Comments on Accountability and the Higher Education Act

As the Senate works on its version of a Higher Education Act reauthorization bill, the Health, Education, Labor and Pensions Committee has made accountability one of its key areas of discussion in recent hearings. Committee chairman Lamar Alexander asked the higher education community for comments regarding accountability, so I sent my comments along to the committee. They are reprinted below.

February 15, 2018

The Honorable Lamar Alexander

The Honorable Patty Murray

Senate Health, Education, Labor and Pensions (HELP) Committee

Dear Senators:

My name is Robert Kelchen and I am an assistant professor of higher education at Seton Hall University.[i] I have been closely following the Higher Education Act reauthorization hearings in the Senate HELP Committee and am pleased to see the committee beginning to work on writing a comprehensive, bipartisan piece of legislation. Accountability is a crucial issue for both protecting students and taxpayers alike, and is such it is essential to design a system that encourages institutional improvement while discourages colleges from trying to game the system to remain eligible to receive federal Title IV financial aid dollars.

I have spent the last several years researching higher education accountability in all of its forms, including efforts by the federal government, state governments, accrediting agencies, the private sector, and colleges themselves.[ii] In this letter, I share four key points from my research on how to design accountability systems that have the highest likelihood of success and provide the best possible information for students, their families, and policymakers.


Point 1: Avoid all-or-nothing accountability systems. Three key federal accountability policies—cohort default rates (CDRs), the 90/10 rule, and gainful employment—grant institutions access to federal Title IV financial aid if they pass a certain threshold. Although the gainful employment metrics are too new to actually take away programs’ financial aid, it is clear from CDRs and the 90/10 rule that very few colleges are affected. In the most recent year of data, only ten small colleges faced the loss of either federal student loans or all Title IV aid for high CDRs and no colleges were subject to the loss of federal student aid for failing the 90/10 rule in two consecutive years.[iii]

Setting minimum performance floors sounds like an appealing idea, and the idea of ‘bright line’ standards has been proposed with respect to recognizing accreditors.[iv] But actually following through and pulling the plug on the lowest-performing colleges by denying them access to federal financial aid is a much more difficult task. As the majority staff’s white paper notes, just eleven colleges have lost access to Title IV funds due to high CDRs since 1999—even though approximately ten colleges per year should have lost aid if federal laws were strictly followed. Congress and the Department of Education have shown a lack of willingness to effectively close colleges (particularly public or private nonprofit institutions), as shown by the Senate Majority Leader’s recent efforts to exempt a Kentucky technical college from losing Title IV aid and the previous administration’s alteration of CDR calculations just prior to release in 2014 that protected an unknown number of colleges.[v]

A more effective way to hold a larger number of colleges accountable for their outcomes is to use a gradually increasing set of sanctions for lower-performing institutions. In theory, risk-sharing proposals for federal student loan repayment can provide that sort of incentive. However, the PROSPER Act’s loan repayment metric would create the same all-or-nothing incentive that would be subject to both institutional gaming and intense lobbying efforts. The Student Protection and Success Act creates a sliding scale to some extent (although retaining a minimum eligibility threshold), and academics’ risk-sharing proposals also are based on sliding scales.[vi] It is also important to note that risk-sharing proposals should include incentives for institutional improvement as well as sanctions, similar to what the ASPIRE Act would do.


Point 2: Both institution-level and program-level accountability policies are important. Policy conversations are rapidly moving toward holding individual programs within colleges accountable for their performance. Gainful employment regulations already seek to do that for a subset of programs, but the PROSPER Act would use program-level loan repayment rates for all institutions of higher education. This proposal makes intuitive sense, but program-level data collection efforts have some important limitations.

In general, programmatic outcomes data make the most sense when students enter a college or university with a particular field of study in mind. This is less of a concern for vocationally-oriented programs at the undergraduate level and for graduate and professional education in which students are generally admitted to study in particular programs. But not all students enter associate or baccalaureate degree programs with a declared major, and roughly one-third of first-time college students changed their declared major at least once within three years of starting college.[vii] This means that attributing student outcomes to a particular program becomes a concern. Additionally, if undeclared students are ignored for the purposes of program-level accountability metrics, colleges suddenly have an incentive to restrict when students can officially declare majors. Waiting until two years into a bachelor’s degree program to declare a major restricts the pool of students to those with a higher likelihood of success, meaning that dropouts are less likely to be counted.

Another potential option is to restrict program-level accountability only to students who graduate, as is the case with the current gainful employment regulations. But this obscures important data for students, since only 54% of students at four-year colleges and 32% of students at two-year colleges graduated from that same institution within eight years of initial enrollment.[viii] It is likely the case that reported program-level outcomes are far better for graduates than dropouts, thus providing an overly rosy picture of lower-performing programs.[ix]

Given the interest in program-level outcomes data alongside the difficulty in fully relying on program-level accountability measures in certain sectors of higher education, a more reasonable solution would be to use a combination of program-level and institution-level data for accountability purposes. It may be worthwhile to consider tougher performance measures for entire institutions than individual programs due to the difficulty in accurately measuring programmatic outcomes in non-vocational fields and due to concerns about small cell sizes for certain programs of study.


