Higher Education’s Return on Investment: Methodology

The FREOPP research project Does College Pay Off? A Comprehensive Return on Investment Analysis calculates the net financial gains that students can expect from tens of thousands of postsecondary degree programs across America. The project incorporates data from the College Scorecard, the American Community Survey, the National Longitudinal Survey of Youth, the National Survey of College Graduates, the National Postsecondary Student Aid Study, and the Integrated Postsecondary Education Data System to build an interactive database to help prospective students make better choices around higher education — including whether to go, where to go, and what to study.

This page describes the methodology behind the construction of the return on investment (ROI) database. It explains the process of calculating ROI for bachelor’s degrees, along with graduate degrees and subbaccalaureate credentials. The ROI calculation combines four elements:

• Realized earnings: The amount that a student who graduates with a particular postsecondary credential can expect to earn over the course of her life.
• Counterfactual earnings: The amount that the same student would have earned over the course of her life had she not pursued the credential. This number includes the income that the student would have earned while she was in school.
• Costs of education: The amount that the student spends on tuition, required fees, and other education-related expenses such as books and equipment. It does not include living expenses such as rent and food, as students must spend money on these regardless of whether they attend college.
• Completion rate: The chance that the student will finish her degree program and, therefore, realize the earnings benefits of the credential.

ROI is equivalent to realized earnings minus the sum of counterfactual earnings and college costs. My preferred estimates of ROI incorporate the possibility that the student will not complete her program, and hence fail to attain the realized earnings associated with that degree. The following sections describe how each component of the ROI calculation was constructed, and how it figures into the overall estimate.

All cash flows are discounted at a real three percent rate to the year in which the student begins the degree or certificate program. For undergraduate credentials, I assume this is the year in which the student turns 18; for graduate credentials, the starting year varies by degree type. Unless otherwise noted, I adjust all figures to 2022 dollars.

Realized earnings

My estimates of realized earnings combine two sources of data: the U.S. Department of Education’s College Scorecard by Field of Study (Scorecard) and the U.S. Census Bureau’s American Community Survey (ACS). Data for the latter were accessed through the IPUMS USA database at the University of Minnesota.

Scorecard provides information on the median earnings of graduates of tens of thousands of higher education programs. It reports data at the institution-program level: For example, Scorecard records not just the median earnings of University of Maryland graduates, but also the median earnings of University of Maryland graduates with a degree in economics.

Scorecard earnings are defined as the sum of wage and salary (Form W-2) income and positive self-employment income for individuals who received federal Title IV financial aid. Those who were not employed are excluded from earnings cohorts. Institution-program observations with small cohort sizes are suppressed for privacy reasons, so earnings data are not available for all degrees and certificates listed in Scorecard.

I use Scorecard’s earnings estimates for a cohort of students who graduated in the 2015–16 and 2016–17 academic years. Scorecard records median earnings for this cohort of graduates in the first, second, third, and fourth calendar years after graduation. Roughly 55,000 programs with graduates in this cohort have at least one earnings data point in Scorecard, while 34,000 programs have earnings data for all four years.

As long as Scorecard records at least one earnings data point for a program, I use regression analysis to impute the program’s expected earnings for the missing years. The regression predicts the missing earnings data points based on the observed earnings data points for the same program, as well as the evolution of earnings for programs at the same credential level and in the same field of study.

However, the Scorecard’s central limitation is its short time horizon; it only measures student earnings in the first four years after graduation. Since wage profiles tend to be steep in the early career, Scorecard earnings are likely to underestimate the lifetime returns to a postsecondary credential. Therefore, I augment the Scorecard figures with data from ACS.

Conducted by the U.S. Census Bureau, the ACS surveys a random one percent sample of the U.S. population every year, collecting basic demographic data along with information on individuals’ earnings. Since 2009, ACS has also asked bachelor’s degree holders about their undergraduate college major. ACS does not, however, ask respondents where they attended college.

To estimate life-cycle earnings for Scorecard programs I use a subsample of the ACS datasets for the years 2013 through 2022, excluding 2020 as disruptions to data collection during the COVID-19 pandemic make that sample unreliable. To align with the cohort definition in Scorecard, I exclude those currently enrolled in school or college, people who are unemployed or not in the labor force, noncitizens, and people with real earnings below $12,000 per year.

