How Do Colleges Respond to Accountability Pressures? Examining the Relationship between Cohort Default Rates and College Pricing

Transcription

1 How Do Colleges Respond to Accountability Pressures? Examining the Relationship between Cohort Default Rates and College Pricing Robert Kelchen 1 Assistant Professor, Department of Education Leadership, Management and Policy Seton Hall University robert.kelchen@shu.edu March 1, 2016 Prepared for the Association for Education Finance and Policy annual conference, Denver, CO DRAFT PAPER Comments welcome (but please don t cite without contacting the author) Keywords: Cohort default rates, accountability, cost of attendance Abstract: High-stakes accountability regimes prevalent in K-12 education have become more common in higher education. Given the rising price of higher education and growth in student debt, reducing student loan default rates is a primary focus of federal accountability policy. Using a regression discontinuity framework, I examine whether colleges at risk for federal sanctions due to high default rates respond by reducing tuition or indirect costs such as living expenses in an effort to reduce the risk of future defaults. I find that for-profit colleges with default rates do reduce posted tuition prices relative to colleges with lower default rates, but nonprofit colleges subject to potential sanctions due to high default rates increase both tuition prices and living allowances. I conclude by discussing the implications of these findings and suggesting future research in this area Jubilee Hall, 400 South Orange Avenue, South Orange, NJ Phone (973) I would like to thank Sara Goldrick-Rab and four anonymous reviewers for helpful comments on earlier versions of this paper, Braden Hosch for sharing his expertise on cost of attendance components, and Olga Komissarova for her assistance in preparing the manuscript. 1

2 As outstanding student loan debt has exceeded $1.2 trillion amid concerns about the longterm implications of debt on students life outcomes (Federal Reserve Bank of New York, 2016; Houle & Berger, 2015), colleges are facing increasing pressure to demonstrate that their former students are able to repay their obligations to the federal government. This pressure was heightened following the release of the revised College Scorecard dataset in 2015, which showed that 41% of students at the average college failed to repay at least $1 in loan principal within three years of entering repayment (Kelchen, 2015). Colleges receiving federal student loan dollars are held accountable for their performance via the cohort default rate (CDR) metric, which subjects colleges to sanctions up to and including the loss of all federal financial aid eligibility if the percentage of borrowers defaulting on their loans within a given period of time exceeds a given threshold. In 2014, ten colleges faced the loss of all federal grant and loan aid for having three-year default rates over 30% for three consecutive years, while an additional ten colleges were also subject to the loss of federal loan eligibility for having the most recent cohort default at a rate over 40% (Federal Student Aid, 2014). 2 A substantial amount of federal student aid funds flow through colleges with default rates close to the threshold for potential sanctions. Between the and academic years, the percentage of federal student loans disbursed to students attending colleges with default rates over 15% rose from 11% to 22%, while the percentage of grants flowing through these colleges rose from 24% to 44% (Jaquette & Hillman, 2015). Approximately $35 billion in federal student aid goes to students attending colleges with default rates over 15%, potentially placing some 2 Not all of these colleges will end up losing federal aid eligibility after going through an appeal process; Congressional Research Service data show that only 11 colleges have actually lost eligibility since 1999 (Senate Committee on Health, Education, Labor, and Pensions, 2015). 2

3 institutions at risk of additional oversight or losing funds if economic conditions limit students ability to repay loans. The sharp rise in student loan debt and the amount of loans in default has led to concerns about students overborrowing for college. 3 Current federal law does not give colleges the ability to summarily restrict loan offers to what they deem necessary, but both the professional association representing financial aid administrators (Drager, McCarthy, & McClean, 2013) and the powerful chairman of the U.S. Senate Health, Education, Labor, and Pensions Committee (Alexander, 2015) have called for colleges to get the ability to restrict student borrowing in the impending Higher Education Act reauthorization. Federal regulations limit the amount of financial aid that a student can receive to the total cost of attendance (COA), which includes tuition and fees, room and board, books and supplies, and a category of miscellaneous expenses such as transportation, personal care, and child care (Federal Student Aid, 2015). As a result, the maximum amount of loans a student can take is limited to the COA less any grant aid or work-study funds received. Public colleges in many states do not have full authority to set tuition and fee amounts, as this responsibility is often shared with systems, state higher education coordinating boards, and/or legislatures (Carlson, 2013). However, colleges have a great deal of leeway in how they determine other components of the cost of attendance, as long as they fall within broad federal guidelines (NASFAA, 2014). Research has shown a wide amount of variation in the estimated living expenses portion of COA within given metropolitan areas and regions, which contributes to variations in student borrowing limits across otherwise similar colleges (Kelchen, Hosch, & Goldrick-Rab, 2014). 3 This is in spite of a body of literature suggesting that underborrowing or loan aversion may be greater concerns (e.g., Cunningham & Santiago, 2008; Cadena & Keys, 2013; Goldrick-Rab & Kelchen, 2015). 3

