Accounting for lack of trust in randomized response models
- UNCG Author/Contributor (non-UNCG co-authors, if there are any, appear on document)
- Pujita Sapra (Creator)
- Institution
- The University of North Carolina at Greensboro (UNCG )
- Web Site: http://library.uncg.edu/
- Advisor
- Sat Gupta Gupta
Abstract: This study addresses a key assumption made while using traditional Randomized Response models in survey sampling when the question being asked pertains to a sensitive topic. It is traditionally assumed that under a randomized response framework, survey participants have no further reason to lie due to privacy concerns. We demonstrate that if this assumption is not true and even if a small proportion of respondents do not trust the RRT model being used in a survey, we get considerably biased estimates. We also propose alternative binary and quantitative models that account for respondents’ lack of trust in traditional RRT models. These proposed models are mixtures of traditional RRT models and in one particular case mixture of an RRT model with an encryption technique, commonly used in the computer science domain. We also incorporate optionality into these models which helps improve the model efficiency. We evaluate the overall model performance using a combined measure of privacy and efficiency. Both theoretical and empirical results confirm that accounting for lack of trust helps us obtain more reliable results when survey respondents may not trust the RRT model used. Simulation studies have also been conducted to verify theoretical results. For sensitive mean estimation, we also propose estimators that utilize the auxiliary information and are more efficient compared to the ordinary mean estimator that does not utilize the auxiliary information.
Accounting for lack of trust in randomized response models
PDF (Portable Document Format)
2486 KB
Created on 8/1/2023
Views: 232
Additional Information
- Publication
- Dissertation
- Language: English
- Date: 2023
- Keywords
- Lack of trust in RRT, Randomized response models, Respondent privacy, Unified measure of efficiency and privacy
- Subjects
- Sampling (Statistics)
- Surveys $x Statistical methods
- Trust $x Simulation methods