Survival and Financial Literacy in Investment Decisions Later in Life

We implement two information treatments: a survival literacy treatment (S) and a financial literacy treatment (F), delivered through short animated videos (approximately 2 and 6 minutes, respectively).

Participants are recruited online via Prolific and complete a Qualtrics survey. They are randomly assigned with equal probability to one of four groups in a 2 × 2 factorial design: (i) control group, (ii) survival literacy treatment only, (iii) financial literacy treatment only, and (iv) both treatments.

The survival literacy treatment provides information on age- and gender-specific survival probabilities derived from life tables, with the aim of improving individuals’ awareness of longevity risk. The financial literacy treatment provides information on the functioning of financial investments, including risk/return trade-offs and long-term investment considerations.

The control group is exposed to a placebo video (approximately 3 minutes) on the advantages and disadvantages of different payment methods, ensuring a comparable level of engagement across experimental conditions.

Following the intervention, participants complete survey modules eliciting subjective survival expectations, portfolio allocation decisions in a hypothetical investment task, and expectations about long-term care needs.

This design allows us to estimate the causal effects of each intervention separately, as well as their combined effect.

2026-05-04

2026-07-31

The primary outcomes are:
1. Survival belief bias.
2. Risky asset share (age-50 vignette).

Survival belief bias is a constructed outcome. For each respondent, subjective survival probabilities are elicited using survey questions modeled on the English Longitudinal Study of Ageing (ELSA) survey. Objective survival probabilities at the relevant target age are assigned using age- and gender-specific life tables. The main outcome is constructed as:

subjective survival probability − benchmark survival probability

where the target age depends on the respondent’s age. This measure captures the deviation between individuals’ subjective expectations and benchmark survival probabilities, which may reflect both differences in information and idiosyncratic individual characteristics.

As an alternative measure, we construct a standardized discrepancy index based on an accuracy measure defined as the ratio between subjective and objective survival probabilities, censored at 1. The discrepancy index is defined as one minus accuracy, standardized by its standard deviation.

2. Risky asset share is constructed from respondents’ answers to a hypothetical portfolio allocation task. Respondents are asked how to allocate £50,000 between a safe asset (e.g., savings account or government bonds) and a risky asset (e.g., stock index fund). The outcome is defined as the amount allocated to the risky asset divided by £50,000, yielding a continuous variable bounded between 0 and 1.

Respondents are also randomly assigned to a version of the vignette with or without a participation cost for investing in the risky asset. The primary outcome pools across these randomized conditions.

The portfolio-related outcome is defined pooling across the randomized participation-cost conditions. We will also examine heterogeneity in portfolio choices by whether the vignette includes participation costs.

The secondary outcomes are:
1. Risky asset participation, defined as an indicator for any positive investment in the risky asset.
2. Subjective probability of needing long-term care.

1. Risky asset participation is constructed as a binary variable equal to one if the respondent allocates any positive amount of the £50,000 endowment to the risky asset, and zero otherwise.
2. Subjective long-term care (LTC) risk is elicited through a survey question asking respondents to assess the probability of ever needing professional care on a continuous basis, either in a nursing home or at home (excluding informal care). Responses are recorded on a 0–100 scale.
All portfolio-related outcomes are defined pooling across the randomized participation-cost conditions. We will also examine heterogeneity in portfolio choices by whether the vignette includes participation costs.

The study is an online randomized controlled trial administered to UK residents aged 50–70 recruited via Prolific. Randomization takes place at the individual level.

Participants are assigned with equal probability to one of four groups in a 2 × 2 factorial design: placebo control, survival literacy treatment (S), financial literacy treatment (F), or both treatments (S+F). The control group views a placebo video on payment methods.

After treatment exposure, participants complete survey modules eliciting subjective survival expectations, portfolio allocation choices, and expectations about long-term care needs.

Within the portfolio module, participants are additionally randomized to a vignette with or without participation costs for the risky asset. All participants provide allocation decisions for both a 50-year-old and a 70-year-old hypothetical individual.

The design identifies the separate and combined effects of the two information interventions, as well as heterogeneity with respect to the randomized participation-cost condition.

Not available

Randomization is implemented at the individual level through the Qualtrics survey platform, using its built-in random assignment functionality to allocate participants with equal probability across the four treatment groups. Independent randomization is also used within the survey to assign participants to the portfolio vignette with or without participation costs.

Individual-level randomization.

No

Not applicable.

Approximately 3,600 individuals.

Approximately 900 individuals in each of the four groups: control, survival literacy treatment only, financial literacy treatment only, and combined treatments.

Power calculations assume two-sided tests, a significance level of α = 0.05, and power of 80%. Randomization is at the individual level, and no clustering adjustment is required. Under equal allocation across the four experimental groups, the main treatment effects are estimated by pooling two groups versus two groups within the 2 × 2 factorial design. Main outcome 1: Survival belief bias The outcome is measured in probability units (0–1). We assume a standard deviation of approximately 0.23. The minimum detectable effect size for the main effect of the survival literacy treatment is 0.0285 (2.85 percentage points), which requires approximately 511 observations per experimental cell (total N ≈ 2,044). The minimum detectable effect for the financial literacy treatment is 0.057 (5.7 percentage points), requiring approximately 128 observations per cell (total N ≈ 512). The assumed effect sizes are guided by prior evidence on subjective survival expectations and their responsiveness to information. In particular, Hurwitz, Mitchell, and Sade (2022, Journal of Economic Behavior & Organization) show that longevity beliefs can be meaningfully updated in response to informational interventions. Main outcome 2: Risky asset share The outcome is defined as the fraction of a hypothetical endowment allocated to a risky asset (ranging from 0 to 1). We assume a standard deviation of 0.25, consistent with the dispersion observed in survey and experimental data. The minimum detectable effect size for the main treatment effects is 0.03 (3 percentage points), which requires approximately 545 observations per experimental cell (total N ≈ 2,180). Larger effects of 0.05 (5 percentage points) would be detectable with approximately 196 observations per cell (total N ≈ 784). The assumed effect sizes are informed by prior experimental evidence. In particular, Billari, Favero, and Saita (2023, Journal of Banking & Finance) document increases in equity exposure of approximately 3 percentage points following interventions combining financial and demographic information. The assumed range of 3–5 percentage points reflects conservative benchmarks based on this literature. Interaction effects in the 2 × 2 design require substantially larger samples and are therefore treated as exploratory. For example, detecting an interaction effect of 0.0285 in survival belief bias would require more than 2,000 observations per experimental cell (total N > 8,000). The planned sample size of approximately 3,600 individuals provides sufficient power to detect main effects for both primary outcomes within the range of effect sizes considered.