, the effect of having a double major becomes larger.

Table 6.
Effects of a Double Major on Earnings (Exclude Graduate School Attendees)
Compared With Higher Paying MajorCompared With Lower Paying Major
 OLS PSW OLS PSW 
Dependent: 2009 earnings (BA recipients who never enrolled graduate degree programs as of 2012) 
Double major −0.006 −0.012 0.080** 0.088*** 
 (0.036) (0.039) (0.037) (0.034) 
N 5,820 5,820 5,820 5,820 
Dependent: 2012 earnings (BA recipients who never enrolled graduate degree programs as of 2012) 
Double major 0.071 0.080 0.157*** 0.140*** 
 (0.045) (0.049) (0.043) (0.051) 
N 5,560 5,560 5,560 5,560 
Compared With Higher Paying MajorCompared With Lower Paying Major
 OLS PSW OLS PSW 
Dependent: 2009 earnings (BA recipients who never enrolled graduate degree programs as of 2012) 
Double major −0.006 −0.012 0.080** 0.088*** 
 (0.036) (0.039) (0.037) (0.034) 
N 5,820 5,820 5,820 5,820 
Dependent: 2012 earnings (BA recipients who never enrolled graduate degree programs as of 2012) 
Double major 0.071 0.080 0.157*** 0.140*** 
 (0.045) (0.049) (0.043) (0.051) 
N 5,560 5,560 5,560 5,560 

Notes: Estimations are based on the respondents who had never enrolled a post-bachelor's (BA) degree program (doctoral, master's, and professional degrees) as of 2012 and who were employed in 2009 and 2012, respectively. Control variables include demographic and family background, academic performance, institutional control and selectivity, and postgraduate experiences. Propensity scores are estimated from the preferred logit model. OLS = ordinary least squares; PSW = propensity score weighting.

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

Double majoring may also affect the accumulation of human capital through different course selections. We compared the number of credits taken in all fields against the number of credits taken within the field of the first major in academic disciplines with the most double major students: business/economics, biology, psychology, education, and philosophy/religion/history (see table A.5). We found that although students with double majors or minors generally have taken more credits overall, single major students take more credits within the major field of study. This suggests that if students had not pursued a double major, they might have taken additional classes within the field of the single major. Perhaps depth rather than breadth of knowledge is valued in the labor market. Meeting course requirements for double majors and minors may decrease the amount of time available to develop depth of knowledge in either field.

Finally, we examined occupational choice by double major status to illustrate the effect of double majoring. Ideally, we would like to compare the probability of choosing each occupation between graduates with double majors and graduates with single higher (or lower) paying major; however, this comparison would require a large number of observations. Consequently, we grouped all of the majors into three categories of roughly equal size based on earnings power (i.e., high, middle, and low) and further generated a total of six combinations of double majors: high-high (HH), high-middle/middle-high (HM/MH), high-low/low-high (HL/LH), middle-middle (MM), middle-low/low-middle (ML/LM), and low-low (LL). Similarly, we created three roughly equal-sized groups of occupations based on their earnings power12 (i.e., high, middle, and low). In table 7, we tabulated the percentages of respondents in each major combination working in occupations with high, middle, and low earnings power. For example, compared with students with a single high-paying major, a higher (lower) proportion of double major graduates with at least one high-paying major (e.g., HL/LH, HM/MH, and HH) worked in low- (high-) paying occupations. In 2009, just 32.7 percent of those with a double major who had pursued one high-paying major and one low-paying major (LH/HL) worked in high-paying occupations versus 59.7 percent of graduates with the single high-paying major (H). The simple tabulation in table 7 suggests that graduates with a double major who pursue one low-paying major and one high-paying major are more likely to choose lower (higher) paying occupations than their peers with the single higher (lower) paying major, which is probably the reason for the negative (positive) effect of a double major on wages relative to these reference groups.