Point 3: Defining the loan repayment rate is perhaps the most important accountability-related issue in HEA reauthorization. Very few academics consider CDRs to be a tremendously valuable measure of institutional performance due to their ability to be manipulated by colleges, the presence of income-driven repayment programs, and their relatively short time horizon. The student loan repayment rate that was included in the 2015 release of the College Scorecard represented a more comprehensive look at how former students are managing their loans, and painted a completely different pictures than CDRs (especially after an unfortunate coding error was finally fixed in early 2017).[x]

Both the definition of student loan repayment rates in the College Scorecard and the types of loans included are decisions that have substantial accountability implications. The Scorecard definition (repaying at least $1 in principal at 1, 3, 5, and 7 years after entering repayment) includes federal subsidized and unsubsidized loans, omitting Parent and Grad PLUS loans. Approximately $21 billion of the $94 billion in federal loans during the 2016-17 academic year was in the form of PLUS loans, yet this is entirely missing from the Scorecard repayment rate (and CDRs, as well).[xi] It may be worth considering a separate loan repayment rate metric for Parent PLUS loans, but Grad PLUS loans should be included with other student loans for accountability purposes.

There are two other potential definitions of loan repayment rates that are worth considering. The first is the percentage of dollars that are repaid during a certain period of time. This is similar to the initial definition of repayment rates that was used in the 2010 negotiations regarding gainful employment.[xii] This is a more taxpayer-focused metric, as it captures overall risk of nonrepayment instead of the percentage of borrowers who are struggling to pay down principal. The second definition is the percentage of students who are on track to repay their loans within a fixed window of time. The challenge with this definition is that students can choose from a menu of loan repayment plans, with extended payment plans being particularly common among students with larger amounts of debt.

A further complicating factor is the growth of income-driven repayment plans, which now represent about 40% of all outstanding federal student loan dollars.[xiii] These plans often result in outstanding balances rising in early years of repayment (when incomes are low) before principal is paid down later. An analysis of recent bachelor’s degree recipients found that only about 25% of students in income-driven plans paid off any principal within five years of entering repayment, while about 75% of students not in income-driven repayment had repaid principal within one year of entering repayment.[xiv] How to address students in income-driven repayment plans is a key concern regarding student loan repayment rates, as the federal government could simultaneously encourage students to enroll in income-driven plans while penalizing colleges where students take up such plans.


Point 4: Free the higher education data! Students and their families are currently being asked to make one of the biggest financial decisions of their lives based on relatively little objective information. The College Scorecard was a helpful step forward, as was including part-time and Pell recipient graduation rates in the Integrated Postsecondary Education Data System in 2017. A student unit record data system would certainly be helpful in making better data available in the college choice process, but there are things that the Department of Education can do (with the support of Congress) without overturning the ban. A few of the most important data points are the following:

  • Program-level outcomes for all Title IV institutions. Regardless of how the gainful employment negotiated rulemaking panel turns out, HEA reauthorization should encourage data to be released for all programs of study (with the caveats as noted earlier in this letter). Some programmatic accreditors are starting to require institutions to release this sort of information, but ED can do so fairly easily for all students receiving federal financial aid. It would be nice to include all students attending Title IV-participating institutions, but program-level data for federal aid recipients would be a good start.
  • Separate data for undergraduate and graduate students. Although graduate and professional students represent just 17% of all federal student loan borrowers, they make up 38% of all federal student loan dollars.[xv] Yet loan repayment and debt data are not presented separately for graduate students—and current data do not even include Grad PLUS loans.
  • Include outcomes for Parent PLUS loans. Although only five percent of students had a Parent PLUS loan in the 2011-12 academic year, this is still an important financing source for families; borrowing rates are higher (13%) at HBCUs and average loan amounts among borrowers are over $12,000.[xvi] Yet the only information available on PLUS loan outcomes is a set of sector-level default rates that ED released following a negotiated rulemaking panel in 2014.[xvii] It is more difficult to envision a high-stakes accountability policy based on parent outcomes instead of student outcomes, but making institution-level data public would be a valuable service to families.
  • Provide data on incomedriven repayment plan usage by institution. Because repayment rates are affected by income-driven repayment plans, it would be helpful to provide information on the percentage of borrowers from each institution who are enrolled in income-driven repayment plans, ideally at the undergraduate and graduate level. A lower repayment rate at a college that graduates a large percentage of students into public service may be more acceptable due to income-driven repayment plans, while a similar repayment rate at a college where students are enrolled in more lucrative majors would be a greater cause for concern.[xviii]

I would like to thank the HELP Committee for holding a series of hearings on Higher Education Act reauthorization and for actively engaging with the research community throughout the process. As the committee is drafting the bill over the next few months, so I encourage Senators and staff members to continue reaching out to researchers while considering potential policy proposals and legislative text. I am more than happy to talk with any committee members during this process and I wish you all the best of luck in working on a much-needed overhaul of the Higher Education Act.


[i] All views and opinions expressed in this letter are mine alone and do not necessarily reflect my employer.

[ii] See my new book Higher Education Accountability, which is now available through Johns Hopkins University Press.

[iii] Author’s calculations using Federal Student Aid data on cohort default rates (FY14 cohort) and proprietary school revenue percentages (2015-16). Additionally, I appreciate the majority staff referencing one of my blog posts on the topic—that made my day!

[iv] Lederman, D. (2017, June 21). Where winds are blowing on accreditation. Inside Higher Ed.

