New Data on Long-Term Student Loan Default Rates

In recent years, more data have come out on how well students are able to manage repaying their loans beyond the three-year window currently used for federal accountability purposes (via cohort default rates). A great 2015 paper by Adam Looney and Constantine Yannelis used tax records merged with data from the National Student Loan Data System (NSLDS) to show longer-term trends in default in repayment. Two days later, the release of the College Scorecard provided college-level data on student loan repayment rates going out seven years (even though the repayment rates were initially calculated incorrectly).

Thanks to a lot of hard work by the data folks at the U.S. Department of Education and their contractor RTI, there are new data available on long-term student loan default rates. ED and RTI used NSLDS data going through 2015 to match records from the Beginning Postsecondary Students studies of cohorts beginning college in 1995-96 and 2003-04. This allowed a 20-year look at student loan default and payoff rates for the 1995-96 cohort and a 12-year look at the 2003-04 cohort, as detailed in this useful report from the National Center for Education Statistics.

Thanks to NCES’s wonderful PowerStats tool, I took a look at the percentage of students in the 2003-04 entering cohort (my college cohort) who had defaulted on at least one of their federal student loans within 12 years. Many of the news headlines focused on the high default rates of students at for-profit colleges (about 52%!), but this isn’t entirely a fair comparison because for-profit colleges tend to serve more economically-disadvantaged students. So in this post, I focused on racial/ethnic differences in default rates by type of college attended to give a flavor of what the data can do.

As the below chart shows, nearly half of all black students (49%) defaulted on at least one loan within 12 years—more than twice the rate of white students (20%) and more than four times the rate of Asian students (11%). The differentials are still present across sector, with more than one-third of black students defaulting across all sectors while a relatively small percentage of Asian students defaulted across all nonprofit sectors. Default rates at for-profit colleges are high for all racial/ethnic groups, with almost half of white students defaulting alongside nearly two-thirds of black students.

An advantage of the PowerStats tool is that it allows users to run regressions via NCES’s remote server. This allows interested people to analyze the relationship between long-term default rates and attending a for-profit college after controlling for other characteristics. However, PowerStats is overwhelmed by requests by my fellow higher education data nerds at this point, so I gave up on trying to run the regression after several hours of waiting. But if someone wants to run some regressions using the new loan repayment data in the BPS once the server calms down, I’m happy to feature their work on my blog!

Examining Trends in Student Loan Repayment Rates

It’s been a good week for higher education data nerds. The Department of Education released updated student loan cohort default rates on Wednesday afternoon (see my summary here), followed by an update to the massive College Scorecard dataset on Thursday morning. This is the third update to the Scorecard, with this year’s update also featuring a nice new comparison tool on the student-facing version of the site.

In this post, I focus on trends in student loan repayment rates (defined as the percentage of students who have repaid at least $1 in principal) at various periods entering loan repayment. I present data for colleges with unique six-digit Federal Student Aid OPEID numbers (to eliminate duplicate results), weighting the final estimates to reflect the total number of borrowers entering repayment. Additionally, I use the January 2017 data release for the 2012-13 Scorecard data because there appears to be an error in that year’s dataset that results in very few colleges having loan repayment rates.

I begin by show the trends in the 1-year, 3-year, 5-year, and 7-year repayment rates for each cohort of students with available data.

Repayment cohort 1-year rate (pct) 3-year rate (pct) 5-year rate (pct) 7-year rate (pct)
2006-07 61.8 63.5 64.6 66.6
2007-08 53.0 54.2 56.1 59.7
2008-09 46.1 47.9 52.0 56.0
2009-10 41.0 43.2 48.7 N/A
2010-11 36.6 40.7 46.3 N/A
2011-12 32.2 38.1 N/A N/A
2012-13 33.0 38.3 N/A N/A

There are two clear trends from this table. First, repayment rates have steadily dropped for more recent cohorts of students. The one-year repayment rate for students entering repayment in 2006-07 (before the Great Recession) was 61.8%, while the most recent cohort of students had a one-year repayment rate of just 33.0%. Much of this decline is likely due to the growth of income-driven repayment plans (which can allow students to be current on their payments while not making a dent in the overall principal). But economic circumstances also likely play a role here.