Table 7.
Percentage of Graduates Working in Different Occupation Groups by Major Combinations
2009 Occupations (1 Year Post-BA)2012 Occupations (4 Years Post-BA)
Major CombinationsLowMiddleHighTotalLowMiddleHighTotal
Single low (L) 53.24 37.6 9.17 100 52.18 33.11 14.71 100 
Single middle (M) 47.31 34.65 18.04 100 38.19 35.26 26.55 100 
Single high (H) 13.72 26.61 59.67 100 15.2 27.31 57.49 100 
Low-low major (LL) 57.6 31.2 11.2 100 65.83 23.33 10.83 100 
Low-middle/middle-low (LM/ML) 63.57 22.48 13.95 100 46.75 32.47 20.78 100 
Low-high/high-low (LH/HL) 28.57 38.78 32.65 100 34.48 27.59 37.93 100 
Middle-middle (MM) 52.5 26.25 21.25 100 33.33 37.25 29.41 100 
Middle-high/high-middle (MH-HM) 22.41 18.97 58.62 100 20.51 28.21 51.28 100 
High-high (HH) 16.38 25.86 57.76 100 18.84 28.99 52.17 100 
Total 35.37 31.62 33.01 100 32.94 31.21 35.86 100 
2009 Occupations (1 Year Post-BA)2012 Occupations (4 Years Post-BA)
Major CombinationsLowMiddleHighTotalLowMiddleHighTotal
Single low (L) 53.24 37.6 9.17 100 52.18 33.11 14.71 100 
Single middle (M) 47.31 34.65 18.04 100 38.19 35.26 26.55 100 
Single high (H) 13.72 26.61 59.67 100 15.2 27.31 57.49 100 
Low-low major (LL) 57.6 31.2 11.2 100 65.83 23.33 10.83 100 
Low-middle/middle-low (LM/ML) 63.57 22.48 13.95 100 46.75 32.47 20.78 100 
Low-high/high-low (LH/HL) 28.57 38.78 32.65 100 34.48 27.59 37.93 100 
Middle-middle (MM) 52.5 26.25 21.25 100 33.33 37.25 29.41 100 
Middle-high/high-middle (MH-HM) 22.41 18.97 58.62 100 20.51 28.21 51.28 100 
High-high (HH) 16.38 25.86 57.76 100 18.84 28.99 52.17 100 
Total 35.37 31.62 33.01 100 32.94 31.21 35.86 100 

Notes: The earnings power of occupations is derived by the same strategy used for major earnings power. We ran the wage function based on the single major sample only; the independent variables include occupations (33 categories), demographic and family background, academic performance, institutional control and selectivity, and work experience and its squared. The coefficients for occupations extracted from this model can be interpreted as their earnings power. BA = bachelor's degree.

This “downward” mobility could be due to several reasons. First, students choose double majors based on personal interests, which also guide their occupational choices, regardless of monetary benefits. Second, having a double major may influence and broaden graduates’ occupational choices. For example, a graduate with a double major in economics and education may be more likely to choose an occupation related to education than a graduate with a single major in economics. Third, pursuing a second lower paying major may be perceived unfavorably in the labor market. Although we were unable to test this possibility directly, for this mechanism to work at all there must be some ability sorting associated with choosing a double major. Subsequently, we compared the normalized college GPAs of double and single major graduates for subjects in which both of them earned credits (see figure A.1 in the appendix). Students double majoring in a lower paying major had lower GPAs relative to their peers with a single high-paying major. In contrast, double major students had higher GPAs compared with those with a single low-paying major. These results are at least partially consistent with the possibility of ability sorting in the decision to pursue a double major.

