Respondents to survey questions involving sensitive information, such as sexual behavior, illegal drug usage, tax evasion, and income, may refuse to answer the questions or provide untruthful answers to protect their privacy. This creates a challenge in drawing valid inferences from potentially inaccurate data. Addressing this difficulty, non-randomized response approaches enable sample survey practitioners and applied statisticians to protect the privacy of respondents and properly analyze the gathered data.
Incomplete Categorical Data Design: Non-Randomized Response Techniques for Sensitive Questions in Surveys is the first book on non-randomized response designs and statistical analysis methods. The techniques covered integrate the strengths of existing approaches, including randomized response models, incomplete categorical data design, the EM algorithm, the bootstrap method, and the data augmentation algorithm.
A self-contained, systematic introduction, the book shows you how to draw valid statistical inferences from survey data with sensitive characteristics. It guides you in applying the non-randomized response approach in surveys and new non-randomized response designs. All R codes for the examples are available at www.saasweb.hku.hk/staff/gltian/.
Guo-Liang Tian is an associate professor of statistics in the Department of Statistics and Actuarial Science at the University of Hong Kong. Dr. Tian has published more than 60 (bio)statistical and medical papers in international peer-reviewed journals on missing data analysis, constrained parameter models and variable selection, sample surveys with sensitive questions, and cancer clinical trial and design. He is also the co-author of two books. He received a PhD in statistics from the Institute of Applied Mathematics, Chinese Academy of Science. Man-Lai Tang is an associate professor in the Department of Mathematics at Hong Kong Baptist University. Dr. Tang is an editorial board member of Advances and Applications in Statistical Sciences and the Journal of Probability and Statistics; associate editor of Communications in Statistics-Theory and Methods and Communications in Statistics-Simulation and Computation; and editorial advisory board member of the Open Medical Informatics Journal. His research interests include exact methods for discrete data, equivalence/non-inferiority trials, and biostatistics. He received a PhD in biostatistics from UCLA.
Introduction Randomized Response Models Item Count Techniques Non-Randomized Response Models Scope of the Rest of the Book The Crosswise Model The Warner Model A Non-Randomized Warner Model: The Crosswise Model Bayesian Methods for the Crosswise Model Analyzing the Induced Abortion Data An Experimental Survey Measuring Plagiarism The Triangular Model The Triangular Design Comparison with the Warner Model Asymptotic Properties of the MLE Bayesian Methods for the Triangular Model Analyzing the Sexual Behavior Data Case Studies on Premarital Sexual Behavior Sample Sizes for the Crosswise and Triangular Models Precision and Power Analysis Methods The Triangular Model for One-Sample Problem The Crosswise Model for One-Sample Problem Comparison for the Crosswise and Triangular Models The Triangular Model for Two-Sample Problem An Example The Multi-Category Triangular Model A Brief Literature Review The Survey Design Likelihood-Based Inferences Bayesian Inferences Questionnaire on Sexual Activities in Korean Adolescents The Hidden Sensitivity Model Background The Survey Design Likelihood-Based Inferences Information Loss and Design Consideration Simulation Studies Bayesian Inferences under Dirichlet Prior Bayesian Inferences under Other Priors Analyzing HIV Data in an AIDS Study The Parallel Model The Unrelated Question Model A Non-Randomized Unrelated Question Model: The Parallel Model Comparison with the Crosswise Model Comparison with the Triangular Model Bayesian Inferences An Example: Induced Abortion in Mexico A Case Study on College Students' Premarital Sexual Behavior at Wuhan A Case Study on Plagiarism at The University of Hong Kong Discussion Sample Size Calculation for the Parallel Model Sample Sizes for One-Sample Problem Comparison with the Crosswise Model Comparison with the Triangular Model Sample Size for Two-Sample Problem An Example The Multi-Category Parallel Model The Survey Design Likelihood-Based Inferences Bayesian Inferences A Special Case of the Multi-Category Parallel Model Comparison with the Multi-Category Triangular Model An Example Discussion A Variant of the Parallel Model The Survey Design and Basic Properties Statistical Inferences on ? Statistical Inferences on ?? Bootstrap Confidence Intervals Bayesian Inferences Comparison with the Crosswise Model Comparison with the Triangular Model The Noncompliance Behavior An Illustrative Example of Sexual Practices Case Studies on Cheating Behavior in Examinations Discussion The Combination Questionnaire Model The Survey Design Likelihood-Based Inferences Bayesian Inferences Analyzing Cervical Cancer Data in Atlanta Group Dirichlet Distribution Appendix A The EM and DA Algorithms Appendix B The Exact IBF Sampling Appendix C Some Statistical Distributions References Author Index Subject Index