Second, repayment rates steadily rise for a given cohort as they have more time in the labor market after college. In the 2008-09 repayment cohort, the seven-year repayment rate was 56.0%, 9.9% higher than the one-year rate. These trends still suggest that it will be a long time before students repay their loans, but this is a step in the right direction.

I also show the distribution of colleges’ repayment rates for the 2008-09 cohort across all of the repayment periods by the type of college (public, private nonprofit, and for-profit). In general, private nonprofit colleges have higher repayment rates than both public and for-profit colleges (in part because private nonprofit colleges are primarily four-year institutions), but all sectors see slight improvements between the one-year and seven-year repayment rates.

Finally, a programming note: I’ll be getting the final page proofs for my book shortly and have to do final checks and put together an index during the month of October. I’ll try to write a couple of short blog posts when the new National Postsecondary Student Aid Study and full IPEDS Outcomes Measures survey come out; otherwise, stay tuned for some exciting new research that I’ll be unveiling in early November.

It’s Time to Move Beyond Cohort Default Rates

Today marked the annual release of data on cohort default rates—representing the percentage of students at a given college who default on their federal student loans within three years. The newest data show that 11.5% of students who entered repayment in Fiscal Year 2014 defaulted during this period, which is up slightly from 11.3% for those who entered repayment in Fiscal Year 2013.

Cohort default rates (CDRs) have been used for decades as an accountability metric by the federal government, with colleges posting CDRs of over 40% in a given year losing access to federal student loans for a two-year period and colleges with CDRs above 30% in three consecutive years losing access to all federal financial aid for two years. This year, six colleges posted default rates high enough to lose all Title IV aid and four more had default rates high enough to lose loan access.

Yet CDRs suffer from two key concerns that make them almost toothless from an accountability perspective—and show the need for better accountability metrics. I discuss the two key points in brief below (and if you like this topic, you’ll love my book on higher education accountability that will come out in January!).

Point 1: Default rates are an almost meaningless indicator of student outcomes. The availability of income-driven repayment programs means that no student should ever default on their obligations (although these programs are still clunky and some students simply don’t ever want to repay their loans). But for students who are able and willing to jump through the hoops of income-driven programs and have very low incomes, they can be current on their loans while making zero payments. Many colleges also adopt default management programs that can encourage students to either enroll in income-driven plans or to defer their obligations beyond the three-year accountability window.

In a recent article (a summary is available here), Amy Li of the University of Northern Colorado and I explored the relationship between default and repayment rates (as defined as paying down at least $1 in principal over a given period of time). We showed that although reported default rates stayed low, the percentage of students failing to repay any principal—a key question for taxpayers—was far higher.

Point 2: Default rate sanctions affect almost no colleges. Ben Miller of the Center for American Progress summed up how few colleges faced the loss of federal aid:

The all-or-nothing nature of potential sanctions gives colleges a tremendous incentive to make sure they aren’t affected. In 2014, the Obama Department of Education agreed to a controversial last-minute change to CDRs that allowed some colleges to sneak just below the 30% threshold. In 2017, a provision appeared in the FY 2018 budget that would effectively void CDR sanctions for colleges in economically distressed areas:

It turns out that Senator Mitch McConnell (R-KY) inserted the provision, likely to help out Southeast Kentucky Community and Technical College—one of the six institutions that is at risk of losing all federal financial aid due to high default rates. It pays to have friends in high places, I reckon.