5.  Discussion and Conclusion

Although a large number of college students graduate with a double major, our knowledge of the effects of double majoring is insufficient. In this paper, we used data from the B&B:08/12 to estimate the effects of a double major by applying a PSW approach. Our analyses yielded several important results regarding postgraduate earnings. First, graduates with a double major earn less on average than those with the single higher paying major, but more than those with the single lower paying major constituting the double major combination. Second, the relative standing of graduates with a double major improved between one year after graduation and four years after graduation. In other words, the disadvantage relative to those with the single higher paying major decreased, and the advantage relative to those with the single lower paying major increased. Third, the effect of a double major varies across major combinations. When combining two majors with a larger earnings gap, the disadvantage relative to those with the single higher paying major and the advantage relative to those with the single lower paying major both increase. Finally, a minor tends to result in a lower economic return than a double major when compared with the same reference group.

The initial earnings penalty of having a double major compared with having the single higher paying major is significantly mitigated by positive effects on a variety of employment-related outcomes, including labor force participation and employment status four years after graduation, relative to having the single lower paying major. In addition, having a double major can significantly increase the probability of enrolling in a graduate school program within four years after graduation, suggesting that some benefits of double majoring are realized through increased educational attainment and related monetary or nonmonetary returns. As B&B:08/12 only followed up with respondents four years after graduation, about 20 percent of respondents were still enrolled in graduate school programs, including doctoral and professional degree programs (e.g., medicine, law) which typically are associated with substantially higher earnings after graduation. The estimated returns in this paper, therefore, likely underestimate the benefits of double majoring in the long run. We encourage researchers to examine the long-term effects of double majoring once follow-up data become available.

The improvement of the relative standing of graduates with a double major between one and four years after college graduation is intriguing. Interestingly, in studying the time-varying effects of college quality on graduate earnings, Thomas and Zhang (2005) reached a similar finding that graduates from higher-quality colleges enjoyed a greater rate of growth in earnings during their early career. One possible explanation is that graduates with double majors, especially those with relatively low initial earnings, are more likely to attend graduate school than their peers with a single major. Another possibility could be that graduates with double majors may have quite different career paths than their peers with a single major. For example, they could be more likely to get a promotion or land a new job. Future research may examine the economic returns and career paths of graduates with double majors over a longer run.

In conclusion, we offer some suggestions to students who are considering pursuing a double major. The findings of our analyses suggest that, while on average students with a double major compare favorably to those with a single major, they could earn less than those with the single higher paying major when there is a large earnings gap between the higher and lower paying majors in their double major combination. In other words, having a double major does not necessarily always yield positive economic returns. It only makes economic sense to choose a second major with higher earnings power than the first major. Pursuing a second major with low market value does not help, and in fact may even dampen one's economic prospects. Likewise, because in many cases the earnings power of a double major, on average, falls somewhere between the earnings of the two single majors, it may make sense to simply switch to a higher paying major rather than pursue it as a second major. Our results also suggest that students who pursue a double major might want to take more classes in the higher paying major because deeper knowledge might make them more competitive in the labor market. As double majors continue to be a trending phenomenon in colleges and universities, administrators should provide more guidance on major choice and course selection for students who want to pursue a double major to help them obtain interdisciplinary learning experience while developing specialized knowledge with enough depth to make them valuable in the labor market.

Notes

1. 

It is not common for students to claim more than two majors. For instance, in the Baccalaureate and Beyond Longitudinal Study 2008--2012, only thirty out of 1,460 students with at least two majors claimed three or more majors. In this paper, we focus on students with double majors and exclude respondents with more than two majors.

3. 

All sample sizes reported in this study are rounded off to the nearest ten, according to Institute of Education Sciences requirements.

4. 

We coded majors based on their order on the transcript (e.g., Bachelor of Arts, English, Psychology would have been coded as First Major = English, Second Major = Psychology). Therefore, the order of majors depends on how institutions identify double majors. Most institutions define the first major as the one that is declared first.

5. 

The B&B:08/12 measured earnings as the annual salary associated with a respondent's primary job at the time of data collection. If earnings were reported as a monthly, biweekly, weekly, daily, or hourly rate, the B&B used the appropriate multiplier to convert earnings to annual salaries. A respondent's primary job as his or her current or most recent position lasting longer than three months; if more than one job met these criteria, the job with the most hours worked per week was selected as the primary job. Self-employed respondents are included in the sample.