4 The literature on responses to accountability policies suggests that institutions may try to game performance metrics instead of fundamentally changing their actions. For example, studies of K-12 accountability policies have found evidence of institutions focusing only on students close to the passing threshold, lowering the bar for satisfactory performance, or attempting to exclude certain students from the tested sample (e.g., Darling-Hammond, 2004; Figlio & Getzler, 2006; Neal & Schanzenbach, 2010). Colleges have responded to high-stakes accountability systems at the state level in similar ways, such as changing admissions policies to enroll students with a higher probability of completion (e.g., Dougherty, Jones, Lahr, Natow, Pheatt, & Reddy, 2014; Umbricht, Fernandez, & Ortagus, forthcoming). Colleges could potentially respond to receiving a high cohort default rates by gaming the system in ways that make them look better. A college could potentially reduce its cost of attendance by reducing allowances for indirect costs such as living expenses for students living off-campus, which could reduce student borrowing and thus possibly reduce future default rates. This is a particularly advantageous strategy for colleges to consider, as they do not necessarily receive any revenue for off-campus students living expenses while they receive all of the revenue generated by direct costs such as tuition and fees. However, by reducing students access to credit, this could reduce persistence and completion rates among students from low-income families. In this paper, I explore whether colleges are engaging in this sort of strategic behavior by examining whether colleges close to the threshold for facing sanctions for their default rates reduce their allowances for living expenses, tuition and fees, or the total cost of attendance. 4

5 To briefly preview the results, I find little evidence that colleges at risk of facing sanctions reduce living allowances for off-campus students. For-profit colleges subject to sanctions saw slightly lower tuition prices than those not subject to sanctions, while nonprofit colleges raised their tuition. Instead of reducing living allowances, nonprofit colleges with high default rates raised their allowances, a counterintuitive finding that I attempt to explain later in this paper. An overview of cohort default rates Cohort default rates represent one of the primary metrics that the U.S. Department of Education uses to hold colleges accountable for their performance. A former student of a college is considered to have defaulted if he or she does not make a payment on their federal subsidized and/or unsubsidized loans for a 270-day period (Federal Student Aid, 2014). 4 Colleges were evaluated using a two-year CDR for accountability purposes through the release of the outcomes for the cohort of students entering repayment in Fiscal Year 2011 in 2013, with a switch made to three-year CDRs starting with the 2009 through 2011 repayment cohorts in an effort to capture a longer period of post-college outcomes. Colleges with a CDR over a certain percentage (25% using the two-year CDR and 30% using the three-year CDR) in a given year must adopt a comprehensive default management plan that must be reviewed by the U.S. Department of Education, and a default rate of over 40% in one year results in the potential loss of federal student loans. Three consecutive default rates over 25% (two-year CDR) or 30% (three-year CDR) result in the potential loss of federal grant and loan eligibility for a period of three years. 4 The CDR metric excludes student and parent PLUS loans, which tend to have lower default rates than subsidized or unsubsidized loans. This is likely because PLUS loans are subject to passing a credit check, while other federal loans are not. 5

6 Table 1 below shows trends in cohort default rates by institutional sector and year. Twoyear CDRs peaked at 22.4% for the cohort entering repayment in 1990, before steadily falling to under six percent by Default rates then remained around five percent in the early and mid- 2000s before rising to 10% in 2011, the final cohort for which a two-year rate was calculated. [Insert Table 1 here] There is substantial variation in the typical default rate by institutional sector and type. For the 2011 cohort, community colleges had the highest two-year CDRs (15%), although only about 37% of community college students take out federal loans (author s calculation using data from the National Postsecondary Student Aid Study). For-profit colleges had slightly lower default rates than community colleges, with a 13.6% CDR for the 2011 cohort; however, forprofit students make up 44% of all loan defaults due to their high borrowing rate (Turner, 2015). Both four-year public and four-year private nonprofit colleges had CDRs well below 10%, reflecting the greater level of financial resources possessed by their student bodies and the larger average returns to attending a four-year college versus a two-year college or a for-profit institution (e.g., Hout, 2012). Prior research has shown that both student-level factors such as race, gender, and family income and institutional-level factors such as level and sector are associated with default rates (Gross, Cekic, Hossler, & Hillman, 2009; Hillman, 2014; Webber & Rogers, 2014). Of particular relevance to this study is research by Hillman (2015), who examined trial three-year CDRs from the cohort of students leaving college in He showed that for-profit colleges, minorityserving institutions, and colleges serving large percentages of black and low-income students are 6

