Navigating Higher Education Insurance: An Experimental Study on Demand and Adverse Selection

Finance and Economics Discussion Series

Federal Reserve Board, Washington, D.C.

ISSN 1936-2854 (Print)

ISSN 2767-3898 (Online)

Navigating Higher Education Insurance: An Experimental Study

on Demand and Adverse Selection

Sidhya Balakrishnan, Eric Bettinger, Michael S. Kofoed, Dubravka Ritter,

Douglas A. Webber, Ege Aksu, and Jonathan S. Hartley

2024-024

Please cite this paper as:

Balakrishnan, Sidhya, Eric Bettinger, Michael S. Kofoed, Dubravka Ritter, Douglas A. Web-

ber, Ege Aksu, and Jonathan S. Hartley (2024). “Navigating Higher Education Insurance:

An Experimental Study on Demand and Adverse Selection,” Finance and Economics Dis-

cussion Series 2024-024. Washington: Board of Governors of the Federal Reserve System,

https://doi.org/10.17016/FEDS.2024.024.

NOTE: Staﬀ working papers in the Finance and Economics Discussion Series (FEDS) are preliminary

materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth

are those of the authors and do not indicate concurrence by other members of the research staﬀ or the

Board of Governors. References in publications to the Finance and Economics Discussion Series (other than

acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Navigating Higher Education Insurance: An Experimental

Study on Demand and Adverse Selection

Sidhya Balakrishnan

†

, Eric Bettinger

‡

, Michael S. Kofoed

, Dubravka Ritter

Douglas A. Webber

, Ege Aksu

, and Jonathan S. Hartley

††

February 21, 2024

Abstract

We conduct a survey-based experiment with 2,776 students at a non-proﬁt university to analyze in-

come insurance demand in education ﬁnancing. We offered students a hypothetical choice: either a

federal loan with income-driven repayment or an income-share agreement (ISA), with randomized fram-

ing of downside protections. Emphasizing income insurance increased ISA uptake by 43%. We observe

that students are responsive to changes in contract terms and possible student loan cancellation, which is

evidence of preference adjustment or adverse selection. Our results indicate that framing speciﬁc terms

can increase demand for higher education insurance to potentially address risk for students with varying

outcomes.

The authors appreciate the helpful comments and feedback of Neil Bhutta, Julia Cheney, Barry Cynamon, Andrew Hertzberg,

Jerome Hodges, Caroline Hoxby, Rajeev Darolia, Robert Hunt, Jeff Larrimore, Joseph Marchand, Lois Miller, Kevin Mumford,

Marshall Steinbaum, and Joshua Price. We are also grateful for participants at seminar and conference presentations including the

American Economic Association, Appalachian State University, Association for Education Finance and Policy, Association for Policy

Analysis and Management, Brigham Young University, National Bureau of Economic Research (Economics of Education), Providence

College, Southern Economic Association, United States Air Force Academy, and University of Tennessee, Knoxville.

†

Sidhya Balakrishnan is the director of research at the Jain Family Institute, email: [email protected]g

‡

Eric Bettinger is the Conley DeAngelis Family Professor of Education at Stanford University and a research associate at NBER,

email: [email protected].

Michael S. Kofoed is an assistant professor of economics at the University of Tennessee, Knoxville and Research Fellow at IZA,

and corresponding author: [email protected]

Dubravka Ritter is a senior advisor and research fellow at the Consumer Finance Institute, Federal Reserve Bank of Philadelphia,

email: [email protected].org. This Philadelphia Fed working paper represents preliminary research that is being circulated for

discussion purposes. The views expressed in these papers are solely those of the authors and do not necessarily reﬂect the views of the

Federal Reserve Bank of Philadelphia or the Federal Reserve System. Any errors or omissions are the responsibility of the authors.

No statements here should be treated as legal advice.

Douglas A. Webber is a senior economist at the Board of Governors of the Federal Reserve, email: [email protected]v.

The analysis and conclusions in this paper are those of the author and should not be interpreted as reﬂecting the views of the Board of

Governors or the Federal Reserve System.

Ege Aksu is a PhD candidate at CUNY Graduate Center and fellow at the Jain Family Institute, email:

[email protected]g

††

Jonathan S. Hartley is a PhD candidate at Stanford University, email: [email protected]

1 Introduction

Insurance products are important tools employed by individuals to hedge risks in their ﬁnancial lives. Insur-

ance markets allow individuals to pool risk against unexpected, negative outcomes and are well developed in

many contexts, like healthcare or real estate.

However, risk-hedging opportunities are not readily available

in post-secondary education, even though college is an increasingly uncertain investment (Webber, 2022)

made only once in a lifetime. While returns to college are positive on average (Lovenheim and Smith,

2022), their distribution is more nuanced (Webber, 2016, Broady and Hershbein, 2020). Financial out-

comes for students, for example, vary across institution types (e.g. selective vs. non-selective; four-year vs.

two-year), ﬁelds of study (e.g. education vs. engineering vs. economics), and macroeconomic conditions

upon graduation (Rothstein, 2023). Perhaps more importantly, returns vary within each of these segments

given unobservable student skill – which is potentially difﬁcult for the student and/or the institution to iden-

tify – and uncertain labor market conditions. Particularly for younger students and those entering longer

degree programs, there is uncertainty both in the expected level of income and in its variability.

Students, educational institutions, and government entities understand and bear these risks to different

degrees. For example, work by Stange (2012) shows that many students treat attending their ﬁrst year of

college akin to purchasing an options contract—completing an initial year so as to develop a better sense of

their likely returns, after which they decide whether to exercise the option for a second year. Policymakers

and advocates often work to transfer the riskiness of the return to taxpayers, e.g. via the free college

movement, ﬁnancial aid policy (both grants and loans), or the Covid-19 student loan repayment pause for

loans guaranteed by the federal government.

Interestingly, individual educational insurance policies where students pay a premium to protect them-

selves from income risk are either not well developed or are non-existent.

One reason for the low preva-

lence of educational insurance in post-secondary markets may be low demand. There is evidence that stu-

dents can be over-optimistic about future earnings (e.g. Baker et al. (2018)), failing to adequately consider

insurance risk at the time of enrollment and ﬁnancing because of difﬁculty in predicting future incomes ac-

cording to major (Arcidiacono et al., 2012; Baker et al., 2018; Conlon, 2021). On the supply side, another

explanation could be the presence of adverse selection (Einav et al. (2023)) and moral hazard (Zweifel and

Manning, 2000) in insurance markets.

Closest to our setting, individuals purchase insurance to mitigate ﬁnancial losses (e.g. Arrow, 1963), buffer against income

shocks (e.g. Chetty and Szeidl (2007)), and for a variety of other reasons. Guiso and Paiella (2008) document the increasing propensity

of households to hedge against labor income risk in particular, indicating a rising awareness of employment/income uncertainties. In

the case of ﬁnancial markets, diversiﬁcation, including the use of derivative instruments like futures and options, remains a primary

strategy for risk management in the face of uncertain economic outcomes (e.g. Bodie (1994); Goyal and Welch (2007)).

Throughout this paper, we will assume that the primary form of insurance in post-secondary education is against the risk of low

or uncertain income, and will refer to this as "low income insurance" or "educational insurance."

Our paper makes a unique contribution to our understanding of student demand for educational insur-

ance and the potential relevance of adverse selection in the viability of low-income insurance in education

markets. Students who are unsure about their prospects may demand insurance to protect themselves against

downside labor market risks. Adverse selection is related but different: the insured use information that they

can easily conceal from the insurance company to take advantage of the insurance’s downside protections.

Our major contribution is a framing experiment used to tease out insurance demand; our survey allows us

to see concealed information unavailable to the hypothetical insurer to test for adverse selection. Using

the experiment coupled with the survey, we can test how framing affects a student’s demand for education

insurance and see if students use concealed information to take up the protections at different rates.

We partnered with a large, non-proﬁt university (hereafter, The University) that typically serves non-

traditionally aged students who are often working adults. We conducted a randomized survey lab experi-

ment with 2,776 students to understand their preferences over different educational ﬁnancing choices.

In the survey, students were asked to choose between a hypothetical federal student loan with the option

of an income-driven repayment (IDR) plan and a hypothetical income-share agreement (ISA). Both options

provided information on monthly loan payments that are waived for very low incomes and otherwise are

capped to a ﬁxed share of an individual’s income, with markedly different implementation and paths for

satisfying the loan obligation.

In the experiment, students were randomized into two equal groups and, similar to Abraham et al.

(2020), the presented hypothetical options differed in terms of level of detail provided for each of the

choices. The ﬁrst group was shown descriptions of the student loan with option of IDR and the ISA with a

risk-neutral framing that explained the differences in monthly payments, general structure of the loan, the

baseline payment terms, and the source of funding. The terms of the student loan with IDR and the ISA

were set to be actuarially equivalent. We exposed the second group – our treatment group – to the same

descriptions of the student loan with option of IDR and the ISA, but with an additional emphasis on the

insurance features (nature of the income contingency and maximum repayment term) of the two ﬁnancing

options.

There are many differences between federal student loans with IDR and ISAs, and many reasons why

different borrowers might prefer one over the other. With federal student loans, borrowers who do well

in the labor market will pay less in total by paying ﬁxed monthly payments for the minimum number of

years (120 payments, or 10 years with no gaps in payment). Since there is no prepayment penalty for

federal student loans, they can also be paid off faster than scheduled and may be particularly attractive to

students who expect consistently high earnings after college. To access the income contingency, borrowers

must follow a series of administrative hurdles in order to qualify for reduced monthly payments capped at a

certain percentage of their income, paying nothing if their income falls below a set threshold, but potentially

extending their term up to 20 years compared with the standard repayment plan.

With an ISA, borrowers’

monthly payments are set as a pre-agreed share of income by design, and the repayment term is typically

extended to a lesser degree than IDR due to months of non-payment, making an ISA a potentially attractive

proposition for borrowers with persistently low or variable earnings. On the other hand, there is no way

to “reﬁnance” out of an ISA and borrowers who end up earning high incomes will pay up to a multiple

of the original loan amount, described in our experiment. Finally, borrowers may have preferences over

borrowing from the government versus a private lender.

We ﬁnd that students have a signiﬁcant preference for the built-in income insurance in the ISA and

that our insurance framing greatly increases the demand for the hypothetical ISA – by about 10 percentage

points, or 43 percent. Importantly, there is limited heterogeneity in treatment to be found along demo-

graphic, academic, or ﬁnancial lines for students in our sample.

The insurance framing has, by far, the

largest effect on take-up. Our results suggest that students are not necessarily thinking about income risk

or about the potential beneﬁts of educational insurance when they choose how to ﬁnance their studies, but

that educational and loan providers can help make the potential need for educational insurance salient for

borrowers by thoroughly explaining the costs and beneﬁts of such insurance.

Our survey and follow-up questions also allow us to characterize how adverse selection may enter the

educational insurance market (Herbst and Hendren, 2021).

The survey allows us to solicit information

from the student that is unavailable to the ISA originator. Students may be conﬁdent about their income

potential but can easily conceal this information from ﬁnancing providers who do not have the ability to

price discriminate (i.e. must charge the same interest rate or income share to all). In such an environment,

students expecting low incomes will sort into the ISA while students expecting high incomes will opt for a

traditional loan. Educational insurance that looks more like an ISA will not be a sustainable policy choice

if students who achieve signiﬁcant returns to college systematically choose student loans. Overall, there

is less evidence suggestive of adverse selection across a variety of variables than we supposed ex ante.

Employment uncertainty, for example, does not appear to inﬂuence take-up. However, we do ﬁnd strong

suggestive evidence of adverse selection based on likelihood of future incomes being low.

To further test for adverse selection, we ask several follow-up questions to investigate how students

Recent policy changes around the Saving on a Valuable Education (SAVE) IDR plan simplify some of these processes for

federal student loans, but the income protection is still far from built in. Although some currently available IDR plans offered by the

Department of Education extend the repayment term to up to 25 years, we wanted to keep the comparison simple for borrowers and

chose the (modal) maximum term of 20 years.

The preregistered baseline variables for heterogeneous treatment effects included race/ethnicity (Black, Hispanic, white), gender

(indicator for female respondent/recipient), household size, age (median split), marital status, risk aversion.

Since we cannot follow students after the survey, we cannot shed light on potential moral hazard from the availability of insurance

for students in the treatment group.

might change their answers if the offer terms were modestly different. After their initial choice between a

student loan and an ISA, we offered students who originally selected a student loan with IDR an actuarially-

equivalent alternative ISA with a lower income share and a longer term. If a student maintained their

original choice of student loan in the second round as well, we then offered them another alternative ISA in

the third round – this time with a higher income share and a shorter term than the original ISA. To students

who originally selected the hypothetical ISA, we separately offered both alternative ISAs at the same time.

We ﬁnd that both original ISA choosers and original student loan choosers were equally likely to switch

to the longer term ISA, with 18% of respondents selecting the longer term ISA over their original choice.

Interestingly, respondents who originally chose the ISA were considerably more likely (61%) to choose the

shorter term ISA compared with the respondents who chose the student loan with IDR in both the ﬁrst and

second rounds (17%).

We further ﬁnd that treated students who originally chose the student loan were 5.1 percentage points

(17%) more likely to choose the ISA option with a lower share and longer term. The treatment effect for

those offered the alternative ISA with shorter term and higher share is not statistically signiﬁcant. For

students who chose the original ISA in our base experiment, treated students were 8.6 percentage points

(15%) more likely to choose the alternative ISA option with a higher share and shorter term. We ﬁnd no

signiﬁcant treatment effects on switching toward the lower share and longer term ISA for students who

originally chose the ISA over the student loan with IDR. Overall, our results suggest that the insurance

framing helped reinforce student preferences over shorter maximum repayment term (12 years for ISA v.

20 years for student loan), and that students selected alternative ISA contracts in our follow-up questions in

a way that reﬂected their stated preference between the initial choice.

