A Comparison of Prevalence Estimates of the Self-Protective Response Behaviour Obtained with Different Randomized-Response Models

Abstract

Randomized response is an interview technique that provides confidentiality when answering sensitive questions. The response of the respondent depends on a randomizing device, making the true response of the respondent misclassified with predetermined probabilities. Despite this protection not all of the respondents comply with the rules of the randomized response method, consequently some of the respondents answer consistently no despite the outcome of the randomizing device. This response pattern is known as the self-protective-no response pattern. In this report four models used to handle randomized response data, namely the profile likelihood, the log-linear randomized response model, the item randomized response model and the zero-inflated poisson model, are used to discuses the behaviour of an additional parameter for the self-protective-no response pattern.