7 more likely to have CDRs above the 30% cutoff that would subject institutions to additional federal oversight or sanctions. Some colleges have responded to the threat of sanctions by opting out of offering federal student loans in order to protect federal Pell Grant dollars. The majority of colleges opting out are community colleges, where a relatively low percentage of students borrow, and small, lessexpensive for-profit colleges (Hillman & Jaquette, 2014). Community colleges serving approximately one million students opted out of offering federal loans in the academic year, and these colleges had a higher percentage of minority students than other community colleges (Cochrane & Szabo-Kubitz, 2014). Although cohort default rates are not a significant predictor of opting out of offering federal loans after controlling for other characteristics, race/ethnicity and socioeconomic status remain important predictors of opting out (Hillman & Jaquette, 2014). There is also some evidence that students respond to the possibility that a college might lose its eligibility for federal student financial aid due to high CDRs. Darolia (2013) analyzed for-profit colleges close to the 25% two-year default rate cutoff for facing sanctions using data from 1990 to He found that colleges just above the threshold has fall enrollment levels about 17% to 18% lower than colleges just below the sanctioning threshold, with the enrollment decline driven by fewer new students rather than an increase in the number of students transferring away from the college. However, there is no research regarding whether colleges facing sanctions lower their price in an effort to recruit additional students or limit borrowing, a topic which I address in this paper. 7

8 Data, sample, and methods To examine whether there is a relationship between colleges facing pressures to reduce federal cohort default rates and any reductions of cost of attendance components, I used a 14- year panel dataset consisting of cost of attendance components, cohort default rates, and local economic characteristics that could influence default rates. In the following section, I detail the data, sample selection procedures, and analytic methods. Data I used two-year cohort default rates for the cohorts of students entering repayment in Fiscal Year 1998 through Fiscal Year 2011, the final cohort for which two-year CDRs were calculated. These data, from the U.S. Department of Education s Office of Federal Student Aid and compiled through the College Scorecard, reflect the number of students who defaulted within the two fiscal years following leaving college and beginning the repayment period. For example, the FY 2009 cohort of students, who entered repayment between October 1, 2008 and September 30, 2009, were tracked through the end of FY 2010 (September 30, 2010). Colleges are typically informed of their default rate between February and April of the following year in order to allow them an opportunity to verify the calculated CDR and appeal any inaccuracies. Default rates are then released to the public between September and December of that year; for example, the 2009 CDR data were released to the public on September 12, 2011 (U.S. Department of Education, 2011). During this period, a 25% default rate in any given year required colleges to develop a default management plan and submit it to the U.S. Department of 8

9 Education for review, while three consecutive CDRs above 25% resulted in the potential loss of federal financial aid dollars. The default rate data were matched with data on cost of attendance (COA) components from the Integrated Postsecondary Education Data System (IPEDS) from the academic year following the initial release of CDRs to colleges, meaning that data from the through academic years were used to match up with the FY 1998-FY 2011 repayment cohorts. For example, I used the COA components to match up with the cohort of students entering repayment in FY2009 and tracked through the end of FY2010 (September 30, 2010), as colleges received the draft default rates in early 2011 and had several months to adjust their pricing before the start of the following academic year. The COA components of interest included tuition and fees, books and supplies, room and board (for students living away from home only), and a miscellaneous expense category including transportation, laundry, and entertainment. 5 I focused on changes in the posted COAs for full-time students living off-campus with their family (living at home ) and those living offcampus without their family, excluding COAs for students living on-campus because only 13% of students live on campus and fewer than one-third of colleges report on-campus COAs (Kelchen, Hosch, & Goldrick-Rab, 2014). Even fewer colleges at risk of sanctions for high default rates have students living on-campus, making this measure less relevant. I used two different measures to examine whether colleges changed cost of attendance components in response to default rate pressures. First, I used the logged value of each of the COA components, adjusted for inflation using the Consumer Price Index. This explores whether 5 Colleges are not required to report room and board allowances for students living off-campus with their family to IPEDS, and as such data are not available for these students. 9

10 colleges with a default rate subject to sanctions increased their allowances at lower rates than colleges not subject to sanctions. To reduce the influence of outlying values, I trimmed these changes to the 1 st and 99 th percentiles using Winsorization (e.g. Tukey, 1962). Second, I considered an indicator of whether a college decreased the living allowance (in nominal dollars) between two years. The action of cutting living allowances in a period in which the cost of living was generally increasing is a likely indicator of a college trying to reduce student budgets. Adjusting for inflation here would result in colleges with no changes for valid reasons (such as surveying students regarding textbook expenditures every two years instead of every year) as having decreases due to a lack of action instead of an explicit action. Colleges reported COA data in two different ways. Nearly all public and private nonprofit four-year colleges and community colleges granting associate s degrees reported COA components for the length of an academic year typically nine months. However, the majority of for-profit colleges and vocationally-oriented public community and technical colleges reported COA components for the length of the largest program offered. The majority of the programs lasted between 9 and 18 months, and I excluded any colleges with programs shorter than six months or larger than 48 months due to likely data reporting issues. For the purposes of consistency, I also excluded data from colleges if the largest program (as evidenced by its twodigit CIP code) changed from the code reported in before normalizing all program reporter data to a nine-month length to compare alongside academic year reporters. 6 I also included county-level annual wages and unemployment rates from the Bureau of Labor Statistics as control variables in my model to account for the influence of local economic conditions in 6 Prior to 2006, the length of a program was only measured in hours, while it was measured in both hours and years between 2006 and I estimated the length in months in prior years by taking the ratio of hours per month in 2006 and applying that to earlier data if the length of the program (in hours) was within 20% of the 2006 length. 10

