Sample Survey Design On Dichotomous Sensitive Questions Investigation

Objective: When a question in a sampling survey is sensitive or highly personal, it is likely to lead either to refusals or to untruthful answers by using the traditional method of direct interview because of the respondent's fear of revealing their privacy, which makes it difficult to acquire the real character of the population. By ingenious use of a randomizing device, Warner (1965) showed that it is possible to estimate the proportion without the respondents revealing their personal status with respect to the sensitive questions and thus introduced a new method for the sensitive questions survey—randomized response technique(RRT). Over the past few decades, a number of modifications of Warner's method as well as several other new methods have been emerged in the literature of randomized response. By now, most of the RR procedures available in the literature are developed and studied with the restriction that the sample is selected by simple random sampling. In the applications of RRT on sensitive questions, the formulas for simple random sampling are abused when the sample is selected by stratified sampling, cluster sampling or other relatively complicated sampling methods. What's more, the study on assessing the reliability and validity of the investigation on sensitive questions with RRT is seldom reported. In this regard, we select four RRT methods of Warner model, Simmons model, Greenberg model and the improved RRT model, and aim to explore the feasibility of the methods to investigate sensitive issues with the sample selected by stratified sampling, cluster sampling or stratified duster sampling. Meanwhile, the reliability and stability of the methods are assessed by the test-retest reliability and the congruence reliability of the four RRT models.Method: Total probability formula and the theory of RRT was employed to deduce the formula for the estimator of the population proportion and its variance when the four RRT methods of Warner model, Simmons model, Greenberg model and the improved RRT model are applied to investigate sensitive issues with the sample selected by simple random sampling. Then we extend the formula to the situation when the sample is selected by stratified sampling, cluster sampling or stratified cluster sampling according to the theory of sampling techniques. In the following premarital sexual behavior investigation of the college students in the Dushu Lake campus of soochow university, we adopt the sample method of stratified cluster sampling and select 12 clusters of undergraduates and 8 clusters of postgraduates at the first step, then applied four RRT models to collect data of sensitive characters, and finally calculate the estimator of the population proportion and its variance with the formula we deduced. In addition, the test-retest reliability and the congruence reliability of the four RRT models are analyzed by SPSS so as to evaluate the reliability and feasibility of the methods.Conclusion: Combining the theory of RRT and that of sampling technique, we provide the method at the first time to calculate the estimator of the population proportion and its variance in sensitive issue survey under the situation of relatively complicated sample method such as stratified sampling, cluster sampling and stratified cluster sampling. The results of the four RRT models are consistent on the whole. The estimated populated proportions of those who have the sensitive character got by the four RRT models are 17.03%, 18.32%, 17.83%, 16.88%, with high test-retest reliability and congruence reliability, which indicate that the results of the methods are reliable. The randomized response technique has an extensive application in sensitive issue investigation on a large scale.