Ethical and Social Issues Around Data Collection

The section's core safeguard is to collect only what the feature actually needs. Privacy risk scales with how much is held, not with how carefully the program computes, so the amount collected is itself a design decision.

• B is not correct: Being publicly available says nothing about whether users gave permission or how the data will be used, and the section warns not to treat data collection as automatically neutral.
• C is not correct: Longer retention does not fix missing values, inconsistent definitions, measurement errors, or duplicate records, and the section says to keep data only as long as it is needed.
• D is not correct: Personal data also includes location information, visit history, and device information, so stripping names and student IDs still leaves records that can expose sensitive visits.

The section asks you to check collection bias before drawing conclusions, because a collection method can systematically exclude or underrepresent a group. Two data sets on the same topic can support different conclusions purely because of their sources.

• A is not correct: Size does not protect against collection bias; a large collection can still miss entire groups, and the section says more data does not automatically produce a better decision.
• B is not correct: File format is not one of the checks the section lists; the questions that matter are relevance, coverage, quality, and collection bias.
• D is not correct: Public availability does not supply the clear permission the section calls for, and reuse without approval is exactly the unapproved-purpose risk it describes.

The section pairs obtaining clear permission with explaining why the data is collected. Permission is only meaningful when the person granting it understands the purpose and can decide before the data leaves them.

• A is not correct: A description that arrives after submission gives the user no chance to decide, because the data has already been collected by the time the purpose is stated.
• C is not correct: Asking for permission without naming the purpose leaves the user agreeing to an unknown use, which drops the explanation half of the section's guidance.
• D is not correct: Treating app use as blanket approval is how data ends up used for a purpose the user did not approve, which the section names as a privacy harm.

Algorithmic bias means systemic, repeated unfair outcomes caused by data or design choices that disadvantage a specific group. The unfairness comes from who is missing or how labels were assigned, not from an arithmetic mistake.

• A is not correct: Processing every stored record identically does nothing about the people who were never recorded, so uniform treatment of a skewed collection still yields unfair outcomes.
• C is not correct: Adding records does not restore a group the collection method excluded; a larger collection missing the same groups stays unrepresentative.
• D is not correct: Removing names and student IDs is a privacy step, and it leaves untouched the gaps in who was collected and how labels were assigned.

The section says to keep data only as long as it is needed and to limit who can access it. Once the monthly totals are computed, the detailed histories are data whose only remaining role is to be exposed if records are breached.

• A is not correct: Holding every trip record for a possible future use keeps data past the point where it is needed, leaving trip histories available to a breach or unauthorized viewer.
• B is not correct: Handing histories to every employee who asks is the opposite of limiting access, and unauthorized access to stored records is named as a way privacy is harmed.
• C is not correct: Adding location detail collects more than the feature needs, and stored location history is exactly what can expose sensitive visits.

Missing readings and duplicates are two of the data-quality problems the section tells you to look for before drawing conclusions. Duplicates also overrepresent whatever they duplicate, which distorts any rate computed from the records.

• A is not correct: Nothing repairs these records on its own; the section places the responsibility on checking for and addressing the problems before the data is used.
• B is not correct: Missing readings remove information and duplicates repeat records that are already present, so neither one brings in users the collection did not cover.
• D is not correct: Cleaning up missing and duplicate readings has nothing to do with privacy, since the stored records can still be shared, leaked, or used without approval.

The section states directly that a program can process every value correctly and still create a privacy problem for its stakeholders. Correctness is a property of the computation; approval and purpose are separate design commitments.

• A is not correct: Correct calculation of progress is unrelated to whether students agreed to the second use, so it cannot make the unapproved reuse acceptable.
• B is not correct: Showing everyone the same advertisement does not obtain permission from anyone, and the records were still repurposed without informing the students.
• C is not correct: Stakeholders include the people who provide the data and experience the program's decisions, and the section says correct processing does not rule out harm to them.

Relevance asks whether the data set contains variables directly tied to the question. A smaller, relevant, accurate data set can be more useful than a larger collection filled with unrelated fields.

• A is not correct: Monthly visitor totals cannot separate workshop attendees from anyone else, matching the section's example of totals with no indicator for the thing being studied.
• B is not correct: A list of topics carries neither attendance nor checkout records, so it contains no variable that could connect the workshop to later borrowing.
• D is not correct: Titles and publication dates are unrelated fields, and the section notes that a large collection of unrelated fields is not the better choice.

The section warns against a data set that covers only a short window and is therefore too small to generalize. The variables here are the right ones; it is the span of time they cover that does not match the question.

• A is not correct: Relevance fails when no variable is tied to the question, but attendance records are exactly the variable the school wants to study.
• C is not correct: Quality concerns missing values, inconsistent definitions, measurement errors, and duplicates, and nothing here says the spring records are defined differently.
• D is not correct: Permission is not what the shortfall describes; the gap is the time period the records cover, and the section's four fit checks are relevance, coverage, quality, and collection bias.