11 contributing to cohort default rates (e.g., Dynarski, 1994). In order to include these county-level measures, I excluded one online college (American International University Online) that met my other sample inclusion criteria. Sample I began with 7, 200 colleges that participated in federal Title IV student loan programs and reported a cost of attendance for off-campus students at least once during the period of my study (default rates for the cohorts of students matched with cost of attendance components from through ). I then restricted the sample to colleges with a cohort default rate of at least 25% during at least one year of the panel in order to exclude colleges that did not face additional oversight for high default rates during the period of the panel. This eliminated over 90% of Title IV-participating institutions, reflecting how accountability policies based on default rates currently affect only a small percentage of colleges. The resulting sample consisted of 567 institutions (423 that were ever for-profit and 150 that were ever nonprofit), with 4,685 observations during the period of the panel. 7 These institutions represent 454 unique OPEIDs, which are the level at which default rates are calculated. The summary statistics for those colleges active in the academic year can be found in Table 2. [Insert Table 2 here] Most colleges subject to sanctions for high default rates were small and primarily granted associate s degrees or certificates. The average number of full-time equivalent students at forprofit colleges in my sample was just 331 students, compared to 1,796 students at nonprofit 7 A small number of colleges switched from for-profit to nonprofit status (or vice versa) during the period of the study, and they are included in both columns as appropriate. 11

12 (public or private) colleges. Just four percent of for-profit colleges and one-fourth of nonprofit colleges that faced sanctions were classified as four-year institutions. The average default rate in the cohort of students entering repayment in Fiscal Year 2011 was 20%, with nearly one-third of colleges in the sample having a default rate over 25% in that year. The average college in the sample had a listed nine-month cost of attendance of $23,076 for students living off-campus without their family and $14,901 for students living off-campus with their family; the difference in costs between students living with and without their family is due to room and board allowances for students living at home not being included in the official COA estimates. The average tuition rate was $10,898, with for-profit colleges charging higher prices than nonprofits ($12,779 vs. $6,031, respectively). This differential in tuition explains the differential in total COAs across sectors, as values for living allowances are similar between forprofit and nonprofit colleges. One in four for-profits and 15% of nonprofits reduced their total COA for off-campus students between and , while 16% of for-profits and 6% of nonprofits reduced their listed tuition price. Room and board allowances, which are $7,771 for nine months at the average college and vary little across sector, make up nearly one-third of the total cost of attendance for students at for-profit colleges living off-campus with their family and more than 40% of the COA for students at nonprofit colleges. Only nine percent of colleges cut the room and board allowance between and , with similar trends across sector. A higher percentage of colleges reduced the allowance for other living expenses (such as transportation, laundry, and personal care expenses), with about 20% of for-profit and 13% of nonprofit colleges reducing these allowances for students living both with and without their families. In both sectors, these allowances are close to $4,000 per year. Finally, the average allowance for textbooks and 12

13 supplies was $1,133 in , with 13% of colleges reducing the allowance from the prior year. Methods I used regression discontinuity techniques to test for whether colleges with two-year CDRs at or close to the cutoff for facing federal sanctions reduced cost of attendance components in the year following the release of default rate data for the most recent cohort compared to colleges farther from the cutoff. As was previously discussed, colleges received the draft cohort default rate between February and April of the year following the end of the twoyear repayment period that was used during the period of this study. As this is before most colleges would set tuition for the following academic year, I assume that colleges facing the potential of either heightened oversight or the loss of federal financial aid would reach quickly to their draft default rate. Because colleges get a copy of their draft cohort default rate and then have the potential to challenge any potential inaccuracies before the final default rate is released to the public the following fall, it is possible that colleges that were just over the 25% threshold to face sanctions under the draft default rates were able to successfully lower their default rates to below 25% after appealing to the Department of Education. As the data I use in this analysis are from the final release, regression discontinuity models and the relationship between the draft and final default rates could be a concern if colleges are able to cluster just below the 25% default rate threshold. There is no published evidence that colleges systemically were able to lower their default rates during the length of the panel, as the first large-scale change to draft default rates 13

