The Impact of Group Size on the Discovery of Hidden Profiles in Online Discussion Groups

Abstract

Individuals and organizations are turning to crowds to help make important decisions in diverse areas from health to software to data science. Though usually drawn from large populations, the number of users who contribute to a specific discussion may vary widely. Because crowds’ advantages – diverse knowledge and expertise – derive from their scale, we might expect that crowd performance should increase with size. On the contrary, this experiment reveals a task in which mid-sized groups outperformed both larger and smaller groups. Specifically, we compared information sharing and performance on a hidden profile problem across several crowd sizes. We found that individuals in medium-sized discussions performed the best, and we suggest that this represents a tradeoff in which larger groups tend to share more facts, but have more difficulty than smaller groups at resolving misunderstandings. We discuss design implications and ways to extend information exchange theories developed for traditional small groups to crowd discussions.