Incomplete Categorical Data Design

'Preface Acquirement of sensitive information is often needed in a broad range of statistical applications. For instance, some behavioral, epidemiological, public health and social studies may need to solicit information on reproductive history, sexual behavior, abortion, human immunode ciency virus, acquired immune de ciency syndrome, illegal drug usage, family violence, income, child abuse, employee theft, shoplifting, social security fraud, premature sign-o s on audits, in delity, driving under in uence, having a baby outside marriage, tax evasion, and cheating in university examinations. When being directly asked these sensitive survey questions, some respondents may refuse to answer and some may even provide untruthful answers in order to protect their privacy. The problem becomes even more complicated with surveys in diverse populations because of the interaction of sensitivity and respondent diversity. It is therefore difficult to draw valid inferences from these inaccurate data that include refusal bias, response bias and perhaps both. It has long been a challenge to obtain such information while having the privacy of the respondent protected and the resulting data analyzed properly. Although there are a number of methods (see, e.g., Barton, 1958) for asking embarrassing questions in non-embarrassing ways, the rst ingenious interviewing technique to overcome the above di culties is the randomized response approach, proposed by Warner (1965), that aims to encourage truthful answers from respondents. The randomized response technique is designed to ask a sensitive question according to the outcome of a randomizing device while the interviewer is blind to the outcome'--