Separately, we asked all students whether they would select a student loan if there were a 20% chance

that the $10,000 loan they borrowed would be forgiven. We ﬁnd that students who switched from a loan

to an ISA in the second round were 6.4 percentage points more likely to switch back to a loan when of-

fered the chance of student loan forgiveness. Conversely, when students who picked the original ISA were

offered a loan with a chance of future debt forgiveness, the prospect of future balance reduction decreased

the willingness/likelihood of switching back to the federal student loan. The students who switched to an

ISA in the second round are marginally attached to the ISA and may be easily induced to switch between

the two ﬁnancing choices given relatively small changes to terms, while the original ISA choosers appear

to be more set in their ISA preference. Taken together with our results on student preferences over alterna-

tive ISA variations compared with a student loan with IDR, our study contributes to the understanding of

optimal design of loan products with income insurance features with regard to both upside protections (like

maximum term or total payment amount) and downside protections (like income share).

Though focused on educational insurance markets, our paper contributes to other lines of research,

including the literature on ﬁnancial aid, education ﬁnance, and student debt. As the cost of higher education

has risen and the purchasing power of public subsidies have fallen along with public ﬁnancial support to

universities (Webber, 2017), families have had to incur debt or forego consumption to afford post-secondary

education. Recent research has emphasized the burden that student loans place on students (Chakrabarti et

al., 2020) - including on their other consumer spending (e.g. Mezza and Sommer, 2015), "life milestones"

(Mezza et al., 2020), and educational outcomes (e.g. Black et al., Forthcoming, Denning and Jones, 2021).

Because the monthly payment is proportional to income, education insurance such as IDR and ISAs can

hedge against the adverse effects of student loans. As such, take-up of such products, particularly among

populations where student loans have had adverse effects (e.g. students at technical or public regional

colleges with higher variance in college outcomes), is important.

Many families do not apply for aid – meaning they do not complete the required ﬁnancial aid forms –

because of a lack of information or uncertainty of eligibility (e.g. Kofoed (2017), Bettinger et al. (2012)).

Even for families that apply for aid, the protection of some assets within the Pell eligibility formula leads

to less ﬁnancial aid eligibility for students from disadvantaged families (Levine and Ritter, 2023) resulting

in higher student loan burdens and decreased access to selective institutions. Forms of built-in educational

insurance are potentially more attractive to these students, as they automatically reallocate some of the risk

from the student to the provider and may reduce uncertainty around ﬁnancial aid eligibility.

Additionally, our research contributes to the behavioral literature on student take-up of ﬁnancial aid

programs under varied framing.

Abraham et al. (2020) and Marx and Turner (2019) demonstrate that

framing matters for government-sponsored IDR plans and traditional student loan take-up, respectively. In

this literature, researchers manipulate the students’ norms, the terms, risk, and the costs communicated to

students with respect to speciﬁc ﬁnancial instruments. Our paper contributes to this discussion by showing

that students prefer income contingent ﬁnancing when we emphasize the built-in educational insurance of

the ISA, with treatment effects for the insurance framing comparable in magnitude to Abraham et al. (2020).

The lessons from our study are applicable to the design of any income-contingent education ﬁnancing

product and are particularly salient to ongoing policy discussions around the Department of Education’s

income-driven repayment plans for federal student loans.

Our paper is organized as follows. Section 2 reviews students’ college ﬁnancing and the prospect

for education insurance in ﬁnancing education. Section 3 details the experimental design of our research

questions. Section 4 lays out our empirical strategy and elaborates on the data collection. Section 5 provides

Cox et al. (2020) examine why students don’t choose IDR when they are worried about future income expectations. They

conduct a laboratory experiment where they provide information about IDR and default students into the plan. They ﬁnd that extra

information and correct defaulting does increase enrollment.

empirical results. Section 6 offers concluding remarks and policy considerations.

2 College Financing and Educational Insurance

This section describes borrower labor market expectations and available (student loan with IDR) and rel-

atively novel (ISAs) higher education ﬁnancing options with income insurance that motivated our exper-

iment. In Section 2.1, we describe borrowers’ income/employment trajectories and potential risks and

disruptions to future income and employment. We then proceed to explain the mechanics of an IDR option

for a traditional student loan in Section 2.2 and the typical ISA in Section 2.3. In Section 2.4, we discuss

how the features of the two ﬁnancing options might inﬂuence choices between them and which types of

borrowers might respond to which incentives. Finally, Section 2.5 discusses the motivation behind our

experiment and the comparison we offered to students in our study.

2.1 Students’ Educational Risks, Earnings Trajectories, and Repayment Shocks

Until recently, students and parents typically repaid government student loans in ﬁxed monthly payments

over a given repayment period, akin to a traditional mortgage loan (Karamcheva et al., 2020). This payment

remained constant regardless of age, income, employment status, or family situation. Most non-government

lenders have offered private loans with a standard, mortgage-style payment schedule, though select lenders

are beginning to offer or contemplate alternative options. Some 95% of outstanding student debt is guaran-

teed by the federal government, so repayment plans designed and offered by the Department of Education

dominate the set of choices available to students.

It is important to consider risk to the returns to college attendance (including borrowing for that atten-

dance) when thinking about future income and employment prospects for students (Webber, 2016; Balakr-

ishnan and Cynamon, 2018; Hendricks and Leukhina, 2018; Akers, 2021). Perhaps the largest risk factor

to the repayment of educational debt that students face involves the risk of non-completion. Historically,

6-year completion rates have hovered around 60%, and are even lower for nontraditional college students

and non-selective institutions (Bowen et al., 2009). Financial circumstances, lack of academic prepared-

ness, and a host of academically orientated challenges may put students’ ﬁnancial investment in college at

risk. Additionally, there are “life” risks that students face, including emergencies arising from physical to

emotional health to family circumstances. Adult learners, in particular, report that child-care emergencies,

children’s health, and even transportation emergencies can derail their educational careers (Markle, 2015).

Even for completers, risks to income and employment are many. The average ﬁnancial return for the

median graduate of a 4-year college or university is large, and handily outweighs the implicit and explicit

costs of attending college, which is why enrollment in a postsecondary program of study makes sense for

most students ex-ante. But ex-post, returns are heterogeneous across many dimensions including major,

institution type, and institution prestige. Since student populations particularly at risk of low or negative

returns to college enrollment tend to skew toward vulnerable groups, addressing the riskiness of college

attendance with product/program design and effective public policy is imperative.

Borrowers also face a variety of income shocks during repayment, such that repayment burdens can vary

widely for individuals with variable or uncertain income and/or employment (Chapman and Lounkaew,

2015; Chapman and Dearden, 2017). Borrowers may face temporary repayment challenges (e.g. due to

periods of unemployment or underemployment) or chronic repayment struggles due to low incomes (e.g.

because of degree non-completion, or degree/major with poor ﬁnancial return on investment).

Fixed payments over a 10-year period for borrowers who recently completed or dropped out of a pro-

gram of study may not be optimal given typical earnings trajectories, either. For most borrowers, student

loan debt service ratios (i.e., scheduled payments as a share of a borrower’s income) are typically greater

early in the repayment term, when a borrower’s income is lower. This is particularly true for student loan

borrowers with little work experience upon entering repayment, for borrowers who typically begin repay-

ment in lower paid early-career positions but ultimately earn substantial amounts (e.g., medical doctors),

and for borrowers with degrees in majors that traditionally have steep earnings trajectories (e.g., biology).

Considering all of these factors, standard repayment plans with ﬁxed scheduled monthly payments may

be burdensome for borrowers and poorly suited for a setting with myriad risks to income and employment.

2.2 Income-Driven Repayment Plans for Federal Student Loans

In response to these earnings patterns, repayment shocks, and increasing debt service ratios, the Depart-

ment of Education expanded income-contingent repayment programs (sometimes referred to as income-

driven repayment, or IDR) for federal student loans. IDR plans reduce student loan payments to 5-15%

of discretionary income, deﬁned as the amount of adjusted gross income (AGI) above a multiple of the

Federal Poverty Level (FPL). As of the time of our experiment, the dominant IDR plan (and one on which

our student loan with IDR option was modeled) was Revised Pay As You Earn (REPAYE) plan. Under

REPAYE, borrowers owed 10% of income over 150% of the FPL for up to 20 (undergraduate loans) or

25 (graduate loans) years. The newest IDR plan introduced in 2023, the Saving for a Valuable Education

(SAVE) IDR plan, sets scheduled payments to 5-10% of income over 225% of the FPL. As of 2019, about

35% of borrowers with federal Direct student loans enrolled in one of the several IDR plans offered by

the Department of Education (Trends in Student Aid, 2019), with the rest ineligible for IDR (e.g., parent

borrowers) or preferring to remain with the standard plan.

A borrower’s payment in an IDR plan is deﬁned below:











0 if DI

it−1

≤ 0

rDI

it−1

if DI

it−1

> 0

(1)

where last year’s discretionary income is DI

it−1

= AGI

it−1

− mF P L

(n), n is the applicant’s family

size, m ∈ {150%, 225%}, and r ∈ {0.05, 0.10, 0.15}. The IDR obligation is satisﬁed when a) the original

balance and accumulated interest are repaid, or b) the borrower has reached the maximum repayment period

of 20 or 25 years (depending on the particular IDR plan), whichever comes ﬁrst. By design, an IDR plan

retains a balance-tracking feature, such that borrowers whose IDR payment is less than the monthly interest

experience interest capitalization and can see their balances grow (even balloon), as frequently reported in

the popular press.

Again, the hypothetical student loan with IDR in our experiment is modeled after the REPAYE plan,

such that m = 150%∗F P L(n), r = 0.10, and the maximum repayment period is 20 years. We note that the

equation above determines the maximum IDR payment, and that a borrower’s payment in a particular month

may amount to less than this maximum payment, such that the IDR option acts as insurance against low

income at the potential cost of a higher aggregate payment amount and/or an extended repayment period

relative to the 10 years in the standard plan.

Yet, enrollment rates in these repayment plans remain low for myriad reasons such as administrative

hurdles, poor design, and servicer misconduct (e.g. Mueller and Yannelis, 2022). Herbst, 2023 uses ran-

domized variation in loan servicer outreach to ﬁnd that enrollment in an IDR plan reduces student loan

delinquencies by 22 percentage points and decreases outstanding balances within a year of take-up. Many

students do not qualify for federal student loans — perhaps because they have exhausted the federal life-

time undergraduate borrowing limit (approximately $30,000), or because they are ineligible for federal

loans (e.g. due to parental income/assets, due to attending a non-eligible institution, or due to not being US

citizens). Other students (and their parents) take out loans that are ineligible for IDR altogether.

2.3 Alternative Income-Contingent Financing: Income Share Agreements

One alternative to student loans are ISAs, or income share agreements. Both states and institutions have

experimented with ISAs in the past decade. ISA providers claim to ﬁll funding gaps faced by students

and to push programs to align incentives with students because institutions receive repayment as graduates

succeed in the labor market because repayment rates are dependent on students’ post-college employment

and income (Ritter and Webber, 2019). Over the past decade, ISAs have covered the spectrum of post-

secondary education, from large public systems and private non-proﬁt colleges to certiﬁcate programs and

small, for-proﬁt coding bootcamps (Ritter and Webber, 2019, Zaber et al., 2023).

ISAs are neither a traditional loan product nor a traditional ﬁnancial aid instrument. They are intention-

ally designed to share some of the risks of pursuing post-secondary education between the student and the

ISA provider, including risks to graduation or adverse life events. Under an ISA agreement, students pledge

a proportion of their future earnings in lieu of paying tuition in the present. ISAs typically include, as built-

in features of the ﬁnancial product, downside protections (a minimum income threshold below which no

payments are due, a maximum repayment window, and/or and maximum number of payments) and upside

protections (maximum cap for the amount that can be repaid before the obligation is satisﬁed).

A borrower’s payment in an ISA is deﬁned as below:











0 if I

≤ I

min

a ∗ s ∗ I

if I

> I

min

(2)

where P

is the scheduled payment under the ISA plan for borrower i in time t, I

min

is the minimum

income threshold, a is the total amount borrowed, and s is the income share per dollar borrowed. Differently

from the maximum IDR payment speciﬁed in Equation 1, which is determined irrespective of the amount

borrowed, the ISA payment scales with the amount borrowed.

In terms of the exit criteria from the obligation, ISAs do not track balances, by design. Instead, the

provider tests whether the borrower’s cumulative amount of payments is above a maximum cap or multiplier

of the borrowed amount (e.g. 2X original borrowed amount) in each period, or whether the borrower has

reached the maximum number of payments or the maximum term length. In other words, the total amount

of payments is subject to the following restriction:

CumulativeP ayments

≤ c ∗ a (3)

s.t. NumP ayments < MaxNumP ayments and t < T imeCap.

As before, a is the total amount borrowed, while c is the multiplier or maximum cap. In other words,

an ISA obligation is satisﬁed if a) the borrower reaches the maximum cumulative payment amount, b)

the borrower makes the required number of non-zero payments, or c) the borrower reaches the maximum

repayment window. In our experiment, the hypothetical ISA presented to the students closely resembles

the one considered by The University - an income share of 2.3% of monthly income, maximum number of

nonzero payments of 120 (10 years), maximum term of 13 years, and a max cap of 2X the original borrowed

amount.

From a consumer protection perspective, ISAs are currently not as tightly regulated as traditional stu-

dent loans, with a hodgepodge of state-level approaches to enforcing consumer protection and prudential

regulations and an emerging federal regulatory framework for ISAs. The Consumer Financial Protection

Bureau considers ISAs private student loans for regulatory purposes, but has provided limited guidance

regarding how existing regulations apply to ISAs. For example, the Truth in Lending Act governs the

disclosures of APR, among things, which is a distinctly loan-speciﬁc term and does not straightforwardly

apply to an ISA-style contract. However, although these important consumer protections concerns exist, we

have chosen to abstract from them in our experiment as the typical borrower would be highly unlikely to be

aware of any regulatory or consumer protection distinctions between the two ﬁnancing options.

2.4 Borrower Choice Between Financing Options

While many comparable insurance protections are presently available under IDR plans for federal student

loans, they are not the default option in the United States. Borrowers have to bear the time and effort cost

of enrolling and often fulﬁll many requirements in order to remain enrolled, including burdensome annual

recertiﬁcation processes for the most generous IDR plans. Additionally, the menu of IDR programs has

changed multiple times since the introduction of the ﬁrst repayment program of this style – as recently as

summer of 2023 – meaning that the availability of a particular repayment plan for federal loans is uncertain,

at best. Therefore some students may ﬁnd attractive the ex-ante agreement to repayment terms that will be

in effect for the duration of the contract inherent with an ISA. Monthly payments vary with income by

design, and there is no interest rate, per se, which could be welcome (as payments will never be overly

burdensome) or unwelcome (because payments may be seen as unpredictable) for borrowers.

