Striking to be Productive: An Experimental Study of Collective Exit Threats and Centralized Redistribution

2026-07-27

2026-08-07

1. Worker productivity per conducted working phase: number of correctly completed sliders per worker per working phase in which work took place (excluding strike rounds).
2. Total group output per round: sum of worker-generated ECU per group per round, averaged over all rounds including strike rounds (in which output is zero), before the employer multiplier.

Outcome 1 captures the effect of the strike threat on effort provision; Outcome 2 captures the net productivity effect including the output loss from realized strikes. Because the number of rounds is stochastic, both outcomes are normalized per round/phase.

Total group surplus per round including the employer multiplier (welfare measure; the redistributed pool); worker share of redistributed income per round and the employer's self-allocated share; cumulative payoffs by role; within-group payoff inequality (employer-to-mean-worker payoff ratio; Gini coefficient of cumulative payoffs across all 4 members); employer redistribution style measured by three continuous indices (within-round Spearman correlation between worker effort and allocation; coefficient of variation of worker allocations; average self-allocated share); strike rate, passed strike votes, individual voting, strike duration and timing (treatment only); share of worker-rounds below the 20-slider minimum; share of rounds in which the employer's clicker threshold is met; worker effort in round t as a function of the share received in round t−1; time trends and first-round effects. Post-experimental questionnaire (age, gender, field of study, general risk attitude 0–10) is used for covariates and balance checks.

Between-subject laboratory experiment with two conditions (Baseline, Collective Exit), one session per condition (12:00 and 16:00 on the same day), 60 participants (15 groups of 4) each. Within sessions, participants are randomly assigned to groups and to roles (1 employer, 3 workers), fixed for the entire session. Groups are the independent unit of observation; because each condition is run in a single session, treatment effects cannot be statistically separated from session-level effects, and inference relies on the 15 independent groups per condition. The experiment is computerized in oTree. The number of rounds is stochastic (minimum 10, then 20% ending probability per round). After the game ends, participants complete a short questionnaire (age, gender, field of study, general risk attitude).

Not available

Treatment is assigned at the session level in a fixed order: one Baseline session and one Collective Exit session, conducted sequentially on the same day at the WISO Research Laboratory of the University of Hamburg. Participants register via the laboratory's standard recruitment system without knowledge of the experimental condition. Within sessions, the oTree software randomly assigns participants to groups and roles.

Session (condition); individual (group and role assignment within session).

Yes

30 groups (15 per condition)

120 participants (90 workers, 30 employers); approximately 1,260 worker-round and 420 group-round observations at an expected game length of aprox. 14 rounds.

60 participants (15 groups) Baseline; 60 participants (15 groups) Collective Exit.

Unit of analysis: group-level means, 15 independent groups per arm (two-sided t-test, alpha = 0.05, power = 0.80), yielding a minimum detectable effect of approximately d = 1.06 standard deviations for both primary outcomes. Using standard deviations estimated from the pilot (4 groups; imprecise — one pilot group was a strong outlier in productivity, so these SDs are likely overestimates and the stated MDEs conservative): Outcome 1 (worker effort per conducted working phase), SD ≈ 24 sliders around a mean of ≈ 52, MDE ≈ 25 sliders (≈ 48% of the pilot mean); Outcome 2 (group output per round including strike rounds), SD ≈ 65 ECU around a mean of ≈ 136, MDE ≈ 69 ECU (≈ 51% of the pilot mean). Panel regressions at the worker-round level (≈ 1,260 observations, group-clustered standard errors) provide additional power for the effort and mechanism hypotheses.