14 before their release to the public did not occur until 2014 after the move to three-year CDRs as the official accountability metric (Field, 2014). I explored this possible threat to validity by examining the density of the running variable (e.g., McCrary, 2008). The two panels of Figure 1 show histograms of the density of default rates around the 25% threshold for facing sanctions, with Figure 1a showing the broader distribution of CDRs and Figure 1b zoomed in around the 25% threshold. These histograms show a jump in the number of observations at a default rate of exactly 25.0%, which still results in sanctions. There were 69 observations with a default rate between 24.0% and 24.9%, 86 with a default rate of exactly 25.0%, and 95 with a default rate between 25.1% and 25.9%. This distribution suggests that colleges are unable to systemically manipulate their default rates to fall below the 25% threshold, a result also found by Darolia (2013) using observations from students entering repayment between 1990 and The analytic model was the following: Outcome jt = β 0jt + β 1jt CDRFlag j(t 1) + β 2jt CDR j(t 1) 25 + β 3jt CDR j(t 1) β 4jt Econ j(t 1) + α j + θ t + e jt, (1) where Outcome reflects the two sets of outcomes in this analysis. The first set of outcomes reflects the logged values of each cost of attendance component, with a logarithmic transformation used to both reflect the skewed nature of the outcomes and to enhance the ease of interpreting the coefficients on the outcomes (which can be viewed as percentage point changes). The second set of outcomes is a binary variable for whether there was an annual decline (in nominal dollars) in the cost of attendance components between years t-1 and t. 14

15 The key independent variable is CDRFlag, which reflects whether a college s default rate for the cohort that exited the two-year repayment period in the previous fall was above a certain threshold, meaning that cohort default rates from students entering repayment in FY and exiting repayment in FY are matched with cost of attendance measures from to I then added first-order and second-order polynomials for the cohort default rate (recentered around 25%) to allow for a nonlinear flexible form while excluding higher-order polynomials (e.g., Gelman & Imbens, 2014). Econ is a set of county-level variables for wages and unemployment rates, as the majority of institutions with high cohort default rates draw their students from a local area and thus the default rates could be affected by local labor market conditions. These economic conditions are measured in year t-1 to reflect the conditions in place when students are in the midst of repayment. Finally, α j is a set of year-level fixed effects, θ t is the institutional-level fixed effect, and e jt is an idiosyncratic error term. I ran a Hausman test to determine whether fixed or random effects were appropriate for my analyses, and the results (available upon request) reject the null hypothesis that the error terms are uncorrelated. As a result, I used fixed effects in this paper. In all model specifications, standard errors are clustered at the OPEID level to reflect that Federal Student Aid reports default rates at the OPEID level instead of the IPEDS UnitID level at which cost of attendance components are available. 8. This is particularly salient among for-profit colleges, as about one-fourth of UnitIDs do not have their own unique OPEID and instead share default rates with other branches of the same institution. Fewer than ten nonprofit colleges do not 8 I consider an OPEID to be unique if the last two digits are 00, as that designates a parent institution for reporting purposes. For further details on the difference between IPEDS UnitIDs and Federal Student Aid OPEIDs, see Jaquette & Parra (2014). 15

16 have a unique OPEID, as most multicampus institutions had default rates below the 25% threshold throughout the period of the panel. I conducted two sets of subgroup analyses to see whether the relationship between receiving a default rate that would subject a college to sanctions and colleges living allowances varied by institutional type and potential severity of the sanctions. I first examined whether forprofit colleges and nonprofit colleges appeared to react differently to facing a CDR close to the 25% cutoff for facing federal sanctions by running the model separately by sector. Because forprofit colleges do not have the typical shared governance structure of nonprofit colleges that results in greater deliberation in the decision making process, for-profits might be expected to respond quicker to any external pressures to cut costs (e.g., Deming, Goldin, & Katz, 2012). However, this might not be true if changes in COA components can be approved in an expedited process at nonprofit colleges. In the second subgroup analysis, I examined colleges responses based on whether they had a default rate over 25% in the prior year. This allowed for separate comparisons of colleges responses to the first year of crossing the default threshold (where they are subject to additional oversight) and their second year of crossing the threshold (where one more year of high default rates could result in the loss of federal financial aid eligibility). However, due to a lack of statistical power, I am unable to examine colleges responses to facing a 25% default rate conditional on having high default rates in the prior two years. I do not present results from an interaction model testing for whether colleges respond differently to the second year of high default rates than the first year because coefficients are generally insignificant, although full results are available upon request from the author. 16

17 As an additional test for whether regression discontinuity models are appropriate, I tested for whether there was a discontinuity at the 25% default rate threshold using outcome measures from one year following the cohort of students entering repayment. For example, I used cost of attendance components for students who entered repayment in FY 2010, as draft default rates would not have released to colleges until early 2012 and released to the public in late The results from quadratic regressions estimated using interactions between the 25% default rate threshold and the distance from the threshold along with year fixed effects are presented in Table 3 by institutional type, with graphics of the regression discontinuity checks available upon request from the author. The regressions reveal only one statistically significant difference (at p<.10) at the 25% default rate threshold prior to a college learning of its newest cohort s default rate. This suggests that colleges are generally not manipulating living expenses prior to learning about their newest default rate. [Insert Table 3 here] Limitations The cohort default rate measure has been broadly criticized for its limitations, such as the ability of colleges to manipulate default rates by pushing students into deferment or forbearance statuses that result in rising balances but fewer defaults within the loan window (Miller, 2015) and the modest correlation between reported cohort default rates and the percentage of students repaying at least some principal (Kelchen, 2015). Although these limitations may disproportionately affect colleges close to the 25% default rate threshold, they may affect colleges with lower default rates as colleges face incentives from their constituents and the federal government to keep default rates as low as possible. For example, colleges that have 17

