We study the value of garbled survey methods as a tool to monitor harassment. Theory predicts that randomly switching reports that no harassment took place to reports that harassment did take place can improve information transmission by guaranteeing participants plausible deniability in the event they file an incriminating report. We evaluate this prediction in a phone-based survey of workers at apparel manufacturing plants in Bangladesh. We vary the survey method (direct or garbled), the degree of personally identifiable information (team id) associated with the report, as well as the
degree of rapport built with respondents. We find that garbling increases reporting of sexual harassment by about 306%, physical harassment by 295%, and threatening behavior by 56%. We also find a negative effect of attaching team id to the report. We use the improved data to assess policy-relevant aspects of harassment: How prevalent is it? What share of managers is responsible for the misbehavior? How isolated are victims? How do harassment rates compare for men and women? Based on the answers to these questions, we draw implications for decision-makers.