So what can be done to improve federal accountability policies on student loans? I offer two simple ideas to start. First, move from default rates to repayment rates in order to get a better idea of students’ post-college circumstances. Second, move from an all-or-nothing sanction system to gradual sanctions. I go into both of these points in more depth in a paper I wrote in 2015 on the idea of “risk sharing” for student loans. It is essential to move away from CDRs as quickly as possible, even though some in higher education community may prefer the CDR system that affects relatively few colleges.

Trends in Student Fees at Public Universities

Out of all the research I have done during my time as an assistant professor, I get more questions from journalists and policymakers about my research on student fees than any other study. In this study (published in the Review of Higher Education in 2016), I showed trends in student fees at public four-year institutions and also examined the institutional-level and state-level factors associated with higher levels of fees. Yet due to the time it takes to write a paper and eventually get it published, the newest data on fees in the paper came from the 2012-13 academic year. In this blog post, I update the data on trends in fees at public universities for in-state students to go through the 2016-17 academic year.

It’s quite a bit harder than it appears to show trends in student fees because of the presence of fee rollbacks—colleges resetting their fees to a lower level and increasing tuition to compensate. Between the 2000-01 and 2016-17 academic years, 89 public universities reset their fees at least once (as measured by decreasing fees by at least $500 and increasing tuition by a larger amount). This includes most public universities in California, Massachusetts, Minnesota, and South Dakota, as well as a smattering of institutions in other states. Universities that reset their fees had a 115.3% increase in inflation-adjusted tuition and fees since 2000-01 (from $4,286 to $9,228), compared to an 83.7% increase for the 441 universities that did not reset their fees (from $4,936 to $9,068). With the caveat that I can’t break down consistent increases in tuition and fees for some of the colleges with the largest price increases, I present trends in tuition and fees for the other 441 institutions below.

The first figure shows average tuition (dashed) and fees (solid) levels for each year through 2000-01 through 2016-17. During this period, tuition increased from $3,999 to $7,183 in inflation-adjusted dollars (a 79.6% increase). Fees went up even faster, with a 106.7% increase from $912 to $1,885.

The second figure shows student fees as a percentage of overall tuition and fees. This percentage increased from 18.6% in 2000-01 to 20.8% in 2016-17.

This increase in fees is particularly important in conversations about free public college. Many of the policy proposals for free public higher education (such as the Excelsior Scholarship in New York) only cover tuition—and thus give states an incentive to encourage colleges to increase their fees while holding the line on tuition. It’s also unclear whether students and their families look at fees in the college search process in the same way they look at tuition, meaning that growing fee levels could surprise students when the first bills come due. More research needs to be done on how students and their families perceive fees.

A Peek Inside the New IPEDS Outcome Measures Dataset

Much of higher education policy focuses on “traditional” college students—those who started college at age 18 after getting dropped off in the family station wagon or minivan, enrolled full-time, and stayed at that institution until graduation. Yet although this is how many policymakers and academics experienced college (I’m no exception), this represents a minority of the current American higher education system. Higher education data systems have often followed this mold, with the U.S. Department of Education’s Integrated Postsecondary Education Data System (IPEDS) collecting some key success and financial aid metrics for first-time, full-time students only.

As a result of the 1990 Student Right-to-Know Act, all colleges were required to start compiling graduation rates (and disclosing them upon request) for first-time, full-time students and a smaller group of colleges were also required to collect transfer-out rates. Colleges were then required to submit the data to IPEDS for students who began college in the 1996-97 academic year so information would be available to the public. This was a step forward for transparency, but it did little to accurately represent community colleges and less-selective four-year institutions. Some groups, such as the Student Achievement Measure, have developed to voluntarily provide information on completion rates for part-time and transfer students. These data have shown that IPEDS significantly understates overall completion rates even among students who initially fit the first-time, full-time definition.

