First-hand information for everyone
This paper studies how organizations should use peer monitoring when people work on different but related tasks. The authors build a simple model in which a manager (the principal) assigns agents to points along a line of tasks. Each agent can pay a cost to learn the true state of the task they work on and then report it. The manager herself knows the state at one point on the line—her own area of expertise—but not the others. Nearby tasks tend to have similar states, so an agent’s work can help judge nearby peers.
The authors ask what monitoring rules make every agent choose to work, no matter what others do. They focus on “robust” contracts that make effort the unique rational outcome. In plain terms, that means designing pay so that shirking is never a better choice after iteratively ruling out strictly bad strategies. In the model, an agent’s report can affect a peer’s pay, and the manager picks which reports count toward which wages. Local states are linked by a simple statistical rule so that reports from close-by tasks are more informative.
A key result is that robust contracts create a monitoring hierarchy. Agents closest to the manager’s expertise are easiest for the manager to check directly. Those agents are therefore given the strongest authority to evaluate others. Information flows downward in a chain: the manager disciplines an agent near her, that agent’s report becomes credible, and it is then used to discipline a more distant agent, and so on. The chain structure avoids circular dependence in incentives and makes the cost of keeping everyone working add up in a simple, local way.
That chain has consequences for how tasks should be assigned. Spreading agents out would maximize the value of the team’s information, because different locations cover more ground. But peer monitoring works better when agents are close to the people who evaluate them. The robust requirement therefore pulls task assignment inward: an organization over-invests in tasks near the manager’s core expertise and under-explores novel or distant tasks. In other words, the need for a credible hierarchy biases knowledge production toward familiar ground.
The paper also shows one way to ease this bias. If task assignments are kept confidential and random, agents do not know the realized monitoring order for sure. That uncertainty lets the manager promise rare but large bonuses tied to possible monitoring links. With low probability of those rare monitoring states, expected payments fall toward the minimum needed for incentives, while information coverage approaches the ideal. Finally, the authors contrast their robust-design approach with standard equilibrium-based contracts: allowing cycles or mutual monitoring can lower payments but risks multiple, fragile equilibria where work may fail to be produced.
Important caveats follow from the paper’s setup. The results come from a stylized model: tasks lie along a one-dimensional line, local states follow a Markov-style correlation, and the main solution concept is unique rationalizability (iterated elimination of dominated strategies). These assumptions drive the chain outcome and the specific trade-offs. The TeX excerpt provided may be truncated, so further technical details and extensions in the full paper could add nuance. Still, the analysis highlights a clear tension between producing diverse knowledge and maintaining credible peer evaluation.