How does the communication of scientific uncertainty in times of changing evidence affect people's trust over time?

The survey experiment will be set up as follows: First, all participants will receive the following introductory information:

"Suppose that it is the year 2030 and the world is faced with a new global pandemic that poses significant health risks to those infected with the virus. In the early phase of this new pandemic, the World Health Organization issues the following information:"

Participants will then be randomised into one of two health messages, one not acknowledging scientific certainty and one explicitly acknowledging scientific uncertainty.

Group 1 - Not acknowledging uncertainty: "The World Health Organization currently recommends that everyone should only drink bottled water or boil tap water for at least one minute to make it safe to drink. Sticking to these drinking water guidelines can best protect you from infection with the new virus, which transmits primarily through contaminated water."

Group 2 - Acknowledging uncertainty: "The World Health Organization currently recommends that everyone should only drink bottled water or boil tap water for at least one minute to make it safe to drink. Sticking to these drinking water guidelines can best protect you from infection with the new virus, which transmits primarily through contaminated water.\\\\It is possible that this recommendation may change because we are learning about the new pandemic in real time. This is expected since science is all about discovery, testing, learning, and adapting. We will do our best to update you with emerging evidence and what the new information means for you."

In the next step of the experiment, participants first receive the following information:

"Three months after this first recommendation, the spread of the virus is still not contained. At this point, the World Health Organization issues a new recommendation."

Following this information, we will show a second health message but with behavioural recommendations that are different from those outlined in the first health message. Again, the updated information will either not acknowledge or acknowledge scientific uncertainty. Specifically, participants will be randomised into seeing one of the following two messages:

Group 1 - Not acknowledging uncertainty: "The World Health Organization has updated their previous recommendation and currently recommends that people should wash their hands with soap and water for at least 20 seconds and use hand sanitisers that contain at least 60% of alcohol. We also recommend sanitising high-touch surfaces such as doorknobs and handrails. Practicing hand and surface hygiene can best protect you from infection with the new virus, which transmits primarily via direct contact from hands to a person’s nose or mouth and not via contaminated water."

Group 2 - Acknowledging uncertainty: "The World Health Organization has updated their previous recommendation and currently recommends that people should wash their hands with soap and water for at least 20 seconds and use hand sanitisers that contain at least 60% of alcohol. We also recommend sanitising high-touch surfaces such as doorknobs and handrails. Practicing hand and surface hygiene can best protect you from infection with the new virus. This change was due to new scientific evidence showing that the virus transmits primarily via direct contact from hands to a person’s nose or mouth and not via contaminated water. It is possible that this recommendation may change as we learn about the pandemic in real time. We will do our best to let you know if new information comes along that may lead us to change our recommendation and what the change means for you."

For each of the messages, the survey will be programmed so that participants will be unable to progress to the next screen for twenty seconds to ensure they take sufficient time to read through and process the presented information.

2024-01-30

2024-04-30

The experiment has two main trust-related outcomes: (1) participants' trust in the content/information of the recommendations; and (2) participants' trust in the sender/people formulating the recommendation.

The variables will be measured with the following questions:

Trust in content: "How trustworthy do you think this (updated) recommendation is?"
Trust in the sender: "How trustworthy do you think the organization that formulated this (updated) recommendation is?"

Both questions will have options ranging from 1 (not at all trustworthy) to 7 (very trustworthy). We will ask these questions after each round of the experiment to measure the initial trust and change in trust (i.e. updated trust).

We will capture two secondary outcomes: (1) intent to adopt the recommended behaviours; and (2) intent to share the received information.

The variables will be measured by means of the following questions:

Behavioural intent: "How likely is it that you would adopt the recommendation of drinking only bottled or boiled water?" (for first message) / "How likely is it that you would adopt the recommendation of sanitising hands and high-touch surfaces?" (for second message)

Intent to share information: "How likely is it that you would share this information with your friends and family?"

Both questions will have options ranging from 1 (Not at all likely) to 7 (very likely). Similar to the primary outcomes, we will ask these questions after each round of the experiment.

In the first stage of the experiment, we will randomise 2/3rds of the participants into the certainty communication arm and 1/3rds into the uncertainty communication arm. In the second stage, we will randomise those who initially received the certainty communication into either updated certainty communication or updated uncertainty communication in a 1:1 ratio. Those who initially received the uncertainty communication will all receive the updated uncertainty communication in the second stage. This leads to three experimental groups in a 1:1:1 ratio: (i) certainty in the first and certainty in the second message (the reference group), (ii) uncertainty in the first and uncertainty in the second message, and (iii) certainty in the first and uncertainty in the second message.

Not available

Computer-based randomization in survey software Qualtrics

Individual survey participants

No

NA

4000-5000 individuals per country

1000-2000 individuals per experimental arm

Our main outcomes are Likert measures of trust that range from 1 to 7. To be conservative, we powered our study under a dichotomised version of the Likert response with the aim of detecting a 5 percentage point effect. Thus, with a binary outcome, a control proportion of 50\% (the most conservative assumption), a two-tailed alpha=0.05, and 80% power, we would require a sample of 1565 individuals per experimental arm to detect a 5 percentage point effect. This would lead to a required total sample size of 4695 individuals per country.