18 default rates less than 15% in each of the three most recent cohorts can disburse loan dollars to students in one payment at the beginning of the semester instead of having to wait to disburse half of the funds until the middle of the academic term (Federal Student Aid, 2015). One key limitation of my dataset is that colleges with default rates at or near the threshold for facing sanctions may choose to exit the federal student loan program in order to preserve access to Pell Grant funds. Colleges that had decided to drop out of federal student loan programs as a result of a high CDR prior to 2000 are not in the dataset, and colleges that dropped out during the length of my panel are only observed for a portion of the period. As most of the colleges that opt out of offering loans are community colleges and small for-profit colleges with low borrowing rates (Hillman & Jaquette, 2014), these types of institutions are likely underrepresented in my dataset. Future research should examine whether there is a relationship between leaving the federal student loan program and living allowances. Additionally, I am unable to observe colleges draft cohort default rates, which are what institutions likely use to set tuition prices and living allowances for the following year. However, as I detailed earlier, it does not appear that colleges were generally able to get their default rates revised to fall below the 25% threshold for facing sanctions. Results I first examined whether all colleges above the 25% cohort default rate threshold for facing potential sanctions either changed their cost of attendance components by different amounts or were more likely to reduce student allowances than colleges below the 25% threshold. The results (summarized in Table 4) showed that when measuring allowances one year after students exited the repayment window, there were no statistically significant differences in 18

19 how the full sample of colleges above or below the sanctioning threshold changed tuition or living allowances. This suggests a lack of systemic responses to receiving a high default rate across all institutions. [Insert Table 4 here] I then explored whether responses appeared to be different based on whether a college had received a CDR below 25% in the prior year (making this year the first year in which it was subject to sanctions) or whether the college had also been at or above 25% in the prior year (reflecting the second year of sanctions, with one more year of high default rates subjecting the college to the potential loss of federal financial aid). Using separate regressions for these two groups of colleges, I found that colleges that were subject to sanctions in the previous year and the current year had a total cost of attendance for off-campus students living with family of approximately 6.8% lower than colleges with a lower default rate in the current year (p<.05). The same coefficient was also statistically significant among colleges not subject to sanctions in the prior year (2.9%, p<.10), but none of the individual COA components were statistically significant for either group of colleges. Colleges subject to sanctions in the prior year were 21.5 percentage points less likely to reduce their room and board allowance for off-campus students (p<.05), while colleges not subject to sanctions in the prior year were 12.1 percentage points more likely to reduce their allowance upon crossing the 25% CDR threshold for the first time. This is a surprising finding that merits additional investigation in future research. Because for-profit and nonprofit colleges may be able to respond differently to accountability pressures, I then ran separate models by institutional control (Table 5) as well as an interaction model testing for differences between for-profit and nonprofit colleges. As a 19

20 whole, for-profit colleges generally reacted by either not changing living allowances or slightly reducing them. For-profit colleges with sanctionable default rates had posted tuition and fee prices 2.9% lower than non-sanctioned colleges (p<.10), while the total cost of attendance for students living off-campus with their families was 3.4% lower at sanctioned colleges (p<.05). There is no evidence that for-profit colleges systemically reduced their living allowances (in nominal dollars); rather, any effects appear to be due to slowing the rate of increase. [Insert Table 5 here] Nonprofit colleges (both public and private) reacted differently than for-profit colleges to these accountability pressures. The positive coefficient on tuition (3.5%) is in some sense not surprising, as financially-struggling nonprofit colleges may seek to increase their revenue to stay fiscally solvent or to fund additional services such as default rate management. This is the opposite pattern than what is exhibited by for-profit colleges facing sanctions, as evidenced by the interaction term between for-profit status and a 25% CDR being statistically significant at p<.01. Neither for-profit nor nonprofit colleges responded by altering the allowance for books and supplies. There was a clear pattern of results regarding living allowance components of the cost of attendance at nonprofit colleges. Institutions with a default rate at or above 25% had higher living allowances than those with default rates below 25%, with the coefficients on the room and board and other expenses categories generally falling between seven and nine percentage points (p<.05). These findings are surprising, as colleges with a high default rate could be expected to reduce living allowances in an effort to reduce borrowing and thus potentially reduce future default rates. Unlike tuition dollars, colleges do not typically capture the majority of funds for 20

