How Sharing Information Can Garble Experts’ Advice

Executives often employ experts who draw on rich information to recommend an appropriate action. For instance, a commander-in-chief is advised by several intelligence experts about how to respond to potential threats. When such interactions are discussed in policy circles, a common worry is that there is not enough sharing of information among experts, resulting in bad advice and inefficient outcomes. On the one hand, it is clear that combining information can result in better decisions. On the other hand, integrating experts’ capacities could potentially interact with their incentives in ways that harm a principal. In this article, we focus on the latter force, and describe a class of examples in which a principal is worse off when experts share information. Consider a situation where there is an unknown binary state of interest (e.g., whether a terrorist attack will be attempted). There is a correct decision for each state (e.g., whether to take costly preemptive measures). Each of several experts (e.g., the directors of the CIA and the FBI) chooses from a set of procedures, which generate information about the state. They then use that information to recommend an action to the executive, who decides which action to take. Each expert finds it costly to recommend the ex post incorrect action. The key payoff variables are the costs of two kinds of error: predicting the unusual event when How Sharing Information Can Garble Experts’ Advice †