Behavioral Nudges for Credit Card Debt Repayment in the Dominican Republic

In partnership with the Superintendencia de Bancos (SB) and Innovations for Poverty Action (IPA), the research team will implement and evaluate behavioral “nudge” messages intended to increase repayment behavior among ProUsuario users with revolving credit card balances. Messages will be delivered digitally and repeatedly to test whether low-cost information and reminders can generate sustained improvements in repayment and debt reduction.

2026-04-21

2026-10-30

Full balance payment (dummy)
Proportion of debt paid (payment-to-balance ratio)
Revolving debt payment (dummy)
Minimum payment compliance (dummy)
Less than minimum payment (dummy)
Statement balance net of payments, as a proportion of statement balance

Full balance payment (dummy): Indicator equal to 1 if the user's total payments during the grace period cover the full statement balance.
Proportion of debt paid (payment-to-balance ratio): Ratio of total payments to prior-period statement balance, aggregated across all cards held by the user.
Revolving debt payment (dummy): Indicator equal to 1 if the user made payments above the minimum but below the full balance, i.e., the user revolves debt while making a partial repayment effort.
Minimum payment compliance (dummy): Indicator equal to 1 if the user's total payments meet or exceed the minimum payment due across all cards.
Less than minimum payment (dummy): Indicator equal to 1 if the user made a positive payment that falls below the minimum payment due across all cards.
Statement balance net of payments, as proportion of statement balance: Captures net debt reduction or increase of balance in the analyzed cycle

Number of months of revolving debt
The existence of interest charges (dummy)
Days past due
Dummies for 1, 2 or 3 missed payments
Interest amount
Debt level
Difference in payment behavior across cards (if the person has more than 1 card)

Number of months of revolving debt: Count of billing cycles during the follow-up period in which the user carries a revolving balance, aggregated across all cards at the user level.
Existence of interest charges (dummy): Indicator equal to 1 if the user incurs any interest charges in a given billing cycle, aggregated across all cards at the user level.
Days past due: Number of days beyond the payment due date that the user's account remains in arrears, aggregated across all cards at the user level.
Dummies for 1, 2 or 3 missed payments: Indicator variables (3) equal to 1 if the user has missed their payment in the past 1, 2 or 3 months.
Interest amount: Total revolving interest accrued during the billing cycle.
Debt level: Total outstanding statement balance at the end of the billing cycle, aggregated across all credit card accounts held by the user.
Consecutive months of full balance payment
Consecutive months of null balance on account.
Difference in payment behavior across cards: For people with more than one card, in different banks, where we can observe the different payment levels per card, we will compare the payment behavior across cards.

We target credit card users in the Dominican Republic whose accounts are serviced by financial institutions supervised by the Superintendencia de Bancos (SB). Eligibility requires: (i) the account carried a non-zero revolving balance (i.e., was not fully paid) in at least two of the twelve months prior to the baseline month; and (ii) the statement balance in the baseline month (January 2026) was at least 500 DOP (approximately 8 USD). These criteria identify revolvers — users who habitually carry debt and are therefore most likely to benefit from repayment nudges and most likely to show detectable behavioral responses.

From this eligibility frame, we selected all accounts satisfying the criteria in the baseline billing cycle (January 2026), yielding a randomized sample of approximately 65,400 users. The unit of observation is a user–billing-cycle pair. We define a billing cycle for month X as comprising all accounts whose statement date falls within month X, regardless of whether the corresponding due date falls in a subsequent month (as is common). All outcome measurements follow this billing-cycle definition throughout the intervention and follow-up periods.

Randomization was conducted using stratified random assignment. The stratification variables are: (i) number of cards held (1 card vs. 2+ cards) and (ii) gender (female vs. male), yielding four strata. Within each stratum, users were allocated with equal probability (25% each) to the four experimental arms. A fixed random seed was set prior to randomization to ensure full reproducibility; the randomization script and seed are archived and available upon request.

We will implement a 4-arms RCT. The arms are defined as follows:

- Control
- Treatment 1 — Single message type: Participants receive one message per billing cycle. The message type is fixed across all 6 intervention months, assigned at randomization. Within T1, participants are sub-randomized to receive either an interest-focused message (M-I) or a debt-focused message (M-D).
- Treatment 2 — Alternating messages: Participants receive one message per billing cycle, alternating between M-I and M-D each month across the 6 intervention months. Within T2, participants are sub-randomized to start with either M-I or M-D in month 1.
- Treatment 3 — Combined message: Participants receive one message per billing cycle containing both debt and interest salience in a single communication. Within T3, participants are sub-randomized to receive either an interest-then-debt (M-I-D) or debt-then-interest (M-D-I) ordering across all 6 months.

Messages are delivered through three channels — email (Mailchimp), SMS (SendIU), and push notification (Firebase) — depending on the contact information available for each user. One message per billing cycle per user is delivered.

