Content practice questions.

A hospital replaces patient names in a research extract with random codes and keeps the code-to-name key in a separate restricted file, so that patients can be re-contacted later. Under the GDPR this extract counts as

• anonymous data, because the names have been removed from the records
• pseudonymised data, and therefore still personal data
• aggregated data, because the records no longer describe individuals
• encrypted data, because identifiers were replaced by a coded value

Answer: pseudonymised data, and therefore still personal data

While a key exists that can reverse the substitution the people remain identifiable, so the full obligations continue to apply; only irreversible removal makes data anonymous. Encryption is a third thing again, reversible with a secret rather than with a lookup table.

A running app records several kinds of data about one user. Which of them belongs to that user's passive digital footprint rather than the active one?

• The profile photograph that the user chose to upload during registration
• The review the user wrote about a running route
• The accelerometer readings logged while the phone sits in a pocket
• The display name the user picked for a public leaderboard

Answer: The accelerometer readings logged while the phone sits in a pocket

The active footprint is what a person decides to publish, while the passive footprint is what the device produces as a by-product, and it is usually larger, more continuous and less remembered. Sensor and log data are the reason that deleting a post does not delete a person's record.

A supermarket sells loyalty-card purchase histories to a firm that combines them with other datasets and resells risk scores to insurers. The firm is best described as

• a data controller acting on explicit consent given by each shopper
• a data processor, which only stores data on the retailer's instruction
• a data protection officer, appointed to supervise the retailer's practice
• a data broker, which trades in data it did not collect itself

Answer: a data broker, which trades in data it did not collect itself

The defining feature of a broker is distance: it never met the people in its files, which is why they rarely know it exists and struggle to exercise rights against it. A processor acts only on instruction, whereas the broker sets its own purposes.

A team adds ten times more records to a training set, drawn by exactly the same collection method as before. Using Source B, identify the limitation of the data that the extra records leave untouched.

• Groups the collection method never reaches stay absent from the data
• The size of the sampling error around the accuracy measured on a held-out test set
• The number of rare cases from the recorded groups seen during training
• The influence that one mislabelled record has on the fitted model

Answer: Groups the collection method never reaches stay absent from the data

Volume cannot repair coverage: group D never entered the collection process, so ten times more of the same records only confirm the gap, while sampling error, the count of rare cases and the weight of any single bad record all move with the number of records. Representativeness is a property of how data were gathered, not of how much there is.

A city police service directs patrols using a model trained on the past five years of recorded arrests. After two years, the neighbourhoods the model flags as highest risk are the same ones that received the most patrols at the start. The best explanation of this pattern is that

• arrests record where officers were sent, so the model retrains on its own effects
• the model was built without any labelled training examples to learn from
• an unsupervised method was chosen where a supervised one was needed
• the developers wrote a rule that explicitly named those neighbourhoods as high risk

Answer: arrests record where officers were sent, so the model retrains on its own effects

Arrest records measure police activity as well as crime, so the model's own deployment produces the data that confirms it — a closed feedback loop. Digital Society treats this as a data and design problem, not as evidence that the flagged neighbourhoods cause more crime.

A video platform changes its ranking system so that clips are promoted mainly by the watch time they generate. An implication is a wider or longer-term consequence that goes beyond the effect being measured. Which statement describes an implication of the change?

• The share of recommendation slots filled by clips under three minutes falls by twelve per cent
• Creators redesign what they make to fit a target they cannot inspect
• Average watch time per session rises across the whole platform
• Views recorded on short explainer channels fall within the first month

Answer: Creators redesign what they make to fit a target they cannot inspect

The other three statements are quantities the platform can read off its own dashboard; the implication is that editorial control passes from creators to an objective none of them can see. Treating impact and implication as synonyms is the most common vocabulary error in the course.