I then divide the ACS sample according to the respondent’s level of education, corresponding to each degree level in Scorecard (undergraduate certificates, associate degrees, bachelor’s degrees, master’s degrees, and doctoral or professional degrees). I further divide the ACS sample of bachelor’s degree holders into 21 categories based on field of study. These subcategories yield estimates of the distribution of earnings for people with various types of postsecondary education over the course of their lifetimes.

The following chart shows the median earnings for ACS respondents by credential level and age, demonstrating how earnings evolve over the course of a graduate’s career. I will use these figures to estimate how much graduates of Scorecard programs will earn at later stages in life. (Note: I assume the category “1+ years college credit, no degree” corresponds to undergraduate certificates in Scorecard, though the overlap is not perfect as the former category includes some college dropouts.)

One challenge concerns individuals who earn a higher degree after receiving a lower degree, especially bachelor’s degree completers who later earn a graduate degree. I must disaggregate the portion of their earnings attributable to the graduate degree versus the bachelor’s degree. Simply excluding them from the analysis would bias my estimates, since individuals who choose to go to graduate school may have higher earnings potential than their peers with the same major who earn only a bachelor’s degree.

The solution is to estimate what people with a graduate degree would have earned had they instead stopped out of school with only a bachelor’s degree. I turn to the National Longitudinal Survey of Youth, 1997 cohort (NLSY) to tease out the portion of the graduate degree earnings premium (over earnings for people with just a bachelor’s degree) that is attributable to the graduate degree itself, versus the portion that is attributable to selection bias, or the fact that people who pursue graduate degrees may have higher earnings potential than their peers.

NLSY collects a rich suite of variables on individuals’ cognitive ability, personality traits, and family background; all of which are correlated with the decision to attend graduate school and future earnings potential. By controlling for these variables when I regress earnings on graduate school attendance, I observe how the graduate degree earnings premium shrinks when accounting for selection bias. The portion of the graduate degree earnings premium attributable to selection bias averages around 15 percent, though it varies by field of study.

More details on the NLSY and how it can be used to account for selection bias are available in the section of this report that deals with counterfactual earnings.

Armed with this decomposition of the graduate degree earnings premium, I adjust the earnings of graduate degree holders downward by the amount of the graduate degree premium not attributable to selection bias. I call the result counterfactual lower-degree (CLD) earnings. I include the CLD earnings of graduate degree holders in my ACS subsample of individuals with only a bachelor’s degree.

Similarly, I compute CLD earnings for bachelor’s degree holders — i.e., what bachelor’s degree holders would have earned had they stopped out with an associate degree — and include these data points in the ACS subsample of individuals with only an associate degree. In other words, bachelor’s degree holders enter two different ACS subsamples: their CLD earnings enter the subsample of associate degree holders, while their true (observed) earnings enter the ACS sample of bachelor’s degree holders. The same is true for graduate degree holders: their CLD earnings are in the ACS subsample of bachelor’s degree holders, while their true earnings are in the ACS subsample of graduate degree holders.

A further, related challenge concerns the inconsistent manner in which Scorecard reports earnings for graduates who go on to earn a higher credential than the one being measured. For the cohort of graduates that I use in the analysis, Scorecard excludes individuals who earned a higher credential when computing median earnings one and two years after graduation but includes individuals who earned a higher credential when computing median earnings three and four years after graduation. (Individuals currently enrolled in higher degree programs are always excluded from median earnings.)

Ideally, the Scorecard median earnings figures would include true earnings for individuals who stop out with the credential being measured, along with CLD earnings for individuals who go on to earn a higher credential. This would yield a more precise measure of the value of each credential: it would adequately reflect the earnings potential for the subset of individuals who go on to higher degrees but would not unduly credit the lower credential for earnings, which are attributable to the higher credential.

To solve this problem, I use ACS to create two earnings distributions for people with at least a bachelor’s degree. The first distribution includes the true earnings of individuals with a bachelor’s degree or higher but includes CLD rather than true earnings for individuals with a graduate degree (the “CLD” distribution). The second distribution includes the true earnings of all individuals with a bachelor’s degree or higher (the “observed” distribution).

