Mechanisms for Making Accurate Decisions in Biased Crowds

This paper studies procedures for identifying the true answer to a binary question using the opinions of potentially-biased individuals. It’s common and natural to side with the majority opinion, but the majority may make the wrong choice when the agents are biased. Taking majority rule as a baseline, I study peer-prediction decision rules, which ask agents to predict the opinions of others in addition to providing their own. This extra information enables us to potentially recognize the correct answer even when the majority is wrong. I first show that peer-prediction rules cannot be more accurate than the majority when we require them to satisfy the same symmetry conditions as majority rule and to be incentive-compatible for agents who intend to push the final decision towards their own opinion. Realistically though, not all agents distort their information strategically. I provide a simple decision rule based on the median agent’s prediction that matches majority rule when all agents are strategic and makes more accurate decisions than majority rule when some agents are honest.