6. 

Employment status consists of out of the labor force, unemployed (not working but looking for work), part-time employed, and full-time employed. Respondents with multiple jobs were coded as full-time employed if any jobs were full-time; if all jobs were part-time, respondents were coded as part-time employed. We measured working hours based on the number of self-reported hours worked per week.

7. 

Post-bachelor's degrees include master's, doctoral, and professional (e.g., medicine, law) degrees.

8. 

Among the observations, 1.2 percent were missing “first-generation status,” 17.3 percent were missing “SAT score,” 13.4 percent were missing “earned college credits in high school,” 0.7 percent were missing “number of Advanced Placement credits,” and 0.2 percent were missing “number of total credits.” We used multivariate normal regression to impute the missing values. Predictors in the imputation included gender, age, race, family income, and institutional control and selectivity.

9. 

Independent variables in this model include nineteen dummies for major categories, gender, age, age squared, race/ethnicity, family income, first-generation college graduate, institutional control and selectivity, GPA, total credits, and work experience and its square term.

10. 

The coefficients in these log earnings regressions are in log points, which are similar to percent change when coefficients are small. For large values, the transformation [exp(b) − 1] is used.

11. 

The largest earnings power gap in our data is 0.5, which comes from respondents with double majors in health-related clinical sciences (0.443) and education (−0.057).

12. 

For a list of occupations in each group, see Appendix table A.4. The earnings power of an occupation is derived by the same strategy used for the earnings power of a major. We ran the earnings function based on the single major sample only; the independent variables include occupations (thirty-three categories), gender, age, age squared, race/ethnicity, family income, first-generation college graduate, institutional control and selectivity, GPA, total credits, and work experience and its squared term. The coefficients extracted from this model can be interpreted as the earnings power of each occupation.

Acknowledgments

We would like to thank participants at the 42nd Annual Conference for the Association for Education Finance and Policy and two anonymous reviewers for their valuable questions and comments.