Turning to the monthly payment incentives to choose a student loan with IDR versus an ISA, there

are several factors which may lead to differing payment levels considering the parameters introduced in

Equations 1 and 2. This is true both at a point in time and in terms of the cumulative payment amount deﬁned

above, which could lead borrowers to prefer one product over the other. For one, the total time obligation of

each potential loan product is potentially different. Under an ISA, the length is determined based on (1) the

number of non-zero monthly payments, and (2) the total number of months elapsed including months where

no payment is required. For simplicity, in the hypothetical ISA we offered to students in our experiment

these terms were both set at 10 years.

Therefore borrowers who have a strong preference against long

maximum terms might prefer the ISA.

In practice, while the maximum term length and maximum number of payments could align, most ISA contracts differentiate

between total payment length and the number of non-zero payments.

From an income trajectory perspective, the shorter obligation period of ISA would be preferable to

borrowers who believe they will fall below the income threshold early on (for instance if they have private

knowledge they will either be out of the labor market), but may be substantially above the threshold in later

years. Apart from the length of obligation, the hard discontinuity in ISA payment calculations close to

the threshold can also create a strong incentive to prefer one contract over the other. IDR payment shares

only apply to discretionary income, e.g. the marginal income above the 1.5*FPL threshold (for a single

borrower, in the neighborhood of $14,000 in 2022). And although the ISA minimum income threshold of

$36,000 per year borrower is higher than the IDR minimum income threshold, a borrower making a single

dollar above the ISA minimum income threshold will make a payment based on their entire income, rather

than the marginal amount above the threshold. Thus, borrowers expecting to earn incomes between the two

thresholds might be particularly motivated to choose the student loan with IDR.

Finally, we consider borrowers who may anticipate doing particularly well in the labor market and

beneﬁting from the upside protections in either ﬁnancing option. Borrowers who might like to have the

option of paying their loans back more quickly than the initial repayment schedule might prefer the student

loan because the repayment term in the standard plan is less than 10 years for borrowers making excess

payments above scheduled amount. Similarly, borrowers expecting to potentially earn high amounts may

be attracted by the lower cumulative amount paid on the standard plan ($12,720 for $10,000 borrowed

given typical interest rates) compared with the relatively higher maximum cap on the ISA ($20,000 per

$10,000 borrowed in our hypothetical ISA). Of course, this potentially lower cumulative amount paid given

a high income outcome is traded off against the possibility of a low income outcome where the borrower

experiences income capitalization and may be required to make payments for up to 20 years.

Ultimately, students’ decisions between ﬁnancing options are ultimately about earnings expectations

(both level and time path), preferences over monthly payments and upside/downside protections, risk toler-

ance, and risk aversion, as described above. They are very similar to the decisions made in contemplating

other types of insurance.

2.5 Motivation for the Experiment

Student choices between available ﬁnancial options could provide information useful not only to educa-

tional programs and students, but also to regulators as they decide how to apply or augment regulations in

the coming years. Given the similarities between ISAs and IDR plans for federal student loans, insights

into student preferences over different features of income-contingent ﬁnancing can be highly relevant to

plans for the expansion or redesign of IDR. As mentioned previously, the Department of Education has

recently undertaken the design of a new version of an IDR plan for Direct student loans, the SAVE IDR

plan launched in August 2023, so insights from our paper are particularly timely.

Despite the recent proliferation of these ﬁnancing products, there is a surprising dearth of evidence

regarding student preferences over features of income-contingent ﬁnancing generally or ISAs in particular.

Currently, regarding the United States, there are no studies exploring experimentally why students may

choose an ISA or a product like it over a more traditional student loan. Even globally, there are limited

studies (i.e. Herbst et al., 2022) discussing which types of students may beneﬁt from an ISA, or addressing

the effects an ISA might have on students’ educational success and ﬁnancial well-being. Understanding the

preferences and characteristics of students likely to take up an ISA sheds light on which of them can be

helped or harmed by educational insurance and why.

Finally, ISAs offer a useful laboratory for applying lessons on income-contingent ﬁnancing to public

policy on student loans. Our results shed light on IDR plans offered by the Department of Education and

other entities – especially given that federal loans generally and IDR plans in particular are not available to

many students and programs. Thus, the experimental evaluation we conduct in the present paper has high

practical importance.

3 Research Questions and Experimental Setting

3.1 Research Questions

We study the following research questions. First, how can framing or emphasizing a ﬁnancing product as

educational insurance affect student take-up? Second, what factors inﬂuence a student’s decision to take up

educational insurance? Third, what is the nature of adverse selection in educational insurance markets?

The ﬁrst question involves behavioral testing of whether explicitly emphasizing the low income insur-

ance features of the ISA affects take-up relative to a federal student loan with the option of IDR. Many

students are unaware of the differences in upside and downside risk protections that IDRs and ISAs offer.

In the experimental portion of our paper, we randomize students into an educational insurance framing of

the ISA to see if insurance related attributes increase interest in the program and draw students with private

information into the risk pool.

The second question examines take-up moderators. For example, ISAs are typically offered by univer-

sities to help students manage the risks of non-completion, unemployment, and low earnings; as we show

below, students with high risks for these situations are precisely the students who stand to beneﬁt the most

For more information on the SAVE IDR plan, see: https://studentaid.gov/announcements-events/save-plan.

from the built-in educational insurance and are most likely to take-up the ISA. More generally, we measure

how take-up and the efﬁcacy of an insurance framing varies by student characteristics, speciﬁcally those

associated with the risks that ISAs insure against.

The third question aims to understand the nature of adverse selection and how personal perceptions of

risk affect students’ decisions to participate in hypothetical ISA programs. In education markets, adverse

selection arises when students use information which is not observable (or simply not permissible in a lend-

ing decision) about their academic ability, possible choice of major, and potential labor market outcomes

to sort into an education ﬁnancing product with insurance features such as an ISA or IDR in a way that

is correlated with repayment rates. Given the rich administrative data provided by the University, and the

questions from our survey related to risk tolerance and income expectations, we can show which types of

students appear to be interested in educational insurance ex ante.

3.2 Survey and Experimental Settings

The survey targeted a sample of 8,000 students enrolled in programs that The University considered offering

possible ISA funding through a third-party provider. The University received a 35% response rate for

the survey. Eligible students were undergraduates who were either juniors or seniors (or approximately

two years from degree completion) and who have completed a FAFSA form. The University focused on

students majoring in pre-license and post-license nursing, health information management, cyber security,

or business administration.

Potential study participants received an invitation from the University offering them the opportunity to

help the University better understand the student experience with ﬁnancial aid. The University offered a

$10 Amazon gift cards to compensate students for their time and effort, informing them that they could at

any time withdraw their consent to be a part of the study. The University branded the study with its logo

and name as opposed to the institutions of the research team or the ISA lender.

The survey experiment asked students to choose between a federal student loan with the option of an

IDR plan and a hypothetical ISA. Monthly loan payments capped to a ﬁxed share of an individual’s income –

or waived for very low incomes – are available under both options, with markedly different implementation

and paths for satisfying the loan obligation. In the experiment, students were randomized into two equal

groups and, similar to Abraham et al. (2020), the presented options differed in terms of level of detail

provided for each of the choices. The ﬁrst group was shown descriptions of the student loan with option of

IDR and the ISA, with a risk-neutral framing that explained the differences in monthly payments, general

structure of the loan, the baseline payment terms, and the source of funding. The terms of the student loan

with IDR and the ISA were set to be actuarially equivalent. We exposed the second group – our treatment

group – to the same descriptions of the student loan with option of IDR and the ISA, but with an additional

emphasis on the insurance features (nature of the income contingency and maximum repayment term) of

the two ﬁnancing options.

We randomized students into two different framings that compared the terms of a hypothetical ISA to a

federal student loan with an option of IDR. First, we randomized half of the participants into the treatment

arm that emphasized the ISA as a form of insurance that protects against downside employment risk, with

the stress placed on ﬁxed repayment terms as opposed to a balance. This treatment measured the effects of

emphasizing the possible need of “education insurance" given the potential downside risk of costly human

capital investment. The control group consisted of half of the students, receiving a “risk neutral” framing

which focused on monthly payments and repayment terms. In both the treatment and control groups,

students were asked their preference and the intensity of their preference between the ISA and the student

loan with IDR. This survey experiment serves as our primary comparison throughout the paper. Figure 1

displays a ﬂowchart that describes the randomization in more detail.

Second, given that students’ risk and time preferences may shed light on the nature of insurance take-

up, we created a subsequent survey module varying the terms of the students’ funding choice. If a student

chose a federal student loan with the option of IDR in either of the treatment arms, we provided them with

alternative options of ISA terms, varying income share percentage and term length to do so, while keeping

the internal rate of return (IRR) constant. These alternative ISA options and the students’ selections provide

a revealed preference measure of discount rates, which could be an important marker of potential adverse

selection. Of course, an important caveat is that the sequential nature of the experiment induces selection at

each stage, such that the estimates are not representative of the survey population as a whole. We describe

this in greater detail in Section 5.4.

Details of the research design and analysis were preregistered on the Social Science Registry (Balakr-

ishnan et al., 2023). We discuss our ﬁndings below.

4 Data and Econometric Model

4.1 Data Sources

For our analysis, we combine data from two unique sources: our primary survey and administrative data

from the University.

Our ﬁrst data source comes from the survey of students that we conducted in June-July 2022. Out of

8,000 students invited to complete the survey, the University received 2,776 responses, a 35% participation

rate. The survey data includes detailed information about household composition, current living situation,

personal income as well as household income, ﬁnancial stability

, current education ﬁnancing, and ﬁnancial

literacy. Within the survey, we randomized how recipients learned about a hypothetical ISA, framing it as

either an insurance option or a risk-neutral option in comparison to a federal student loan with an option

of IDR. Our dependent variable is whether they selected the ISA or the federal student loan. Given the

importance of career expectations on the desirability of insurance, we also ask questions on career, income,

and employment expectations, perceptions on graduation, and risk preferences. Our methodology for all

derived controls can be found in Appendix A.

The administrative data includes detailed information about social and demographic characteristics such

as age, gender, race/ethnicity, marital status, state of residency, household size, highest education level,

employment status, and household income. Along with academic records showing students’ majors, we

received information on how many semesters they have completed, attempted and completed number of

credits, transfer credits if any, and credits needed to graduate. The sample includes 5,465 observations for

2,776 students – our survey sample – and when available all variables were observed in both June 2022

and July 2022 to match the time frame of the survey’s administration. The administrative data also includes

variables on student ﬁnance. The ﬁnancial aid data include information on students’ academic careers

such as their enrollment status (active, dropped, taking a term break, etc.), grade level (undergraduate

or graduate), academic standing, receipt of ﬁnancial aid or award, lifetime federal loan borrowings, and

proximity to aggregate federal loan limits. The ﬁnancial aid sample includes 7,028 observations with 2,776

unique students. The data are structured as an unbalanced panel with annual observations of each student

for ﬁnancial aid years on ﬁle with the University, between Fall 2018 and Fall 2022. These properties help

us track the ﬁnancial aid history and academic progress of the students.

4.2 Econometric Model

To estimate the determinants for ISA take-up, we estimate Equation 4:

= β

+ β

+ γX

+ ϵ

(4)

where y

, the outcome variable of interest, denotes student i’s preference for a hypothetical ISA (coded as

1) or a federal student loan (coded as 0). T is a treatment indicator, randomly assigned to respondents,

that represents either the insurance-based framing of an ISA (coded as 1) or a risk-neutral framing (coded

See Appendix A for deﬁnition.

as 0). C is a vector of the key student characteristics associated with risk and potentially adverse selec-

tion. Speciﬁcally, we include the following characteristics: current income and employment status, career

expectations including both income and employment expectations after graduation, perceived barriers to

graduation, and risk preferences.

X is a vector of demographic controls including age, gender, race/ethnicity, and marital status. In the

survey administered to students, the University also collected data on household size and current educational

status. We also include variables that we solicited in our survey before students observed the education

insurance framing. These variables include risk preferences, ﬁnancial stability, measures for income and

employment uncertainty, ﬁnancing choices for their education, and barriers to graduation.

Unlike in many RCTs, where respondent characteristics and control variables are primarily included to

reduce bias or increase precision, we do have a ﬁrst-order interest in the coefﬁcients on every variable, as

we wish to assess both the existence and extent of adverse selection. In other words, we test for effects of

each variable on ISA preference.

We also test for heterogeneous effects by estimating Equation 5:

= β

+ β

× H

+ β

+ δC

+ γ

′

+ ε

(5)

where H

is an indicator variable for the dimension of heterogeneity including race/ethnicity, gender,

household size, age, marital status and risk aversion. As in Equation 4, T is a treatment indicator, C is a

vector of the key student characteristics associated with risk, and X is a vector of demographic variables.

In Section 5.2.1, we additionally consider differences in ISA take-up in the control and treatment groups by

employment uncertainty and income uncertainty.

4.3 Balance

First, we present summary statistics for our sample and show covariate balance as evidence that our random-

ization was successful. We regress each of our demographic variables on the treatment indicator. Table 1

summarizes the characteristics of our sample overall and also checks for balance between treatment and

control groups. We do not see any imbalance between our two randomized groups across pre-treatment co-

variates that include age, sex, race/ethnicity, marital status, household size, education level, current income

and employment, ﬁnancial stability, future income and employment expectations, risk aversion, current ﬁ-

nancing choice, and barriers to graduation. The average student in the survey sample is in their late 30s

(37.8), employed, and ﬁnancially stable. A greater fraction of students are white (55%) and female (71%).

Half of the students are married. Most students have an associate’s degree (64%) or a bachelor’s degree

(20%). For half of the students, their annual total personal income is between $40,000 and $90,000 whereas

almost one-third of the students have an annual total household income that is $120,000 or more. Students

in our sample believe that they are highly likely to be employed full-time within 5 years of graduation and to

earn an income around the average of the individual middle-income spectrum at $80,000. Among barriers to

graduation, they see work responsibilities (79%), family constraints (58%), and ﬁnancial constraints (41%)

as the main reasons for not getting as much out of their educational experience as they would like to. In

terms of enrollment balance, Columns 3 and 4 show that the students randomized into the treatment group

are overall very similar to those in the control group, conﬁrming that we executed the random assignment

across framing narratives correctly.

