Adaptive lasso for analyzing sensitive survey data

Randomized response techniques (RRT) are widely used in surveys on sensitive topics to reduce response bias and protect respondent privacy. This paper proposes a novel adaptive lasso method for analyzing unrelated-question RRT data in high-dimensional settings. The sensitive behavior is modeled via logistic regression, and the proposed estimator is shown to possess the oracle property, with explicit covariance estimation. Recognizing that the indicator of whether a respondent answered the sensitive question is latent under the RRT design, we develop a tailored shooting algorithm embedded within an expectation-maximization (EM) framework for efficient computation. Simulation studies demonstrate the method’s effectiveness, and its practical utility is illustrated through an analysis of extramarital sex data from the Taiwan Social Change Survey (TSCS).