- M-D: Hola [NOMBRE], reduce intereses hoy: paga más de tu tarjeta el Banco Lopez de Haro si tienes balance pendiente.
- M-I: Hola [NOMBRE], rompe el ciclo de deuda hoy: paga más de tu tarjeta el Banco Lopez de Haro si tienes balance pendiente.
- M-D-I: Hola [NOMBRE], reduce intereses y rompe el ciclo de deuda hoy: paga más de tu tarjeta del Banco Lopez de Haro si tienes balance pendiente.
- M-I-D: Hola [NOMBRE], rompe el ciclo de deuda y reduce intereses hoy: paga más de tu tarjeta del Banco Lopez de Haro si tienes balance pendiente

The expected timeline for the project is defined as follows:
Baseline – January 2026: Eligibility and sample selection
Pre-intervention – Feb-March 2026: Randomization, protocol finalization and setup
Intervention – April-September 2026: Six monthly messages delivered to treatment arms
Follow up – October 2026 - March 2027: Three post-intervention billing cycles, no messages will be sent.

Our main specification is an OLS regression with standard errors clustered at the user level. We follow participants across multiple billing cycles from the start of the intervention, covering both the six intervention months and the six post-intervention follow-up months. We employ two complementary approaches to recover treatment effects over time. First, we estimate separate cross-sectional regressions for each billing cycle, yielding a period-specific ITT estimate without imposing any trajectory on the treatment effect. Second, we estimate a pooled regression with treatment × billing-cycle interactions, which allows us to formally test for effect decay, growth, or stabilization over time. We do not impose a priori restrictions on the treatment effect trajectory, as it is theoretically ambiguous; both specifications let the data speak to this directly. Pre-specified baseline covariates (age, gender, number of cards held, number of financial institutions, interest rates, and app login activity as a proxy for engagement) are included throughout to improve estimation precision. The unit of analysis is the user–billing cycle observation. Where a user holds multiple credit cards, debt balances and payment variables are aggregated across all their accounts prior to analysis..

Not available

We will conduct stratified random assignment through computer assisted software. The software used for the randomization is R from Posit. The randomization will stratify in the following two variables:
1. Gender of card holder: Male or female
2. Number of credit cards: One card vs two or more credit cards.

As a result we have 4 possible stratas.

The randomization will be conducted at user level

No

Approximately 65,400 users

Approximately 65,400 users

- Control: 16,300 users
- Treatment 1: 16,300 users
- Treatment 2: 16,300 users
- Treatment 3: 16,300 users

Power calculations use January 2026 administrative baseline data (N = 65,464 users; 16,366 per arm across 4 arms). Binary outcomes are treated as linear probability models using two-sample t-tests; continuous outcomes use empirical SDs.All tests are two-sided at α = 0.05. We also report Bonferroni-corrected p-values to account for multiple testing across six primary outcomes. Power calculations are conducted at the Bonferroni-adjusted threshold (α_adj = 0.05/6 ≈ 0.0083) to ensure adequate power under the stricter correction, with an 80% power target. Full balance payment is defined as pagos_p2 ≥ 95% of the statement balance, with a 5% buffer to account for rounding.. - Full Balance Payment (baseline: 7.8%; SD: 0.269): At 16,366 users per arm, the study achieves 77% power for a 1.0 pp increase and >99% power for 1.5 pp and above. A 1.0 pp MDE requires 17,459/arm — marginally above our sample, placing power just below 80% at that threshold. Effects of 1.5 pp and above are detected comfortably (requires 7,761/arm). - Payment-to-Balance Ratio (baseline mean: 0.214; SD: 0.300): At 16,366 users per arm, the study achieves 65% power for a 1.0 pp improvement and 97% power for 1.5 pp. A 1.0 pp MDE requires 21,730/arm — above our sample — placing power below 80% at that threshold. Effects of 1.5 pp and above are well-powered (requires 9,659/arm). - Revolving Debt Payment (baseline: 49.9%; SD: 0.500): The near-50% baseline implies maximum variance. At 16,366 users per arm, the study achieves 84% power for a 2.0 pp change and 97% for 2.5 pp. A 2.0 pp MDE requires 15,139/arm and is comfortably within sample; a 1.5 pp MDE would require 26,912/arm, above sample. - Minimum Payment Compliance (baseline: 61.3%; SD: 0.487): At 16,366 users per arm, the study achieves 86% power for a 2.0 pp increase and 98% for 2.5 pp. A 2.0 pp MDE requires 14,369/arm and is comfortably powered; a 1.5 pp MDE would require 25,543/arm, above sample. - Less than Minimum Payment (baseline: 33.4%; SD: 0.472): At 16,366 users per arm, the study achieves 88% power for a 2.0 pp reduction and 98% for 2.5 pp. A 2.0 pp MDE requires 13,467/arm and is comfortably powered; a 1.5 pp MDE would require 23,940/arm, above sample. - Net Balance Ratio (baseline mean: 0.764; SD: 1.165): This continuous outcome has high variance and is substantially underpowered for small effects at our sample size. At 16,366 users per arm, the study achieves approximately 80% power only at a 4.5 pp MDE (requires 16,232/arm); effects below 4.0 pp yield power below 68%. This outcome is included as a secondary measure of debt-level dynamics; causal interpretation of small effects should be made with caution.