REFERENCES

Altonji
,
Joseph G.
,
Peter
Arcidiacono
, and
Arnaud
Maurel
.
2016
. The analysis of field choice in college and graduate school: Determinants and wage effects.
In Handbook of the economics of education
Vol.
5
,
edited by
Eric A.
Hanushek
,
Stephen
Machin
, and
Ludger
Woessmann
, pp.
305
396
.
Amsterdam
:
North Holland
. https://doi.org/10.1016/B978-0-444-63459-7.00007-5.
Altonji
,
Joseph G.
,
Erica
Blom
, and
Costas
Meghir
.
2012
.
Heterogeneity in human capital investments: High school curriculum, college major, and careers
.
NBER Working Paper No. 17985
. https://doi.org/10.3386/w17985.
Altonji
,
Joseph G.
,
Todd E.
Elder
, and
Christopher R.
Taber
.
2005
.
Selection on observed and unobserved variables: Assessing the effectiveness of Catholic schools
.
Journal of Political Economy
113
(
1
):
151
184
.
Arcidiacono
,
Peter
.
2004
.
Ability sorting and the returns to college major
.
Journal of Econometrics
121
(
1–2
):
343
375
. https://doi.org/10.1016/j.jeconom.2003.10.010.
Austin
,
Peter C.
2009
.
Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples
.
Statistics in Medicine
28
(
25
):
3083
3107
.
Carnevale
,
Anthony P.
,
Ban
Cheah
, and
Jeff
Strohl
.
2012
.
Hard times: College majors, unemployment and earnings: Not all college degrees are created equal
.
Washington, DC
:
Center on Education and the Workforce
.
Del Rossi
,
Alison F.
, and
Joni
Hersch
.
2008
.
Double your major, double your return
?
Economics of Education Review
27
(
4
):
375
386
. https://doi.org/10.1016/j.econedurev.2007.03.001.
Del Rossi
,
Alison F.
, and
Joni
Hersch
.
2016
.
The private and social benefits of double majors
.
Journal of Benefit-Cost Analysis
7
(
2
):
292
325
. doi:10.1017/bca.2016.14.
Demnati
,
Abdellatif
, and
Jnk
Rao
.
2004
.
Linearization variance estimators for survey data
.
Survey Methodology
30
(
1
):
17
26
.
DuGoff
,
Eva H.
,
Megan
Schuler
, and
Elizabeth A.
Stuart
.
2014
.
Generalizing observational study results: Applying propensity score methods to complex surveys
.
Health Services Research
49
(
1
):
284
303
.
Guo
,
Shenyang
, and
Mark W.
Fraser
.
2015
.
Propensity score analysis: Statistical methods and applications
. 2nd
edition, volume 11
.
Thousand Oaks, CA
:
SAGE Publishing
.
Hemelt
,
Steven W.
2010
.
The college double major and subsequent earnings
.
Education Economics
18
(
2
):
167
189
. https://doi.org/10.1080/09645290802469931.
Hirano
,
Keisuke
, and
Guido W.
Imbens
.
2001
.
Estimation of causal effects using propensity score weighting: An application to data on right heart catheterization
.
Health Services and Outcomes Research Methodology
2
(
3
):
259
278
. https://doi.org/10.1023/A:1020371312283
Nichols
,
Austin
.
2008
.
Erratum and discussion of propensity-score reweighting
.
Stata Journal
8
(
4
):
532
539
.
Pitt
,
Richard N.
, and
Steven A.
Tepper
.
2012
. Double majors: Influences, identities, and impacts.
Nashville, TN
:
The Curb Center for Art, Enterprise & Public Policy at Vanderbilt University
.
Robst
,
John
.
2007
.
Education and job match: The relatedness of college major and work
.
Economics of Education Review
26
(
4
):
397
407
. https://doi.org/10.1016/j.econedurev.2006.08.003.
Rosenbaum
,
Paul R.
, and
Donald B.
Rubin
.
1983
.
The central role of the propensity score in observational studies for causal effects
.
Biometrika
70
(
1
):
41
55
. https://doi.org/10.2307/2335942.
Rumberger
,
Russell W.
, and
Scott L.
Thomas
.
1993
.
The economic returns to college major, quality and performance: A multilevel analysis of recent graduates
.
Economics of Education Review
12
(
1
):
1
19
. https://doi.org/10.1016/0272-7757(93)90040-N.
Taylor
,
Paul
,
Rick
Fry
, and
Russ
Oates
.
2014
.
The rising cost of not going to college
.
Available
https://www.pewsocialtrends.org/2014/02/11/the-rising-cost-of-not-going-to-college/.
Accessed 1 June 2020
.
Thomas
,
Scott L.
, and
Liang
Zhang
.
2005
.
Post-baccalaureate wage growth within four years of graduation: The effects of college quality and college major
.
Research in Higher Education
46
(
4
):
437
459
. https://doi.org/10.1007/s11162-005-2969-y.
Zafar
,
Basit
.
2012
.
Double majors: One for me, one for the parents
?
Economic Inquiry
50
(
2
):
287
308
.

Appendix

Table A.1.
Sample Size by Employment and Enrollment Status
20092012
Initial sample in 2008 15,720 15,720 
Analytical sample in follow-ups 13,700 13,280 
Enrolled 3,540 2,620 
Not enrolled 10,160 10,660 
Out of labor force 400 1,010 
Unemployed 950 900 
Employed 8,810 8,750 
Part-time 2,450 840 
Full-time 6,360 7,910 
20092012
Initial sample in 2008 15,720 15,720 
Analytical sample in follow-ups 13,700 13,280 
Enrolled 3,540 2,620 
Not enrolled 10,160 10,660 
Out of labor force 400 1,010 
Unemployed 950 900 
Employed 8,810 8,750 
Part-time 2,450 840 
Full-time 6,360 7,910 

Notes: The initial sample in 2008 limits to respondents who received bachelor's degrees from a four-year institution between July 2007 and June 2008 for whom major information and transcript data are available. The analytical sample limits the initial sample by excluding respondents with no information on employment and enrollment status in 2009 and 2012, respectively. All sample sizes are rounded off to the nearest ten per Institute of Education Sciences requirements.

