When Tax Enforcement Changes: Social Learning and Compliance

Taxpayers rarely observe audit probabilities and must infer changes in enforcement from personal and social experience. We study this process in a laboratory tax-reporting experiment with 568 participants. Each participant faces hidden audit probabilities of 5 percent and 25 percent in randomized order, and we vary peer information across sessions from none to one or two preceding audit outcomes in a sparse network and three in a dense network. Before any peer outcome is transmitted, assignment to either network raises compliance by about 20% relative to no peer information. Once outcomes circulate, compliance in the dense network is approximately twice as responsive to enforcement as without peer information and about 60% more responsive than in the sparse network. This amplification is directional: relative to no peer information, the dense network raises compliance by about 12% after enforcement strengthens but lowers it by about 25% after enforcement weakens. Elicited belief distributions show that broader information reach improves learning about the changed enforcement environment. Higher perceived audit probabilities predict greater subsequent compliance, and the dense network's advantage comes from accumulating more peer signals rather than weighting each signal more heavily. Counterfactual policy exercises show that broader diffusion can reinforce deterrence under strong enforcement but erode it under weak enforcement; a model-based exercise suggests that full disclosure of the audit probability can reduce compliance under both weak and strong enforcement.