[v] Douglas-Gabriel, D. (2018, February 8). McConnell attempts to protect two Kentucky colleges in budget deal. The Washington Post. Stratford, M. (2014, September 24). Reprieve on default rates. Inside Higher Ed.

[vi] Kelchen, R. (2015). Proposing a federal risk-sharing policy. Indianapolis, IN: Lumina Foundation. Webber, D. A. (2017). Risk-sharing and student loan policy: Consequences for students and institutions. Economics of Education Review, 57, 1-9.

[vii] Leu, K. (2017). Beginning college students who change their majors within 3 years of enrollment. Washington, DC: National Center for Education Statistics Report NCES 2018-434.

[viii] Ginder, S. A., Kelly-Reid, J. E., & Mann, F. B. (2017). Graduation rates for selected cohorts, 2008-13; outcome measures for cohort year 2008; student financial aid, academic year 2015-16; and admissions in postsecondary institutions, fall 2016. Washington, DC: National Center for Education Statistics Report NCES 2017-150rev.

[ix] There is an argument that students want outcomes of graduates only instead of combining graduates and dropouts, given common complaints about College Scorecard data and the common trend of students overstating their likelihood of graduation. But for an accountability system tied to federal financial aid instead of consumer information, the proper sample may differ.

[x] Kelchen, R., & Li, A. Y. (2017). Institutional accountability: A comparison of the predictors of student loan repayment and default rates. The ANNALS of the American Academy of Political and Social Science, 671, 202-223.

[xi] Baum, S., Ma, J., Pender, M., & Welch, M. (2017). Trends in student aid. New York, NY: The College Board.

[xii] Belfield, C. R. (2013). Student loans and repayment rates: The role of for-profit colleges. Research in Higher Education, 54(1), 1-29.

[xiii] Author’s analysis using Federal Student Aid data.

[xiv] Conzelman, J. G., Smith, N. D., & Lacy, T. A. (2016, July 11). The tension between student loan accountability and income-driven repayment plans. Brown Center Chalkboard.

[xv] Baum, S., & Steele, P. (2018). Graduate and professional school debt: How much students borrow. West Chester, PA: AccessLex Institute.

[xvi] Goldrick-Rab, S., Kelchen, R., & Houle, J. (2014). The color of student debt: Implications of federal loan program reforms for black students and historically black colleges and universities. Madison, WI: Wisconsin HOPE Lab.

[xvii] Stratford, M. (2014, April 3). Default data on Parent PLUS loans. Inside Higher Ed.

[xviii] It would be helpful to have data on Public Service Loan Forgiveness interest by institution, but these data would be incomplete because students do not have to signal any intent to use the program until they officially apply after making 120 qualifying payments.

Examining Long-Term Student Loan Default Rates by Race and Ethnicity

Since the release of long-term student loan default data in the Beginning Postsecondary Students Longitudinal Study last fall, one finding that has gotten a great deal of attention is the large gap in default rates by race and ethnicity. Judith Scott-Clayton of Teachers College, Ben Miller of the Center for American Progress, and I all highlighted the high percentage of African-American students who began college in the 2003-04 academic year and defaulted on their loans by 2015. As the chart below shows, black students were more than twice as likely to default on their loans than white students (49% versus 20%), with some differences by institutional type.

But since some of the difference in default rates could be due to other factors (such as family resources and the type of college attended), I ran logistic regressions using the handy regression tools in PowerStats. The first regression just controls for race/ethnicity, while the second regression adds in other control variables of interest. The results are presented as odds ratios, meaning that coefficients larger than 1 reflect a higher likelihood of default and coefficients smaller than 1 reflect a lower likelihood. (Here’s a good primer on interpreting odds ratios.)

In the first regression, the odds ratios for black (3.69), Hispanic (2.09), multiracial (2.56), and American Indian/Alaska Native students (2.45) were all significantly higher than white students at p<.05, while Asian students (0.48) had significantly lower default rates.

Results of logistic regression predicting student loan default rates by 2015 (with controls).
Variable Odds Ratio Lower 95% Upper 95% p-value
Race/ethnicity (reference group: white)
  Black or African American 3.6890 3.0490 4.4620 <0.01
  Hispanic or Latino 2.0870 1.5770 2.7600 <0.01
  Asian 0.4750 0.3170 0.7120 <0.01
  American Indian or Alaska Native 2.4540 1.1220 5.3680 0.03
  Native Hawaiian / other Pacific Islander 0.7170 0.1640 3.1330 0.66
  Other 1.2200 0.7610 1.9560 0.41
  More than one race 2.5640 1.6800 3.9140 <0.01
Source: Beginning Postsecondary Students Longitudinal Study.

After adding in control variables, the coefficients for underrepresented minority students were somewhat smaller. But for African-American (2.56) and multiracial (2.45) students, they were still significantly higher than for white students after adding other controls. This means that black students were about 150% more likely to default than white students—an enormous gap after taking a number of other important factors into account. The coefficients for Hispanic and American Indian students were no longer significant, and Asian students were still less likely to default than white students (an odds ratio of 0.42).