Table A.2.
Descriptive Statistics of Independent Variables by Double Major Status
Single MajorDouble Major
MeanSDMeanSD
Demographic and family background     
Female 0.574 0.495 0.604 0.489 
Age 25.16 6.990 23.12 4.293 
White 0.726 0.446 0.750 0.433 
Black or African American 0.0938 0.292 0.0607 0.239 
Hispanic 0.0936 0.291 0.0843 0.278 
Asian 0.0559 0.230 0.0642 0.245 
Other race/ethnicity 0.0309 0.173 0.0403 0.197 
Family income ($1,000) 72.85 64.44 84.02 70.72 
First generation 0.456 0.498 0.366 0.482 
Academic performance     
SAT score 1072.8 185.5 1144.0 198.4 
Earned college credits in high school 0.434 0.496 0.516 0.500 
Number of Advanced Placement credits 1.007 3.704 2.099 6.971 
Normalized GPA (0—4) 3.250 0.476 3.331 0.484 
Total credits earned 123.1 31.35 130.8 35.69 
Institutional control and selectivity     
Public 0.630 0.483 0.589 0.492 
Private not-for-profit 0.321 0.467 0.394 0.489 
Private for-profit 0.0490 0.216 0.0165 0.127 
Most competitive 0.0559 0.230 0.134 0.341 
Highly competitive 0.109 0.311 0.118 0.322 
Very competitive 0.239 0.427 0.286 0.452 
Competitive 0.389 0.488 0.334 0.472 
Less competitive 0.0753 0.264 0.0469 0.211 
Noncompetitive 0.0265 0.161 0.0310 0.173 
Special focus 0.0150 0.121 0.00368 0.0606 
Single MajorDouble Major
MeanSDMeanSD
Demographic and family background     
Female 0.574 0.495 0.604 0.489 
Age 25.16 6.990 23.12 4.293 
White 0.726 0.446 0.750 0.433 
Black or African American 0.0938 0.292 0.0607 0.239 
Hispanic 0.0936 0.291 0.0843 0.278 
Asian 0.0559 0.230 0.0642 0.245 
Other race/ethnicity 0.0309 0.173 0.0403 0.197 
Family income ($1,000) 72.85 64.44 84.02 70.72 
First generation 0.456 0.498 0.366 0.482 
Academic performance     
SAT score 1072.8 185.5 1144.0 198.4 
Earned college credits in high school 0.434 0.496 0.516 0.500 
Number of Advanced Placement credits 1.007 3.704 2.099 6.971 
Normalized GPA (0—4) 3.250 0.476 3.331 0.484 
Total credits earned 123.1 31.35 130.8 35.69 
Institutional control and selectivity     
Public 0.630 0.483 0.589 0.492 
Private not-for-profit 0.321 0.467 0.394 0.489 
Private for-profit 0.0490 0.216 0.0165 0.127 
Most competitive 0.0559 0.230 0.134 0.341 
Highly competitive 0.109 0.311 0.118 0.322 
Very competitive 0.239 0.427 0.286 0.452 
Competitive 0.389 0.488 0.334 0.472 
Less competitive 0.0753 0.264 0.0469 0.211 
Noncompetitive 0.0265 0.161 0.0310 0.173 
Special focus 0.0150 0.121 0.00368 0.0606 

Notes: Based on unimputed data. Panel and transcript weights are applied, and variance are estimated by Taylor linearized method. “Age” is measured in 2008. “Family income” is family income (dependents’ parents or independents) in 2006. “First generation” is defined as neither of parents has a bachelor's degree and beyond. Institutional selectivity index is from Barron's College Competitiveness Index 2008. Financial variables are adjusted to 2012 dollars using the Consumer Price Index. SD = standard deviation; GPA = grade point average.

