主 题： Recent Advances for Non-randomized Response Techniques:
The Parallel Model, A Variant and An Extension
主讲人： 田国梁博士 （香港大学统计与精算学系副教授）
研究领域：缺失数据分析、约束参数模型、变量选择、敏感性问题抽样调查、癌症临床试验与设计
主持人： 张琳教授
时
间：
2013
年
5
月
8
日下午
3
：
00
地 点： 湖南大学北校区信用研究中心
主讲内容摘要： Since the randomized response model to solicit sensitive information was proposed by Warner in 1965, it has been used in a broad range of statistical applications for surveys involving sensitive questions. However, the Warner model is limited in several ways including (i) a lack of reproducibility; (ii) a lack of trust from the interviewees; (iii) a higher cost due to the use of randomizing devices; and (iv) narrow range of applications. Recent developments of the non-randomized approach have shown the promise to alleviate or eliminate these limitations. Following a brief introduction of the Warner model and other randomized response model, we review the non-randomized crosswise model and the non-randomized triangular model. However, the crosswise and triangular models cannot be applied to situation where both {Y=0} and {Y=1} are sensitive. In addition, the triangular model still has a lower efficiency for some cases. Therefore, this article proposes a new non-randomized response model called the parallel model and corresponding statistical analysis methods. Theoretical and numerical comparisons show that the randomized parallel model is more efficient than the triangular model for some cases. A variant of the parallel model and a multi-category parallel model are also developed.