To adjust the Scorecard earnings of students three and four years after graduation, I compare the observed and CLD distributions for bachelor’s degree holders aged 25 and 26, respectively. I measure the percentile rank of each Scorecard program on the observed distribution of ACS respondents in the same field of study. I then record the earnings of ACS respondents with the same percentile rank on the comparable CLD distribution; this becomes my measure of earnings for graduates of that program three and four years after graduation.

For instance, the fourth-year Scorecard earnings of graduates of Swarthmore College’s economics program are $112,100 (in 2022 dollars). This corresponds to the 80th percentile of earnings on the observed earnings distribution of economics majors aged 26. The 80th percentile of earnings on the CLD distribution of economics majors aged 26 is $109,400. Therefore, I assign Swarthmore economics graduates a fourth-year earnings figure of $109,400, as this does not unduly credit the bachelor’s degree program with earnings that should be attributed to higher degrees.

The logic behind this adjustment is that the CLD and observed earnings distributions represent the same population at different levels of earning potential. The observed earnings distribution reflects some people with bachelor’s degrees only and some people with graduate degrees; the CLD distribution reflects the same groups, but records CLD rather than true earnings for those with graduate degrees. It follows that the 80th percentile on the CLD distribution should approximate the 80th percentile on the observed distribution if graduate degree holders’ earnings were adjusted for the impact of the higher degree. I apply a similar adjustment for associate degree programs.

These adjustments do not make a large difference to Scorecard earnings: less than five percent for the vast majority of programs. The reason is that most college graduates in the first few years after completion do not earn advanced degrees, so there is little change to the median earnings for most programs.

After generating satisfactory estimates of earnings for students in the first four years after graduation, I turn my attention to estimating earnings for the rest of graduates’ careers. I start with each Scorecard program’s fourth-year percentile rank on the applicable earnings distribution. For the Swarthmore economics program, recall that this program is at the 80th percentile for ACS economics majors aged 26.

I use the fourth-year percentile rank to generate earnings estimates for the rest of students’ careers, using the applicable earnings distribution for ACS respondents at later stages of life. This requires an assumption that the median graduate of each program stays at roughly the same relative position on the earnings distribution throughout their entire careers. For the Swarthmore economics example, I estimate that fifth-year earnings are equal to the 80th percentile of CLD earnings for ACS economics majors aged 27 ($117,200). Tenth-year earnings are equal to the 80th percentile of CLD earnings for ACS economics majors aged 32 ($162,000). And so on, until the conclusion of the cohort’s career at age 65.

Estimating career earnings for people with graduate degrees presents additional complications. Since ACS does not record graduate degree holders’ graduate field of study, I must determine this based on their undergraduate field of study. I use the National Survey of College Graduates (NSCG), which does record both graduate and undergraduate fields, to determine the proportions of undergraduate majors within each graduate field. I use these figures from NSCG to weight the ACS subsample that I use to estimate the earnings distribution for each graduate degree.

For individuals with a graduate degree in medicine, for instance, NSCG reports that 55 percent majored in biology and life sciences as undergraduates, 14 percent majored in health fields, 8 percent majored in the physical sciences, and the rest were split across various majors. Therefore, my earnings estimates of people with a medical degree use an ACS sample of people with professional degrees that is weighted 55 percent on undergraduate biology majors, 14 percent on undergraduate health majors, and so forth.

I assume that people with undergraduate credentials start school at age 18 and graduate on time given the length of their program (this assumption will be relaxed later on). Associate degree students graduate at age 20, and bachelor’s degree students graduate at age 22. However, the starting ages for graduate students vary.

I use the 2019–20 National Postsecondary Student Aid Study (NPSAS) to estimate the median age of graduation for each credential type. For instance, medical students have a median graduation age of 27, so I assume their first full year of earnings is at age 28 and their fourth full year of earnings is at age 31. Therefore, I compare the fourth-year earnings of medical programs in Scorecard to the NSCG-weighted ACS distribution of people aged 31 with a professional degree. Medical programs’ percentile rank on this earnings distribution becomes the percentile rank I use to estimate medical students’ earnings for the rest of their careers.

