How does differential privacy protect analytics results?

Differential privacy is a mathematical approach that limits how much the output of an analysis can change because one person’s information was included or excluded. It commonly works by adding carefully calibrated randomness to calculations.

For example, a website could release a noisy count of visitors who used a feature rather than publishing an exact count that might expose activity within a very small group.

Three implementation choices matter:

• The unit being protected, such as a person rather than a single event.

• The privacy parameters, which control the strength of the guarantee.

• The cumulative information disclosed through repeated analyses.

Adding arbitrary noise is not enough to establish differential privacy. The mechanism and its privacy accounting must support the claimed guarantee.

The ICO’s differential privacy guidance explains these trade-offs. It can support privacy-preserving analytics, although original records retained behind the analysis may still be personal data.