Information Elicitation from Teams of Privacy-Conscious Experts

Abstract

Companies often inform their decisions by eliciting votes from teams of internal experts – for example, their employees. However, such experts may vote against their true beliefs if pressured by powerful stakeholders, whose interests may be misaligned with the company’s best interest. The experts' desire to protect their true beliefs from such stakeholders presents a privacy concern and makes eliciting their information a challenge. This paper explores how such privacy concerns can be addressed by designing appropriate vote elicitation processes. We build a parsimonious economic model of a firm making a binary decision based on binary votes from a team of experts. The latter observe private noisy signals about an unknown state of the world and decide whether to vote truthfully based on their signals. The experts weigh (i) the benefit of making the firm’s decision more accurate by voting truthfully against (ii) the cost of losing privacy by exposing their true beliefs. We show that without intervention from the firm’s side, privacy concerns undermine the conventional “wisdom of crowds” logic – in which having a larger pool of experts strictly improves the quality of the firm’s decisions. Instead, when experts are privacy-conscious, the firm prefers a finite team size, since larger teams unravel due to free-riding by the experts. Because of this issue, the firm cannot always use the insights of all available experts. To alleviate this problem, we propose combining two mechanisms – adding noise to experts’ votes and paying bonuses for vote accuracy. Our analysis shows that using these mechanisms may restore truthful voting and ensure that the firm can make use of all available experts. Our numerical extensions further support the practicality of our approach, suggesting that in certain settings, even complete anonymization – a seemingly simpler and more practical option – is outperformed by the mechanisms we propose.

Keywords: differential privacy, information elicitation, privacy preservation, voting, wisdom of crowds

JEL Classification: C72, D81, D82, M20, M10