Estimated earnings are then discounted at a three percent rate to the year in which the student begins school. Earnings are assumed to be zero while the student is enrolled.

Counterfactual earnings

With earnings estimates in hand, I turn to the calculation of counterfactual earnings: what graduates of each higher education program would have earned had they never attended the program. For undergraduate programs, counterfactual earnings are what students would have earned with only a high school degree. For graduate programs, counterfactual earnings are what students would have earned with only a bachelor’s degree.

However, simply comparing the earnings of college graduates and high school graduates is problematic. Those who choose to go to college — and especially those who complete college — differ in key ways from those who never earn a higher degree. In the academic literature, rigorous causal estimates of the financial return to college tend to lag the raw earnings gap between college graduates and high school graduates, though the causal return to college is still substantial. In other words, if college graduates had not gone to college, they would have still earned significantly more than the average high school graduate.

To calculate counterfactual earnings, I return to ACS. To begin the counterfactual earnings calculation for undergraduate programs, I take a subsample of individuals with at least a high school diploma or less than a year of college with no postsecondary degree. To align with the ACS sample of college graduates, I exclude those currently enrolled in school or college, people who are unemployed or not in the labor force, noncitizens, and people with real earnings below $12,000 per year. For graduate programs, I use a similar ACS subsample of individuals with exactly a bachelor’s degree.

I use the ACS sample to build an OLS regression model that predicts an individual’s log personal income based on age, sex, race, ethnicity, state, and metropolitan area. I use this model to predict counterfactual earnings for each Scorecard observation based on the racial and gender breakdown of the program and the location of the college.

The Integrated Postsecondary Education Data System (IPEDS) at the National Center for Education Statistics collects data on the racial and gender breakdown of program completers. My model adjusts counterfactual earnings based on the demographic composition of each program’s graduates. The model also adjusts counterfactual earnings based on the school’s location (state and metropolitan area). Different regions of the country have different labor market opportunities for high school graduates, and thus counterfactual earnings should differ by geography.

The Swarthmore economics example illustrates how the model estimates counterfactual earnings. The model estimates earnings for a high school graduate living in the Philadelphia-Camden-Wilmington metropolitan area, where Swarthmore is located. This fictional individual’s earnings are weighted based on gender and race: the economics program’s graduates are 39 percent white male, 15 percent white female, 8 percent black male, and so on. The model yields counterfactual earnings for a graduate’s entire lifetime, accounting for the fact that wages tend to grow as students age.

I generally follow the same procedure to estimate counterfactual earnings for graduate degrees, except that the ACS subsample fueling the regression model consists of people with exactly a bachelor’s degree. I also account for undergraduate major by using the National Survey of College Graduates to determine which undergraduate majors feed into which graduate degrees. For instance, since medical students are more likely to have biology as an undergraduate major, counterfactual earnings for medical students are weighted more heavily towards ACS respondents who majored in biology.

This model provides an estimate of counterfactual earnings after adjusting for basic demographics and geography. But there are many other factors that influence counterfactual earnings, including ability and family background. The ACS, however, does not collect data on most of those factors.

Instead, I turn to the National Longitudinal Survey of Youth, 1997 cohort (NLSY97). The NLSY is a nationally representative sample of 8,984 individuals born between 1980 and 1984. These individuals were first surveyed in 1997, when most were in high school, and periodically re-contacted afterwards. In addition to basic data on demographics, education, and income, NLSY collects comprehensive information on many other aspects of each individual’s background and identity.

NLSY allows me to account for the impact that differences in cognitive ability, motivation, and other factors may have on the wage premium of college graduates over high school graduates, otherwise known as selection bias. Decomposing the college wage premium into the portion attributable to higher education and the portion attributable to selection bias allows me to adjust my estimates of counterfactual earnings to account for selection bias. Moreover, I can use NLSY to examine how selection bias differs for different subpopulations of students, particularly students who choose different courses of study, and students of different socioeconomic backgrounds.