After years of technical review panels and discussions about how to best collect data on part-time and non-first-time students along with a one-year delay to “address data quality issues,” the National Center for Education Statistics released the first year of the new Outcome Measures survey via College Navigator earlier this week. This covers students who began college in 2008 and were tracked for a period of up to eight years. Although the data won’t be easily downloadable via the IPEDS Data Center until mid-October, I pulled up data on six colleges (two community colleges, two public four-year colleges, and two private nonprofit colleges in New Jersey) to show the advantages of more complete outcomes data.

Examples of IPEDS Outcome Measures survey data, 2008 entering cohort.
Institution 6-year grad rate 8-year grad rate Still enrolled within 8 years Enrolled elsewhere within 8 years
Community colleges
Atlantic Cape Community College
First-time, full-time 26% 28% 3% 27%
Not first-time, but full-time 41% 45% 0% 29%
First-time, part-time 12% 14% 5% 20%
Not first-time, but part-time 23% 26% 0% 38%
Brookdale Community College
First-time, full-time 33% 35% 3% 24%
Not first-time, but full-time 36% 39% 2% 33%
First-time, part-time 17% 18% 3% 25%
Not first-time, but part-time 25% 28% 0% 28%
Public four-year colleges
Rowan University
First-time, full-time 64% 66% 0% 20%
Not first-time, but full-time 82% 82% 1% 7%
First-time, part-time 17% 17% 0% 0%
Not first-time, but part-time 49% 52% 5% 21%
Thomas Edison State University
Not first-time, but part-time 42% 44% 3% 29%
Private nonprofit colleges
Centenary University of NJ
First-time, full-time 61% 62% 0% 4%
Seton Hall University
First-time, full-time 66% 68% 0% 24%
Not first-time, but full-time 67% 68% 0% 18%
First-time, part-time 0% 0% 33% 33%
Not first-time, but part-time 38% 38% 0% 38%

There are several key points that the new data highlight:

(1) A sizable percentage of students enrolled at another college within eight years of enrolling in the initial college. The percentages at the two community colleges in the sample (Atlantic Cape and Brookdale) are roughly similar to the eight-year graduation rates, suggesting that quite a few students are transferring without receiving degrees. These rates are lower in the four-year sector, but still far from trivial.

(2) New colleges show up in the graduation rate data! Thomas Edison State University is well-known for focusing on adult students (they only accept students age 21 or older). So, as a result, they didn’t have a first-time, full-time cohort for the traditional graduation rate. But TESU has a respectable 42% graduation rate of part-time students within six years, and another 29% enrolled elsewhere within eight years. On the other hand, residential colleges may just have a first-time, full-time cohort (such as Centenary University) or small cohorts of other students for which data shouldn’t be trusted (such as Seton Hall’s tiny cohort of first-time, part-time students).

(3) Not first-time students graduate at similar or higher rates compared to first-time students. To some extent, this is not surprising as students enter with more credits. For example, at Rowan University, 82% of transfer students who entered full-time graduated within six years compared to 64% of first-time students.

(4) Institutional graduation rates don’t change much after six years. Among these six colleges, graduation rates went up by less than five percentage points between six and eight years and few students are still enrolled after eight years. It’s important to see if this is a broader trend, but this suggests that six-year graduation rates are fairly reasonable metrics.

Once the full dataset is available in October, I’ll return to analyze broader trends in the Outcome Measures data. But for now, take a look at a few colleges and enjoy a sneak peek into the new data!

Beware OPEIDs and Super OPEIDs

In higher education discussions, everyone wants to know how a particular college or university is performing across a range of metrics. For metrics such as graduation rates and enrollment levels, this isn’t a big problem. Each freestanding college (typically meaning that they have their own accreditation and institutional governance structure) has to report this information to the U.S. Department of Education’s Integrated Postsecondary Education Data System (IPEDS) each year. But other metrics are more challenging to use and interpret because they can cover multiple campuses—something I dig into in this post.

