Abstract
Asking sensitive questions by a direct survey method causes non-response bias and response bias. Non-response bias arises from interviewees refusal to respond and response bias arises from giving incorrect responses. To rectify these biases, Warner (1965) introduced a randomized response model which is an alternative survey method for socially undesirable or incriminating behavior questions. The randomized response model is a procedure for collecting the information on sensitive characteristics without exposing the identity of the respondent. Many survey researchers have proposed diverse variants of the Warner randomized response model and applied their model to collect the information of sensitive questions. Using an optimal allocation, we proposed three-stage stratified randomized response technique which is an extension of the Kim and Elam (2005) two-stage stratified randomized response technique. In this study, we showed that the estimator based on the proposed response model is more efficient than Kim and Elam (2005). But by adding one more survey step to the Kim and Elam (2005), our proposed model may have relatively less privacy protection compared to the Kim and Elam (2005) model.
직접면접으로 민감한 질문을 할때 발생하는 무응답이나 거짓응답의 문제를 개선하고자 Warner (1965)가 최초로 제안한 확률화응답모형에 관한 연구는 이후 많은 연구자들에 의해 개선, 발전되어 오고 있다. 본 연구에서 표본은 층화임의복원추출법에 의해 추출되었으며, 표본배분은 최적배분법에 의해 배분되었다. 한편, Kim과 Elam (2005)의 2단계 층화확률화응답모형을 확장한 3단계 층화확률화응답모형을 사용하였다. Kim과 Elam (2005)의 2단계 층화확률화응답모형과 상대효율을 비교한 결과 본 논문에서 제시한 3단계 층화확률화모형의 효율성이 상대적으로 높다는 결과가 도출되었다. 그러나 2단계확률화응답모형을 3단계로 확장함으로써 상대적으로 효율성은 증대되지만 반대로 조사과정의 어려움이 예상된다.