Table A.3.
Descriptive Statistics of Post-Baccalaureate Outcomes by Double Major Status
2009 (1 Year Post-BA)2012 (4 Years Post-BA)
Single MajorDouble MajorSingle MajorDouble Major
MeanSDMeanSDMeanSDMeanSD
Annualized salary, dollarsa 39,126.8 461.9 37,576.0 1,200.5 45,204.5 30,535.7 45,270.9 27,974.8 
Labor force participation rateb 0.967 0.178 0.965 0.183 0.906 0.291 0.932 0.252 
Employment ratec 0.909 0.288 0.913 0.283 0.909 0.287 0.947 0.224 
Full-time job ratea 0.729 0.444 0.742 0.438 0.907 0.290 0.925 0.264 
Working hours per weeka 32.37 17.72 33.08 18.46 39.84 11.37 41.11 11.88 
Work experiencea 14.63 5.505 14.16 4.975 44.89 15.23 44.48 14.14 
Job not related to majora 0.269 0.443 0.259 0.438 0.230 0.421 0.227 0.419 
Job somewhat related to majora 0.267 0.443 0.286 0.452 0.335 0.472 0.356 0.479 
Job closely related to majora 0.464 0.499 0.456 0.498 0.435 0.496 0.416 0.493 
Graduate school enrollment rated 0.216 0.411 0.254 0.435 0.317 0.465 0.412 0.492 
Have a post-bachelor's degreed 0.0216 0.145 0.0299 0.170 0.195 0.396 0.242 0.428 
2009 (1 Year Post-BA)2012 (4 Years Post-BA)
Single MajorDouble MajorSingle MajorDouble Major
MeanSDMeanSDMeanSDMeanSD
Annualized salary, dollarsa 39,126.8 461.9 37,576.0 1,200.5 45,204.5 30,535.7 45,270.9 27,974.8 
Labor force participation rateb 0.967 0.178 0.965 0.183 0.906 0.291 0.932 0.252 
Employment ratec 0.909 0.288 0.913 0.283 0.909 0.287 0.947 0.224 
Full-time job ratea 0.729 0.444 0.742 0.438 0.907 0.290 0.925 0.264 
Working hours per weeka 32.37 17.72 33.08 18.46 39.84 11.37 41.11 11.88 
Work experiencea 14.63 5.505 14.16 4.975 44.89 15.23 44.48 14.14 
Job not related to majora 0.269 0.443 0.259 0.438 0.230 0.421 0.227 0.419 
Job somewhat related to majora 0.267 0.443 0.286 0.452 0.335 0.472 0.356 0.479 
Job closely related to majora 0.464 0.499 0.456 0.498 0.435 0.496 0.416 0.493 
Graduate school enrollment rated 0.216 0.411 0.254 0.435 0.317 0.465 0.412 0.492 
Have a post-bachelor's degreed 0.0216 0.145 0.0299 0.170 0.195 0.396 0.242 0.428 

Notes: BA = bachelor's degree; SD = standard deviation.

aBased on the respondents who were employed and not enrolled.

bBased on the respondents who were not enrolled.

cBased on the respondents who were in the labor force and not enrolled.

dBased on all respondents who had participated 2009 and 2012 interviews and had valid transcript data.