In the 2015-16 academic year, there were 7,409 individual colleges (excluding administrative offices) in the 50 states and Washington, DC that reported data to IPEDS and were uniquely identified by a UnitID number. A common mistake that analysts make is to assume that all federal higher education (or even all IPEDS) data metrics represent just one UnitID, but that is not always the case. Enter researchers’ longtime nemesis—the OPEID.

OPEIDs are assigned by the U.S. Department of Education’s Office of Postsecondary Education (OPE) to reflect each postsecondary institution that has a program participation agreement to participate in federal student aid programs. However, some colleges within a system of higher education share a program participation agreement, in which one parent institution has a number of child institutions for financial aid purposes.

Parent/child relationships can generally be identified using OPEID codes; parent institutions typically have OPEIDs ending with “00,” while child institutions typically have OPEIDs ending in another value. These reporting relationships are fairly prevalent, with there being approximately 5,744 parent and 1,665 child institutions in IPEDS in the 2015-16 academic year based on OPEID values. For-profit college chains typically report using parent/child relationships, while a number of public college and university systems also aggregate institutional data to the OPEID level. For example, Penn State and Rutgers have parent/child relationships while the University of Missouri and the University of Wisconsin do not.

In this case of a parent/child relationship, all data that come from the Office of Federal Student Aid or from the National Student Loan Data System are aggregated up across a number of colleges. This includes all data on student loan repayment rates, earnings, and debt from the College Scorecard as well as student loan default rates that are currently used for accountability purposes. Additionally, some colleges report finance data out at the OPEID level on a seemingly chaotic basis—which can only be discovered by combing through data to see if child institutions do not have values. For example, Penn State always reports at the parent level, while Rutgers has reported at the parent level and the child level on different occasions over the last 15 years. Ozan Jaquette and Edna Parra have pointed out in some great research that failing to address parent/child issues can result in estimates from IPEDS or Delta Cost Project data being inaccurate (although trend data are generally reasonable).

If UnitIDs and OPEIDs were not enough, the Equality of Opportunity Project (EOP) dataset added a new term—super-OPEIDs—to researchers’ jargon. This innovative dataset, compiled by economists Raj Chetty, John Friedman, and Nathaniel Hendren, uses federal income tax records to construct social mobility metrics for 2,461 institutions of higher education based on pre-college family income and post-college student income. (I used this dataset last month in a blog post looking at variations in marriage rates across four-year colleges.) However, the limitation of this approach is that the researchers have to rely on the names of the institutions on tax forms, which are sometimes aggregated beyond UnitIDs or OPEIDs. Hence, the super-OPEID.

The researchers helpfully included a flag for super-OPEIDs that combined multiple OPEIDs (the variable name is “multi” in the dataset, for those playing along at home). There are 96 super-OPEIDs that have this multiple-OPEID flag, including a number of states’ public university systems. The full list can be found in this spreadsheet, but I wanted to pull out some of the most interesting pairings. Here are a few:

–Arizona State And Northern Arizona University And University Of Arizona

–University Of Maryland System (Except University College) And Baltimore City Community College

–Minnesota State University System, Century And Various Other Minnesota Community Colleges

–SUNY Upstate Medical University And SUNY College Of Environment Science And Forestry

–Certain Colorado Community Colleges

To get an idea of how many colleges (as measured by UnitIDs) have their own super-OPEID, I examined the number of colleges that did not have a multiple-OPEID flag in the EOP data and did not have any child institutions based on their OPEID. This resulted in 2,143 colleges having their own UnitID, OPEID, and super-OPEID—meaning that all of their data across these sources is not combined with different institutions. (This number would likely be higher if all colleges were in the EOP data, but some institutions were either too new or too small to be included in the dataset.)

I want to close by noting the limitations of both the EOP and Federal Student Aid/College Scorecard data for analytic purposes, as well as highlighting the importance of the wonky terms UnitID, OPEID, and super-OPEID. Analysts should carefully note when data are being aggregated across separate UnitIDs (particularly when different types of colleges are being combined) and consider omitting colleges where aggregation may be a larger concern across OPEIDs or super-OPEIDs.

