6.8 Chapter Review
6.8.1 Chapter summary
Responsible data practice protects people and organizations throughout the data lifecycle. Analysts should use only the data and tools that are authorized and necessary, recognize that de-identification does not eliminate all risk, document access and transformation decisions, and follow the controlling policy for generative AI and external services.
6.8.2 Common mistakes
- assuming that access permission allows every possible use or disclosure;
- collecting or retaining fields that are not necessary for the analytical purpose;
- treating removed names as proof that re-identification is impossible;
- uploading confidential data to an external service without explicit authorization; and
- documenting security decisions only after analysis is complete.
6.8.4 Exercises
Exercise 1. Classify every NVRW file by access level, approved use, retention period, and permitted storage location, even though the supplied case is synthetic.
One Possible Answer
The published synthetic CSV files may be classified for instructional use in the approved project repository, while analysis outputs may be shared according to the project instructions. A real version of the same project would likely place member-level records in a restricted category, limit use to the approved purpose and team, store them only in authorized locations, and apply the client’s retention and disposal schedule.
Exercise 2. Prepare a pre-use checklist for an external generative AI service. Include confidentiality, authorization, data minimization, output verification, citation, and retention.
One Possible Answer
Before use, confirm that course and client policies permit the service; determine whether any content is confidential, personal, or restricted; submit only the minimum authorized information; remove unnecessary identifiers; review service retention and training settings; verify every factual, analytical, and code output; document the use and required citations; and preserve only approved outputs in authorized storage.
Exercise 3. Examine whether a small program-by-location table could create indirect disclosure in a real client dataset. Propose a reporting treatment that protects confidentiality without inventing data.
One Possible Answer
A responsible treatment might combine compatible categories, suppress a small cell, report a range, remove an unnecessary geographic breakdown, or restrict the detailed output to an authorized audience. The selected rule should follow the data agreement and be disclosed in a table note.
Exercise 4. Draft a data-handling plan for a project containing member demographics, transaction histories, and free-text comments. Identify access levels, storage locations, approved tools, retention periods, and disposal procedures.
One Possible Answer
Member-level demographics, transactions, and comments should be restricted to named project members and stored only in the approved encrypted location. Analysis should use the minimum fields needed, and free text should receive additional review for identifying details. Approved tools, export restrictions, backup rules, access review dates, the client’s retention period, and secure deletion responsibilities should all be documented before analysis begins.
Exercise 5. Evaluate three proposed uses of generative AI in an applied project: debugging synthetic code, summarizing public documentation, and analyzing confidential client records. Explain what authorization and safeguards each use would require.
Check Your Work
Debugging code with fully synthetic inputs may be permitted if the governing policy allows it and the output is verified. Summarizing public documentation still requires source checking and appropriate citation. Uploading confidential client records is not acceptable without explicit authorization, an approved service and account configuration, a compatible data-use agreement, minimum-necessary fields, and all required privacy and security safeguards. Course and client policy always controls.