Variable Odds Ratio Lower 95% Upper 95% p-value
Race/ethnicity (reference group: white)
  Black or African American 2.5587 2.0370 3.2139 <0.01
  Hispanic or Latino 1.2606 0.9526 1.6683 0.10
  Asian 0.4249 0.2629 0.6869 <0.01
  American Indian or Alaska Native 1.7371 0.7307 4.1299 0.21
  Native Hawaiian / other Pacific Islander 0.3473 0.0473 2.5505 0.30
  Other 0.8835 0.4458 1.7509 0.72
  More than one race 2.4492 1.5499 3.8704 <0.01
Parents’ highest level of education (reference group: high school grad)
  Did not complete high school 0.7242 0.5085 1.0315 0.07
  Some college or associate degree 0.7883 0.6404 0.9702 0.02
  Bachelor’s degree 0.6181 0.4821 0.7923 <0.01
  Graduate/professional degree 0.5502 0.4242 0.7135 <0.01
Income as percent of poverty level 2003-04 0.9981 0.9976 0.9987 <0.01
Dependency status 2003-04 (reference group: dependent)
  Independent 1.4552 1.0738 1.9719 0.02
Gender (reference group: female)
  Male 1.3553 1.1491 1.5984 <0.01
Age first year enrolled 0.9893 0.9709 1.0081 0.26
First institution sector 2003-04 (reference group: community colleges)
  Public 4-year 0.7858 0.6403 0.9644 0.02
  Private nonprofit 4-year 0.7756 0.6025 0.9985 0.05
  Private nonprofit 2-year or less 1.4838 0.6498 3.3880 0.35
  For-profit 2.1968 1.7624 2.7384 <0.01
Source: Beginning Postsecondary Students Longitudinal Study.

Additionally, the regression also shows the importance of parental education, family income, and sector of attendance in predicting the likelihood of long-term default. Notably, students who began at a for-profit college were about 120% more likely to default on their loans than community college students, while four-year students were less likely. Men were 36% more likely to default on their loans than women, an interesting finding given men typically earn more money than women.

Much more needs to be done to dig deeper into factors associated with long-term student loan default rates. But at this point, it appears clear that other demographic and institutional characteristics available in the BPS do relatively little to explain the large gaps in default rates between black and white students. It would be helpful to have measures of family wealth available given large black-white differences in wealth to see how much of the variation in default rates is due to family resources.

Examining Average Student Loan Balances by State

In a blog post last month, I used newly-available data from the U.S. Department of Education’s Office of Federal Student Aid to look at the amount of student loan dollars in income-driven repayment plans by amount of debt. In that post, I showed that students with more debt were far more likely to use IDR than students with less debt, with students having over $60,000 in debt being about twice as likely to use IDR as those with between $20,000 and $40,000 in debt.

In this post, I want to highlight some other new data that provides interesting insights into the federal student aid portfolio. I looked at state-level data (based on current residence, not where they went to college) that shows outstanding balances and the number of borrowers for both all Direct Loans (the vast majority of federal student loans at this point) and for those enrolled in income-driven plans. I then estimated the average loan value by dividing the two. The data are summarized in the table below.

 All Direct Loans  Loans in IDR plans
State  Balance ($bil) Borrowers (1000s)  Avg loan  Balance ($bil) Borrowers (1000s)  Avg loan
AL 15.9 522.2        30,400 5.4 100.0        54,000
AK 1.7 59.9        28,400 0.6 10.3        58,300
AZ 21.2 711.9        29,800 7.7 137.1        56,200
AR 8.5 312.9        27,200 3.0 62.0        48,400
CA 102.8 3307.3        31,100 36.7 600.2        61,100
CO 20.4 662.1        30,800 7.7 133.6        57,600
CT 12.1 414.6        29,200 3.4 59.6        57,000
DE 3.1 101.2        30,600 1.0 17.4        57,500
DC 5.0 102.2        48,900 2.4 25.7        93,400
FL 65.7 2063.1        31,800 26.4 473.1        55,800
GA 45.8 1350.2        33,900 16.6 279.2        59,500
HI 3.1 104.5        29,700 1.1 18.2        60,400
ID 5.3 191.7        27,600 2.1 41.6        50,500
IL 45.7 1439.7        31,700 14.9 247.9        60,100
IN 21.6 794.7        27,200 7.3 152.3        47,900
IA 10.4 405.8        25,600 3.3 67.8        48,700
KS 9.2 339.5        27,100 2.9 57.6        50,300
KY 13.8 507.1        27,200 4.9 102.6        47,800
LA 14.1 499.1        28,300 4.9 95.2        51,500
ME 4.4 158.8        27,700 1.5 30.0        50,000
MD 25.1 707.2        35,500 8.3 123.5        67,200
MA 23.1 783.7        29,500 7.0 114.3        61,200
MI 37.9 1262.4        30,000 13.2 243.4        54,200
MN 19.9 709.9        28,000 6.7 124.4        53,900
MS 10.6 360.7        29,400 3.8 75.2        50,500
MO 21.0 707.2        29,700 7.6 143.5        53,000
MT 3.0 106.5        28,200 1.2 23.1        51,900
NE 5.7 216.9        26,300 1.9 37.8        50,300
NV 7.5 262.9        28,500 2.8 51.6        54,300
NH 4.7 165.2        28,500 1.4 26.2        53,400
NJ 29.7 999.5        29,700 8.4 145.1        57,900
NM 5.3 189.3        28,000 2.1 39.7        52,900
NY 67.9 2113.1        32,100 24.0 387.8        61,900
NC 32.7 1065.5        30,700 11.8 213.6        55,200
ND 1.8 75.1        24,000 0.6 12.0        50,000
OH 45.4 1577.1        28,800 16.0 313.8        51,000
OK 10.3 383.0        26,900 3.6 71.5        50,300
OR 14.9 475.8        31,300 6.1 107.7        56,600
PA 46.1 1539.3        29,900 15.1 275.3        54,800
RI 3.3 119.6        27,600 1.0 18.8        53,200
SC 18.2 584.7        31,100 6.8 123.5        55,100
SD 2.6 98.9        26,300 0.9 17.9        50,300
TN 21.3 700.7        30,400 7.8 146.0        53,400
TX 76.5 2772.1        27,600 26.1 516.4        50,500
UT 6.9 256.8        26,900 2.7 47.3        57,100
VT 2.1 66.4        31,600 0.8 13.1        61,100
VA 29.9 913.8        32,700 10.1 166.0        60,800
WA 20.1 674.8        29,800 7.4 128.0        57,800
WV 5.4 200.3        27,000 1.8 37.0        48,600
WI 17.0 646.6        26,300 5.7 114.5        49,800
WY 1.2 45.3        26,500 0.4 8.0        50,000