To exploit the NLSY97’s rich suite of variables, I follow a strategy used by Doug Webber (2014). First, I regress the log of respondents’ income on basic demographic information and a set of binary variables representing several mutually exclusive categories of students who completed a two-year or four-year college degree. High school diploma holders are the reference group. The coefficients on these binary variables represent the “unadjusted” college earnings premium associated with each student category.

I then run a second regression of income on the demographic and educational attainment variables, but I also include a rich set of controls for standardized test scores, high school characteristics, personality traits, self-perception, and family background. The coefficients on the educational attainment variables represent the “adjusted” earnings premium associated with each education category; in other words, the portion of the return to education that cannot be explained by measured ability, family background, or other observable factors.

The difference between the educational attainment coefficients between the first and second regressions represents the portion of the education premium that is attributable to selection bias. Because NLSY likely does not capture all student characteristics that are correlated with both college attendance and future earnings potential, measured selection bias likely understates true selection bias. Nevertheless, incorporating selection bias into my estimates of counterfactual earnings should improve their accuracy.

I calculate the selection bias adjustment separately for five categories of associate degree and 21 categories of bachelor’s degree, broken out by field of study. This breakout tells me how selection bias differs for engineering majors versus English majors, and so forth.

Within each degree category, I further calculate the selection bias adjustment separately based on whether the student would have qualified for a Pell Grant in 2019–20, based on their family’s socioeconomic characteristics. Lower-income students are likelier to qualify for a Pell Grant. This tells me how selection bias differs for students from different socioeconomic backgrounds. This is a significant augmentation of the methodology compared with FREOPP’s previous estimates of ROI.

I calculate the selection bias adjustment factors in log terms; they are listed in the table below for 26 undergraduate degree categories, for both Pell Grant and non-Pell Grant students. Selection bias adjustments for Pell Grant students are generally quite small and occasionally negative, but much larger for non-Pell Grant students. This means that students who would not qualify for a Pell Grant tend to have much higher preexisting earnings potential than the typical high school graduate, even without college, while the preexisting earnings potential of Pell Grant students is roughly in line with the typical high school graduate.

I use these estimates to calculate selection bias adjustments for each individual program. First, I match each undergraduate program to the appropriate field of study. (Undergraduate certificates, which are not identified in NLSY, are assumed to have a selection bias adjustment factor of zero; a reasonable assumption given the low adjustment factors for associate degrees.) Then, I calculate a selection bias adjustment factor for each program by averaging the adjustment factors for Pell and non-Pell students in that field of study, weighted by the actual share of Pell Grant students among each institution’s completing cohort for that degree level in 2019–20. This gives me a unique selection bias adjustment factor for each program, calibrated according to the Pell Grant share at its institution.

For bachelor’s degrees in economics, the adjustment factor is 0.03 for Pell Grant students and 0.22 for non-Pell Grant students. Swarthmore College’s Pell Grant share (among Title IV students who complete a bachelor’s degree) is 32 percent, so the selection bias adjustment factor for the Swarthmore economics program is equal to 0.03*0.32 + 0.22*(1–0.32), or 0.16.

Therefore, I take the “base” counterfactual earnings estimates for Swarthmore economics students, which are based on my ACS subsample of high school graduates, and adjust them upwards by 0.16 log points. I apply the same adjustment factor across the life cycle, which requires assuming that the selection bias adjustment factor remains constant throughout the individual’s life. At age 19, Swarthmore economics students’ counterfactual earnings are $32,300. By age 45, however, counterfactual earnings have risen to $62,400.

The process for adjusting graduate students’ counterfactual earnings for selection bias is similar. However, I do not separately calculate adjustment factors for Pell and non-Pell students because graduate students are ineligible for Pell Grants. Instead, I calculate a base estimate of counterfactual earnings from my ACS subsample of bachelor’s degree holders and adjust these estimates upwards by the appropriate selection bias adjustment factor for each graduate degree level and field of study. Selection bias adjustment factors for graduate degrees are generally small.

Counterfactual earnings are discounted at a real three percent rate to the year in which the student begins school. Note that I create counterfactual earnings estimates for both the period while students are enrolled and the period while they are in the workforce; the former represents foregone earnings while pursuing a degree, and the latter represents the base upon which the “boost” in earnings attributable to college is calculated.

