On accuracy and anonymity of privacy-preserving negative survey (NS) algorithms

摘要

The traditional method of conducting a survey is to ask each respondent to select the one that best applies from a set of options. This makes surveys on sensitive topics challenging. Apart from misreporting, interviewees who are surveyed on a sensitive topic may simply refuse to answer the questions due to privacy concerns. Negative surveys provide useful statistical information while being resistant to data disclosure. Respondents of negative surveys answer each question with random false answer(s). The real statistical data about the population being surveyed can be rather accurately inferred from the collective false information using probabilistic algorithms. The first negative survey approach is the one-select negative survey (1-NS). Multi-select negative survey (MNS) was later proposed to further improve accuracy of the estimate. While MNS does improve estimate accuracy to some extent, it is at the cost of reduced individual user anonymity (IUA). This runs counter to the intention of protecting user privacy. In this paper, we propose a new NS algorithm - Two-Question Negative Survey (TQNS). TQNS provides greater IUA while still producing an unbiased estimate. A comprehensive analysis of different NS algorithms is undertaken of both estimate accuracy and the level of IUA. The goal is to provide with adequate information about different NS algorithms so surveyors can choose the approach that suits their needs.