5 Results

5.1 Impact of Insurance Framing on Take-up

First, we explore whether framing an ISA as a form of educational insurance and putting emphasis on its

protection against downside labor market risk is effective in increasing interest in the program. Table 2

summarizes our ﬁndings. We present various speciﬁcation in the following way: Column (1) captures the

baseline effect of treatment with no additional controls. Column (2) adds additional controls such as current

personal income, employment status, and ﬁnancial stability (not reported) to the control set. Column (3)

adds controls for the expectations for post-graduation career plans (not reported; available upon request),

income, and employment. Column (4) adds risk preferences. In Column (5), we incorporate the current

ﬁnancing choice of the student. Model (6) adds controls for barriers to graduation. Finally, Column (7)

additionally controls for demographic variables – age, gender, race/ethnicity, marital status, household size,

and educational status (we report signiﬁcant coefﬁcients only). Column (7) reﬂects the speciﬁcation listed

in our preregistration. Except for Column (1), all speciﬁcations include ﬁxed effects for students’ major.

Table 2 contains corresponding columns for each of our proposed model speciﬁcations.

Our results indicate that the insurance framing increases the likelihood of taking up an ISA, compared

to a federal loan with the option of IDR, by roughly 10 percentage points. This coefﬁcient is robust to our

model choice and statistically signiﬁcant. Given a mean of 23 percentage points ISA take-up for the control

group, this effect corresponds to a 43% relative increase in the take-up rate for ISAs in the treatment group.

The large baseline effect is striking, and emphasizes the value that individuals place on the ISA’s built-in

insurance against adverse employment/income outcomes and the shorter maximum repayment term. Our

treatment effects for the insurance framing are comparable in magnitude to Abraham et al. (2020) and is

stable across observable dimensions and demographic characteristics.

5.2 Heterogeneous Treatment Effects

Various demographic groups or individuals with different risk preferences may have disparate susceptibili-

ties to default risk and therefore may be affected differently by the framing of a ﬁnancing contract. We test

for heterogeneous treatment effects along several dimensions including race/ethnicity, gender, household

size, age, marital status and risk aversion. We use the Coefﬁcient of Relative Risk Aversion (CRRA), a

measure for risk preferences following a similar method to Kimball et al. (2008). We focus on students’

characteristics where we would expect to see a differential impact ex ante. But we do not ﬁnd any statisti-

cally signiﬁcant heterogeneous treatment effects under the context of any of these characteristics. Table 3

presents the heterogeneous treatment effects. For all subsamples we investigate, we observe that the insur-

ance framing is the driving force behind the ISA take-up.

One might have expected that some aspects of

adverse selection (e.g. future income risk) could lead to heterogeneous impacts; however, we do not ﬁnd

any evidence of adverse selection in this context.

One possible explanation for the lack of differential effects may be related to the imperfect nature of

the risk metrics and small subgroup cell sizes that may leave us with little power to precisely estimate any

differential effect (Balakrishnan et al., 2023). Additional analysis, including visual evidence and machine-

learning methods, is presented in the next section.

5.2.1 Additional Analysis

We further investigate the heterogeneous treatment effects based on future expectations and uncertainty

over employment and income to uncover adverse selection, if any. Figure 2 looks at the percentage of

students choosing an ISA given degrees of employment uncertainty and their treatment group. One can

observe that treatment increases the percentage of students choosing ISA in every category of employment

uncertainty. For example, among the people who are very certain that they would get a job in the future,

22% of students in the control group in this category chose ISA whereas the number is 32% for the treated

students. We investigate whether these percentages are signiﬁcantly different from each other within and

across categories.

The conﬁdence intervals drawn on the graph answer that question. We see that the percentage of stu-

dents choosing ISA is signiﬁcantly higher in the treatment group than the control group for both those

We preregistered corrections for multiple comparisons using the false discovery rate. However, since our dependent variable is

a dummy variable indicating ISA take-up, we did not need to use this correction.

See Table A1 to Table A6 in Appendix B for detailed results. In addition to the variables in Appendix B, we considered

heterogeneous effects based on income uncertainty and income uncertainty, but similarly did not ﬁnd any statistically signiﬁcant

differences; we omit those results for brevity and focus on the analysis in 5.2.1 for these two variables.

who are very certain and those who are moderately uncertain of their employment prospects. One can also

conclude that the percentage of students choosing the hypothetical ISA for treatment and control groups

separately across categories are not statistically different from each other at the 5% signiﬁcance level.

Yet

there appears to be some evidence that those who have very uncertain employment forecasts have a slightly

higher preference for the ISA. That may be why the treatment effect in that subgroup is not statistically

signiﬁcant. Therefore it could be that an insurance "message" for a group that does not have much concern

about employment uncertainty is where there is potential for behavior change.

Figure 3 carries out a comparable analysis for (high) income uncertainty. Similarly to what we ob-

serve for employment uncertainty, the treatment increases the percentage of students choosing ISA in every

category of uncertainty over earning a higher income in the future. The gap is smallest for those with the

highest level of uncertainty. For people who are very/moderately certain about their income prospects being

high, the treatment signiﬁcantly increases the percentage of students taking up the hypothetical ISA.

In-

terestingly, the ISA preference in the control group shows considerable heterogeneity, rising 15 percentage

points between the very certain and very uncertain categories (with the difference being borderline signiﬁ-

cant). The treatment effect for the two groups is almost the same, so it is the difference in the baseline that

explains the signiﬁcance of the treatment effect (or lack thereof).

Figure 2 and Figure 3 reveal that the results are most clear for (high) income expectations, likely because

this category does not have the added complexity of full-time/part-time expectations. As would be expected,

there is a stronger preference for the ISA contract among those with very uncertain beliefs about positive

future income and employment prospects. Among the most uncertain respondents, there is no discernible

treatment effect. However, as the certainty expressed by respondents increases, the framing of the ISA

contract becomes much more important. It is these respondents with relatively certain/secure beliefs about

the future that appear to be driving the baseline results.

One possible mechanism driving this result is the salience of downside risk among different types of

respondents. Those who initially express signiﬁcant uncertainty about their future prospects being favorable

are already prone to imagining negative outcomes, and hence they did not need the extra framing to view

an ISA in terms of its insurance properties. In other words, for those with considerable uncertainty over

high future incomes, perhaps no additional messaging about insurance is necessary because that aspect of

the ISA is already salient to them. For those with little uncertainty, though, the priming from the treatment

may have a greater opportunity to affect salience. One should also, once again, note the smaller sample size

We note that in the treatment group, the percentage of students choosing an ISA is statistically different at the 10% signiﬁcance

level between those who are moderately uncertain and very certain, and moderately uncertain and very uncertain about their future

employment.

One can also note that for the control group, the percentage of students choosing ISA is statistically different at the 10% signiﬁ-

cance level between those who are very certain and very uncertain about their future income being high.

– 103 students in the “Very uncertain of high income” category – and the fact that the uncertainty measures

in the survey are deﬁned based on Likert scales, which capture risk in limited ways.

As a robustness check and to incorporate the other dimensions of risk, we also employ the data-driven

approach by Athey and Imbens (2016) to estimate treatment effect heterogeneity. We investigate several

risk factors including perceived barriers to graduation, current ﬁnancing choices, risk aversion, future in-

come/employment expectations, and demographics

as candidates for sources of heterogeneity. The sug-

gested methodology builds on regression tree methods from the prediction-based machine learning literature

(Breiman et al., 1984; Breiman, 2001). The sample is ﬁrst divided into two parts: training and estimation

data. We use the training data to partition the population according to covariates, and then we use the

estimation data to obtain treatment effects for each subpopulation. This approach provides an advantage

for cases where there are many covariates by letting the data tell the researcher which relevant subgroups

one should look to for heterogeneity. By separating the datasets used to select the model structure and to

estimate, Athey and Imbens (2016) mitigate the possible bias in machine learning methods where spurious

correlations between covariates and outcomes affect the selected model. They argue that although the re-

duced sample size in each step through partition leads to loss of precision, the estimates are unbiased for

every subpopulation.

In this exercise, we simulate with different cut-off points in the sample to get the training and the

estimation data. In general, all simulations return housing insecurity and other ﬁnancial constraints, college-

related risks, work responsibilities, having government versus private student loans, and risk aversion as the

main sources of heterogeneity. We present evidence that many of these subsamples have signiﬁcant effects;

however, when we test the symmetry of the treatment effects, we ﬁnd few signiﬁcant forms of heterogeneous

treatment effects.

Figure A1 in Appendix C summarizes our ﬁndings when we use 40% of our sample as the training data

and the rest as our estimation sample.

5.3 Adverse Selection in Take-up

Adverse selection is a key concern to the stability of any insurance market. In particular, preference for

a ﬁnancing contract that depends on a prediction of future income is difﬁcult to navigate for a lender. If

the degree of selection is too large, the market can quickly unravel. Historical examples include dental

insurance, certain annuity markets, and even lifetime airfare products (Einav et al., 2023).

There are several dimensions along which adverse selection could present itself in an ISA context. First,

The demographics include age, gender, marital status, race and ethnicity, household size, and educational attainment.

the current ﬁnancial conditions of individuals (high income, stable employment, etc.) and/or future employ-

ment/income expectations may play a signiﬁcant role in whether respondents wish to pledge a percentage

of future income in exchange for downside risk protection. These student characteristics may not be ob-

servable to the lender or may be observable but not permissible to use in loan decisioning (consistent with,

for example, non-discrimination laws). If these characteristics are correlated with loan repayment rates,

then adverse selection is present in the market. Contrary to our expectations, we do not ﬁnd that current

income or employment status is predictive of ISA takeup in our student sample, many of whom report

income/employment concurrent with their studies.

Similarly, individuals may have private information about their future income/employment prospects,

irrespective of their current income/employment, which could substantively inﬂuence the decision to buy

insurance and is likely to be correlated with eventual repayment rates. A student may believe they are likely

to have a high income for any number of reasons. For instance, they may have an internship with a virtually

guaranteed job offer, or their degree may open up a promotion in their current job. In such cases, this

private information would lead to these students selecting out of the insurance market and opting to repay a

student loan as quickly as possible at the least amount of total cost. Conversely, students who believe they

are unlikely to have incomes above the ISA income threshold would be disproportionately likely to opt for

insurance, as it would render their tuition bill effectively free. Such students may include those anticipating

being out of the workforce, working part-time, taking low-paying jobs with high nonpecuniary beneﬁts, or

others. The measurement of adverse selection is the key determinant of whether a given insurance market

is viable, and is thus an important aspect of our current investigation.

To examine the potential for adverse selection, we asked students to rate the likelihood (based on a

5-point Likert scale) that they would be employed 5 years after graduating. We presented a comparable

question around expectations of future income being low, moderate, or high. Of particular note, at the time

students reported their beliefs about future income/employment, they were unaware of the hypothetical edu-

cation ﬁnancing choices they would be presented later in our study. As with the current income/employment

variables, uncertainty over future employment prospects does not appear to drive ISA preferences, indicat-

ing a somewhat surprising lack of adverse selection along this dimension. However, higher uncertainty over

future income being high does appear to have an economically large and marginally signiﬁcant impact on

student choice of ﬁnancing.

Employment and wage risk conditional on degree completion are not the only risks for providers of

education ﬁnancing. Another important dimension where adverse selection might emerge relates to the

likelihood that students do not complete their college degree. Students who believe they are less likely

to complete based on private information should be more likely to wish to insure against future negative

outcomes. We see some evidence of this in the "college-related risks" variable in Table 2.

We also investigate other risks to employment/income, including housing security, family health risks,

and current work responsibilities/risks. We ﬁnd no evidence that these risk factors seem to be driving

take-up. We do ﬁnd that certain student characteristics are strongly associated with take-up. Race, marital

status, and the speciﬁc way in which students are already ﬁnancing their education all have some power in

predicting take-up.

Namely, students who have previously ﬁnanced their education with federal student

loans are considerably more likely to choose a student loan with the option of IDR.

Overall, there is less evidence suggestive of adverse selection across a variety of variables than we

might have imagined ex ante. Future employment uncertainty, for example, does not seem to be associated

with higher ISA take-up. However, we do ﬁnd strong suggestive evidence of adverse selection based on

uncertainty around high future incomes. Given that income uncertainty is the focal point for educational

insurance decisions, this ﬁnding is an important contribution of our study.

5.4 Change in Preferences in Response to Changes in Price

Another dimension we investigate is whether students’ preferences change when offered alternative ISA

contracts with different income share percentages and term lengths, while keeping the offers actuarially

equivalent (comparable internal rate of return). As mentioned in section 3.2, students who chose a federal

student loan with an option of IDR against the original ISA contract (2.3% share for 10 years) in either

of the treatment arms were asked if they would switch to an ISA contract where they would pay smaller

installments over a longer period (2.2% share for 12 years). Those who still chose the federal student

loan option in this group were then asked to choose between the same federal student loan and another

actuarially equivalent ISA contract where they would pay larger installments over a shorter period (2.8%

share for seven years). Separately, students who chose the original ISA contract against the federal student

loan were then presented with both the alternative ISA options at the same time. Figure 4 and Figure 5

depict the ﬂow of the follow-up questions on alternative ISA options depending on the student’s initial

choice.

Out of the 2001 students who initially chose the federal student loan, 365 (18% of the sample) switch to

the longer term ISA contract. For those who still preferred the federal student loan over the original and the

longer-term ISAs, 285 out of 1636 (17%) would now switch to the actuarially equivalent ISA contract with a

shorter term and higher income share. On the other hand, among 775 students who originally chose the ISA,

162 (21%) would still choose the original ISA, 137 (18%) switch to the longer term ISA contract where they

We reiterate that there can be adverse selection on variables – such as current employment/income and most demographics –

that are "observable" to the lender but may not legally be used in decision-making. Consequently, such variables can be a source of

adverse selection even though they are "observable."

pay a lower share of their income over a longer time period, while 475 (61%) opt for the shorter term ISA

with higher income share. While a comparable share of students switch to the longer term ISA compared

with the original student loan choosers, a much higher share (61%) of original ISA choosers switch to a

shorter term ISA compared with the original student loan choosers (17%). Several factors may explain

this result. Our student sample is more representative of working adults (re)entering higher education mid-

career, as the average age of our sample is 38 and most students are employed while attending college.

