活动

We consider the organizational challenge of motivating experimentation when facing communication frictions. A principal relies on the information obtained through delegated experimentation to make decisions. Biased agent(s), who always favour higher decisions, strategically design the experiment and can privately tamper with its outcome at some cost when reporting it. The principal can (i) separate experimental design and reporting by allocating it to two different agents, or integrate both tasks in one agent, and (ii) set the intensity at which the outcome is audited. We show how these two organizational levers contribute more informative experimentation. First, we give conditions so that if different task allocations lead to experiments ranked according to their informativeness, then the principal always prefers to separate tasks. Second, an imperfect audit can, regardless of task allocation, lead the agent in charge of design to revert to a more informative experiment as it increases the approval probability when the outcome is unaudited. If the principal only needs to approve a project, then separation is always (weakly) optimal and the principal commits to an imperfect audit, even if she could perfectly audit the outcome at no cost. If instead the principal can deviate from a status quo by going to one of two extreme options, then integration of tasks can be optimal. This model provides a rationale for lax auditing of information and initiatives to centralize data acquisition in organizations.