Abstract: | I study a principal–agent model in which the agent collects information and then chooses a verifiable action. I show that the principal can find it desirable to constrain the agent's action set even though there is no disagreement about the ranking of actions ex post . The elimination or penalization of „intermediate” actions, which are optimal when information is poor, improves incentives for information collection. I characterize optimal action sets when the agent is infinitely risk averse with respect to income shocks and optimal incentive schemes when the agent is risk neutral. |