Follow Up - Survival and Financial literacy in investment decision later in life

Two information treatments were administered in Survey 1: a survival-literacy treatment (S) and a financial-literacy treatment (F), delivered through short animated videos. Survey 1 participants were randomly assigned with equal probability, in a 2 × 2 factorial design, to: (i) control, (ii) S only, (iii) F only, or (iv) S+F. The control group viewed a placebo video on payment methods.

The present wave administers no new intervention and performs no new randomization. It re-contacts the same cohort via Prolific (using the stored Prolific ID), about two months after the original intervention, to estimate whether the causal effects of each treatment—separately and combined—persist over time on survival expectations, financial choices, and LTC perceptions.

2026-07-01

2026-08-15

1. Survival belief bias — the signed gap between subjective and objective own-survival probability at the closest target age (10 years).
2. Persistence of treatment knowledge of financial-literacy concepts, and financial choices.

1. Survival belief bias. We elicit subjective survival probabilities at 10-, 20- and 30-year horizons, both for the respondent and for a person of the same age and sex in the UK. Objective benchmarks are assigned from age- and gender-specific UK life tables at the relevant target age. Consistent with Survey 1, accuracy is built from the signed gap (subjective − objective) and, as an alternative, a standardized discrepancy index based on the ratio of subjective to objective probabilities (censored at 1), defined as one minus accuracy and standardized by its standard deviation. The own-survival measure is the main outcome; the population-survival measure is a complementary outcome. The primary test is whether the improvement in accuracy induced by the survival-literacy treatment in Survey 1 persists about 2 months later—i.e., whether respondents assigned to S (and S+F) retain more accurate beliefs than control.

2. Persistence of treatment knowledge of financial-literacy concepts, and financial choices. Retention of the financial-literacy content is measured with four comprehension vignettes covering: the risk–return trade-off and expected returns; the role of the investment horizon; the non-guaranteed nature of past returns; and volatility as a measure of risk. The knowledge outcome is the share/number of correct answers. The primary test is whether respondents exposed to the financial-literacy treatment (F and S+F) in Survey 1 still answer these items more accurately than the remaining groups after about 2 months. Financial choices are captured by self-reported portfolio composition and market participation and by stated likelihood, over the next 12 months, of opening/topping up a Stocks & Shares ISA, investing in a mutual/index fund, seeking professional advice, and reviewing retirement allocation.

1. Demand for financial literacy
2. Long Term Care needs and perception

1. Demand for financial literacy. Captured by three sets of measures. (a) A revealed-preference behavioral measure: whether the respondent asks to receive the link to a free online personal-finance course. (b) Stated interest and willingness to invest time: willingness to participate; expected confidence with vs. without completing the course (an expected-benefit measure); the willingness-to-spend-time schedule. (c) Information-acquisition behavior: frequency of monitoring financial information and the set of information sources used in recent months. We test whether Survey 1 treatment assignment predicts higher demand for financial literacy about 2 months later.

2. Long-Term-Care needs and perception. Subjective probability of experiencing loss of autonomy at 10-, 20- and 30-year horizons. New to this wave, we elicit the subjective uncertainty around lifetime LTC risk: a min–max realistic range and a 20-ball distribution over five probability intervals. We also record exposure to LTC within the respondent's social network. We examine whether the treatments affect LTC central beliefs and/or perceived uncertainty.

The randomization comes from Survey 1, an online RCT administered to UK residents aged 50–70 recruited via Prolific in May 2026, with individual-level assignment, equal probability, to a 2 × 2 factorial design: placebo control, survival literacy (S), financial literacy (F), or both (S+F); control viewed a placebo video on payment methods. The present study re-contacts the same cohort about 2 months later and uses the Survey 1 assignment to identify the persistence of the treatments on survival expectations (own and population), financial knowledge/choices, and LTC needs and uncertainty.
Because treatment was randomly assigned, the across-arm comparisons at this wave identify the causal persistence of the treatment effects.

Not available

Randomization was implemented in Survey 1 at the individual level via Qualtrics' built-in random assignment, equal probability across the four groups. The present study relies entirely on that randomization (Dal Bianco et al. 2026, doi:10.1257/rct.18209-1.1); no new randomization is performed.

Individual-level randomization ((inherited from Survey 1)

No

Not applicable.

Approximately 3,200 individuals (subject to re-contact response).

Approximately 800 individuals in each of the four groups selected among Survey 1 respondents: control, survival literacy treatment only, financial literacy treatment only, and combined treatments.

Power assumes two-sided tests, α = 0.05, power = 0.80, individual-level randomization (no clustering). Main treatment effects pool two arms vs. two arms, giving approximately 1,280 vs. 1,280 at the planned analytic sample (640/arm assuming a retation rate of 80%). Survival (primary 1). Immediately after the video, the survival treatment moved the raw gap by about 2 percentage points. Comparing arm means alone, we could detect persistence only if roughly 87% of that effect survived (MDE approximately 1.7 pp) — i.e., only a barely-attenuated effect. Adjusting for the pre-treatment population belief, a covariate that strongly predicts the outcome, removes enough noise to bring the threshold down to about 1.4 pp (planning scenario), so we remain powered even if the effect has faded to roughly two-thirds to three-quarters of its original size. A more optimistic covariate assumption or the 800/arm ceiling lowers the MDE slightly further; the control-arm SD as a conservative input raises it to about 1.5 pp — none of which changes the conclusion. The same logic applies to the population-survival gap, where the pre-treatment baseline is of the same construct and if anything stronger. The MDE refers to the pooled main effect; a single-arm contrast (e.g., survival vs. control alone) is powered only for the full 2 pp effect. Financial-literacy knowledge (primary 2). The outcome is the share of correct answers on the four comprehension vignettes. Because these items are new to the follow-up, there is no same-construct Survey 1 baseline to adjust for, so — unlike the survival outcome — no covariate reduction is applied (a deliberate, stated asymmetry between the two primary outcomes). Using a provisional standard deviation of 0.30, the design detects a main effect of about 3.3 percentage points in the share correct. This value is provisional; the items were written to be harder than the post-treatment attention checks (which approximately 93% of treated respondents passed) to avoid a ceiling effect that would erode power, and the standard deviation will be updated from the pilot. Most-biased subgroup (best-powered persistence test). The immediate effect is highly concentrated: among the third of respondents with the most downward-biased pre-treatment population belief (bottom tertile of that variable), the Survey 1 effect is 7.56 pp — nearly four times the full-sample average — with a within-arm SD of approximately 17. Restricting to this tertile cuts sample size by roughly three (213 per arm at 80% retention, 426 vs. 426 in the main-effect contrast), which raises the MDE to about 3.3 pp. Because the effect grows far more than the MDE, this is the study’s most detectable persistence test: it remains significant even if only about 44% of the immediate effect survives, versus about 70% for the full-sample primary. This makes it the analysis most tolerant to the two-month attenuation we are most concerned about, and we pre-specify it accordingly. Two points are noted: the subgroup is defined on a pre-treatment variable, so randomization holds within it and the contrast is an unbiased causal effect; and re-contact must be tracked within the tertile, since the most mis-calibrated respondents may also attrit differentially. Interaction effects in the 2 × 2 design require far larger samples and are exploratory.