Compared with traditional-age undergraduate students, they don’t perceive a signiﬁcant employment risk,

as we see in Table 2. Moreover, as these students are further along in their age-earnings proﬁle than

traditional-age students, they may reasonably expect to earn more than most graduates in subsequent years.

They may thus prefer a shorter-term ISA to a longer-term income-contingent repayment contract during

high-income years.

Table 4 considers the treatment effect on the switching behavior, separately for original student loan

choosers (Panels A and B) and original ISA choosers (Panels C and D). Panel A shows that treatment

increases the likelihood of switching from student loan to ISA with a longer term by 5.1 percentage points

(33% of control mean) in our preferred model. The effect is robust across all speciﬁcations and statistically

signiﬁcant. In Panel B, however, we do not observe a signiﬁcant treatment effect when we consider the

likelihood of switching from federal student loan to the shorter term ISA contract. Given that treated

students were exposed to our insurance framing, part of which emphasized the maximum repayment term

for a student loan with IDR (20 years) compared with the ISA (12 years), and picked the student loan

anyway, it makes sense that the treated students were much more likely to select the longer term hypothetical

ISA over the original hypothetical ISA compared with the control group. Our ﬁndings are consistent with

Mumford (2020).

On the other hand, we see contrary results for those who chose the ISA contract in the ﬁrst round. Panel

D shows that treatment increases the likelihood of switching to a shorter-term contract signiﬁcantly for

original ISA choosers – by 8.6 percentage points (15% of control mean) in our preferred model. We do

not ﬁnd a similar effect for switching to a longer-term ISA from the original ISA in Panel C. In contrast to

our results for the original student loan choosers, among original ISA choosers the treatment group is more

likely to switch to a shorter-term ISA after consistently being exposed to our insurance framing of lower

maximum repayment term for the hypothetical ISA (12 years) compared with the student loan with IDR (20

to 25 years) and potentially selecting the ISA in the ﬁrst round in part because of that factor. When given

the opportunity, in follow-up questions to pick an actuarially equivalent shorter term ISA, these students

are more likely to do so than the control group.

5.5 Proposed Student Loan Cancellation

Finally, in a separate follow-up question to all students in our survey, we tested whether respondents would

prefer the student loan with IDR if the student loan option incorporated a possibility of student loan can-

cellation. In the question, we informed respondents of a 20% chance that the federal government would

forgive the $10,000 they were borrowing via the federal student loan with the option of IDR. Students could

then respond by saying they were either: (a) much more likely to favor a student loan, (b) somewhat more

likely to favor a student loan, or (c) that the new information had no effect on their preference for a student

loan.

For ease of interpretation, we create an indicator variable that takes the value of 1 if a student expresses

any additional interest in student loans (the ﬁrst two options) and the value of zero if the student expresses

no change in preference. Table 5 shows results from this question, with model speciﬁcations comparable to

those in Table 2, apart from the dependent variable being increased preference for a student loan instead of

ISA take-up.

In Panel A, we consider all students who participated in the experiment and highlight whether a student

chose an ISA in the ﬁrst round as an explanatory variable. We ﬁnd that students who originally chose the

ISA are 7.7 percentage points less likely to express an increased preference for student loans when presented

with the chance of student loan forgiveness, compared with students who chose the student loan with IDR.

Essentially, students who chose the ISA in the ﬁrst round appear to reinforce their original reluctance to take

out a student loan. Taken with the other results in our study, this is consistent with the idea that students

choosing the ISA are committed to the ISA or especially loan averse, even if there is a chance that the

federal government will cancel any loans they may take out.

In Panel B, we consider those students who chose a loan in the ﬁrst round and to whom we subsequently

offered the smaller-share but longer-term hypothetical ISA in the second round. We show that students who

switched from the student loan with IDR to the longer-term, smaller-share ISA are 6.4 percentage points

more likely (in our preferred speciﬁcation) to prefer the student loan with IDR if it comes with a 20% chance

of loan forgiveness. These students are marginally attached to the ISA (since they changed their choice to

an ISA in the second round) and may be easily induced to switch between the two ﬁnancing choices given

relatively small changes to terms. This effect is comparable in size and direction for the group who switched

to the shorter-term ISA (Panel C); further evidence for our hypothesis of marginal attachment.

Finally, in Panel C, we only consider students who still chose the student loan in the previous two rounds

and to whom we offered a higher-share but shorter-term ISA in the third round. We ﬁnd that the likelihood

of student loan forgiveness induces a signiﬁcant increase in expressed preference for student loans by 8.9

percentage points, similar to the previous group. Again, this result is evidence that those students marginally

attached to an ISA (being switchers to the ISA in the third round) are willing to switch back to a loan if

given the prospect of loan forgiveness.

Our results show that offers of student loan forgiveness do little to change the student loan preference of

students who initially preferred the hypothetical ISA. However, for students who did switch their preference

due to our alternative offers of income share and repayment term, the potential for student loan forgiveness

has a large effect on respondents’ preferences changing back to a student loan with IDR. These and our

previous results provide suggestive evidence that offering a lower income share that is longer-term may

drive more adverse selection to the ISA, that student loan forgiveness could potentially draw some of that

adverse selection back into the federal loan program.

6 Conclusion

Income contingent education ﬁnancing is designed to lessen the burden of repaying federal student loans

with an option of IDR by tying the monthly payments to the borrower’s income. Students’ preferences, the

risk factors that may affect their take-up and hence the potential adverse selection in educational insurance

markets are of interest for both policymakers and researchers. We address these questions by randomizing

the messaging of hypothetical ﬁnancing contracts for 2,766 students at a large, non-proﬁt university to un-

derstand how the framing of ﬁnancial contracts, along with students’ characteristics and future expectations,

can inﬂuence decisions.

The results from this study have key implications for how student ﬁnancing contracts should be de-

signed, and they offer crucial insights into the behavioral responses of students. First, student choices

depend on the presentation of contract terms. The single biggest result from our intervention is the degree

to which information and framing impacts students’ preferences for how to ﬁnance their education. The fact

that emphasizing the insurance aspects of an ISA leads to a roughly 43% increase in hypothetical take-up

underscores how important messaging is for both private market participants like lenders and servicers, and

for the Department of Education.

Our study clearly shows that providing more details about the nuances of ﬁnancial contracts, and how

repayments would be affected by employment and ﬁnancial circumstances, enables students to make more

informed choices about their ﬁnancing decisions. The distinguishing feature of the hypothetical ISA is

its proportional payment structure with built-in insurance features. Speciﬁcally, the ISA contract requires

repayments proportional to income while providing contingencies in periods of low income; and it caps

the maximum amount to repayment. Students do perceive downside income protection risk as an insurance

while also creating an income smoothing effect. Our study shows that information about these features has

a strong effect on student preference.

Second, uncertainty about future earning potential contributes to adverse selection in the ISA insurance

market. The evidence we ﬁnd is consistent with a connection between the take-up of ISAs and uncertainty

about earning a higher income in the future. While current income and future unemployment status are not

related to take-up, income uncertainty matters. Other risks, with the exception of college completion risks,

do not seem to predict ISA take-up.

Third, contract choice may depend on familiarity with existing ﬁnancial contracts: A signiﬁcant barrier

that ISA providers (or any alternative form of ﬁnancing higher education) encounter is that the student loan

system is well-known, even if not well-liked, by many students. Indeed, our survey ﬁnds that students

currently ﬁnancing their education with federal student loans are signiﬁcantly less likely to prefer ISAs.

Given the pervasiveness of participation in the federal student loan market, this descriptive result illustrates

the uphill climb that any potential competitor to the traditional student loan model faces. This is especially

true given the recent expansion of favorable repayment terms via the SAVE IDR plan and the continuing

possibility of loan forgiveness being expanded to more federal student loan borrowers. Indeed, the Uni-

versity largely abandoned their potential ISA offering shortly after the Biden administration’s lowering of

mandatory monthly payments.

Finally, there are indications that students prefer different ﬁnancial contracts when they are presented

with different versions of monthly payments and contract lengths. Speciﬁcally, students who chose the

federal student loan with the option of IDR changed their preference to ISA when they were given a longer

contract with a lower income share percentage. In contrast, students who chose the original ISA contract

preferred a shorter term contract with larger monthly payments. Conventional wisdom suggests that bor-

rowers are averse to the long-time horizon for forgiveness offered by previous IDR plans. This proposition

is one of the justiﬁcations both for system-wide student loan forgiveness and also the recent redesign of

IDR. Our results cast some doubt on this assertion, as we ﬁnd that hypothetical borrowers who currently

borrow in the student loan system prefer lower payments stretched out over a longer time horizon. This

result is likely more useful for the design of future government repayment programs than for ISAs, as the

government is in a unique position to be able to forego repayment over a longer period of time.

Our study examines why educational insurance remains underdeveloped or absent despite the income

and employment risk to student loan borrowers. We randomized students into a framing that presented a

risk neutral choice between a student loan with IDR or an ISA, or a framing that emphasized the insurance

attributes of an income share agreement. We ﬁnd signiﬁcant evidence for adverse selection among students

who had the highest uncertainty about higher future incomes. We also ﬁnd that the marginal ISA participant

prefers a plan with a lower share and higher term; perhaps suggesting that those participants have higher

discount rates or present biased preferences. However, we do ﬁnd that this group will switch back when

offered the possibility of student loan forgiveness.

These results show that while there is signiﬁcant demand for educational insurance, adverse selection

among students with higher expected income uncertainty or present biased discount rates may make it

difﬁcult for traditional ﬁrms to provide such insurance in the private market. Furthermore, we ﬁnd that

offers of student loan cancellation may increase the amount of adverse selection present in federal loan

programs. These ﬁndings are important for our understanding of the nature of insurance in higher education

markets and how policymakers could help students make more efﬁcient decisions.

References

Abraham, K. G., Filiz-Ozbay, E., Ozbay, E. Y., & Turner, L. J. (2020). Framing effects, earnings expec-

tations, and the design of student loan repayment schemes. Journal of Public Economics, 183,

104067.

Akerlof, G. A. (1970). The market for "lemons": Quality uncertainty and the market mechanism. The Quar-

terly Journal of Economics, 84(3), 488–500.

Akers, B. (2021). Making college pay: An economist explains how to make a smart bet on higher education.

Currency.

Arcidiacono, P., Hotz, V. J., & Kang, S. (2012). Modeling college major choices using elicited measures of

expectations and counterfactuals. Journal of Econometrics, 166(1), 3–16.

Athey, S., & Imbens, G. (2016). Recursive partitioning for heterogeneous causal effects. Proceedings of the

National Academy of Sciences, 113(27), 7353–7360.

Baker, R., Bettinger, E., Jacob, B., & Marinescu, I. (2018). The effect of labor market information on

community college studentsâ major choice. Economics of Education Review, 65, 18–30.

Balakrishnan, S., Bettinger, E., Kofoed, M. S., Ritter, D., Webber, D., Aksu, E., & Hartley, J. S. (2023). Pre-

Analysis Plan for: An Evaluation of Income Share Agreements: The Effect of Education Insurance

Framing and the Nature of Adverse Selection. https://doi.org/10.1257 /rct.8936-1

Balakrishnan, S., & Cynamon, B. Z. (2018). Borrowing arrangements and returns to college education.

Available at SSRN 3263639.

Bettinger, E. P., Long, B. T., Oreopoulos, P., & Sanbonmatsu, L. (2012). The role of application assistance

and information in college decisions: Results from the h&r block fafsa experiment. The Quarterly

Journal of Economics, 127(3), 1205–1242.

Black, S. E., Denning, J. T., Dettling, L. J., Goodman, S., & Turner, L. J. (Forthcoming). Taking it to the

limit: Effects of increased student loan availability on attainment, earnings, and ﬁnancial well-

being. American Economic Review.

Bodie, Z. (1994). On the risk of stocks in the long run. Capital Markets eJournal. https : / / api

.semanticscholar.org/CorpusID:153602556

Bowen, W., Chingos, M., & McPherson, M. (2009). Crossing the ﬁnish line: Completing college at amer-

ica’s public universities. Princeton University Press.

Breiman, L. (2001). Random forests. Machine learning, 45, 5–32.

Breiman, L., Friedman, J., Stone, C. J., & Olshen, R. A. (1984). Classiﬁcation and regression trees. CRC

press.

Broady, K., & Hershbein, B. (2020). Major decisions: What graduates earn over their lifetimes. Brookings

Institution Papers. https://www .brookings .edu /articles /major -decisions

-what-graduates-earn-over-their-lifetimes/

Brodeur, A., Cook, N., Hartley, J., & Heyes, A. (2022). Do pre-registration and pre-analysis plans reduce

p-hacking and publication bias? Available at SSRN.

Chakrabarti, R., Gorton, N., & Lovenheim, M. F. (2020). State investment in higher education: Effects on

human capital formation, student debt, and long-term ﬁnancial outcomes of students (Working

Paper No. 27885). National Bureau of Economic Research.

Chapman, B., & Dearden, L. (2017). Conceptual and empirical issues for alternative student loan designs:

The signiﬁcance of loan repayment burdens for the united states. The ANNALS of the American

Academy of Political and Social Science, 671(1), 249–268.

Chapman, B., & Lounkaew, K. (2015). An analysis of stafford loan repayment burdens. Economics of

Education Review, 45, 89–102.

Chetty, R., & Szeidl, A. (2007). Consumption commitments and risk preferences. The Quarterly Journal of

Economics, 122(2), 831–877.

Conlon, J. J. (2021). Major malfunction: A ﬁeld expirement correcting undergradaute’s beliefs about salaries.

Journal of Human Resources, 56(3), 922–939.

Cox, J. C., Kreisman, D., & Dynarski, S. (2020). Designed to fail: Effects of the default option and infor-

mation complexity on student loan repayment. Journal of Public Economics, 192, 104298.

Denning, J. T., & Jones, T. R. (2021). Maxed out? the effect of larger student loan limits on borrowing and

education outcomes. Journal of Human Resources, 56(4), 1113–1140.

Eide, E. R., Hilmer, M. J., & Showalter, M. H. (2016). Is it where you go or what you study? the relative

inﬂuence of college selectivity and college major on earnings. Contemporary Economic Policy,

34(1), 37–46.

