ABSTRACT
Consider, as in Chapter 1, the problem of estimating the proportion πA of population members of a sensitive group A. Closer examination reveals the following distinctive feature of the Warner model as applied to such a situation. Essentially, each respondent is supplied with two questions, of which he or she selects one at random and replies only to the question chosen. In the process the respondent’s privacy is protected since the interviewer is unaware of the question to which his or her response relates. In order that the procedure be relevant for the basic problem, that is, estimation of TEA, one of these two questions should relate to membership in A. In the Warner model, the other question relates to membership in the complementary group Ā. Thus the two questions relate to groups that are perfectly negatively associated.