For example, earnings data from the College Scorecard would be fine for the University of Maryland-College Park (as the dataset just reflects those earnings), but social mobility data would include a number of other institutions. Users of these data sources should also describe their strategies in their methods discussions to an extent that would allow users to replicate their decisions.

Thanks to Sherman Dorn at Arizona State University for inspiring this blog post via Twitter.

How Acela Corridor Educational Norms Look to an Outsider

Education policy discussions in the United States tend to be dominated by people living in the Acela Corridor—the densely-populated, highly-educated, and high-income portion of the United States that is served by Amtrak’s fast train from Boston to Washington, DC. Since moving from the Midwest to New Jersey four years ago to start on the tenure track at Seton Hall, I have probably logged 50 trips to Washington via Amtrak’s slower (and less-expensive) Northeast Regional train. (It sure beats driving, and Amtrak’s quiet car is a delight!)

Many of the suburban communities in northern New Jersey have median household incomes of well over $100,000 per year, which is roughly the top 20% of American families. The top 20% is notable because that is the cutoff that the Brookings Institution’s Richard Reeves uses in his new book, Dream Hoarders, to highlight how upper-income individuals have taken steps to make sure their children have every opportunity possible—typically at the expense of other families. The sheer concentration of high-income families within much of the Acela Corridor has created a powerful set of social norms regarding education that can leave outsiders flabbergasted.

Yet in spite of having two parents with bachelor’s degrees, a PhD in education, and being one half of a two-income professional household, I find myself confused by a number of practices that are at least somewhat common in the Acela Corridor but not in other parts of the country. This was highlighted by a piece in Sunday’s New York Times on affirmative action. The reporter spoke with two students at private boarding schools in New Jersey, of which there are apparently a fair number. My first reaction, as a small-town Midwesterner, was a little different than what many of my peers would think.

Here are some other things that have surprised me in my interactions with higher-income families in the Acela Corridor:

  • K-12 school choice debates. Unlike some people in the education world, I don’t have any general philosophical objections to charter schools. But in order for school choice to work (barring online options), there needs to be a certain population density. This is fine in urban and suburban areas, but not so great in rural areas where one high school may draw from an entire county. A number of Republican senators from rural states have raised concerns about school choice as a solution for this reason.
  • SAT/ACT test preparation. I attended a small-town public high school with about 200 students in my graduating class. The focus there was to get students to take the ACT (the dominant test in America as a whole, with the coasts being the exception), while also encouraging students to take the PLAN and PSAT examinations. But I never saw a sign advertising ACT prep services, nor was I even aware that was I thing people do. (I took the practice ACT that came with the exam the night before the test—that was it.) In the Northeast, there seem to be more signs on the side of the road advertising test prep than any other product or service.
  • The college admissions process. Going to a four-year college is the expectation for higher-income families in the Acela Corridor, and families treat the college choice process as being incredibly important. Using private college counselors to help manage the process, which often includes applying to ten or more colleges, is not uncommon. A high percentage of students also leave the state for college, which is quite expensive. (In New Jersey, about 37% of high school graduates head to other states to attend college.) Meanwhile, in much of the country, the goal is to get students to attend college at all rather than to get students to attend a slightly more prestigious institution. I can think of just one of my high school classmates who went out of state, and a large percentage of the class did not attend college immediately after high school.
  • Private tutoring while in college. I supplemented my income in graduate school by tutoring students in economics, typically charging between $25 and $40 per hour to meet with one or two students to help them prepare for exams. (I paid for an engagement ring using tutoring income!) I was never aware of anyone paying for private tutoring when I was an undergraduate at Truman State University, but this was a common practice at the University of Wisconsin-Madison. Nearly all of these students came from the suburbs of New York City or Washington, DC and were used to receiving private tutoring throughout their education. I got very few tutoring requests from in-state students, but they were typically paying for their own college (and thus got a substantial discount from my normal rates).

