Changing the choice architecture to reduce student borrowing

Loan process
Experiment Type
Field Experiment
Improve budgeting & borrowing
Reduce debt
Focus Areas
Behavioral Concepts
Default bias Friction
Duke Personal Finance
Partner Type

What Happened

It worked. There was a significant reduction in the odds of taking on all offered debt when the loan process was expanded in the experimental condition. Additionally, those in the experimental condition borrowed 8.4% less than those in the control.

Lessons Learned

Defaulting students to only take out the COA and making them enroll again for living cost related expenses decreased student borrowing.


Student loans are the second largest form of US debt. While the focus has centered on undergraduate borrowing, 40% of America’s $1.3 trillion student loan debt has financed graduate degrees, and that number is growing. In fact, from 2008 to 2016, the share of graduate students with over $100,000 in loans doubled.

We partnered with Duke’s Office of Student Loans and Personal Finance to reduce the amount graduate students borrow, particularly for living expenses. While tuition, fees, and health insurance are fixed, direct costs, “living expenses” (i.e. housing, food, transportation, and other similar expenses) can vary greatly based on student lifestyle choices. Nevertheless, initial analyses of Duke’s financial aid data showed that most students simply accept Duke’s full loan offer.

Key Insights

Duke students learn about their financial aid options after being accepted to their chosen graduate program. Students receive an email and/or letter about loans and scholarships for which they are eligible. After accepting Duke’s offer, they accept their aid package via a PeopleSoft lending portal. We began by auditing this process, which brought to light several key insights into how graduate students usually accept loans:

  • Duke’s PeopleSoft user interface made accepting all loans look like the default behavior to most students, and the fact that students could reduce the loans in their offer was not clear.

  • Loans for the full Cost of Attendance (living expenses plus fixed costs like tuition) are all awarded together, making it impossible for students to distinguish how much they are borrowing for direct vs. indirect costs.


Drawing on these insights, we hypothesized that we could reduce student borrowing if we partitioned the decision about borrowing into two decisions.

Working with the Duke Personal Finance Office, admitted students within six Duke graduate programs were randomly assigned to receive either a loan offer for the full COA (standard operating procedure within most financial aid offices) or an offer for direct costs only. In this case, we defaulted students into covering direct costs but made the default for living expenses to borrow nothing. Students who received the second offer just covering direct costs were prompted to use a tool that let them estimate their likely monthly living expenses to generate a custom living expenses loan. Their financial aid officers received an email with that loan amount and then students could log back into the Duke Lending portal and accept it. We measured whether students reduced their initial borrowing behavior and are continuing to track whether they borrow less over the full school year. We also tracked how students used the living expenses estimation tool.


In total, our experiment launched to 439 graduate students. We found that in the control condition, 28% of students accepted the full loan amount while in our experimental condition only 13% did – a signficant, 65% reduction in the odds of taking on all offered debt. Furthermore, among students who did choose to take on some debt (N=289), those in the experimental condition borrowed 8.4% less than those in the control. For the average student, in our experimental condition, this could mean a $5,472 decrease in their overall debt and an over $2,000 savings in interest over the life of the loan.