Steve Yeh

Managers often have little discretion over pay, but they routinely decide who works on which projects. I study a manager who staffs projects by capping each worker’s effort on each project. Workers prefer some projects to others and work harder when their teammates do, but find spreading effort across projects costly. Optimal assignments are characterized by a single ranking of projects, and two worker-specific cutoffs on it determine where each worker contributes freely, is held below his desired effort, or is kept off a project team. Because the ranking is common to all workers, project teams are nested. Standard network centrality can get the ordering of generalists and specialists backwards because the links that make a worker valuable to one team also make spreading him across several teams costly. Workers instead sort by a ratio of two centralities that captures their comparative advantage in spreading versus concentrating effort. Assignment selects the network through which shocks propagate, and an improvement at one project can degrade every other project in the organization.

A principal assigns agents to tasks in a multi-task project. Agents tend to shirk at their tasks and rely on others’ efforts because effort on one task substitutes for effort on similar tasks. Assigning the same agent to more tasks mitigates free-riding, but convex effort costs prevent him from completing all of his tasks. For a class of projects, an optimal task assignment assigns each agent to a module comprising a task central to the project and all locally related tasks. For general projects, modular task assignments are approximately optimal with a performance guarantee that depends on a project's task structure.

In conventional network games, each player chooses one action that impacts all of her neighbors. We study players that select neighbor-specific efforts and face a convex total effort cost. Our main result highlights a striking reversal in comparative statics: When bilateral efforts are strategic complements (substitutes), shocks propagate through the network in a pattern befitting strategic substitutes (complements) in standard network games. This reversal, together with other fundamental changes in player behavior, suggests that retaining the single effort assumption when it is not appropriate may lead to poor policy recommendations.

I propose a stability concept for network formation under incomplete information. A network is plausibly pairwise stable if players’ beliefs are consistent with common knowledge of network stability. I provide a general condition under which plausibly pairwise stable networks exhibit clustering. In a simple model of vertical differentiation, I characterize minimal plausibly pairwise stable networks. I show that incomplete information can stabilize segregated and sparse networks that are unstable under complete information. I discuss how incomplete information resolves the tension between stability and efficiency in the presence of negative linking externalities.