For Swarthmore’s economics program lifetime realized earnings (discounted) sum to $3.7 million; lifetime counterfactual earnings (discounted) sum to $1.3 million. I therefore estimate the lifetime boost in earnings attributable to a Swarthmore economics degree as the difference between the two: $2.4 million.

College costs

The next element of the ROI calculation is the cost of higher education: how much the student and their family must pay to receive the estimated earnings boost. I calculate tuition after all sources of grant aid — federal, state, local, and institutional — but before student loan aid. I include the cost of books and supplies in college costs. However, I exclude living expenses, as students must pay for the basic costs of living regardless of whether they attend college, and the idea behind the ROI calculation is to assess how pursuing higher education changes a student’s financial position.

Data on tuition, required fees, and the cost of books and supplies is available from the Integrated Postsecondary Education Data System (IPEDS). For the most part, IPEDS collects tuition data at the institution level; in other words, one price is reported for all undergraduate programs, and another price is reported for all graduate programs. Occasionally, IPEDS collects tuition data for individual programs, especially vocational programs at the undergraduate level and professional programs such as law and medicine at the graduate level. Where a program-level price is available, I use that as my measure of tuition and fees; otherwise, I use the institution-level price. I use sticker prices for the 2019–20 academic year or the most recent prior year available.

The IPEDS tuition and fees variables are “sticker” prices before financial aid is applied. However, for undergraduate students, IPEDS also reports the average amount of federal, state, local, and institutional aid provided to full-time students in their first year of enrollment who receive Title IV federal financial aid. (A word of caution is in order here, as aid packages for non-Title IV students can differ substantially from aid to Title IV students: this analysis only looks at prices for the latter.) The variable also incorporates only students paying the in-state tuition rate for public universities.

Because many universities slash financial aid grants after freshman year, I reduce estimated first-year aid by a small amount to calculate aid for each subsequent year. According to the 2019–20 NPSAS, the average university reduces financial aid awards by 5 to 10 percent for sophomores, juniors, and seniors. There is also a somewhat larger aid reduction for students pursuing associate degrees.

IPEDS does not report institution-level financial aid awards for out-of-state undergraduate students at public universities or graduate students. Instead, I turn back to NPSAS to measure the average discount (off the sticker price) for these groups of students. Out-of-state, bachelor’s degree-seeking students at public institutions see financial aid packages worth around 38 percent of the sticker price in the first year, and less in subsequent years. I combine these average discount rates with institution-level sticker prices to estimate net tuition for out-of-state undergraduates.

NPSAS also records average discounts for graduate students. I measure these within sectors — public in-state, public out-of-state, private nonprofit, and private for-profit — for various programs. The following table displays the average net price divided by the average sticker price for different degrees within different sectors. These quotients range from 60 percent to 90 percent in most cases. To calculate the net price for a graduate degree, I multiply the sticker price by the appropriate percentage in the table below.

I discount all tuition payments at a three percent rate to the year in which the student begins their higher education program and subtract the sum of tuition payments from the earnings boost estimated in the previous section. The difference is the return on investment for students who graduate on time. For Swarthmore College’s economics program, the four-year discounted net cost is $123,000; I subtract this figure from the $2.42 million earnings boost to yield an ROI estimate of $2.29 million.

Adjustment for completion outcomes

An investment in higher education generally only pays off if the student actually receives a degree. But less than half of students graduate on time, and many don’t finish at all. For students uncertain about their chances of graduation, college is, therefore, a risky proposition. While the above calculations assume that the student’s likelihood of on-time graduation is 100 percent, a more comprehensive measure of ROI incorporates the differential likelihood of completion in various programs.

IPEDS reports data on completion outcomes for first-time, full-time students. For bachelor’s degree-seeking students, student counts are reported for several different completion outcomes: graduated in four years with a bachelor’s degree, graduated in five years with a bachelor’s degree, graduated in six years with a bachelor’s degree, graduated with a lower credential, transferred out, still enrolled after six years, and dropped out without earning a credential.

