Who the Heck Enrolls in Income-Driven Repayment? Using the Survey of Consumer Finances Database to Find Out
By: Daniel Collier, Dan Fitzpatrick, and Chris Marsicano
Our recent research touches upon a rather controversial topic, Income-Driven Repayment (IDR). In policy and research circles, IDR is often discussed as a potential solution to stem student loan debt defaults and — due to the eventual forgiveness subsidy — as a bane to the federal government (taxpayers as it were). Although IDR is hotly debated, we surprisingly know very little about IDR enrollees — because there are few national data-sets that allow researchers (and the public) to explore who has enrolled in IDR.
This blog entry highlights our study using the Survey of Consumer Finances (SCF) 2016 public use data-set. Yes, these data are a bit old. However, the SCF 2016 data is the most recent nationally-representative data that can be used to examine IDR. As already noted, prior work with SCF or other sources leaves us largely in the dark regarding who enrolls in IDR and what their lives may “look” like compared to those making traditional student loan payments.
SCF data are collected so that analyses can be nationally representative . We analyzed respondents with student loan debt , intending to bring more clarity to who enrolls in IDR. Truthfully, we believe our findings bring complications to a general understanding of who enrolls in IDR. However, our findings provide some guidance for the research community by highlighting the extent to which HOW IDR enrollment is modeled may produce unique results — which leads to dissonance.
We first approached the question of who enrolls in IDR in the same way Collier had approached it in a prior study, expecting that enrollment in IDR may be predicated based on a rational choice that is largely driven by student loan debt loads and wages. We ran five models with loan debt and wages as categorical, continuous, a combination of both, and finally generated a debt-to-income ratio. In none of the models was student loan debt a significant predictor. Most measures of income were also not predictive of IDR participation; yet, we found that earning <$12,500 was negatively correlated — an unfortunate finding, given that these individuals likely most need the protection of IDR. We also found that our generated debt-to-income ratio negatively correlated with IDR enrollment — a counterintuitive finding.
Although student loan debt, earnings, and education level were not consistent links across these five models, in each model we found that female and minority borrowers — and in three models married borrowers — were correlated with increased IDR enrollment. These findings indicate that IDR may provide a crucial financial safety-net for people who we long understood face systemic discrimination, such as lower earnings and hiring or advancement discrimination. See Table 1 below for these models.
We also applied another approach to address the question of IDR enrollment by following Looney and Yannelis (2018) and generated a “high” debt category consisting of borrowers who possessed $50k or more in student loan debt. Whereas our first approach did not reliably suggest loan debt was a factor in IDR enrollment, across these four models possessing more than $50k in loan debt correlated to increased enrollment. Similarly, income (log) correlated with increased enrollment in each model.
In these models, females are generally still more likely to enroll in IDR. However, we no longer observe a pattern of minority borrowers being more likely to participate in IDR and the interactions we formerly ran became non-significant. Further complicating our understanding of this question, our exploratory interactions illustrate female borrowers with high debts are linked to a lower chance of IDR enrollment.
Additionally blurring the picture of who enrolls in IDR, married racial minority females show a very high chance of enrollment in IDR, but only in analyses that also account for having children (and interaction terms for having children). Of importance, these findings should be taken with MUCH caution, due to a limited number of participants matching all five characteristics. Yet, we highlight this finding to call for future analyses.
Both the models in Table 1 and the models in Table 2 explain about 3% of the variance in up-take of IDR. The low R-squared figures suggest that enrollment in IDR may be more chance than not, or that the public use dataset does not allow us access to variables that may be critically important to our understanding of who enrolls in IDR (such as residency zip code or urbanicity, the latter of which Collier found correlates to IDR enrollment).
After running these models, we discussed a joint level of discomfort with simply saying $50K in student loan debt is “high” debt. So, we ran another set of models that creates additional categories of high debt (also above $50K) and we see that possessing between $50–80K, $90–120K, $140–160K are reliable correlates of increased IDR participation. The gaps in the relationship may have produced a lack of significance when the loan debt was a continuous variable. See the table below for more information.
