Generalized Randomized Response Questioning Designs

When questions on sensitive subjects, such as harassment at work, domestic violence, illegal employment, number of abortions, income, or voting behavior, are asked by direct questioning, nonresponse and untruthful answering will occur. As can be seen from Eq. (3.1), in the presence of both, the HT estimator t HT, for instance, is decomposed into three sums: one over the truthful answering set s t of sample s, another over the untruthful answering set s u , and a third over the missing set s m . Hence, such behavior by a respondent may cause serious problems in the analysis of sample and population data because the estimators of population parameters based only on a survey’s available cases may strongly be biased. It is therefore essential for data collectors to not ignore nonresponse or untruthful answering. Before applying such methods as weighting adjustment and data imputation (see Sects. 3.2 and 3.3) to compensate for nonresponse that has already occurred, data collectors should do everything to make the rates of both nonresponse and untruthful answering as small as possible.