Chapter 10 — Research reading and critique
Lesson map
Lesson 10 published. Map, example, EX046.
flowchart TD
N0["Learning outcomes"]
N1["10.1 Reading de-identification research"]
N2["10.2 Reading differential privacy"]
N3["10.3 Reading behavioural research"]
N4["10.4 Writing critique"]
N5["Synthesis"]
N0 --> N1 --> N2 --> N3 --> N4 --> N5
N5 --> EX["Worked example"]
EX --> Q["Exercise EX046"]
Logical drawing
flowchart LR
N0["Learning outcomes"]
N1["10.1 Reading de-identification research"]
N2["10.2 Reading differential privacy"]
N3["10.3 Reading behavioural research"]
N4["10.4 Writing critique"]
N5["Synthesis"]
N0 --> N1 --> N2 --> N3 --> N4 --> N5 --> Q["EX046"]
[Learning outcomes]
|
v
[10.1 Reading de-identification research]
|
v
[10.2 Reading differential privacy]
|
v
[10.3 Reading behavioural research]
|
v
[10.4 Writing critique]
|
v
[Synthesis]
|
v
[Worked example] --> [Exercise EX046]
Learning outcomes
You will (1) critique research on de-identification and consent; (2) judge relevance to practice; (3) write a critical summary.
10.1 Reading de-identification research
Dr. Ward assigns key research. Sweeney (2002) showed that combinations of attributes re-identify people. Narayanan and Shmatikov (2008) showed linkage of sparse behavioural data. Priscila applies these to transaction data, which are highly distinctive: a few purchases with times and places can identify a customer. She concludes that Corvina should never treat transaction-level data as anonymous merely because names are removed, which affects how it shares data with partners. Narayanan and Shmatikov showed that sparse records, such as movie ratings, can be re-identified with little auxiliary information, and transaction histories are similarly sparse.
10.2 Reading differential privacy
Priscila reads Dwork and Roth (2014) on differential privacy. She understands the core idea: outputs should change little whether any individual is included. She critiques its practical use at Corvina: aggregate spending statistics for partners could use differential privacy, but fraud detection requires individual data and cannot. She writes that differential privacy is a strong tool for releases but not a general solution, and that choosing parameters requires expertise. Dwork and Roth explain how noise protects individuals, but the privacy budget must be set and spent carefully.
10.3 Reading behavioural research
Priscila revisits Acquisti and colleagues (2015). She critiques the relevance: the review covers many contexts, mostly online; payment apps share features such as defaults and framing. Her own A/B test supports the review's claim that design shapes choices. She notes that research often lacks the specific context practitioners face, so practitioners must test locally, as she did, rather than assume results transfer exactly. Acquisti and colleagues review evidence that people's privacy choices are uncertain, context-dependent and malleable, which matches what her consent test showed about design influencing acceptance.
10.4 Writing critique
Priscila writes a critical summary for colleagues: what each work claims, its evidence, its limits and its implications for Corvina. She avoids overstating results. Dr. Ward praises the summary for connecting research to decisions and for distinguishing what the research shows from what Corvina must test. Priscila decides to share the summary with the privacy champions as background reading. Her critique template asks four questions of each paper: what was studied, on what data, what was found and what would change if the setting were a Brazilian and Portuguese payments firm. The answers fit on one page per paper.
Worked example
Priscila applies Sweeney and Narayanan and Shmatikov to transaction data, limits differential privacy to releases, relates behavioural research to her A/B test and writes a critical summary for colleagues.
Synthetic case. Corvina's applications are synthetic; the cited works are real.
Errors and limits
Summaries simplify; link to originals.
Exercises
- EX046 — Transaction data. Why are transaction data not anonymous after removing names? Rubric: 3 = few purchases with times and places identify customers;; 2 = mentions distinctiveness;; 1 = anonymous;; 0 = none.
- EX047 — DP limits. Where can Corvina not use differential privacy? Rubric: 3 = fraud detection, which needs individual data;; 2 = mentions fraud;; 1 = everywhere;; 0 = none.
- EX048 — Local testing. Why must practitioners test research claims locally? Rubric: 3 = research contexts differ from their own;; 2 = mentions context;; 1 = trust research;; 0 = none.
- EX049 — Summary elements. What does each entry in Priscila's summary contain? Rubric: 3 = claim, evidence, limits, implications;; 2 = three;; 1 = vague;; 0 = none.
- EX050 — Shared reading. Who receives the critical summary? Rubric: 3 = the privacy champions;; 2 = mentions colleagues;; 1 = board;; 0 = none.
Synthesis
Chapter 10 critiques research. Chapter 11 delivers the applied project.
Continue studying — chapter 10 of 12
← Chapter 09 (draft) · Course chapter index · Chapter 11 (draft) →