Einav, L., Finklestein, A., & Fisman, R. (2023). Risky business: Why insurance markets fail and what to do

about it. Yale University Press.

Friedman, M. (1955). The role of government in education. Economics and the public interest, 2(2), 85–

107.

Goyal, A., & Welch, I. (2007). A Comprehensive Look at The Empirical Performance of Equity Premium

Prediction. The Review of Financial Studies, 21(4), 1455–1508.

Guiso, L., & Paiella, M. (2008). Risk Aversion, Wealth, and Background Risk. Journal of the European

Economic Association, 6(6), 1109–1150.

Hendricks, L., & Leukhina, O. (2018). The return to college: Selection and dropout risk. International

Economic Review, 59(3), 1077–1102.

Herbst, D. (2023). The impact of income-driven repayment on student borrower outcomes. American Eco-

nomic Journal: Applied Economics, 15(1), 1–25.

Herbst, D., & Hendren, N. (2021). Opportunity unraveled: Private information and the missing markets for

ﬁnancing human capital. NBER Working Paper Series, 29214.

Herbst, D., Palacios, M., & Yannelis, C. (2022). Equity and incentives in household ﬁnance. SSRN, (4398141),

1–47.

Karamcheva, N., Perry, J., & Yannelis, C. (2020). Income-driven repayment plans for student loans: Bud-

getary costs and policy options. Congressional Budget Ofﬁce Report.

Kimball, M. S., Sahm, C. R., & Shapiro, M. D. (2008). Imputing risk tolerance from survey responses.

Journal of the American Statistical Association, 103(483), 1028–1038.

Kirkeboen, L. J., Leuven, E., & Mogstad, M. (2016). Field of study, earnings, and self-selection. The Quar-

terly Journal of Economics, 131(3), 1057–1111.

Kofoed, M. S. (2017). To apply or not to apply: Fafsa completion and ﬁnancial aid gaps. Research in Higher

Education, 58, 1–39.

Levine, P. B., & Ritter, D. (2023). The racial wealth gap, ﬁnancial aid, and college access. Journal of Policy

Analysis and Management, forthcoming.

Lovenheim, M. F., & Smith, J. (2022). Returns to different postsecondary investments: Institution type,

academic programs, and credentials. Hanbook of Economics of Education, 6, 187–318.

Markle, G. (2015). Factors inﬂuencing persistence among nontraditional university students. Adult Educa-

tion Quarterly, 65(3), 267–285.

Marx, B. M., & Turner, L. J. (2019). Student loan nudges: Experimental evidence on borrowing and educa-

tional attainment. American Economic Journal: Economic Policy, 11(2), 108–141.

Merton, R. C. (1971). Optimum consumption and portfolio rules in a continuous-time model. Journal of

Economic Theory, 3(4), 373–413.

Mezza, A., Ringo, D., Sherlund, S., & Sommer, K. (2020). Student loans and homeownership. Journal of

Labor Economics, 38(1), 215–260.

Mezza, A., & Sommer, K. (2015). A trillion dollar question: What predicts student loan delinquencies?

Journal of Student Financial Aid, 46(3).

Mueller, H., & Yannelis, C. (2022). Increasing enrollment in income-driven student loan repayment plans:

Evidence from the navient ﬁeld experiment. The Journal of Finance, 77(1), 367–402.

Mumford, K. J. (2020). Student selection into an income share agreement. Proceedings. Annual Conference

on Taxation and Minutes of the Annual Meeting of the National Tax Association, 113, 1–42.

Ritter, D., & Webber, D. A. (2019). Modern income-share agreements in postsecondary education: Features,

theory, applications. FRB of Philadelphia Payment Cards Center Discussion Paper, (19-6).

Rothstein, J. (2023). The lost generation? labor market outcomes for post great recession entrants. Journal

of Human Resources, 58(5), 1452–1479.

Stange, K. M. (2012). An empirical investigation of the option value of college enrollment. American

Economic Journal: Applied Economics, 4(1), 49–84.

Thakor, A. V. (2020). Fintech and banking: What do we know? Journal of Financial Intermediation, 41,

100833.

Webber, D. A. (2016). Are college costs worth it? how ability, major, and debt affect the returns to schooling.

Economics of Education Review, 53, 296–310.

Webber, D. A. (2017). State divestment and tuition at public institutions. Economics of Education Review,

60, 1–4.

Webber, D. A. (2022). Decomposing changes in higher education return on investment over time. FEDS

Notes. https : / / www .federalreserve .gov / econres / notes / feds -notes /

decomposing-changes-in-higher-education-return-on-investment-over

-time-20220713.html

Zaber, M. A., Steiner, E. D., Gebremariam, H., Spears, A., Tariq, Z. H., & Carman, K. G. (2023). Income

share agreements: Market structure, communication, and equity implications of a student loan

alternative. RAND Research Report, 1–35.

Zweifel, P., & Manning, W. G. (2000). Moral hazard and consumer incentives in health care. Handbook of

Health Economics, 409–459.

7 Tables

Table 1: Summary Statistics and Balance

(1) (2) (3) (4)

N Overall mean Control mean Treatment

(overall) (std. dev) (std. dev) dummy

Current income 2776 70235.951 69244.060 1985.211

(37762.513) (37023.692) (1433.248)

Currently employed 2315 0.834 0.839 -0.011

(0.372) (0.367) (0.014)

Very uncertain of high future income 2599 0.040 0.040 -0.002

(0.195) (0.197) (0.008)

Very uncertain of future employment 2739 0.128 0.125 0.006

(0.334) (0.331) (0.013)

Risk aversion (CCRA) 2776 3.854 3.848 0.012

(1.095) (1.108) (0.042)

Financially stable 2151 0.775 0.772 0.006

(0.418) (0.420) (0.016)

Current ﬁnancing choice

Cash (self and from family and friends) 1416 0.510 0.500 0.021

(0.500) (0.500) (0.019)

Institutional and federal/state grants 916 0.330 0.328 0.003

(0.470) (0.470) (0.018)

Government and private student loans 1205 0.434 0.437 -0.006

(0.496) (0.496) (0.019)

Other 777 0.280 0.275 0.010

(0.449) (0.447) (0.017)

Barriers to graduation

Housing insecurity 210 0.076 0.080 -0.009

(0.264) (0.271) (0.010)

Difﬁculty doing college-level work 266 0.096 0.091 0.010

(0.294) (0.287) (0.011)

Work responsibilities 2193 0.790 0.790 -0.001

(0.407) (0.407) (0.015)

Family constraints 1595 0.575 0.581 -0.013

(0.494) (0.494) (0.019)

Disability or health concerns (physical or mental) 576 0.207 0.201 0.013

(0.406) (0.401) (0.015)

Financial constraints 1124 0.405 0.400 0.011

(0.491) (0.490) (0.019)

Internet connectivity issues 257 0.093 0.088 0.009

(0.290) (0.283) (0.011)

Issues with academic support services 260 0.094 0.102 -0.016

(0.291) (0.302) (0.011)

Difﬁculty being a transfer student 59 0.021 0.024 -0.005

(0.144) (0.152) (0.005)

Other 123 0.044 0.045 -0.001

(0.206) (0.207) (0.008)

Race/Ethnicity

African-American 196 0.071 0.069 0.003

(0.256) (0.254) (0.010)

Hispanic 417 0.150 0.143 0.014

(0.357) (0.350) (0.014)

Asian 200 0.072 0.068 0.007

(0.259) (0.253) (0.010)

Other race/ethnicity 1666 0.600 0.614 -0.028

(0.490) (0.487) (0.019)

No ethnicity reported 297 0.107 0.105 0.004

(0.309) (0.307) (0.012)

Highest degree earned

High school degree or less 409 0.147 0.138 0.020

(0.355) (0.345) (0.013)

Some college degree 2313 0.833 0.843 -0.020

(0.373) (0.364) (0.014)

Graduate degree 46 0.017 0.017 -0.001

(0.128) (0.130) (0.005)

Age 2776 37.778 37.917 -0.278

(9.188) (9.237) (0.349)

Female 1970 0.710 0.715 -0.011

(0.454) (0.452) (0.017)

Married 1399 0.504 0.511 -0.014

(0.500) (0.500) (0.019)

Household size 2776 2.932 2.952 -0.042

(1.783) (1.717) (0.068)

Notes. For every categorical variable and binary variable, N shows the number of students who

fall into that category. For every other variable, N shows the number of students for whom the

data is available. For details on derived variables including future employment uncertainty, future

uncertainty of high income, and ﬁnancial stability, see Appendix A.

The minimum age is 19 whereas the maximum is 71. The minimum and maximum values for the

household size are 1 and 15, respectively. A continuous current income variable is created from a

12-category categorical variable (less than $20,000, $20,000 - $29,999, ..., $110,000 - $119,999,

$120,000 or more) where the middle point of each category is assigned to students who chose that

category as their current income value. The minimum current income is $10,000 whereas the maxi-

mum is $150,000.

Table 2: Determinants of ISA Take-up

(1) (2) (3) (4) (5) (6) (7)

Treatment 0.099*** 0.099*** 0.104*** 0.105*** 0.103*** 0.103*** 0.102***

(0.017) (0.017) (0.018) (0.018) (0.018) (0.018) (0.018)

Current income -0.007 0.000 0.000 0.001 -0.001 -0.003

(0.014) (0.015) (0.015) (0.015) (0.015) (0.016)

Currently employed 0.020 0.023 0.022 0.027 0.025 0.033

(0.031) (0.032) (0.032) (0.032) (0.032) (0.033)

Very uncertain of high future income† 0.091* 0.091* 0.083* 0.078 0.075

(0.049) (0.049) (0.049) (0.049) (0.049)

Very uncertain of future employment† 0.015 0.015 0.012 0.010 0.014

(0.028) (0.028) (0.028) (0.028) (0.028)

Risk aversion (CCRA) -0.003 -0.001 -0.001 -0.000

(0.008) (0.008) (0.008) (0.008)

Current ﬁnancing choice

Cash (self and from family and friends) 0.018 0.016 0.016

(0.022) (0.022) (0.022)

Institutional and federal/state grants 0.030 0.029 0.033

(0.020) (0.020) (0.021)

Government and private student loans -0.069*** -0.072*** -0.074***

(0.022) (0.022) (0.022)

Other 0.011 0.009 0.008

(0.021) (0.021) (0.021)

Barriers to graduation

Housing security and other ﬁnancial risks -0.016 -0.009

(0.016) (0.016)

College-related risks 0.031* 0.030*

(0.017) (0.016)

Family and health risks 0.013 0.013

(0.016) (0.016)

Current work responsibilities and other risks 0.039 0.036

(0.024) (0.025)

Race/Ethnicity

African-American -0.095**

(0.043)

Hispanic -0.001

(0.038)

Asian -0.012

(0.046)

Other race/ethnicity -0.041

(0.032)

Married 0.034*

(0.020)

Control Mean 0.230 0.230 0.230 0.230 0.230 0.230 0.230

Control Std 0.421 0.421 0.421 0.421 0.421 0.421 0.421

Total number of observations 2776 2776 2599 2599 2599 2599 2594

Adjusted R-Squared 0.012 0.011 0.013 0.013 0.019 0.020 0.022

Notes. Variable deﬁnitions follow the Pre-Analysis Plan; see Appendix A for details on our derived variables. The re-

gression model in all columns follows from Equation 3 in the Pre-Analysis Plan. The dependent variable is the ISA

take-up, which takes the value of 1 if the student chose the hypothetical ISA over a federal student loan with IDR, 0 oth-

erwise. For each independent variable, we report the coefﬁcient and the standard errors in parentheses. The coefﬁcient of

interest, the treatment effect on the ISA take-up, is reported in the ﬁrst row. Except for Model 1, all models include ﬁxed

effects for the student’s major. The preferred regression model (Column 7) also controls for (not reported or statistically

signiﬁcant) age, gender, household size, highest degree earned, career expectations, and ﬁnancial stability. We use log-

transformation for current income.

† Uncertainty measures for future employment and income are deﬁned as indicator variables taking the value of 1 if the

student is very uncertain of their future employment or of earning a high income, and 0 otherwise. See Appendix A for

details on these and other derived variables.

For risk aversion, we use the Coefﬁcient of Relative Risk Aversion (CRRA), a measure for risk preferences following a

similar method to Kimball et al. (2008). Barriers to graduation - originally 10 variables - are reduced to 4 main categories

by maximum likelihood factor analysis. We report the coefﬁcients for the set of standardized factor estimates. College-

related risks include difﬁculty doing college work, being a transfer student, and lack of support from advisor/academic

support services.

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01

Table 3: Heterogeneous Treatment Effects on Hypothetical ISA Take-Up

(1) (2) (3) (4) (5) (6) (7) (8) (9)

White

Treated

Black or

African-American

Treated

Hispanic

Treated

No response for

race/ethnicity

Treated

Female

Treated

Household Size

Treated

Below Median Age

Treated

Married

Treated

Risk Aversion

Treated

ISA take-up -0.016 0.019 0.058 -0.073 0.002 -0.002 -0.020 0.021 0.019

(0.056) (0.080) (0.069) (0.075) (0.039) (0.010) (0.035) (0.035) (0.016)

Current Income Y Y Y Y Y Y Y Y Y

Current Employment Y Y Y Y Y Y Y Y Y

Financial Stability Y Y Y Y Y Y Y Y Y

Career Expectations Y Y Y Y Y Y Y Y Y

Future Income Uncertainty Y Y Y Y Y Y Y Y Y

Future Employment Uncertainty Y Y Y Y Y Y Y Y Y

Risk Aversion Y Y Y Y Y Y Y Y Y

Current Financing Choice Y Y Y Y Y Y Y Y Y

Barriers to Graduation Y Y Y Y Y Y Y Y Y

Race/Ethnicity Y Y Y Y Y Y Y Y Y

Marital Status Y Y Y Y Y Y Y Y Y

Sample composition

55.37%

White

7.06%

African-American

15.02%

Hispanic

10.70%

No response

70.97%

Female —

50%

below-median age

50.40%

married —

Notes. We only report the estimates based on our preferred model, Column 7 in Table 2. Full estimates from other models can be found in Appendix B.