I worry about education policy discussions being dominated by the Acela Corridor regulars because their experiences are so different than what how most Americans experience both K-12 and higher education. If education committee staffers, academic researchers, and think tankers all share similar backgrounds, the resulting policy decisions may not reflect the needs of rural and urban lower-income individuals. It is important to seek out people from other walks of life to make sure policies are best for all Americans.

Not-so-Free College and the Disappointment Effect

One of the most appealing aspects of tuition-free higher education proposals is that they convey a simple message about higher education affordability. Although students will need to come up with a substantial amount of money to cover textbooks, fees, and living expenses, one key expense will be covered if students hold up their end of the bargain. That is why the results of existing private-sector college promise programs are generally promising, as shown in this policy brief that I wrote for my friends at the Midwestern Higher Education Compact.

But free college programs in the public sector often come with a key limitation—the amount of money that the state has to fund the program in a given year. Tennessee largely avoided this concern by endowing the Tennessee Promise program through lottery funds, and the program appears to be in good financial shape at this point. However, two other states are finding that available funds are insufficient to meet program demand.

  • Oregon will provide only $40 million of the $48 million needed to fund its nearly tuition-free community college program (which requires a $50 student copay). As a result, the state will eliminate grants to the 15% to 20% of students with the highest expected family contributions (a very rough proxy for ability to pay).
  • New York received 75,000 completed applications for its tuition-free public college program, yet still only expects to give out 23,000 scholarships. Some of this dropoff may be due to students attending other colleges, but other students are probably still counting on the money.

In both states, a number of students who expected to get state grant aid will not receive any money. While rationing of state aid dollars is nothing new (many states’ aid programs are first-come, first-served), advertising tuition-free college and then telling students they won’t receive grant aid close to the beginning of the academic year may have negative effects such as choosing not to attend college at all or diminished academic performance if they do attend. There is a sizable body of literature documenting the “disappointment effect” in other areas, but relatively little in financial aid. There is evidence that losing grant aid can hurt continuing students, yet this does not separate out the potential effect of not having money from the potential disappointment effect.

The Oregon and New York experiences provide for a great opportunity to test the disappointment effect. Both states could compare students who applied for but did not receive the grant in 2017-18 to similar students in years prior to the free college programs. This would allow for a reasonably clean test of whether the disappointment effect had any implications for college choice and eventual persistence.

Examining Variations in Marriage Rates across Colleges

This piece originally appeared at the Brookings Institution’s Brown Center Chalkboard blog.

Young adulthood is not only the time when most people attend college, but also a time when many marry. In fact, college attendance and marriage are linked and have social and economic consequences for individuals and their families.

When (and if) people get married is an important topic due to the presence of what is known as assortative mating. This phenomenon, in which a person is likely to marry someone with similar characteristics such as education, is a contributing factor to increasing levels of income inequality. In some circles, there is pressure to marry someone with a similar pedigree, as evidenced by the high-profile Princeton alumna who urged women at the university to find a spouse while in college. For people attending less-selective colleges, having the possibility of a second household income represents a key buffer against economic shocks.

In this blog post, I use a tremendous dataset compiled by The Equality of Opportunity Project that is based on deidentified tax records for 48 million Americans who were born between 1980 and 1991. This dataset has gotten a great deal of attention on account of its social mobility index, which examines the percentage of students who move well up in the income distribution by young adulthood.

I use the publicly available dataset to examine marriage rates of traditional-age college students through age 34 based on their primary institution of attendance. In particular, I am curious about the extent to which institutional marriage rates seem to be affected by the institution itself versus the types of students who happen to enroll there. My analyses are based on 820 public and private nonprofit four-year colleges that had marriage rates and other characteristics available at the institutional level. This excludes a number of public universities that reported tax data as a system (such as all four-year institutions in Arizona and Wisconsin).