21 off-campus students housing and miscellaneous expenses making these areas potentially ripe for cuts. I also found no evidence that nonprofit colleges facing default rate sanctions reduced living allowances (in nominal dollars). I finally examined whether for-profit or nonprofit colleges responses to receiving a sanctionable default rate varied based on whether they were sanctioned in the previous year. These results are presented in Appendix 1 for for-profit colleges and Appendix 2 for nonprofit colleges. In general, similar patterns of results are observed among for-profit colleges regardless of their default rate in the prior year. Among nonprofit colleges, the same results generally hold, although the positive point estimates for increases in the total cost of attendance and room and board allowances for off-campus students were slightly larger for colleges that were not sanctioned in the prior year but were sanctioned in the current year. However, interaction tests (not shown, but available upon request from the author) testing for differences between colleges that were above or below 25% in the prior year were generally not statistically significant, in part due to the small number of observations with sanctionable default rates in the prior year. Discussion and Future Work Colleges are facing increasing accountability pressures to improve outcomes and reduce costs from the federal government, state governments, and the general public. While a heightened focus on student success and post-college outcomes is often welcomed by the public, there are concerns that ill-designed accountability systems may result in colleges carefully choosing their student body in a way that undermines the key policy goal of social mobility. In this paper, I build on the body of literature examining how institutions respond to high-stakes 21

22 accountability systems by considering one potential way that colleges could respond to federal accountability pressures. I do this by exploring whether colleges respond to pressure to reduce high student loan cohort default rates by cutting students living allowances for items such as room and board, transportation, and personal care in an attempt to reduce borrowing and thus future default rates. Because colleges have the ability to set these living allowance values and any grant or loan dollars for off-campus students living allowances do not go to the institution, these allowances represent a potential way for colleges to manage default rates while not directly affecting their own finances. In this study, I found limited evidence to support that hypothesis, primarily concentrated among for-profit colleges. For-profit colleges that had a default rate at or above the 25% threshold for facing sanctions in the form of additional oversight from the U.S. Department of Education had slightly lower tuition levels than colleges below the threshold; this slight decline also drove a decline in the overall cost of attendance for students living off-campus with their families. This finding provides some evidence that for-profits subject to sanctions (which are primarily small institutions and not national chains) may charge students slightly lower prices in an effort to reduce borrowing levels. However, there is no evidence that for-profit colleges facing sanctions were more likely to reduce living allowances from the prior year s value. Nonprofit colleges, on the other hand, appear to increase both tuition and fees and nontuition living allowances after receiving a default rate at or above 25%. The increase for tuition and fees makes some sense, as colleges attempt to increase revenue to stay fiscally solvent or provide additional resources to help struggling students. But the sizable increase in non-tuition living allowances among sanctioned colleges is a surprising finding that deserves additional research. Higher living allowances could potentially be used as a recruitment strategy for 22

23 colleges in order to induce students to enroll. However, higher living allowances would also result in a higher average net price of attendance an increasingly important tool for considering a college s affordability that was not available to the public during the first half of my period of study. Another potential explanation for the increase in living allowances at nonprofit colleges with high default rates is because these institutions do not believe that the U.S. Department of Education will follow through and end their access to Title IV financial aid funds after three years of high default rates. The Congressional Research Service reported in 2015 that only 11 colleges had lost full Title IV access due to high default rates (Senate Committee on Health, Education, Labor, and Pensions, 2015), although this small number is partially due to colleges opting out of the federal student loan program to protect Pell Grant access. Nonprofit colleges may feel less threatened by accountability pressures for several reasons. First, colleges with a small percentage of students taking out federal loans are generally exempt from default rate sanctions upon an appeal to the federal government, meaning that a number of community colleges with high default rates may not actually face sanctions. Second, the Department of Education has previously changed default rate calculations just before releasing the data to the public in a way that allowed some colleges to avoid facing sanctions (Stratford, 2014); some nonprofit colleges might expect to not face the loss of federal funds as a result of political pressures while for-profit colleges may not have confidence in policymakers protecting their best interest. Finally, the number of colleges contracting with private companies to develop plans to help students avoid default within the measured time period (e.g., Blumenstyk, 2010) may mean that colleges do not feel such strong pressures to reduce living allowances in an effort to lower future default rates. 23

