Modeling of Sensitive Behavior Using Additive Data Masking: A Simulation-Based Approach for Privacy-Preserving Regression

Abstract

It is difficult to model sensitive survey behaviors because of privacy concerns, stigma, or legal risks that prevent honest disclosure. This paper presents a new regression method that integrates additive data masking into Simple Linear Regression while maintaining privacy. In order to protect respondent privacy while maintaining reliable statistical inference, we use Warner's Randomized Response Technique to introduce additive noise into sensitive responses. We show theoretically that even when the dependent variable is masked, the modified regression estimator is still the Best Linear Unbiased Estimator (BLUE) under Gauss-Markov assumptions. A simulation study is conducted across various sample sizes and masking levels. The results from simulation study support the use of randomized masking techniques in sensitive question regression. This method improves data confidentiality without sacrificing the validity of inferences, which make it especially appropriate for studies in the social sciences, criminology, and public health where privacy is crucial. Despite the required privacy protection mechanism, the method guarantees unbiased estimates.