In these models, we also find that some college (possessing an Associates’ degree or attending but not completing a 4-year school) is a reliable connection to increased IDR enrollment — which was not the case for the prior sets of models. This finding is counter to the common narrative that IDR more strongly subsidizes those with graduate and professional degrees.
So, what does this all mean?
From a research perspective, modeling enrollment in IDR is complicated and we do not have many clear trends. Such complexity may in part be a function of REPAYE not yet being included as an IDR scheme in this data-set (but it should be for the SCF 2019 dataset and this complexity may be eased). However, our work illustrates that it is imperative for other researchers to run many alternative analyses, in order to assess whether their findings are present across various ways of modeling demographics, rather than an illusory “finding” that appears because of the way in which race, income, debt, education, and interactions happen to be modeled.
From a policy perspective, conceptualizations of savvy high-earning borrowers attempting to find loopholes in IDR are not confirmed in these data. Furthermore, based on these models, we are unsure who by levels of earnings are enrolled in IDR.
With a strong degree of confidence, we can say females and those with $50K or more in student loan debt are more likely to be enrolled in IDR. We can suggest that modifications to IDR aimed at generating less favorable terms for borrowers may most affect female borrowers (and likely minority borrowers and those with some college) — which we believe should be strongly considered in any discussion of IDR modification. Next, since borrowers with incomes under $12,500 participate in IDR at about one-third the rate of others, outreach to help the lowest earners participate in IDR could be extremely effective.
Critically, though, without a stronger understanding of the income and finances of those enrolled in IDR, making modifications to IDR now would likely be ineffective and have unintended, possibly unethical, consequences. Therefore, we strongly advocate that any modifications be pushed until we have a better understanding of who the average enrollee in IDR may be.
 The complex structure of the SCF (see Federal Reserve, N.D.) requires accounting for both survey weights and multiple imputation. We make use of the SCFCOMBO package (Pence, 2015; for use, see Nielson, 2015) to produce both correct point estimates and correct standard errors to guide inferences.
 The 2016 SCF allows respondents to report up to 6 student loans. Like Blagg (2018), student loan debt was summed across loans (X7805, X7828, X7851, X7928, X7951) that respondents reported were self or spousal debt (variables X7978, X7883, X7888, X7893, X7898, X7993). Blagg’s report only tabulated federal debt, we aligned with Collier’s (2019) design and tabulated total student loan debt which significantly correlated with enrollment in IDR. Total student loan debt was generated using variables X7805, X7828, X7851, X7905, X7928, X7951. Enrollment in income-driven repayment was determined via variables X9306-X9311. Realigned with Blagg (2018), wage data were tabulated from reported household wages and salary only (X5702).
 In fairness to Delisle in several models found that married couples may be more likely to enroll in IDR — yet, with REPAYE closing the loophole where married couples may file taxes separately, we expect that if this was a function of that lever this finding to evaporate.
*For more methods notes, please contact Dr. Dan Fitzpatrick.
Dr. Daniel Collier is a recent addition to the W.E. Upjohn Institute for Employment Research. Daniel conducts research on enrollment in income-drive repayment schemes, how student loan debt correlates with various post-college outcomes, and on Promise (#TuitionFree) policy and students namely focused on the Kalamazoo Promise.
Dr. Dan Fitzpatrick is an independent researcher and evaluator of educational policy. He has expertise in quantitative methods, the transition from high school to post-secondary education, and research synthesis. He focuses on understanding what version of a policy is most helpful, and has completed work across the pre-K to college continuum.
Dr. Christopher R. Marsicano is a Visiting Assistant Professor of Educational Studies at Davidson College. His research examines the political and financial dynamics of education policymaking with a special focus on interest groups in higher education.