On the acceptance of information avoidance

Design Overview:
We study when and why third parties are willing to intervene to ensure that another person makes an informed decision when their actions can impose a negative externality on another person. Third-party decision makers (“Spectators”) are randomly assigned, in a 2×3 between-subjects design, to evaluate cases involving a second person (“Person 1”).

Factors and Levels:

Factor 1 – Person 1’s information preference
- Person 1 prefers to be informed (Prefers information)
- Person 1 prefers not to be informed (Prefers no information)

Factor 2 – How providing information affects Person 1’s choice
- Information would change the choice in a pro-social direction (Impact—Pro-social)
- Information would not change the choice; the resulting choice is selfish (No impact—Selfish)
- Information would not change the choice; the resulting choice is pro-social (No impact—Pro-social)

By combing the two factors we have 6 treatment arms:
T1: Prefers information × Impact—Pro-social
T2: Prefers information × No impact—Selfish
T3: Prefers information × No impact—Pro-social

T1*: Prefers no information × Impact—Pro-social
T2*: Prefers no information × No impact—Selfish
T3*: Prefers no information × No impact—Pro-social

Primary Focus and Contrasts
Our primary interest is Spectators’ willingness to intervene and provide information against Person 1’s preference to remain uninformed. Accordingly, we focus on the three arms where Person 1 prefers no information (T1*, T2*, T3*) to evaluate research question R1 and R2:

R1. Instrumental effect of information on choice:
T1* vs T2* (does the willingness to impose information increase when it would make the action more pro-social?).

R2. Dependence on the realized choice when impact is zero:
T2* vs T3* (does the willingness to impose information depend on whether the choice is selfish or pro-social?).

The three arms where Person 1 prefers information (T1, T2, T3) mainly serve as benchmarks to gauge any general reluctance to intervene even when the recipient wants the information, and to answer our third research question.

R3. Dependence on Person 1’s information preference
T1 vs T1* / T2 vs T2* / T3 vs T3* (does the willingness to impose information depend on whether the dictator prefers information, within treatments?).

2025-09-05

2025-09-15

The key outcome variable is a dummy variable equal to one if the Spectator choose to intervene and provide information to Person 1 (and zero if deciding not to intervene).

The secondary outcomes can be divided in two.

First we ask a couple of follow-up question to shed light on the decision made by the spectator. For example, there will be an open-ended question where they can outline their reasoning, and there will also be a question eliciting Spectators belief about the choice of Person 1.

Second, to shed light on external validity of the main findings, all participants also take part in a Vignette experiment, after the main experiment, where they are asked whether they would support a hypothetical mandatory information policy or not.

The experiment is consisting of two separate studies, which we here refer to as the "Preference-study" and the "Intervention-study".

The Preference-study is a so-called dictator game where participants are assigned to either the role of the dictator (Person 1) or the recipient (Person 2). In a within-subject design we then elicit Person 1's preference to be informed about the payoff consequences for Person 2, and how being (un)informed affects the allocation decision of Person 1. The sole purpose of the Preference-study is to ensure that we can provide truthful information about a decision that may have real consequences to the decision-maker in the Intervention-study. More details of the Preference-study will be provided in the experimental instructions of the paper.

The Intervention-study is the study that this trial refers to unless anything else is explicitly stated. It is based on a so-called spectator design where decision-makers, in the role of third party observers, are recruited to make a decision that may have real consequences for the final payoffs to participants in the Preference-study. The task for the decision-maker (or "Spectator") is to make a choice of whether to "Intervene" (i.e. inform Person 1 about the payoff consequences for Person 2) or "Not intervene" (i.e. not inform Person 1 about the payoff consequences for Person 2). We will use a 1:20 matching protocol, meaning that 20 spectators is matched to one Person 1, and one of the 20 spectator choices will actually be implemented.

As described in the Intervention section we randomly assign Spectators to one of 6 treatment arms in a 2*3 between-subject design that varies whether the Person 1 in the Preference-study prefers to be informed about the recipient's payoff or not, and whether information impacts the choice of Person 1 (Impact: Pro-social vs. No impact: Selfish vs. No impact: Pro-social).

Both studies are programmed in Qualtrics and use participants from Prolific. We ensure that participants are not taking part in both studies.

Not available

Randomization in Qualtrics.

Individual.

No

We plan to recruit a total of 3900 spectators from the United States. Recruitment is made via Prolific, targeting a sample that is representative of the US population on key background variables such as gender and age.

Note: For the main analysis we will restrict the sample to those participants that correctly answer a set of comprehension questions. Based on a pilot study, about 75-70% of participants answer these correctly, meaning that the total sample used in the main analysis will be at least 2700. As a robustness test we will report the results based on all participants in the appendix of the paper.

See above.

We recruit twice as many observations to our main treatment arms T1*-T3*, relative to T1-T3. Hence we aim for the following sample size per treatment arm:

800 observations (Spectators) per treatment arm T1*-T3*, out of which about 600 will pass the comprehension test and therefore be used in the main analysis.

400 observations (Spectators) per treatment arm T1-T3, out of which about 300 will pass the comprehension test and therefore be used in the main analysis.

Using a two-sided t-test (α = 0.05; 80% power) with a binary outcome (SD = 0.5), the minimum detectable effect sizes for the respective treatment comparisons are: T1* vs T2* and T2* vs T3*: With 600 vs. 600 observations, we can detect a difference of ~8.1 percentage points. T1* vs T1 / T2* vs T2 / T3* vs T3: With 600 vs. 300 observations, we can detect a difference of ~9.0 percentage points. T1 vs T2 vs T3: With 300 vs. 300 observations, we can detect a difference of ~11.4 percentage points.