Alexander Clyde
Aalto University
Narrow Inference and Incentive Design
Abstract:
There is evidence that people struggle to do causal inference in complex multidimensional environments. This paper explores the consequences of this in a principal-agent setting. A principal chooses a mechanism to screen an agent. The agent makes choices on multiple dimensions, and infers the effect of each action separately without properly controlling for the other actions. I fully characterize the principal’s optimal mechanism when facing an agent who does such ‘narrow’ inference, and contrast it with their optimal mechanism when the agent is fully rational. I demonstrate when the principal can exploit narrow inference and in what cases they lose out.