On the Optimality of Diverse Expert Panels in Persuasion Games

We consider a persuasion game between a decision-maker and a panel of biased experts. The decision-maker prefers to take an action in [0, 1] that matches the underlying state but relies on the experts to learn the state. Each expert has his `ideal` action or `agenda` and may conceal unfavorable information. If the decision- maker can select the panel members based on their agendas, what panel would she choose? While common intuition favors diverse panel (as experts would restrict each other`s ability to alter information), Bhattacharya and Mukherjee (2013) presents an example where a `homogeneous` panel (either all have agenda 0, or all have agenda 1) is more conducive to information revelation than a `diverse` panel (where one expert`s agenda is 0 the other`s is 1). We analyze the optimal diversity in expert panels and show that under mild conditions, a homogeneous panel is optimal when the experts observe the state independently of each other. But if the observability of the state is correlated across experts, diverse panel may be optimal. Hence, the diversity of agendas must be considered in conjunction with the diversity of information sources, and it is never optimal to seek diversity in both dimensions.