"Having Confidence in My Confidence Intervals": How Data Users Engage with Privacy-Protected...
summary
Compares Wikipedia pageview releases produced with rounding vs differential privacy, builds dataset documentation refined with five privacy-communication experts, and runs task-based sessions with 15 data scientists on three tasks (max, 10-day mean, equality). Finds baseline internet noise shapes interpretation, only 3 of 15 participants can construct a 95% CI for the mean, many misread DP’s better utility as weaker privacy, and recommends targeted docs plus tooling to auto-generate confidence intervals for privacy-noised downstream analyses.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:07Wikimedia data commitment
- 3:00Privacy-protected datasets
- 4:04Documenting privacy data
- 7:39Documentation as design probe
- 11:35User study tasks
- 14:42Noise and privacy perceptions
- 16:08Rounding vs differential privacy
- 18:58Confidence intervals explained
- 20:04Privacy-utility trade-offs myths
- 21:29Dataset preferences by audience
- 22:32Privacy tooling takeaways
- 24:22Q&A
related talks
06mLessons Learned in Communicating Differential Privacy to Data Subjects
45mMaking Privacy Technology Accessible: Benchmarks and Platforms
10mEvaluating Privacy Perceptions, Experience, and Behavior of Software Development Teams
37mFrom the Trenches Privacy Engineering Pitfalls Successes
40mExtreme Makeover: Privacy Edition
40mCreative Ways to Champion Privacy Engineering Awareness
15mTurning Privacy Risk Assessment Into 20 Questions for Developers
05mDeciding on Personalized Ads: Nudging Developers About User Privacy