Variable deﬁnitions follow the Pre-Analysis Plan; see Appendix A for details on our derived variables. The regression model follows from Equation 3 in the Pre-Analysis Plan. We

report the heterogeneous treatment effect on the ISA take-up for each dimension reported in the column and the standard errors in parentheses. The regression model also controls

for (not reported) age, gender, household size, and highest degree earned, as well as ﬁxed effects for student’s major. We use log-transformation for current income.

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01

Table 4: Responses to Actuarially Equivalent ISA Offers with Different Terms/Shares

(1) (2) (3) (4) (5) (6) (7)

Panel A: Switching from Student Loan to ISA with Longer Term (2.2% share for 12 years)

Treatment 0.057*** 0.060*** 0.051*** 0.052*** 0.051*** 0.051*** 0.051***

(0.017) (0.017) (0.018) (0.018) (0.018) (0.018) (0.018)

Control Mean 0.156 0.156 0.156 0.156 0.156 0.156 0.156

N 2001 2001 1878 1878 1878 1878 1874

Panel B: Switching from Student Loan to ISA with Shorter Term (2.8% share for 7 years)

Treatment -0.017 -0.019 -0.015 -0.014 -0.016 -0.016 -0.013

(0.019) (0.019) (0.020) (0.020) (0.020) (0.020) (0.020)

Control Mean 0.182 0.182 0.182 0.182 0.182 0.182 0.182

N 1636 1636 1543 1543 1543 1543 1539

Panel C: Switching from Original ISA to ISA with Longer Term (2.2% share for 12 years)

Treatment -0.035 -0.032 -0.022 -0.025 -0.022 -0.022 -0.032

(0.028) (0.028) (0.029) (0.029) (0.029) (0.029) (0.029)

Control Mean 0.197 0.197 0.197 0.197 0.197 0.197 0.197

N 775 775 721 721 721 721 720

Panel D: Switching from Original ISA to ISA with Shorter Term (2.8% share for 7 years)

Treatment 0.088** 0.085** 0.083** 0.085** 0.083** 0.081** 0.085**

(0.036) (0.036) (0.037) (0.037) (0.037) (0.037) (0.037)

Control Mean 0.561 0.561 0.561 0.561 0.561 0.561 0.561

N 775 775 721 721 721 721 720

Current Income N Y Y Y Y Y Y

Current Employment N Y Y Y Y Y Y

Financial Stability N Y Y Y Y Y Y

Career Expectations N N Y Y Y Y Y

Future Income Uncertainty N N Y Y Y Y Y

Future Employment Uncertainty N N Y Y Y Y Y

Risk Aversion N N N Y Y Y Y

Current Financing Choice N N N N Y Y Y

Barriers to Graduation N N N N N Y Y

Race/Ethnicity N N N N N N Y

Marital Status N N N N N N Y

Notes. Students who chose a federal student loan with IDR against the original ISA contract in the ﬁrst round (in either of the arms) were presented

with an actuarially equivalent hypothetical ISA contract where they would pay smaller installments over a longer period (2.2% share for 12 years) in

the second round (Panel A). Those who still chose the federal student loan w/ IDR over the longer term ISA contract in the second round were then

given the option to choose between the same student loan w/ IDR and an actuarially equivalent hypothetical ISA contract where they would pay larger

installments over a shorter period (2.8% share for 7 years) (Panel B). Separately, students who chose the original ISA contract (2.3% share for 10

years) against a federal student loan with IDR (in either of the arms) were presented both alternative, actuarially equivalent ISA contracts described

above at the same time. Our results for switching preference to the longer term ISA are presented in Panel C, and results for switching preference to

the shorter term ISA in (Panel D). For further details on how we asked these follow-up questions to the students, please see Figure 4 and Figure 5.

Variable deﬁnitions follow the Pre-Analysis Plan. The regression model in all columns is given by Equation 1 in the Pre-Analysis Plan, with the

exception of the dependent variable. The dependent variable takes the value of 1 if the student switched to the alternative contract offered in the

particular round, 0 otherwise. We only report the coefﬁcient of interest, the treatment effect on the likelihood of switching to one of the alternative

ISAs. Except for Model 1, all models include ﬁxed effects for the student’s major. The preferred regression model (Model 7) also controls for (not

reported) age, gender, household size, and highest degree earned. We use log-transformation for current income. N is the sample size.

The sample is the 2,001 students who chose the student loan w/ IDR over the original ISA offer in round one for Panel A; 1,636 students who chose

the federal student loan w/ IDR over in the ﬁrst round and haven’t switched to the longer term ISA in round two for Panel B; and 775 students who

chose the original ISA over the federal student loan w/ IDR for Panels C and D.

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01

Table 5: Preference for Student Loan w/ IDR if 20% Chance of Loan Forgiveness

(1) (2) (3) (4) (5) (6) (7)

Panel A: Sample = All Respondents (First Round)

Chose ISA = 1 -0.084*** -0.083*** -0.081*** -0.081*** -0.078*** -0.078*** -0.078***

(0.021) (0.021) (0.022) (0.022) (0.022) (0.022) (0.022)

N 2776 2776 2599 2599 2599 2599 2594

Panel B: Sample = Chose Student Loan w/ IDR in First Round (Second Round)

Switched to Longer-Term ISA = 1 0.069** 0.062** 0.069** 0.068** 0.067** 0.064** 0.063**

(0.028) (0.028) (0.029) (0.029) (0.029) (0.029) (0.029)

N 2001 2001 1878 1878 1878 1878 1874

Panel C: Sample = Chose Student Loan w/ IDR in Second Round (Third Round)

Switched to Shorter-Term ISA = 1 0.096*** 0.095*** 0.079** 0.078** 0.082** 0.085*** 0.088***

(0.031) (0.032) (0.032) (0.032) (0.033) (0.033) (0.033)

N 1636 1636 1543 1543 1543 1543 1539

Current Income N Y Y Y Y Y Y

Current Employment N Y Y Y Y Y Y

Financial Stability N Y Y Y Y Y Y

Career Expectations N N Y Y Y Y Y

Future Income Uncertainty N N Y Y Y Y Y

Future Employment Uncertainty N N Y Y Y Y Y

Risk Aversion N N N Y Y Y Y

Current Financing Choice N N N N Y Y Y

Barriers to Graduation N N N N N Y Y

Race/Ethnicity N N N N N N Y

Marital Status N N N N N N Y

Notes. All students, regardless of their choice between a federal student loan w/ IDR and the hypothetical ISA in the core experiment (round 1)

were asked whether a 20% chance that the $10,000 they borrow from the federal government via the student loan may be forgiven in the future

would change their preference for the student loan w/ IDR.

Panel A shows how choosing the original ISA in the core experiment is associated with the likelihood of preferring the hypothetical ISA for the

full sample (round 1). Panel B repeats the same exercise for the students who initially chose the student loan but were then offered an actuarially-

equivalent hypothetical ISA with a lower income share and a longer term (round 2). Finally, Panel C repeats the same exercise for the students

who chose the student loan in round 2 and were then offered an actuarially-equivalent hypothetical ISA with a higher income share and shorter

term in round 3.

Variable deﬁnitions follow the Pre-Analysis Plan; see Appendix A for details on our derived variables. The regression model in all columns

follows from Equation 3 in the Pre-Analysis Plan. The regression models control for all variables shown in Table 2 in the respective models.

The dependent variable is a discrete variable taking the value of 1 if a student reported being more or much more likely to select the federal

student loan if loan forgiveness is available and 0 if they reported no change in their preference. We only report the coefﬁcient of interest, the

association between choosing an ISA and the change in preference for a federal student loan with IDR when loan forgiveness is possible. We use

log-transformation for current income. N is the sample size.

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01

8 Figures

Notes: The survey presented hypothetical ﬁnancing options to students through two framings. The

presented options differed in terms of level of detail provided for each of the choices. Half of the

students were randomized into a "risk neutral" framing of the student loan with option of IDR and

the ISA that explained the differences in monthly payments, general structure of the loan, the baseline

payment terms, and the source of funding. The terms of the student loan with IDR and the ISA were

set to be actuarially equivalent. We exposed the other half of students – our treatment group – to the

same descriptions of the student loan with option of IDR and the ISA, but with an additional emphasis

on the insurance features (nature of the income contingency and maximum repayment term) of the two

ﬁnancing options. Please see Appendix D for the details of the survey.

Figure 1: Core Experiment

Notes: The ﬁgure investigates the share of students choosing the hypothetical ISA given their reported employment uncertainty and their treatment

status. For example, among respondents who are very certain that they’d be employed in the future, 22% of control students in this category chose the

hypothetical ISA whereas the share is 32% for the treated students. Treatment increases the share of students choosing the ISA in every category of

employment uncertainty. Vertical lines show the conﬁdence intervals. The shares of students choosing the ISA are signiﬁcantly higher in the treatment

group than the control group for those who are very certain or who are are moderately uncertain of their future employment prospects. One can also

conclude that the share of students choosing the ISA for treatment and control groups separately across categories are not statistically different from each

other at the 5% signiﬁcance level. See Appendix A for details on our derivation of the employment uncertainty categories.

Figure 2: Hypothetical ISA Take-up by Treatment and Employment Uncertainty

Notes: The ﬁgure investigates the share of students choosing the hypothetical ISA given their reported income uncertainty and their treatment status.

For example, among the people who are very certain that they’d earn a high income in the future, 21% of control students in this category chose the

hypothetical ISA whereas the share is 32% for the treated students. Treatment increases the share of students choosing the hypothetical ISA in every

category of income uncertainty. The gap is the smallest for those with highest uncertainty over high future income. Vertical lines show the conﬁdence

intervals. For students who are very and moderately certain about their high income prospects, treatment signiﬁcantly increases the share of students

taking up the hypothetical ISA. See Appendix A for details on our derivation of the income uncertainty categories.

Figure 3: Hypothetical ISA Take-up by Treatment and (High) Income Uncertainty

Notes: Students who chose a federal student loan with an option of IDR against the original ISA contract (2.3% share for 10 years) in either of

the core experiment arms in the ﬁrst round were asked if they would switch to an ISA contract where they would pay smaller installments over

a longer period (2.2% share for 12 years). Those who still chose the federal student loan option in the second round were then asked to choose

between the same federal student loan and another actuarially equivalent ISA contract where they would pay larger installments over a shorter

period (2.8% share for 7 years). Please see Appendix D for the details of the survey.

Figure 4: Testing for Preferences over Actuarially Equivalent ISAs with Alternative Terms -

Federal Student Loan w/ IDR Choosers in Core Experiment

Notes: Students who chose the original ISA contract (2.3% share for 10 years) against the federal student loan with an option of IDR in either

of the treatment arms in the ﬁrst round were presented with two other ISA contracts at the same time in the second round. The students were

asked to choose between the two actuarially equivalent ISA contracts where they would pay smaller installments over a longer period (ISA

option #1: 2.2% share for 12 years) or larger installments over a shorter period (ISA option #2: 2.8% share for 7 years). Please see Appendix

D for the details of the survey.

Figure 5: Testing for Preferences over Actuarially Equivalent ISAs with Alternative Terms -

ISA Choosers in Core Experiment

Appendix A

Derived Variable Deﬁnitions

Future Employment Uncertainty

For certainty of future employment prospects, we use the following convention to categorize respondents:

• Category 1 (Very Certain of Being Employed): People who answered "Likely" to be fully employed and "Unlikely" to be unemployed

• Category 2 (Moderately Certain of Being Employed): People who answered “Unlikely” to be unemployed and “Likely” or “Neither likely nor

unlikely” to be employed part-time

• Category 3 (Moderately Uncertain of Being Employed): People who answered “Neither likely nor unlikely” to be unemployed

• Category 4 (Very Uncertain of Being Employed): People who answered “Likely” to be unemployed

Our dummy of "Very uncertain of being employed" takes the value of 1 if the student is in the last category, and 0 otherwise.

Future Income Uncertainty

In a similar manner, the degree of certainty over future income/earnings potential is categorized as follows:

• Category 1 (Very Certain of Earning Higher Income): People who answered “Likely” to earn more than US$75,000 and “Unlikely” to earn less than

US$35,000

• Category 2 (Moderately Certain of Earning Higher Income): People who answered "Unlikely" to earn less than US$35,000 and “Likely” or “Neither

likely nor unlikely” to earn between US$55,001 and US$75,000

• Category 3 (Moderately Uncertain of Earning Higher Income): People who answered "Likely" to earn between US$35,001 and US$55,000

• Category 4 (Very Uncertain of Earning Higher Income): People who answered “Likely” or “Neither likely nor unlikely” to earn less than US$35,000

Our dummy of "Very uncertain of high income" takes the value of 1 if the student is in the last category, and 0 otherwise.

Financial Stability

The ﬁnancial stability is assessed using the following question, following the Federal Reserve Board’s Survey of Household Economics and Decisionmaking:

Suppose that your household has an emergency expense that costs $400. Based on your household’s current ﬁnancial situation, how would your

household pay for this expense? If your household would use more than one method to cover this expense, please select all that apply.

• 1. Use a credit card and pay it off in full at the next statement

• 2. Borrow money from a friend or family member

• 3. Use a payday loan, informal loan, deposit advance, or overdraft

• 4. Use money currently in a checking/savings account or use cash

• 5. Use money from a bank loan or line of credit

• 6. Use a credit card and pay it off over time

• 7. Sell something

• 8. Wouldn’t have access to any of the above options

Students are "ﬁnancially stable" if they chose option 1 or option 4 (in other words, can cover the $400 expense with cash or cash equivalent) and "not

ﬁnancially stable" if they did not.

It is, of course, possible to create more categories given the high level of granularity of the data. The categories here are decided in a manner that the

certainty monotonically goes from higher to lower and that there are enough observations to make a meaningful inference in each category.

Appendix B

Heterogeneous Treatment Effects - Full Output

This appendix includes supplementary analysis, with Tables A1-A6 presenting the full results of heterogeneous treatment effects for each dimension that is

summarized in Table 3. We report all coefﬁcients, in addition to the interaction term. These tables conﬁrm that the insurance framing is the primary driving

force behind the ISA take-up.

