Indirect inference of sensitive variables with peer network survey

Abstract

Abstract Misreporting is a common source of bias in population surveys involving sensitive topics such as sexual behaviours, abortion or criminal activity. To protect their privacy due to stigmatized or illegal behaviour, respondents tend to avoid fully disclosure of personal information deemed sensitive. This attitude however may compromise the results of survey studies. To circumvent this limitation, this article proposes a novel ego-centric sampling method (ECM) based on the respondent’s peer networks to make indirect inferences on sensitive traits anonymously. Other than asking the respondents to report directly on their own behaviour, ECM takes into account the knowledge the respondents have about their social contacts in the target population. By using various scenarios and sensitive analysis on model and real populations, we show the high performance, that is low biases, that can be achieved using our method and the novel estimator. The method is also applied on a real-world survey to study traits of college students. This real-world exercise illustrates that the method is easy-to-implement, requiring few amendments to standard sampling protocols, and provides a high level of confidence on privacy among respondents. The exercise revealed that students tend to under-report their own sensitive and stigmatized traits, such as their sexual orientation. Little or no difference was observed in reporting non-sensitive traits. Altogether, our results indicate that ECM is a promising method able to encourage survey participation and reduce bias due to misreporting of sensitive traits through indirect and anonymous data collection.