Incentivizing Teamwork: An Experiment

Participants are matched into two-person teams and play a repeated team production game. In each round each member chooses to work (costly) or shirk (free). Team output is highest when both work, lower when one works, and lowest when neither works; each member's bonus is scaled by team output. The intervention is the employer's incentive scheme, which varies across five between-subject treatments along two dimensions: timing of choices (simultaneous vs. sequential, where the second mover's carried-out action depends on the first mover's) and visibility of bonuses (public vs. privately, randomly drawn).
T1: simultaneous, equal public bonuses (benchmark).
T2: simultaneous, unequal public bonuses.
T3: simultaneous, randomly drawn private bonuses.
T4: sequential, unequal public bonuses; the second mover states a plan conditional on the first mover's action.
T5: move order and bonuses jointly randomly drawn and not revealed before choosing; if the first mover shirks, the second mover's action is set to shirk.

2026-12-01

2027-05-15

1. Frequency of implementing joint work (FIW): the share of team decisions in which both team members work.
2. Cost-effectiveness (CE): average FIW divided by the employer's average realized bonus spending on the same team decisions.
Both outcomes are computed once for each matching group, over the confirmatory rounds.

A team decision is one round of play by one matched pair (raw data contain two records per decision; one is kept). The carried-out choice is the action actually implemented: in T4 the part of the second mover's plan that applies to the first mover's actual choice; in T5 shirk for the second mover if the first mover shirks, otherwise the submitted choice. Realized spending is the sum of both members' bonuses in force, scaled by team output. FIW = joint-work decisions / all decisions; CE = joint-work decisions / total realized spending (equivalently FIW divided by average spending). Treatment values are simple averages over matching groups.

Shares of decisions with exactly one / neither member working; individual work rates by role and own bonus; stated beliefs about the partner's choice; consistency of choice with stated belief (and direction of inconsistency); for T4 first movers, the gap between the two conditional beliefs; changes across rounds.

Shares of one/neither working use carried-out choices (for T5 also submitted choices). Belief consistency: given the member's own bonus and the treatment's payoff rules, the expected earnings from working and shirking are computed using the stated belief as the probability that the partner works. A choice is consistent unless the other action gives strictly higher expected earnings. T4 second movers are not classified. In T5 the belief concerns the partner's submitted choice. The T4 belief gap is the conditional belief that the second mover works if the first mover works, minus that if the first mover shirks.

This is an in-lab experiment with five between-subject treatments. Treatments T2–T5 form a 2×2 (timing × bonus visibility) design; T1 is a benchmark. Each participant takes part in only one treatment. Participants are randomly assigned to matching groups of 8 and to roles within their group, and are re-matched within their group each round. The production technology, cost of working, matching, feedback, number of rounds, and shared wording are held fixed. Belief reports are elicited in unannounced rounds and paid with a binarized scoring rule; social value orientation, risk preferences, and demographics are collected afterwards.

Not available

Treatments are assigned by the laboratory schedule; participants sign up without knowing the treatment. oTree randomly assigns participants to matching groups and roles, and randomly draws the T3 bonus pairs, the T5 branches, and the paid rounds.

Matching group.

Yes

30 matching groups (6 per treatment) of 8 participants each. The matching group is the unit of inference.

240 participants (48 per treatment) in 30 matching groups of 8. Participants are students recruited from the laboratory's subject pool.

T1: 48 participants, 6 matching groups
T2: 48 participants, 6 matching groups
T3: 48 participants, 6 matching groups
T4: 48 participants, 6 matching groups
T5: 48 participants, 6 matching groups

Power was assessed by Monte Carlo simulation (1,000 simulated experiments per scenario, 5% significance level) of the non-parametric tests on matching-group outcomes, with 6 matching groups per treatment. Choices were generated from the logit quantal response equilibrium (QRE) of each treatment's game, with precision parameter lambda (a higher lambda means less noise), plus normally distributed participant and matching-group random effects on the logit scale (baseline standard deviations 0.5 and 0.3). The Holm-corrected comparisons of T1 with each of treatments T2 to T5 have power of at least 0.998 for lambda of 0.3 or higher. The requirement that all three adjacent cost-effectiveness comparisons (T2 versus T3, T3 versus T4, and T4 versus T5) reject after Holm correction has power 0.802 at lambda = 0.4 and at least 0.979 for lambda of 0.5 or higher; the comparison of treatments T4 and T5 is the binding one. At lambda = 0.4, this power falls below 0.8 when matching-group heterogeneity is larger than in the baseline, or when participant heterogeneity is larger and group heterogeneity is at least at baseline. Details are given in the analysis plan.