The first two figures below show the distribution of marriage rates for the 1980-82 and 1989-91 birth cohorts as of 2014 for students who attended public, private religious, and private nonsectarian institutions. Marriage rates for the younger cohorts (who were between ages 23 and 25) were low, with median rates of 12% at public colleges, 14% at religiously-affiliated colleges, and just 5% at private nonsectarian colleges. For the older cohort (who were between ages 32 and 34), marriage rates were 59% at public colleges, 65% at religiously-affiliated colleges, and 56% at private nonsectarian colleges.

There is an incredible amount of variation in marriage rates within each of these three types of colleges. In the two figures below, I show the colleges with the five lowest and five highest marriage rates for both cohorts. In the younger cohort (Figure 3), the five colleges with the lowest marriage rates (between 0.9% and 1.5%) are all highly selective liberal arts colleges that send large percentages of their students to graduate school—a factor that tends to delay marriage. At the high end, there are two Brigham Young University campuses (which are affiliated with the Church of Jesus Christ of Latter-day Saints, widely known as the Mormon church), two public universities in Utah (where students are also predominately Mormon), and Dordt College in Iowa (affiliated with the Christian Reformed Church). Each of these colleges has at least 43% of students married by the time they reach age 23 to 25.

A similar pattern among the high-marriage-rate colleges emerges in the older cohorts, with four of the five colleges with the highest rates in students’ mid-20s had marriage rates over 80% in students’ early-30s.

A more fascinating story plays out among colleges with the lowest marriage rates. The selective liberal arts colleges with the lowest marriage rates in the early cohort had marriage rates approaching 60% in the later cohort, while the 13 colleges with the lowest marriage rates in the later cohort were all either historically black colleges or institutions with high percentages of African-American students. This aligns with the large gender gap in bachelor’s degree attainment among African-Americans, with women representing nearly 60% of African-American degree completions.

Finally, I examined the extent to which marriage rates were associated with the location of the college and the types of students who attended as well as whether the college was public, private nonsectarian, or religious. I ran regressions controlling for the factors mentioned below as well as the majors of graduates (not shown for brevity). These characteristics explain about 55% of the variation in marriage rates for the younger cohorts and 77% of the variation in older cohorts. Although students at religiously-affiliated institutions had higher marriage rates across both cohorts, this explains less than five percent of the overall variation after controlling for other factors. In other words, most of the marriage outcomes observed across institutions appear to be related mostly to students, and less to institutions.

Colleges in the Northeast had significantly lower marriage rates in both cohorts than the reference group of the Midwest, while colleges in the South had somewhat higher marriage rates. The effects of institutional type and region both got smaller between the two cohorts, which likely reflects cultural differences in when people get married rather than if they ever get married.

Race and ethnicity were significant predictors of marriage. Colleges with higher percentages of black or Hispanic students had much lower marriage rates than colleges with more white or Asian students. The negative relationship between the percentage of black students and marriage rates was much stronger in the older cohort. Colleges with more low-income students had much higher marriage rates in the earlier cohort but much lower marriage rates in the later cohort. Less-selective colleges had higher marriage rates for the younger cohort, while colleges with higher student debt burdens had lower marriage rates; neither was significant for the older cohort.

There has been a lot of discussion in recent years as to whether marriage is being increasingly limited to Americans in the economic elite, both due to the presence of assortative mating and the perception that marriage is something that must wait until the couple is financially secure. The Equality of Opportunity project’s dataset shows large gaps in marriage rates by race/ethnicity and family income by the time former students reach their early 30s, with some colleges serving large percentages of minority and low-income students having fewer than one in three students married by this time.

Yet, this exploratory look suggests that the role of individual colleges in encouraging or discouraging marriage is generally limited, since the location of the institution and the types of students it serves explain most of the difference in marriage rates across colleges.