Supply-Side of Consumer Debt Repayment: De-Anchoring Experiment

We will field a survey experiment on Prolific with hypothetical debt repayment scenarios. Our goal is to understand whether anchoring on credit card minimums can help explain repayment behaviors across debt and, in particular, why individuals revolve credit card debt while simultaneously making overpayments on installment debts.

We will field a survey experiment on Prolific with hypothetical repayment scenarios to answer the following questions:
- Does not displaying the minimum as a repayment choice (“de-anchoring”) change the distribution of credit card repayments?
- How do de-anchoring effects vary by whether people overpay another debt, such as a mortgage?
- Does showing information about costs of repayment change repayments?
- Do these interventions change the distribution of repayments on other debts?
- How do interventions affect individuals’ perceptions of the costs of minimum payments?

We provide more details in the experiment design, below.

The second half of the survey contains some questions about own debt repayment more generally. Here we are interested in studying and reporting descriptive statistics about participants’ own repayment behaviors.

2025-07-07

2025-07-14

Outcome 1: Distribution of credit card repayments
- Overall and among borrowers who overpay their mortgage
Outcome 2: “Mistake”: Share who overpay mortgage (by $0 and >$25) and pay less than the card balance
- Overall and among borrowers who make some excess payment
Outcome 3: “Size of Mistake”: Average of min(excess dollars to mortgage, unpaid card balance)
- Among borrowers who make mistake
Outcome 4: Beliefs of cost and time-to-repay if paying the minimum

Below are details on the specific analyses we plan to run:
(1) The distribution of share excess balance repaid on the credit card for each treatment.
(2) Same as (1) but for people who make mortgage overpayments.
(3) Distribution of answers to the costs question, where we ask individuals to estimate the time-to-repay and interest incurred when only paying the minimum.
(4) The following specific summary statistics by treatment arm:
- Share paying the credit card minimum
- Share paying credit card full balance
- Share paying > mortgage minimum
- Share paying > mortgage minimum and < credit card full balance
- Share with misallocation among those making excess repayment
- Share paying > mortgage minimum + $25
- Average misallocation size, conditional on misallocation
- Share underestimating the time and cost to repay

Observations will be excluded if the participants:
- Do NOT meet the eligibility criteria (are under 18, do not have a credit card, or do not have a mortgage)
- Fail to correctly respond "Tuesday and Thursday" in the attention check question about favorite days of the week
- Disclose “yes” to consulting outside resources
- For participants who pay less than the required amount to avoid a late fee in the hypothetical (which suggests they may not have understood the question), we may do cuts with and without these individuals.

When recruiting for participants on Prolific, we will apply the following filters:
- Participants must be in the USA
- Exclude participants from previous pilots
- Owns a credit card
- Holds a mortgage
- Has had at least 100 prior submissions on Prolific
- Has a 98% or higher approval rating
- Can only participate in the survey once

If we exclude any other observations that fall outside of these categories, we will justify the exclusion, and explain why it could not reasonably have been pre-registered.

OLS regressions
Yi = T D_i + X_i’ B + epsilon_i
Yi: The outcomes listed in (4) above, under Primary Outcomes.
Di: I(De-anchor or Box treatment)

Demographics X_i include income and education categories, age, and credit score. We will report heteroskedastic-robust standard errors and use these to conduct statistical tests.

We may also report descriptive statistics or subjective open-ended justifications from survey participants about how they think about their debt repayment.

This is part of a larger paper that includes analysis of administrative data and structural modeling.

There will be three conditions. Each participant will be assigned to one condition, where they will be asked two hypotheticals and one costs question (before proceeding with the second half of the survey).

Hypothetical one: No budget

Participants will be presented with two bills: one credit card and one mortgage, and asked to think about how much they would repay on each debt in this situation.
- Baseline condition: for mortgage, participant sees amount due, overpayment options, and other. For card, participants see minimum, statement balance, and other options.
- De-anchor condition: identical to Baseline, except no minimum option for card (minimum is still on the bill and can be chosen if repaying other)
- Box condition: identical to the Baseline, except with more information provided on the credit card bill

Hypothetical two: Budget

- Participants will be presented with the same credit card and mortgage bills. Now they will be asked to report what they would repay on their credit card if they could only pay $3000 in total on their credit card and mortgage.
- Baseline condition: sees minimum, statement balance, and other options.
De-anchor condition: sees statement balance and other options.
- Box-condition: identical to the Baseline, except with more information provided on the credit card bill

Costs question

Participants will be shown the same credit card bill they saw in the two hypotheticals. They will then be asked to estimate interest costs and time-to-repay the full balance if they only paid the minimum.

Qualtrics randomization: equal probability of being randomized into each of the conditions.

Respondent

No

NA

We will keep the survey open for a week or until we get 1,800 participants, whichever comes first.

We split respondents evenly between treatment arms, so will have up to 600 respondents by arm.

The power calculations below will determine our sample size. We have conducted power calculations for our most important statistics to make sure we collect enough data. Power calculations: Significance (alpha) = 0.05: t-test critical value for 0.05 (two-sided test) = 1.96 Power (1-kappa) = 0.8: t-test critical value for 0.2 = 0.84 Treatment allocation P = ½. This is comparing the baseline and de-anchor treatment, which will be our main specification. MDE and variance of outcome will be from prior pilot. Variance is computed as overall variance across all treatments in the pilot. No budget: Share paying the credit card minimum in control vs. de-anchor treatment MDE = 0.21 Variance of outcome = 0.11 Implies that N should be at least: (0.84 + 1.96)^2 * 4 * 0.11/(0.21)^2 = 78 No budget: Share paying the credit card full balance in control vs. de-anchor treatment MDE = 0.28 Variance of outcome = 0.24 Implies that N should be at least: (0.84 + 1.96)^2 * 4 * 0.24/(0.28)^2 = 96 No budget: Share paying the credit card full balance in control vs. de-anchor treatment, among those who make mortgage overpayments Share who make mortgage overpayment = 0.26 MDE = 0.18 Variance of outcome = 0.16 Implies that N should be at least: (1/0.26) * (0.84 + 1.96)^2 * 4 * 0.16/(0.18)^2 = 596 No budget: Share of excess repayers with misallocation in control vs. de-anchor treatment MDE = 0.057 Variance of outcome = 0.055 Implies that N should be at least: (0.84 + 1.96)^2 * 4 * 0.055/(0.057)^2 = 531 Budget: Share paying more than the mortgage minimum in control vs. de-anchor treatment MDE = 0.11 Variance of outcome = 0.23 Implies that N should be at least: (0.84 + 1.96)^2 * 4 * 0.23/(0.11)^2 = 596 Budget: Share paying more than the mortgage minimum + $25 in control vs. de-anchor treatment MDE = 0.15 Variance of outcome = 0.18 Implies that N should be at least: (0.84 + 1.96)^2 * 4 * 0.18/(0.15)^2 = 251 Budget: Average misallocation size, conditional on misallocation in control vs. de-anchor treatment MDE = 277 Variance of outcome = 120,556 Implies that N should be at least: (0.84 + 1.96)^2 * 4 * 120,556/(277)^2 = 50 Since there are three treatment arms, we need to multiply our numbers by 3/2. We will thus aim for a sample of 1,800 participants (600 in each treatment arm). We will keep the survey open for a week (or until we get 1,800 participants), whichever comes first.