Can causal attribution help prevent adverse events? - Evidence from Lake Victoria

In a lab-in-the-field setting, we manipulate -- in a decision environment of complete, but imperfect information -- the two possible causal pathways, natural and human cause, that lead to material outcomes for participants, depending on their choices.

We test the primary hypothesis that participants that have chosen an action with higher expected reward and that are exposed to an adverse event are more likely to switch to the action with lower expected reward when the same adverse event arose due to a human cause than when it arose due to a natural cause.

In this study, outcomes are chance events co-determined by a participant's decision, the decisions of a designated co-player, and nature in the form of an urn draw. Participants have common knowledge of the game structure, including payoffs, rules, possible strategies, and probabilities, but they do not know all past or current actions of the designated co-player and of nature. We manipulate, through random assignment to co-players and urns, the cause that can lead to an adverse, zero-payoff event for a participant.

2025-05-26

2025-08-31

The main outcome variable is a switch from choosing action H ("flower" in the instructions) in Round 1 to choosing action L ("star") in Round 2 among subjects that chose action H in round 1 and experienced a zero payoff event.

The main outcome variable is a binary variable that is constructed from choice data in Round 1 (conditional on observing the participant choosing action H) and Round 2. The variable 'switch' is set to 1 if the decision in Round 1 was H and the decision in Round 2 was L and is set to zero otherwise.

Secondary outcomes are (1) the choices by participants in Round 1 and Round 2, (2) the subjective beliefs of participants about the causal structure underpinning a future outcome in Round 1 and Round 2; (3) participants' survey responses regarding experiences in the coupled systems, attribution of adverse events and multi-dimensional locus-of- control scale questions.

In the lab-in-the-field experiment, the participants take two decisions, each in the same stochastic environment. This environment consists of two dimensions, the binary choice of another player (called the Designated Player) and the binary draw from an urn. Random matching of participants with other players and the urn leads to a variation in causal histories that lead to events that are identical in terms of material outcomes. This variation enables the identification of the causes that make participants change their decision from round to round, in particular the decision that switch from action H in Round 1 to action L in round 2.

In each session, 21 fishers take part in a classroom-style experiment with a modicum of privacy in taking decisions. There are white balls and orange balls representing natural and human causes of events, respectively, and the balls contain within them the draws from nature (white ball, red or green coin) and the draws from the choices of another player in the room (orange ball, star or flower).
In Round 1, participants, after passing a comprehension test, take their decision on whether to play the action 'star' or 'flower' without knowledge of the urn draw and the choice of the Designated Player, who is the co-player of each participant. Payoffs in the round are determined by a Modified Chicken Game as in Diekert et al. (2025, JAERE), adjusted for the local currency (1 US$ is replaced by 100 KES and so on). After the decision, we elicit subjective beliefs about the choice of the Designated Player vis-à-vis the participant and the outcome of the urn draw for the participant. Participants are provided with a complete feedback after the round: Their own choice, the choice of the Designated Player, and the urn draw. leading to an outcome. There are eight possible histories, five of which result in a zero-payoff ("adverse") event.
An adverse event materializes if at least one of two conditions is fulfilled: (i) a red coin was randomly assigned to the participant (natural cause); (ii) both the participant and the designated co-player choose action H (anthropogenic cause). Under all other circumstances, a favorable event results: The payoffs for a player are KES 100 for the action L and KES 300 for action H. Before moving on to Round 2, participants experience the event, favorable or adverse, of Round 1 and are informed about their own action, that of the designated co-player (action L or H), and the urn draw (green or red coin). For Round 2, a new Designated co-player is determined through a random procedure and a new urn draw applies. Otherwise, Round 2 is identical to Round 1.

There are several randomizations carried out in the course of the experiment:
(1) Participants freely choose a seat with being able to observe the identifier assigned to the seat. The identifiers are randomly assigned to seats.
(2) For each session, the designated players for Round 1 and for Round 2 are randomly chosen by a computer from the set of identifiers.
(3) Computer-based urn draws determine the materialized draw for each participant.
(4) A coin is flipped to determine the payout-relevant round.

The randomization unit is
(1) the individual for identifiers and urn draws,
(2) the experimental session for the selection of the designated co-players and the payout-relevant round.

No

The planned number of individual observations is 638 individuals.

The planned number of individual observations is 638 individuals.

The target sample size for the main comparison is
(1) 76 individuals with action H in Round 1 being treated with the causal pathway HHg of the designated co-player having chosen action H and the urn draw being green and
(2) 76 individuals with action H in Round 1 being treated with the causal pathway HLr of the designated co-player having chosen action L and the urn draw being red.

Our main hypothesis is directional. Our power calculations are therefore based on a one-sided test of proportions. The minimum detectable effect size is a difference of 18 percentage points in the share of switches between causal pathways HHg and HLr (Diekert et al., 2024) at a power of 0.8 and significance level of 0.05.