Nationwide, the average outstanding Direct Loan balance was right at $30,000, with significant variation across states (ranging from $24,000 in North Dakota to $48,900 in Washington, DC). The average outstanding balance in IDR was $55,800, which suggests that many borrowers in IDR attended graduate school in order to accumulate that amount of debt. State-level average IDR balances ranged from $47,800 in Kentucky to an impressive $93,400 in Washington, DC. California, Hawaii, Illinois, Maryland, Massachusetts, New York, Vermont, and Virginia all had average balances over $60,000—and they are all high cost of living states with high percentages of adults obtaining graduate or professional degrees.

Once again, kudos to the Department of Education for slowly releasing more data on the federal student loan portfolio. But there are still quite a few important data points (such as school-level data or anything on PLUS loans) that still aren’t available to the public.

New Research on Brain Drain and Recent College Graduates

As I discussed in my previous post, I believe there is value in education scholars using social media in spite of the concerns that being active on venues like Twitter can raise. One example of this occurred last April, when Doug Webber of Temple University ran some numbers from the American Community Survey looking at the percentage of young college graduates who left New York (in the context of the state’s proposed Excelsior Scholarship program). The numbers got quite a bit of attention in a very nerdy portion of higher ed Twitter and led me to encourage Doug to write up the results.

He then reached out to me about working on the paper with him, which ended up being a lot of fun to write. After going through the peer review process (one substantive and one minor round of changes), our resulting article is now online at Educational Researcher. (And a big kudos to the ER reviewers and editorial team for taking the paper from initial submission to appearing online in just eight months!)

We ended up looking at state-level interstate mobility rates among young (age 22-24) bachelor’s degree recipients using ACS data, focusing on the 2005-2015 period to examine pre-recession and post-recession patterns. Overall mobility rates dropped from 12.7% in 2005 to 10.4% in 2015, which is a surprising finding given that people have historically tended to move at higher rates during economic downturns. We found quite a bit of variation across states in net interstate mobility rates both pre-recession (2005-07) and post-recession (2013-15), as summarized in the table below.

State-level changes in the number of young adults with bachelor’s degrees.
  Gain/loss of young adults w/BAs (pct)
State 2005-2007 2013-2015
Alabama -4.0 -4.6
Alaska 3.9 -5.0
Arizona 4.2 -0.5
Arkansas -1.4 -2.7
California 3.9 3.7
Colorado 0.7 8.0
Connecticut -2.3 -4.1
Delaware -17.5 -7.2
District of Columbia 20.0 19.0
Florida 2.6 1.0
Georgia 6.5 -1.0
Hawaii 7.6 8.1
Idaho -3.9 -10.8
Illinois 3.6 3.4
Indiana -12.9 -7.2
Iowa -5.1 -8.1
Kansas -10.3 -4.6
Kentucky -1.2 -2.8
Louisiana -8.3 3.4
Maine -12.5 -8.7
Maryland 4.9 -1.5
Massachusetts 0.7 2.1
Michigan -8.7 -5.6
Minnesota 1.9 -1.2
Mississippi -2.3 -10.8
Missouri -0.7 -2.6
Montana -23.4 -13.3
Nebraska 3.6 -4.3
Nevada 13.3 10.0
New Hampshire -4.6 -10.0
New Jersey 3.0 -3.4
New Mexico 4.3 2.1
New York -0.2 -0.3
North Carolina 3.6 4.2
North Dakota -9.0 -1.8
Ohio -5.9 -3.5
Oklahoma -5.8 -4.4
Oregon -2.1 1.4
Pennsylvania -6.2 -6.1
Rhode Island -19.1 -11.3
South Carolina -2.7 -2.8
South Dakota -8.0 0.0
Tennessee -1.6 1.9
Texas 3.5 3.4
Utah -12.4 -3.7
Vermont -15.4 -10.9
Virginia 3.6 2.8
Washington 6.2 6.8
West Virginia -12.7 -1.9
Wisconsin -3.3 -0.2
Wyoming 6.1 3.5
(1) The percentages reflect changes over the number of 22-24 year olds with a bachelor’s degree who were in the state in a given year.
(2) These values represent averages across the years referenced above.

