The Dynamic Effects of Audits: An Experimental Assessment

Abstract

Governments and tax administrators around the world rely on the premise that both random and non-random audits will deter tax evasion. This paper is the first in the tax compliance literature to present experimental evidence collected from U.S. residents that this premise may be, at least in part, wrong. Counter-intuitively, I find that random audits encourage taxpayers to cheat more. Where audits were framed as being “random”, participants increased their levels of evasion in the tax periods immediately following the audit. This effect, however, did not plague audits that were framed for participants as being non-random. When a separate group of participants faced audits in which detected evasion could “flag” a participant for one or more future audits, participants cheated less in the periods immediately following the audit. Overall, average compliance in response to non-random audits systematically and significantly dominated average compliance in response to random audits. This result remained robust to the addition of a series of demographic controls as well as the inclusion of participant and time fixed effects. By revealing strong behavioral responses to the way tax audits are presented, this paper (1) opens important new avenues for experimental and controlled field trial research on audit dynamics, and (2) underscores the potential benefits to tax administrators of being more transparent with taxpayers about the nature of audit selection.