Vaccine sharing behaviour in the COVID-19 pandemic: the impact of narratives and peer effects

We introduce a 3x2 factorial design, where subjects are randomly assigned to one of six experimental conditions. In the first dimension, we present our subjects with facts appealing to either altruism or self-interest (plus control group) as a narrative for sharing vaccines. These facts are presented by means of a short video. Subjects then have the opportunity to share any amount of their earnings from the experiment as a donation to the vaccine access initiative COVAX. In the second dimension, we observe peer effects: after learning the average sharing behaviour of their peer group, subjects can revise their pledge. For all subjects, we also collect relevant covariates such as data on vaccine history, exposure to hardship during the pandemic, risk attitudes, inequity aversion, baseline level of altruism, and standard demographics.

2021-12-01

2022-05-31

Dependent Variable 1: Amount of money (in GBP) that subjects pledge towards COVAX, from their earnings on the experiment (initial and final pledge).

Dependent variable 2: Indicator that takes on the value 1 if subjects revise their donation, and zero otherwise.

All treatments + control are introduced to the survey and consent is confirmed. Treatments A, S, PA, and PS (see full pre-analysis plan for reference to treatment names) view informational video with altruism or self-interest narrative. All treatments + control group learn their earnings (8 GBP) and can make a pledge towards COVAX. Treatments P, PA, PS are informed that they are participating in a group (defined by 'within session' and 'within treatment'), and will see the group’s total and average-per-person donation subsequent to their pledge, while their own donation choice will remain confidential. Subjects are not informed that there will be an opportunity to revise the pledge. Treatments P, PA, PS are shown their group’s total and average-per-person donation. All treatments + control can revise their pledge. All treatments + control are asked to complete an unpaid survey.

Randomization done by OTree Software, by a computer.

Randomized unit: individual.

No

0

1500

250 subjects per treatment arm.

We perform a simulation-based, between-subject ANOVA power analysis in order to estimate the minimum sample sized needed to minimise the probability of incurring a Type II error. Our a priori calculation uses Caldwell and Läkens’ Superpower R package, whose algorithm relies on 10,000 Monte Carlo data set simulations with attributes specified by the researcher. We have defined the input parameters, mu = [2,1,4.5,2.5,5.5,4.5] {Each mean corresponds to the following condition, respectively: Control with Peer Effects, Control, Altruism Narrative with Peer Effects, Altruism Narrative, Self-Interest Narrative with Peer Effects, Self-Interest Narrative.} and sigma = 2.89$, based on a linear approximation of Metzger and Günther's (2019) experimental mean and standard deviation whose design is similar to ours. We adapt these numbers to our hypotheses. With a power score of 100 for the narrative factor, 100 for the peer effects factor, and 81.4 for the interaction between these two factors, a sample size of n = 250 per experimental condition suffices to reach the recommended power level (Cohen, 1988), when alpha = 0.05. Cohen’s f estimates of 0.5, 0.23 and 0.08 (respectively) coincide with benchmark big, medium and small effect sizes (0.4, 0.25, 0.1).