This article reflects a great example of how a willingness to share some preliminary data on social media results in a publication that is both (hopefully) policy-relevant and a chance to work with a new collaborator. I can’t say enough great things about working with Doug—and I hope to have more of these types of collaborations in the future!

Who Uses Income-Driven Repayment Plans?

Over the last two years, the U.S. Department of Education’s Office of Federal Student Aid has quietly released additional data on the federal government’s portfolio of nearly $1.4 trillion in student loans. I was on the FSA website today looking up the most recent data on Public Service Loan Forgiveness employment certification forms (up to 740,000 filed as of September 30) for a paper I am currently drafting when a new set of spreadsheets on the income-driven repayment (IDR) programs caught my eye.

Overall, just over $375 billion of the $1.05 trillion in federal Direct Loans is now enrolled across the various types of IDR programs. (The rest of the federal loan portfolio is in the old FFEL program, which does not make new loans.) This is up from $269 billion of loans in IDRs when I last wrote about the topic on my blog in mid-2016, which has implications for both students and taxpayers alike. Here, I summarize some of the new data on the types of borrowers who use IDR, as well as some of the other data elements that would be helpful to have going forward.

It is not surprising that students with more debt are more interested in income-driven repayment plans, as many borrowers with less debt could manage payments under the standard ten-year repayment plan. But I was surprised by how much of the debt is held by a small percentage of borrowers. About 1.9 million of the 35.3 million borrowers (or five percent) have more than $100,000 in debt—and this is primarily due to graduate school attendance (since undergraduates cannot borrow more than $57,500 without resorting to PLUS loans). Yet these borrowers hold $325 billion in Direct Loans, or about 30% of all loans outstanding. About $173 billion of this amount is enrolled in IDR plans—53% of all debt held by those with six-figure debts. On the other hand, less than one-fourth of all debt of borrowers with less than $40,000 outstanding is enrolled in IDR. The table and figure below show the amount of Direct Loans outstanding and the amount enrolled in IDR by debt burden.

(UPDATE 2/1/18: As a commenter noted below, there is a small percentage of loans from the old FFEL program in income-driven repayment plans. But as far as I can tell from the data, this only slightly overstates the percentage of Direct Loans in IDR. I’m confident that the general trends still hold, though.)

Table 1: Direct Loan and IDR volumes by debt burden.
Amount of debt All Direct Loans ($bil) IDR ($bil) Pct of loans in IDR
Less than $5k 16.9 0.9 5.3%
$5k-$10k 45.6 4.0 8.8%
$10k-$20k 110.7 16.7 15.1%
$20k-$40k 220.6 52.9 24.0%
$40k-$60k 154.6 50.9 32.9%
$60k-$80k 110.5 48.2 43.6%
$80k-$100k 71.3 30.3 42.5%
$100k-$200k 191.7 90.5 47.2%
More than $200k 133.5 82.3 61.6%
Total 1055.4 376.7 35.7%

I also examined new FSA data on Direct Loan and IDR volumes by age (Table 2 below) and institution type (Table 3). The data show that younger borrowers (between ages 25 and 49) have a higher percentage of dollars in IDR than older borrowers and that students who attended private nonprofit and for-profit colleges rely on IDR more heavily than students who went to public colleges. The finding by sector matches general patterns in tuition and fees, but it does not suggest that for-profit college students disproportionately turn to IDR to manage their loan burdens.

Table 2: Direct Loan and IDR volumes by age of borrower.
Age All Direct Loans ($bil) IDR ($bil) Pct of loans in IDR
24 or younger 128 9.2 7.2%
25-34 418.3 177.4 42.4%
35-49 337.9 139.8 41.4%
50-61 140.5 39 27.8%
62 or older 30.6 11.3 36.9%
Total 1055.3 376.7 35.7%


Table 3: Direct Loan and IDR volumes by institutional type.
Sector All Direct Loans ($bil) IDR ($bil) Pct of loans in IDR
Public 464.2 143.3 30.9%
Private nonprofit 337 126.5 37.5%
For-profit 176.6 63.6 36.0%
Total 977.8 333.4 34.1%

There are two additional data elements that would be extremely useful in considering the implications of income-driven repayment plans. Ideally, data on IDR takeup would be available at the institutional level (as I have politely requested in the past). But at the very least, a breakdown by undergraduate/graduate student status would be useful information.

And one final request of any journalists or qualitative researchers who may be reading this blog—I would love to know more about how the PSLF approval process is going now that some borrowers have made the 120 monthly payments necessary to qualify for forgiveness. It’s been strange not to hear anything about that process after applications could be submitted as early as October 2017.

Don’t Expect a Wave of Private Nonprofit College Closures

American higher education certainly faces its share of challenges. Overall higher education enrollment has dropped from its post-recession high, students and their families are increasingly skeptical of the value of higher education, and the credit rating agency Moody’s recently downgraded the sector to negative from neutral over revenue concerns. These challenges have led to some doomsday predictions regarding college closures; Clayton Christensen of Harvard predicted back in 2011 that half of all colleges would close within 10 to 15 years and since doubled down on his prediction.

To this point, the data tell a different story. While a sizable number of for-profit colleges merge or close in a given year, nonprofit higher education is remarkably stable (and public colleges rarely ever close). According to the U.S. Department of Education, eight degree-granting private nonprofit colleges closed in 2015-16 (the most recent year of data available). Yet the number of degree-granting private nonprofit colleges increased from 1,672 to 1,701—the largest number in 20 years.

