Research
Working papers:
Abstract: We consider a career concerns model in which an agent's productivity has two components: talent and responsiveness to incentives ("drive"). We show that the extra dimension of heterogeneity changes the behavior of agents and the structure of organizations in significant ways. First, since more driven agents are expected to work harder---and therefore be paid more---than less driven ones, everyone might be induced to work hard to signal their drive. These "drive-signaling incentives" are strongest with intermediate levels of explicit incentives. Over a long horizon, drive-signaling incentives have a tendency to bootstrap themselves, and, if this effect is strong enough, to create significant incentives with little else motivating the agent. On the other hand, signaling one's drive can be detrimental, because past outputs will be taken by the principal to reflect lower ability. Thus, drive-signaling incentives are likely to increase effort early in the career and decrease it later. In an application, we show that to maximize incentives the principal wants to observe an informal measure of the agent's effort (say, his hours worked) early, but not late in the career.
Abstract: In organizational, political, and financial settings, information is collected and reported by experts as it is received over time. This paper studies, in such dynamic situations, the incentives of an expert with reputational concerns to reveal his most recent information and the reporting protocol that induces the most truthful revelation. A principal receives sequential reports from an agent of privately known ability, who privately observes signals about the state of the world. The agent’s signals are of different initial quality and, in contrast to previous work, also of different quality improvement. First, when the talented agent also improves faster, “mind changes” (inconsistent reports) may be a sign of high ability, yet a mediocre agent still tends to repeat his early report. Second, requiring sequential reports creates an incentive to misreport the final, more accurate signal, but requiring a single report can only extract the agent’s final, not interim, opinion. As a result, sequential reports dominate when the principal’s optimal decision is very sensitive to the reports’ accuracy. A single report dominates when either the mediocre agent’s signals improve faster, or when the agent is very unlikely to be talented.
Abstract: This paper analyzes how the gossip process can be manipulated by biased people and the impact of such manipulation on information transmission. In this model, a single piece of information is transmitted via a chain of agents with privately known types. Each agent may be either objective or biased, with the latter type aiming to manipulate the information transmitted toward a given direction. In an indirect impact gossip model where all agents aim to influence a final decisionmaker, the biased type's equilibrium incentive to make up wrong information is independent of their position in the gossip chain. Moreover, adding just a few biased people to the population sharply decreases the amount of information transmitted. In a direct impact gossip model where every biased agent is concerned about influencing his immediate listener, gossip causes initial contamination of data, but eventually dies out as the objective people stop listening.