Website CV
Fields
Public, Law & Crime
Job market paper
Audits as Learning
A principal learns only from what she audits. Targeting low reports deters misreporting but leaves the top unobserved; random audits reveal it, at the cost of weaker deterrence. Under a fixed budget the two experiments are not Blackwell-comparable, so we price information by its purpose: the revenue lost by enforcing with a wrong belief. We derive an exact condition: randomizing pays if and only if one agent revealed at the top is worth more than the several an audit reveals at the edge of deterrence by making them truthful. Whether it holds hinges on whether the principal is uncertain about the scale or the location of types. On U.S. income survey data the condition fails: random audits do not pay.