Table A1: Treatment Effect on Hypothetical ISA Take-Up - by Race

(1) (2) (3) (4) (5) (6) (7)

Treatment 0.092* 0.094** 0.106** 0.107** 0.110** 0.112** 0.109**

(0.048) (0.048) (0.051) (0.051) (0.051) (0.051) (0.051)

White 0.026 0.022 0.015 0.015 0.023 0.028 0.080*

(0.036) (0.036) (0.037) (0.037) (0.037) (0.037) (0.045)

Black or African American -0.020 -0.019 -0.065 -0.064 -0.065 -0.061 0.002

(0.051) (0.052) (0.051) (0.051) (0.051) (0.051) (0.057)

Hispanic 0.014 0.011 0.011 0.011 0.015 0.018 0.076

(0.044) (0.044) (0.046) (0.046) (0.045) (0.045) (0.051)

No response 0.066 0.066 0.063 0.063 0.062 0.069 0.144**

(0.049) (0.049) (0.051) (0.051) (0.051) (0.051) (0.058)

Treatment × White 0.001 -0.002 -0.007 -0.008 -0.015 -0.018 -0.016

(0.053) (0.053) (0.056) (0.056) (0.056) (0.056) (0.056)

Treatment × Black or African American 0.040 0.037 0.026 0.025 0.026 0.023 0.019

(0.078) (0.078) (0.079) (0.079) (0.080) (0.080) (0.080)

Treatment × Hispanic 0.049 0.048 0.059 0.059 0.055 0.053 0.059

(0.065) (0.065) (0.069) (0.069) (0.069) (0.069) (0.069)

Treatment × No response -0.035 -0.041 -0.073 -0.074 -0.075 -0.077 -0.073

(0.072) (0.072) (0.074) (0.075) (0.075) (0.075) (0.075)

N 2776 2776 2599 2599 2599 2599 2594

Variable deﬁnitions follow the Pre-Analysis Plan; see Appendix A for details on our derived variables. The regression model in all columns follows from

Equation 3 in the Pre-Analysis Plan. The regression models control for all variables shown in Table 2 in the respective models. We use log-transformation

for current income. N is the sample size. Out of 2,776 respondents, 1,537 are White, 196 are Black or African American, 417 are Hispanic, 329 identify

with other race/ethnicity categories, whereas 297 declined to answer. Other race/ethnicity categories include Asian (200), American Indian or Alaska Native

(29), Native Hawaiian or Other Paciﬁc Islander (22) and multi-racial (78).

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01

Table A2: Treatment Effect on Hypothetical ISA Take-Up - by Gender

(1) (2) (3) (4) (5) (6) (7)

Treatment 0.102*** 0.100*** 0.098*** 0.098*** 0.100*** 0.102*** 0.100***

(0.032) (0.032) (0.032) (0.032) (0.032) (0.032) (0.033)

Female -0.011 -0.016 -0.013 -0.012 -0.001 -0.002 -0.004

(0.025) (0.028) (0.029) (0.029) (0.029) (0.029) (0.030)

Treatment × Female -0.004 -0.001 0.009 0.009 0.005 0.001 0.002

(0.038) (0.038) (0.039) (0.039) (0.038) (0.039) (0.039)

N 2776 2776 2599 2599 2599 2599 2594

Variable deﬁnitions follow the Pre-Analysis Plan; see Appendix A for details on our derived variables. The regression model in all columns follows from

Equation 3 in the Pre-Analysis Plan. The regression models control for all variables shown in Table 2 in the respective models. We use log-transformation

for current income. N is the sample size.

The coefﬁcient for the interaction term in every model captures the heterogeneous treatment effects.

Out of 2,776 respondents, 1,970 are female.

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01

Table A3: Treatment Effect on Hypothetical ISA Take-Up - by Household Size

(1) (2) (3) (4) (5) (6) (7)

Treatment 0.100*** 0.100*** 0.109*** 0.109*** 0.111*** 0.111*** 0.107***

(0.033) (0.033) (0.034) (0.034) (0.034) (0.034) (0.034)

Household size 0.004 0.005 0.004 0.004 0.005 0.003 0.001

(0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007)

Treatment × Household size -0.000 -0.000 -0.002 -0.002 -0.002 -0.003 -0.002

(0.010) (0.010) (0.010) (0.010) (0.010) (0.010) (0.010)

N 2776 2776 2599 2599 2599 2599 2594

Variable deﬁnitions follow the Pre-Analysis Plan; see Appendix A for details on our derived variables. The regression model in all columns follows from

Equation 3 in the Pre-Analysis Plan. The regression models control for all variables shown in Table 2 in the respective models. We use log-transformation

for current income. N is the sample size.

The coefﬁcient for the interaction term in every model captures the heterogeneous treatment effects.

The mean household size in our sample is 2.93 with a minimum of 1 and a maximum of 15.

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01

Table A4: Treatment Effect on Hypothetical ISA Take-Up - by Age (Median Split)

(1) (2) (3) (4) (5) (6) (7)

Treatment 0.114*** 0.114*** 0.118*** 0.118*** 0.115*** 0.114*** 0.112***

(0.024) (0.024) (0.025) (0.025) (0.025) (0.025) (0.025)

Below median age 0.031 0.030 0.023 0.022 0.015 0.015 0.024

(0.023) (0.023) (0.023) (0.023) (0.024) (0.024) (0.034)

Treatment × Below median age -0.031 -0.031 -0.027 -0.027 -0.024 -0.023 -0.020

(0.034) (0.034) (0.035) (0.035) (0.035) (0.035) (0.035)

N 2776 2776 2599 2599 2599 2599 2594

Variable deﬁnitions follow the Pre-Analysis Plan; see Appendix A for details on our derived variables. The regression model in all columns follows from

Equation 3 in the Pre-Analysis Plan. The regression models control for all variables shown in Table 2 in the respective models. We use log-transformation

for current income. N is the sample size.

The coefﬁcient for the interaction term in every model captures the heterogeneous treatment effects.

The median age in our sample is 37.

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01

Table A5: Treatment Effect on Hypothetical ISA Take-Up - by Marital Status

(1) (2) (3) (4) (5) (6) (7)

Treatment 0.093*** 0.094*** 0.099*** 0.099*** 0.099*** 0.096*** 0.091***

(0.024) (0.024) (0.024) (0.024) (0.025) (0.025) (0.025)

Married 0.029 0.030 0.032 0.032 0.035 0.032 0.023

(0.023) (0.023) (0.023) (0.023) (0.023) (0.023) (0.025)

Treatment × Married 0.013 0.011 0.011 0.011 0.010 0.015 0.021

(0.034) (0.034) (0.035) (0.035) (0.035) (0.035) (0.035)

N 2776 2776 2599 2599 2599 2599 2594

Variable deﬁnitions follow the Pre-Analysis Plan; see Appendix A for details on our derived variables. The regression model in all columns follows from

Equation 3 in the Pre-Analysis Plan. The regression models control for all variables shown in Table 2 in the respective models. We use log-transformation

for current income. N is the sample size.

The coefﬁcient for the interaction term in every model captures the heterogeneous treatment effects.

Out of 2,776 respondents, 1,399 are married.

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01

Table A6: Treatment Effect on Hypothetical ISA Take-Up - by Risk Aversion

(1) (2) (3) (4) (5) (6) (7)

Treatment 0.058 0.058 0.041 0.041 0.030 0.032 0.027

(0.062) (0.062) (0.064) (0.064) (0.064) (0.064) (0.064)

Risk aversion -0.012 -0.011 -0.011 -0.011 -0.010 -0.010 -0.010

(0.010) (0.010) (0.010) (0.010) (0.010) (0.010) (0.010)

Treatment × Risk aversion 0.011 0.011 0.016 0.016 0.019 0.018 0.019

(0.015) (0.015) (0.016) (0.016) (0.016) (0.016) (0.016)

N 2776 2776 2599 2599 2599 2599 2594

Variable deﬁnitions follow the Pre-Analysis Plan; see Appendix A for details on our derived variables. The regression model in all columns follows from

Equation 3 in the Pre-Analysis Plan. The regression models control for all variables shown in Table 2 in the respective models. We use log-transformation

for current income. N is the sample size.

The coefﬁcient for the interaction term in every model captures the heterogeneous treatment effects.

Out of 2,776 respondents, 1,303 (46.94%) are categorized as "Very Risk Averse", 571 (20.57%) as "Moderately Risk Averse", 441 (15.89%) as "Willing to

Take Risk" and 461 (16.61%) as "Very Willing to Take Risk".

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01

Appendix C

Robustness Check: A Data-Driven Approach to Understanding Adverse Selection

As a robustness check and to incorporate every dimension of risk, we employ the data-driven approach by Athey and Imbens (2016) to estimate treatment

effect heterogeneity. We investigate several risk factors including perceived barriers to graduation, current ﬁnancing choices, risk aversion and future

expectations as well as demographics

as candidates for sources of heterogeneity. The suggested methodology builds on regression tree methods from the

prediction-based machine learning literature (Breiman et al., 1984; Breiman, 2001). The sample is ﬁrst divided into two parts: training and estimation data.

We use the training data to create partition of the population according to covariates, and then we use the estimation data to estimate treatment effects for each

subpopulation. The approach provides an advantage for cases where there are many covariates by letting the data tell the researcher the relevant subgroups

where one should look for heterogeneity. By separating the datasets used to select the model structure and to estimate, Athey and Imbens (2016) mitigate the

possible bias in machine learning methods where spurious correlations between covariates and outcomes affect the selected model. They argue that although

the reduced sample size in each step through partition leads to loss of precision, the estimates are unbiased for every subpopulation.

In this exercise, we simulate with different cut-off points in the sample to get the training and the estimation data. Figure A1 summarizes our ﬁndings

when we use 40% of our sample as the training data and the rest as our estimation sample. The training data identiﬁes housing security and other ﬁnancial

risks as the ﬁrst source of heterogeneity. The second iteration on the subsamples based on housing security and other ﬁnancial risks return risk aversion

and college-related risks as second sources of heterogeneity. Having enough sample size in one of the branches, we carry the third iteration to identify

employment uncertainty as another source of heterogeneity. For each subsample, we report the treatment effects and standard errors in brackets based on

the estimation data. Although we show signiﬁcant treatment effects for each subsample, the statistical signiﬁcance only exists for one of the branches within

subsample, i.e., when the index for housing security and other ﬁnancial risks is less than 0.69 but not the other way around. This lack of symmetry prevents

us to have conclusive evidence on heterogeneity and adverse selection. We do, however, present suggestive evidence for potential risk factors that lead to

signiﬁcant heterogeneous treatment effects which may be revealed in larger samples with more power.

The demographics include age, sex, race and ethnicity, household size and educational attainment.

Notes: We use the Coefﬁcient of Relative Risk Aversion (CRRA), a measure for risk preferences following a similar

method as Kimball, Sahm, and Shapiro (2008). The indices for housing security and other ﬁnancial risks and college-

related risks are created via factor analysis based on students’ answers to survey questions on barriers to graduation. The

partition, and hence the p-values, that determines the subsamples where treatment heterogeneity exists are based on the

training data. The numbers in parentheses at the leaves indicate how many data points in the training data belong to each

leaf. All treatment effects are estimated using the estimation data for each subsample with the standard errors shown in

brackets and sample size in parentheses.

Figure A1: Athey and Imbens (2016) regression tree

Appendix D

Core Experiment - Description of the Randomization and Treatment

The survey presented hypothetical ﬁnancing options to students via two arms in what we consider the "core" experiment. Half of the students were randomized

into a "risk neutral" framing. This control arm provided students with limited information on a federal student loan with the option of IDR and a hypothetical

ISA; this information included a basic description of monthly payments, repayment term, and downside protections of the two ﬁnancing options. The other

half of students were randomly selected into the treatment arm that emphasized educational insurance features that would protect the student against downside

employment and income risks, including providing additional information and more detailed examples on repayment caps, income thresholds for no payments

due, and obligation satisfaction conditions for each ﬁnancing method. The exact wording of the choices presented to students is shown below. The additional

insurance emphasis for the treatment arm is highlighted in bold text for the reader (but was not highlighted in any way for survey respondents in the treatment

group).

Figure A2: Presentation of Hypothetical Financing Options in the Core Experiment: Treatment and Control Groups

Alternative Options - Student Loan Choosers in Core Experiment

In subsequent rounds, we tested student preferences over alternative contract terms for the hypothetical ISA. We should note that we will not be able to make

causal inferences with these price/term variations, as the sample who was offered the alternative, actuarially equivalent, contracts was endogenous to the core

experiment. The exact wording of both alternative, actuarially equivalent, hypothetical ISAs is shown below. As before, the additional insurance emphasis

for the treatment arm is highlighted in bold text for the reader (but was not highlighted for survey respondents in the treatment group). The differences in the

income shares and the contract lengths between the ISA alternatives are underlined for the reader (but were not underlined for survey respondents).

In the second round, students who chose a federal student loan with an option of IDR against the original ISA contract (2.3% share for 10 years) in either

of the treatment arms in the ﬁrst round were asked if they would switch to an ISA contract where they would pay smaller installments over a longer period

(2.2% share for 12 years).

Figure A3: Presentation of Hypothetical Actuarially Equivalent ISAs with Alternative Terms -

Original Student Loan w/ IDR Choosers in Core Experiment -

Alternative (Option 3): ISA w/ Longer Term (2.2% Share for 12 Years)

In the third round, those who still chose the federal student loan option in the second round were then asked to choose between the original federal

student loan w/ IDR and another, actuarially equivalent, ISA contract where they would pay larger installments over a shorter period (2.8% share for 7 years).

Figure A4: Presentation of Hypothetical Actuarially Equivalent ISAs with Alternative Terms -

Student Loan w/ IDR Choosers in Second Round -

Alternative (Option 4): ISA w/ Shorter Term (2.8% Share for 7 Years)

Alternative Options - ISA Choosers in Core Experiment

On the other hand, the students who chose the original hypothetical ISA contract over the federal student loan with the option of IDR in the core experiment

were also offered the alternative ISA contracts. They were presented with all three ISA options (Option 2, Option 3, and Option 4) with different pricing

and payment terms described above at the same time, as shown below. As before, the additional insurance emphasis for the treatment arm is highlighted in

bold text for the reader (but was not highlighted for survey respondents in the treatment group). The differences in the income shares and the contract lengths

between the ISA alternatives are underlined for the reader (but were not underlined for survey respondents).

Figure A5: Presentation of Hypothetical Actuarially Equivalent ISAs with Alternative Terms -

Initial ISA Choosers in Core Experiment -

Alternatives: ISA w/ Longer Term (2.2% Share for 12 Years) or ISA w/ Shorter Term (2.8% Share for 7 Years)