Among private nonprofit colleges, there are a few clear risk factors for closure. Small, less-selective institutions with tiny endowments are at a higher risk of closure, particularly if they are located in parts of the country where the pool of traditional-age students is drying up. But these risk factors have existed for decades, yet there is rarely a year in which ten private nonprofit colleges close. (Moody’s expects the number to rise to about 15 per year going forward.)

A recent article published in The Journal of Higher Education helps to provide some data on how resilient small private colleges can be. Melissa Tarrant of the University of West Georgia led a team of researchers who looked back at a 1972 paper by Alexander Astin and Calvin Lee called “The Invisible Colleges.” In that paper, Astin and Calvin identified 491 private, broad-access institutions with fewer than 2,500 students—exactly the type of college that is of greater risk of closure. Yet Tarrant and colleagues showed that 354 of the colleges (more than 70%) were still operating as standalone private nonprofit institutions and only 80 had closed in the following four decades. A failure rate of less than 20% over 40 years does not bode well for predictions that higher education as we know it is going away anytime soon.

A case can be made that the current number of small private colleges is more than would exist if the higher education system were to be designed from scratch to meet the needs of today’s students. But Christensen misses the loyalty of campus communities and alumni (the saga of Sweet Briar College was a great recent example) and the sheer tenacity of institutions as they face extreme financial difficulties. More colleges may consider mergers and strategic alliances, but the rate of college closures in nonprofit higher education is likely to only tick up slightly in the coming decade.

Downloadable Dataset of Marriage Rates by College

I enjoyed reading this recent piece in the Chronicle of Higher Education that looked at the “ring by spring” pressures that students at some Christian colleges face to be engaged by graduation. I looked into factors affecting marriage rates across colleges in a blog post earlier this year and found a nearly six percentage point increase in marriage rates at religiously-affiliated colleges between ages 23 and 25 relative to public institutions, as shown in the figure below.

As a data person—and someone who married his college sweetheart only three years after graduation—I wanted to share a dataset that I had already compiled for that piece so people can look through to their heart’s content. It contains data on 820 public and private nonprofit four-year colleges from the Equality of Opportunity Project, with marriage rates for cohorts ages 23-25 and 32-34 in 2014. The three colleges featured in the Chronicle piece all have higher-than-average marriage rates by age 25, with Cedarville University having a 41% marriage rate, Houghton College having a 34% marriage rate, and Baylor University having a 18% marriage rate.

You can download the dataset here, and have fun exploring the data!

A special thanks to Carol Meinhart for catching a silly error in an earlier version of the dataset, where the two marriage rate column headings were switched. It has since been fixed.

Downloadable Dataset of Pell Recipient Graduation Rates

Earlier this week, my blog post summarizing new data on Pell Grant recipients’ graduation rates at four-year colleges was released through the Brookings Institution’s Brown Center Chalkboard blog. I have since received several questions about the data and requests for detailed data for specific colleges, showing the interest within the higher education community for better data on social mobility.

I put together a downloadable Excel file of six-year graduation rates and cohort sizes by Pell Grant receipt in the first year of college (yes/no) and race/ethnicity (black/white/Hispanic). One tab has all of the data, while the “Read Me” tab includes some additional details and caveats that users should be aware of. Hopefully, this dataset can be useful to others!

A Look at Pell Grant Recipients’ Graduation Rates

This post originally appeared on the Brookings Institution’s Brown Center Chalkboard blog.

The federal government provides nearly $30 billion in grant aid each year to nearly eight million students from lower-income families (mainly with household incomes below $50,000 per year) through the Pell Grant program, which can give students up to $5,920 per year to help pay for college. Yet in spite of research showing that the Pell Grant and similar need-based grant programs are effective in increasing college completion rates, there are still large gaps in graduation rates by family income. For example, among students who began college in the fall 2003 semester, Pell recipients were seven percentage points less likely to earn a college credential within six years than non-Pell students.

In spite of the federal government’s sizable investment in students, relatively little has been known about whether Pell recipients succeed at particular colleges. The last Higher Education Act reauthorization in 2008 required colleges to disclose Pell graduation rates upon request, but two studies have shown that colleges have been unable or unwilling to disclose these data. This means that before now, little has been known about whether colleges are able to graduate their students from lower-income families.[1]

The U.S. Department of Education recently updated its Integrated Postsecondary Education Data System (IPEDS) to include long-awaited graduation rates for Pell Grant recipients, and I focus on graduation rates for students at four-year colleges (about half of all Pell recipients) in this post. I examined the percentage of Pell recipients and non-Pell recipients who graduated with a bachelor’s degree from the same four-year college within six years of entering college in 2010.[2] After limiting the sample to four-year colleges that had at least 50 Pell recipients and 50 non-Pell recipients in their incoming cohorts, my analysis included 1,266 institutions (504 public, 747 private nonprofit, and 15 for-profit).

The average six-year graduation rate for Pell recipients in my sample was 51.4%, compared to 59.2% for non-Pell recipients. The graphic below shows the graduation rates for non-Pell students on the horizontal axis and Pell graduation rates on the vertical axis, with colleges to the left of the red line having higher graduation rates for Pell recipients than non-Pell recipients. Most of the colleges (1,097) had non-Pell graduation rates higher than Pell graduation rates, but 169 colleges (13.3%) had higher Pell graduation rates.

