Metawisdom of the Crowd: Experimental Evidence of Crowd Accuracy Through Diverse Choices of Decision Aids

Abstract

The provision of information can improve individual judgments but also fail to make group decisions more accurate; if individuals choose to attend to the same information in the same manner, the predictive diversity that enables crowd wisdom may be lost. Decision support systems, from search engines to business intelligence platforms, present individuals with decision aids -- relevant information, interpretative frames, or heuristics -- to enhance the quality and speed of decision-making but potentially influence judgments through the selective presentation of information and interpretative frames. We describe decision-making as often containing two decisions: the choice of decision aids followed by the primary decision, and define \textit{metawisdom of the crowd} as any pattern by which individuals' choice of aids leads to higher crowd accuracy than equal assignment to the same aids, a comparison that accounts for the information content of the aids. The theoretical model accounting for aid bias and variance shows that an optimal distribution of aid usage can produce metawisdom based on the characteristics of aids within a collection. Three studies -- two estimation tasks (N=900, 728) and the nowcasting of inflation (N=1,956; across three collections) -- support this claim. Metawisdom emerges from the use of diverse aids, not through widespread use of the aids that induce the most accurate estimates. Thus, the microfoundations of crowd wisdom appear in the first choice, suggesting crowd wisdom can be robust in information choice problems. Given the implications for collective decision-making, the insights warrant future research investigations into the nature and use of decision aids.