24 Although colleges are given some guidance by the Department of Education and the professional association for financial aid administrators on how to set non-tuition portions of the cost of attendance, colleges have a great degree of flexibility in how living allowances are determined. This results in large variations in these components across colleges in the same geographic area (Kelchen, Hosch, & Goldrick-Rab, 2014), which can yield vastly different reported cost of attendance values even if tuition and fees are similar. These differences in COA can affect both the net price of attendance (defined as the COA minus grant aid) and the amount that students can borrow. Congress and the Department of Education should work with colleges to set guidelines for more standardized living expense estimates within the same geographic area in order to put colleges on more equal footing from an accountability perspective. Building on the research examining the ways K-12 schools have tried to game highstakes accountability programs, further research is needed to explore how colleges respond to cohort default rates as a high-stakes accountability system. Little is known about how the COA estimates are developed and approved, and future research should particularly focus on whether the initial estimates developed by financial aid administrators are later changed by administrators or trustees due to accountability pressures. Similar to research in K-12 education finding that accountability efforts induced schools to focus on students on the margin of passing the test, research in higher education should examine how colleges communicate with and counsel students who are viewed as being at a higher risk of defaulting on their loans. Further research is also needed to examine whether colleges respond to high default rates by attempting to recruit students who may be less likely to default, such as those with higher family incomes, are white, and are pursuing more financially lucrative majors. Finally, qualitative research is necessary to 24

25 determine whether colleges are in fact responding to default rate pressures when setting their living allowances or whether other factors are more important. 25

26 References Alexander, L. (2015, May 20). Alexander: If colleges share in the risk of student loan defaults, they can help reduce overborrowing and the cost of college. Accessed 29 February 2016 from Blumenstyk, G. (2010, July 11). Business is up in keeping default rates down. The Chronicle of Higher Education. Accessed 22 February 2016 from Cadena, B. C., & Keys, B. J. (2013). Can self-control explain avoiding free money? Evidence from interest-free student loans. Review of Economics and Statistics 95(4), Carlson, A. (2013). State tuition, fees, and financial assistance policies for public colleges and universities. Boulder, CO: State Higher Education Executive Officers. Cochrane, D. F., & Szabo-Kubitz, L. L. (2014). At what cost? How community colleges that do not offer federal loans put students at risk. Oakland, CA: The Institute for College Access and Success. Cunningham, A. F., & Santiago, D. A. (2008). Student aversion to borrowing: Who borrows and who doesn t. Washington, DC: Institute for Higher Education Policy. Darling-Hammond, L. (2004). Standards, accountability, and school reform. Teachers College Record, 106(6), Darolia, R. (2013). Integrity versus access? The effect of federal financial aid availability on postsecondary enrollment. Journal of Public Economics, 106, Deming, D. J., Goldin, C., & Katz, L. F. (2012). The for-profit postsecondary school sector: Nimble critters or agile predators? Journal of Economic Perspectives, 26(1), Dougherty, K. J., Jones, S. M., Lahr, H., Natow, R. S., Pheatt, L., & Reddy, V. (2014). Implementing performance funding in three leading states: Instruments, outcomes, obstacles, and unintended impacts. New York, NY: Community College Research Center Working Paper No. 74. Drager, J., McCarthy, K., & McClean, M. (2013). Reimagining financial aid to improve student access and outcomes. Washington, DC: National Association of Student Financial Aid Administrators. Dynarski, M. (1994). Who defaults on federal student loans? Findings from the National Postsecondary Student Aid Study. Economics of Education Review, 13(1),

27 Federal Reserve Bank of New York. (2016, February). Quarterly report on household debt and credit. Accessed 29 February 2016 from Q4.pdf.. Federal Student Aid. (2013). National student loan two-year default rates. Accessed 22 February 2016 from Federal Student Aid. (2014). Three-year official cohort default rates for schools. Accessed 22 February 2016 from Federal Student Aid. (2015) Federal Student Aid handbook. Washington, DC: Author. Field, K. (2014, September 24). In 11th-hour move, Education Dept. spares the rod on loan defaults. The Chronicle of Higher Education. Accessed 26 February 2016 from Figlio, D. N., & Getzler, L. S. (2006). Accountability, ability, and disability: Gaming the system? In T. G. Gronberg & D. W. Jansen (Eds.), Improving school accountability: Check-ups or choice, pp San Diego, CA: JAI Press. Gelman, A., & Imbens, G. (2014). Why high-order polynomials should not be used in regression discontinuity designs. Cambridge, MA: National Bureau of Economic Research Working Paper Goldrick-Rab, S., & Kelchen, R. (2015). Making sense of loan aversion: Evidence from Wisconsin. In K. M. Hollenbeck & B. Hershbein (Eds.), Student loans and the dynamics of debt, pp Kalamazoo, MI: W. E. Upjohn Institute for Employment Research. Gross, J. P. K., Cekic, O., Hossler, D., & Hillman, N. W. (2009). What matters in student loan default: A review of the literature. Journal of Student Financial Aid, 39(1), Hillman, N. W. (2014). College on credit: A multilevel analysis of student loan default. The Review of Higher Education, 37(2), Hillman, N. W. (2015). Cohort default rates: Predicting the probability of federal sanctions. Educational Policy, 29(4), Hillman, N. W., & Jaquette, O. (2014). Opting out of federal student loan programs: Examining the community college sector. Paper presented at the Association for Education Finance and Policy, San Antonio, TX. Houle, J., & Berger, L. M. (2015). Is student loan debt discouraging home buying among young adults? Social Service Review, 84(4),