Table 1 below shows five colleges where Pell students graduate at the highest and lowest rates relative to non-Pell students.[3] For example, the University of Akron (which had 3,370 students in its incoming class of first-time, full-time students) reported that just 8.8% of its 1,505 Pell recipients in its incoming class graduated within six years compared to 70.1% of its 1,865 non-Pell students—a yawning gap of 61.3% and the second-largest in the country. Assuming the Pell and non-Pell graduation rates are not the result of a data error that the university made in its IPEDS submission, this is a serious concern for institutional equity. On the other hand, some colleges had far higher graduation rates for Pell recipients than non-Pell students. An example is Howard University, where 79.4% of Pell recipients and just 46.1% of non-Pell students graduated.

Table 1: Colleges with the largest Pell/non-Pell graduation rate gaps.
Name State Number of new students Pell grad rate Non-Pell grad rate Gap Pct Pell
Saint Augustine’s University NC 440 2.7 92.2 -89.5 76.8
University of Akron OH 3370 8.8 70.1 -61.3 44.7
St. Thomas Aquinas College NY 290 20.7 78.3 -57.6 31.7
Southern Virginia University VA 226 20.7 54.3 -33.6 64.2
Upper Iowa University IA 201 27.9 60.8 -32.9 51.7

Ninety-seven of the colleges with at least 50 Pell and 50 non-Pell recipients had graduation rates of over 80% for both Pell and non-Pell students. Most of these colleges are highly selective institutions with relatively low percentages of Pell recipients, but six institutions had Pell and non-Pell graduation rates above 80% while having at least 30% of students in their incoming class receive Pell Grants. All six are in California, with five in the University of California system (Davis, Irvine, Los Angeles, San Diego, and Santa Barbara) and one private institution (Pepperdine). This suggests that it is possible to be both socioeconomically diverse and successful in graduating students.

As a comparison, I also examined the black/white graduation rate gaps for the 499 colleges that had at least 50 black and 50 white students in their graduation rate cohorts. The average black/white graduation rate gap at these colleges was 13.5% (59.0% for white students compared to 45.5% for black students). As the figure shows below, only 39 colleges had higher graduation rates for black students than for white students while the other 460 colleges had higher graduation rates for white students than black students.

Fourteen colleges had higher graduation rates for Pell recipients than non-Pell students and for black students than white students. This group includes elite institutions with small percentages of Pell recipients and black students such as Dartmouth, Duke, and Yale as well as broader-access and more diverse colleges such as CUNY York College, Florida Atlantic, and South Carolina-Upstate. Table 2 shows the full list of 14 colleges that had higher success rates from Pell and black students than non-Pell and white students.

Table 2: Colleges with higher graduation rates for Pell and black students.
Name State Pell grad rate Non-Pell grad rate Black grad rate White grad rate
U of South Carolina-Upstate SC 50.4 34.0 47.3 38.8
CUNY York College NY 31.5 27.3 32.7 28.0
Agnes Scott College GA 71.1 68.3 72.4 62.1
Clayton State University GA 34.0 31.5 33.2 31.0
Duke University NC 96.6 94.3 95.1 95.0
Florida Atlantic University FL 50.6 49.0 50.1 48.5
Wingate University NC 54.5 53.1 60.0 51.4
UMass-Boston MA 45.8 44.7 50.0 40.6
U of South Florida FL 68.1 67.1 68.7 65.5
CUNY City College NY 47.2 46.3 52.8 45.6
Dartmouth College NH 97.2 96.5 97.3 97.1
CUNY John Jay College NY 44.1 43.4 43.5 42.4
Yale University CT 98.2 97.7 100.0 97.6
Stony Brook University NY 72.5 72.3 71.3 70.5

The considerable variation in Pell recipients’ graduation rates across colleges deserves additional investigation. Colleges with similar Pell and non-Pell graduation rates should be examined to see whether they have implemented any practices to support students with financial need. The less-selective colleges that have erased graduation rate gaps by race and family income could potentially serve as exemplars for other colleges that are interested in equity to emulate. Meanwhile, policymakers, college leaders, and the public should be asking tough questions of colleges with reasonable graduation rates for non-Pell students but abysmal outcomes for Pell recipients.

Finally, the U.S. Department of Education deserves credit for the release of Pell students’ graduation rates, as well as several other recent datasets that provide new information on student outcomes. This includes new data on students’ long-term student loan default and repayment outcomes and the completion rates of students who were not first-time, full-time students, along with an updated College Scorecard that now includes a nifty college comparison tool. Though the Pell graduation rate measure fails to cover all students and does not credit institutions if a student transfers and completes elsewhere, it is still a useful measure of whether colleges are effectively educating students from lower-income families. In the future, student-level data that includes part-time and transfer students would be useful to help examine whether colleges are helping all of their students succeed.

[1] Focusing on Pell Grant recipients undercounts the number of lower-income students because a sizable percentage of lower-income students do not file the Free Application for Federal Student Aid, which is required for students to be eligible to receive a Pell Grant.

[2] I calculated the number of non-Pell recipients by subtracting the number of Pell recipients from the total graduation rate cohort in the IPEDS dataset.

[3] This excludes two colleges that reported a 0% or 100% graduation rate for their Pell students, which is likely a data reporting error.

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!