For bachelor’s degree programs, I compute ROI for four separate completion outcomes: finish a bachelor’s degree in four years, finish in five years, finish in six years, and drop out. Students who take five or six years to finish their degrees must pay five or six years’ tuition, stay out of the labor force for five or six years, and shorten their working careers by one or two years relative to someone who graduates on time. I assume dropouts leave school after two years — which is in line with estimates from the Beginning Postsecondary Students Longitudinal Study — and lose out on some labor market experience while enrolled.

I then weight these various completion outcomes by the proportion of students at the institution who actually ended up in those completion categories. When I calculate these proportions, I exclude transfer-out students and students who graduated with a lower credential from the denominator, while students who are still enrolled (generally a small number) are assumed to be non-completers.

The result is a “weighted average” ROI across the four completion outcomes. Using IPEDS data, I estimate that 87 percent of Swarthmore College students graduate in four years, 7 percent graduate in five years, 1 percent graduate in six years, and 5 percent drop out. Recall that ROI for on-time graduates of Swarthmore’s economics program is $2.29 million. ROI for students who graduate in five years is $2.1 million; for graduation in six years, $2.02 million; for dropouts, negative $163,000. Weighting these measures of ROI by Swarthmore’s actual completion rates yields an “adjusted” ROI figure of $2.16 million.

The process is similar for subbaccalaureate credentials, including associate degrees and certificates, except that IPEDS reports completion rate data slightly differently. I create three categories of completion outcomes for subbaccalaureate credentials: graduated within 100 percent of normal time, graduated within 150 percent of normal time, and dropped out. I calculate ROI separately for these three categories and compute a weighted-average ROI for subbaccalaureate credentials.

IPEDS does not report completion rates for graduate programs; instead, I turn to the 2008–18 Baccalaureate and Beyond survey (B&B), which I use to estimate completion rates at the sector-credential level. B&B is a random sample of individuals who earned a bachelor’s degree in the 2007–08 academic year. I use a subsample of B&B respondents who enrolled in a graduate degree program by 2012 and then record the proportion of this sample who had finished their degree program by 2018. This may be interpreted as the share of students who complete their degree programs within 6 to 10 years of enrollment; I assume that individuals who do not finish within this window do not finish at all. B&B-derived completion rates for graduate programs are listed in the table below.

I assume that the completion rate reflects the likelihood of finishing the degree on time and, therefore, securing the associated earnings gains. For students who fail to complete their programs, I assume they leave school halfway through the estimated program length. I calculate ROI separately for these two completion outcomes — graduated on time and dropped out — and compute a weighted average based on the credential type’s typical completion rate.

Adjustment for full cost of education

Most students do not bear the full cost of their education. Government financial aid programs, such as Pell Grants, shift some of the cost to taxpayers. Public colleges and universities receive appropriations from state governments that enable them to charge subsidized tuition. Even at private colleges, gaps can exist between net tuition and the underlying cost of education, plugged by tax-advantaged donations and tuition revenue from graduate programs. While prospective students should be most interested in whether the boost in earnings from their degree justifies the tuition they paid, other stakeholders may be interested in whether this earnings increase justifies underlying spending.

I create new measures of ROI that account for underlying spending. Essentially, these measures of ROI replace net tuition with the college’s education-related spending per full-time equivalent student in the ROI calculation. Education-related spending is the sum of spending on instruction, academic support, student services, and institutional support (administration). It excludes spending on research and public services, as well as spending on auxiliary enterprises such as dormitories and independent operations such as hospitals.

At Swarthmore College, annual net tuition is $33,200, while education-related spending is $94,100. Accounting for underlying spending reduces completion-adjusted ROI for the economics program from $2.16 million to $1.93 million. It is also possible to construct a measure of ROI that accounts for underlying spending but assumes on-time completion (ROI by this measure is $2.07 million for Swarthmore’s economics program).

Conclusion

Given the available data, there is no perfect way to calculate the return on investment of higher education. However, imperfect datasets and tools at our disposal allow a reasonable estimation of ROI, with caveats. As data collection on higher education outcomes improves, it should be possible to assemble some of the missing puzzle pieces and provide even more accurate estimates of ROI.