How to Target Enforcement at Scale? Evidence from Tax Audits in Senegal

Developing economies are characterized by limited compliance with government regulation, such as taxation. Resources for enforcement are scarce and audit cases are often selected in a discretionary manner. We study whether the increasing availability of digitized data help improve audit targeting. Leveraging a field experiment at scale in Senegal, we compare tax audits selected by inspectors to audits selected by a risk-scoring algorithm. We find that inspector-selected audits are more likely to be conducted, similarly likely to detect evasion, and they detect higher amounts of evasion than algorithm-selected cases. On the other hand, algorithm-selected audits are faster and require less manpower. Selection on observables cannot explain the lower execution rate of algorithm-selected audits.