Modeling Self-Protective Responses in Randomized Response

Abstract

Randomized response (RR) is an interview technique developed to ask sensitive questions. Confidentiality is assured, since the answers to the questions depend partly on the respondent's true status and partly on the outcome of a randomizing device. Despite this protection not all respondents answer by the rules of the RR method and consistently give self-protective answers instead. The basic model that takes self-protective answering into account, assumes that this behaviour is independent of the true status of the respondent. Since this model shows a lack of fit, models assuming dependence between self-protective answering and the true status are studied using empirical data from a social welfare study. The results show that the model fit does not improve by using these models. Furthermore, the results show that the models give identification problems. It is concluded that the basic model is a robust model and is therefore preferred for now.