Trust, Autonomy & Paternalism: An Experimental Study

Using an online experiment, we study whether trust influences i) the willingness to accept a paternalistic intervention, or in other words, the willingness to decide and ii) the willingness to paternalistically intervene in another person's freedom of choice.

2025-10-13

2025-12-31

From Part 2 of the experiment:
- Participant A: Willingness to decide
- Participant B: Willingness to intervene

Both primary outcomes are binary variables (yes/no).

From Part 1 of the experiment:
Participant A:
- Amount sent in the trust game (= level of trust)
- Belief in amount returned (as a standard control variable in the trust game)
Participant B:
- Belief in the amount sent (= belief about Participant A’s level of trust)
- Amount returned in the trust game (= level of trustworthiness)
From Part 2 of the experiment:
Participant A:
- Belief about Participant B’s decision to intervene
- Confidence in own (potential) choice
- Confidence in Participant B’s (potential) choice
Participant B:
- Belief about Participant’s A decision to decide
- Confidence in own (potential) choice
- Confidence in Participant B’s (potential) choice

Part 1:
The trustor (=Participant B) will make their choices according to the strategy method.
Beliefs will be incentivized.
Part 2:
Belief about the decision of the other participant will be incentivized.
Confidence will not be incentivized.

We address the research question using a two-part experimental design: Throughout the experiment, participants are matched in groups of two and are randomly assigned the roles of Participant A or Participant B.

In Part 1 of the experiment, the level of trust (for Participants A) and the belief in the level of the other participants’ trust (for Participants B) are measured by decisions and via belief-elicitation in the “Trust Game” (à la Berg et al. 1995).

In Part 2 of the experiment, the paternalism game, Participants A states for a specific decision-making situation whether they want to make their own decision or if they want Participant B to decide for them. Participants B, who has perfect information about the decision-making situation, is asked whether they want to make the decision for Participant A. Participants B receive a fixed payoff independent of their decisions in Part 2. The specific decision-making situation only has payoff-relevant consequences for Participants A.

We implement two treatments between-subjects as induced in Part 1:

After reading the instructions and prior to making their decisions, participants receive the following information on a separate screen depending on the treatment (Control vs. Trust):

Treatment Control: Participants receive general information about the decision situation and the trust game.

Treatment Trust: In addition, participants receive information about the level of trust (for Participants B) or the level of trustworthiness (for Participants A) from prior sessions of the study.

Note: The information is based on two sessions from a pilot study of this study.

Our hypotheses are as follows:
Willingness to decide (Participant A): Trust decreases the willingness to decide.
Willingness to intervene (Participant B): Perceived trust increases the willingness to intervene.

Not available

Randomization done by a computer.
Ex-post matching done by a computer.

Individual level.

No

Treatment assignment is individual, but roles A and B are matched within dyads and exposed to each other. (In the data analysis, we control for the clustering at the dyad level with robust SEs that account for the fact that each A and B pair may be correlated.)

One observation per participant.

1400 participants in total with 700 participants per treatment.

Exclusion Criteria:
- Failed Bot-Check at the beginning of the experiment
- Failed attention checks from the post-experimental questionnaire
- Early termination of the experiment

The participant numbers are based on a priori power analysis with G*Power (Faul et al., 2007). This was calculated for a Fisher’s Exact Test. We assume small to medium effect sizes. We also assume a two-sided alpha level of 0.05 and a power of 1-β = 0.8. The calculation results in a required sample size of at least 1300 participants. Due to possible exclusions, we plan with a sample size of 1400 participants.