Table A.4.
List of Majors by Earnings Power
Major CategoriesEarnings Power
Languages and literatures −0.088 
Visual and performing arts −0.086 
Education −0.057 
Philosophy, religious studies, and history −0.042 
Agriculture and natural resources 0.000 
Communication and related 0.003 
Multi/interdisciplinary studies 0.033 
Psychology 0.046 
Social science (exclude economics) 0.052 
Personal service/family studies/recreations 0.058 
Public administration/social service 0.060 
Security and protective service 0.073 
Biology and biomedical sciences 0.084 
Liberal art and Humanities 0.093 
Physical science and science tech 0.179 
Business/economics 0.194 
Mathematics and statistics 0.257 
Computer science 0.356 
Engineering and engineering technology 0.369 
Health/related clinical sciences 0.443 
Major CategoriesEarnings Power
Languages and literatures −0.088 
Visual and performing arts −0.086 
Education −0.057 
Philosophy, religious studies, and history −0.042 
Agriculture and natural resources 0.000 
Communication and related 0.003 
Multi/interdisciplinary studies 0.033 
Psychology 0.046 
Social science (exclude economics) 0.052 
Personal service/family studies/recreations 0.058 
Public administration/social service 0.060 
Security and protective service 0.073 
Biology and biomedical sciences 0.084 
Liberal art and Humanities 0.093 
Physical science and science tech 0.179 
Business/economics 0.194 
Mathematics and statistics 0.257 
Computer science 0.356 
Engineering and engineering technology 0.369 
Health/related clinical sciences 0.443 

Notes: The earnings power is defined as the coefficient of a major dummy estimated from the earnings function based on single-major graduates after controlling for demographic and family background, academic performance, institutional control and selectivity, and work experience and its square.

Table A.5.
Number of Credits Taken in All Fields and Within the Field of First Major
Business/EconomicsBiological SciencePsychologyEducationPhilosophy/Religion/History
AllWithinAllWithinAllWithinAllWithinAllWithin
FieldsFieldFieldsFieldFieldsFieldFieldsFieldFieldsField
Single major 116.9 18.6 126.3 15.6 116.4 13.3 132.1 17.0 114.4 16.3 
Double up 126.7 12.6 137.0 13.8 129.6 10.5 136.4 12.6 127.7 13.7 
Double down 128.7 13.3 139.5 14.1 125.5 11.5 139.6 12.7 133.1 15.1 
Minor up 127.1 17.8 131.1 15.9 122.6 12.9 128.9 14.2 124.2 14.3 
Minor down 126.3 17.2 132.1 14.4 120.5 13.2 145.2 16.3 128.6 14.5 
Business/EconomicsBiological SciencePsychologyEducationPhilosophy/Religion/History
AllWithinAllWithinAllWithinAllWithinAllWithin
FieldsFieldFieldsFieldFieldsFieldFieldsFieldFieldsField
Single major 116.9 18.6 126.3 15.6 116.4 13.3 132.1 17.0 114.4 16.3 
Double up 126.7 12.6 137.0 13.8 129.6 10.5 136.4 12.6 127.7 13.7 
Double down 128.7 13.3 139.5 14.1 125.5 11.5 139.6 12.7 133.1 15.1 
Minor up 127.1 17.8 131.1 15.9 122.6 12.9 128.9 14.2 124.2 14.3 
Minor down 126.3 17.2 132.1 14.4 120.5 13.2 145.2 16.3 128.6 14.5 

Notes: Double up/down refers to having a second major in a higher/lower paying major; minor up/down refers to having a minor in a higher/lower paying major. Credits taken within the field of first major are credit hours of courses with the same two-digit Classification of Instructional Programs code of their first major.

Figure A.1.

GPA Difference Between Double Majors and Single Majors in Selected Subjects

Notes: “Double Major (HH)-Single Major (H)” represents that students with two highly paid majors are compared with those with a single highly paid major. The positive (negative) difference for a subject represents that double major graduates, on average, had a higher (lower) grade point average (GPA) in the subject than single major graduates.

Figure A.1.

GPA Difference Between Double Majors and Single Majors in Selected Subjects

Notes: “Double Major (HH)-Single Major (H)” represents that students with two highly paid majors are compared with those with a single highly paid major. The positive (negative) difference for a subject represents that double major graduates, on average, had a higher (lower) grade point average (GPA) in the subject than single major graduates.