Table 3.

Regression-estimated impact of describing IDR benefits in % (vs. $) terms and recommending two recommended actions repeatedly (vs. one of two recommended action sequentially) on whether a borrower applied for an IDR plan within 90 d of receiving an intervention email (Model 1), signed up for auto debit within 90 d of receiving an intervention email (Model 2), made at least one loan payment within 90 d of receiving an intervention email (Model 3), and entered into 60-d delinquency within 180 d of receiving an intervention email (Model 4)

	Model 1	Model 2	Model 3	Model 4
IDR benefits described in %	0.00339***	-0.00013	0.00113***	−0.00141***
	(0.00018)	(0.00012)	(0.00026)	(0.00027)
Two-action email	0.00398***	−0.00006	−0.00017	−0.00051+
	(0.00018)	(0.00012)	(0.00026)	(0.00027)
Are controls included?	Yes	Yes	Yes	Yes
Number of observations	8,442,700	9,796,300	10,008,800	10,008,800
Adjusted R-squared	0.02	0.05	0.18	0.13

***P < 0.001, **P < 0.01, *P < 0.05, ⁺P < 0.10

Notes: This table reports the results of four OLS regressions including the subset of borrowers assigned to one of four conditions in our experiment: the “Two-Action, IDR % Benefits Email with Reminder” condition, the “Two-Action, IDR $ Benefits Email with Reminder” condition, the “One-Action, IDR % Benefits Email with Reminder” condition and the “One-Action, IDR $ Benefits Email with Reminder” condition. The dependent variables are a binary indicator for whether a borrower applied for an IDR plan within 90 d of first receiving an intervention email (Model 1), a binary indicator for whether a borrower signed up for auto debit within 90 d of first receiving an intervention email (Model 2), a binary indicator for whether a borrower made at least one loan payment within 90 d of first receiving an intervention email (Model 3), and a binary indicator for whether a borrower lapsed into 60-d delinquency within 180 d of first receiving an intervention email. The primary predictor variables are indicators for whether a borrower received communications that described IDR benefits in % (vs. $) terms and whether a borrower received communications that recommended two actions repeatedly (vs. communications that recommended one of two recommended actions sequentially). For all models, control variables were included for borrower age as of August 2023, an indicator for whether the borrower was male, indicators for education grade level reported on the borrower’s last FAFSA application as of August 2023, an indicator for whether the borrower graduated from their last reported educational institution attended as of August 2023, indicators for the borrower’s state of residence as of August 2023, the number of days that elapsed between when a borrower took out their last loan as of August 2023 and when they first received an intervention email, the total funds loaned to the borrower as of August 2023, the borrower’s current outstanding loan balance as of August 2023, the borrower’s monthly loan payment amount owed as of August 2023, an indicator for a borrower’s loan servicer as of August 2023, an indicator for the borrower’s last loan repayment type prior to the repayment pause as of March 2020 (e.g., standard repayment plan, graduated repayment plan, etc.), indicators for each loan type owned by borrower as of August 2023 (i.e., Parent PLUS loans, consolidation loans), an indicator for ever having a loan payment past due for 60 d or more (60-d delinquency) as of August 2023, an indicator for ever having a loan in default as of August 2023, an indicator for whether the borrower was enrolled in an IDR (income-driven repayment) plan in February 2020 (prior to the repayment pause), an indicator for whether the borrower was signed up for auto debit in February 2020 (prior to the repayment pause), the % of months the borrower ever had a loan payment past due for 30 d or more (30-d delinquency) from January 2015 to March 2020 (prior to the repayment pause), and indicators for whether the participants first missed a payment on all possible missed payment dates (prior to eligibility for study participation being determined and prior to receiving an intervention treatment email). If a control variable contained missing data, we addressed it in two steps: i) for binary or categorical variables, missing values were replaced with zero; for continuous variables, missing values were replaced with the mean that variable took on in nonmissing observations; ii) we added an indicator variable to the regression, coded as one if the original variable was missing and zero otherwise. For Model 1, borrowers who had already signed up for a SAVE plan as of the send date of our first intervention email message and borrowers who were ineligible to apply for a SAVE plan were excluded from our analysis. For Model 2, borrowers who had signed up for auto debit on all their loans as of the send date of our first intervention email message were excluded from our analysis. All SE shown are HC1 robust SE. All sample sizes reported above are rounded to the nearest hundred borrowers in accordance with disclosure policies required by the Department of Education.