Exploring User-Suitable Metaphors for Differentially Private Data Analyses
summary
Researchers generate metaphors for local and central differential privacy, evaluate them analytically and via 30 online interviews (10 per scenario), and find metaphors like the spinner metaphor and noisy combined picture successfully show that added noise gives plausible deniability and a privacy–accuracy trade-off. High-level metaphors frequently produce concrete misconceptions—noise seen as reversible, like pixelation or encryption, or as selective hiding—so metaphors need context, numbers, and clarifying details to avoid wrong expectations.
AI-written summary · about
chapters (11)
- 0:00Intro
- 0:26Objective
- 2:10Differential privacy-models
- 3:14Metaphor for central DP - Scenario 2
- 3:38Our approach
- 4:02Research questions
- 4:25Interviews - design and demographics
- 5:45Information needed for trust and data sharing - RQ1
- 6:34Perceptions of privacy features and the extent of the suitability of metaphors-RQ2/RQ3
- 7:27Misconceptions of DP
- 8:48Challenges and conclusion
related talks
06mLessons Learned in Communicating Differential Privacy to Data Subjects
10mCommunicating Differential Privacy Guarantees to Data Subjects
20mShare data without revealing personal information
45mPractical Experience with Making Synthetic Data Differentially Private
47mDifferential Privacy from a Statistical Perspective–Obtaining Valid Inferences...
01h 43mPerspective
33mPractical Data Privacy
11mMy $5MM Differential Privacy Visualizations