Ostracism and imperfect observability of individual behavior in common pool resources

T0: No ostracism (Baseline). Our experiment treatments start with a simple CPR game as the baseline. This is a standard repeated linear common pool resource game, without the possibility of ostracism.

T1: Ostracism + Perfect observability. This treatment has the same structure as the Baseline treatment, but CPR members may vote to exclude individuals from the group. After all withdrawal decisions are submitted in each round, insiders can vote to remove one individual from the insider group under perfect information about current withdrawals. Excluded individuals only consume their own endowment in each round.

T2: Ostracism + Imperfect observability (High quality). This treatment is the same as T1 except that CPR group members receive a stochastic (high quality) signal of the other CPR group members’ current withdrawals. There is a 75% chance that the signal of an individual’s withdrawal is true; the remaining 25% is distributed uniformly over the other possible withdrawal levels.

T3: Ostracism + Imperfect observability (Low quality). This treatment is the same as T1 except that CPR group members receive a stochastic (low quality) signal of the other CPR group members’ current withdrawals. There is a 50% chance that the signal of an individual’s withdrawal is true; the remaining 50% is distributed uniformly over the other possible withdrawal levels.

T4: Ostracism + Perfect observability + Poaching. This treatment is same as T1, except that excluded individuals may withdraw from the CPR without any consequence for doing so.

T5: Ostracism + Imperfect observability (High quality) + Poaching. This treatment is the same as T4, but with the imperfect observability (high quality) feature of T2.

T6: Ostracism + Imperfect observability (Low quality) + Poaching. This treatment is the same as T4, but with the imperfect observability (low quality) feature of T3.

2026-04-09

2026-06-30

The primary comparisons are the main effects within the 3×2 ostracism treatments:
• the main effect of observability of others’ choices, and
• the main effect of outsider poaching.
In a balanced 3×2 factorial, each main effect compares the average of two treatment cells to the average of two other treatment cells (i.e., pooled marginal means). The effects evaluated will be:
• Exclusion/ostracism choices
• Withdrawal decisions by insiders & outsiders
• insider earnings
The baseline treatment (no ostracism) serves as a reference but does not contribute to identification of these factorial main effects.

outsider earnings
total earnings of insiders + outsiders

This is a laboratory experiment, conducted in a standard university experimental economics lab at the U of Talca, Chile. Students will be recruited from Universidad de Talca and, if needed, from other neighboring universities. Subjects will be randomly assigned to treatments. The experiment consists of a CPR game in which CPR members are paid from a shared account, but individuals may withdraw units from this account for their private benefit. The game is parameterized so that it is individually rational for individuals to withdraw as many units from the shared account as possible, but leaving the shared account intact maximizes group welfare. Except for the baseline CPR experiment, CPR members may vote (by majority rule) to exclude others.

In addition to a baseline treatment, our experiment follows a 3X2 design. The design varies observability by all subjects of current-period individual withdrawals from perfect observability of withdrawals, to subjects receiving high quality signals (described below) of individual withdrawals, to subjects receiving low quality signals of individual withdrawals. The design also varies whether excluded individuals may or may not poach from the resource.

Not available

Students will be recruited with emails, public announcements and classroom visits. They sign up for an available session. The session is randomly assigned a treatment before it begins.

Individual

Yes

7 treatments x 12 groups per treatment = 84 clusters

504 observations at the participant level in the main experiment. Because participants make repeated decisions over 40 rounds, this corresponds to 20,160 participant-round observations in total, although statistical power is based on the 84 independent groups rather than the repeated observations.

12 groups per arm (7 arms total):
• baseline (no ostracism)
• low-quality observability / no outsider poaching
• low-quality observability / outsider poaching
• high-quality observability / no outsider poaching
• high-quality observability / outsider poaching
• perfect observability / no outsider poaching
• perfect observability / outsider poaching
This yields 84 groups total. With 6 participants per group, that corresponds to 504 participants in the main experiment.

The primary outcome is individual harvest per round, measured in resource units on a 0–10 scale. Power calculations account for the grouped and repeated-measures design by treating the 6-person group as the unit of assignment and applying a design-effect adjustment for within-subject and within-group correlation across 40 rounds. Using a pooled planning standard deviation of 2.68 units (scaled from prior related data), two-sided α=0.05, and 80% power, the study is powered to detect an observability main effect of 0.80 units, which corresponds to a standardized effect size of approximately 0.30 SD (0.80/2.68≈0.30). Under these assumptions, the design uses 12 groups per treatment condition.