Intimate Partner Violence, Reporting Bias, and Impact Evaluation: Evidence from Bangladesh

Abstract Survey responses might be biased by enumerator characteristics, which could in turn affect impact evaluation estimates. This study demonstrates this in the context of intimate partner violence (IPV) in Bangladesh, by cross-randomizing the gender of the enumerator eliciting responses, and experimentally varying the rollout of an established anti-poverty program. It finds that respondents significantly under-report (0.51 sd) cases of IPV to males compared to females. Furthermore, the intervention significantly reduces (0.20 sd) incidence of IPV. However, this impact is only discernible when surveyed through females. If only male enumerators are used, the impact of the program is masked due to response bias.