Data ethics deals with the harms that can follow from data work even when every number is correct.
Data about people is collected, combined and acted on at a scale that can affect millions.
The people described rarely see the analysis, so the analyst must consider their interests.
Fairness asks whether decisions treat groups unequally, including through stand-in columns such as postcode.
Consent asks whether people knew and agreed to how their data would be used.
Transparency asks whether people can learn what data is held and how decisions about them were made.
Purpose asks whether data collected for one reason is being used for another that people would not expect.
Harm can come from exposure, such as a leak, or from decisions, such as a loan refused on biased evidence.
Historical data can carry past discrimination, which an analysis then repeats as if it were neutral.
Aggregated results can still harm small groups, such as figures that reveal a rare condition in one village.
A useful test is whether the work would be acceptable if explained on the front page of a newspaper.
Ethics reviews, written guidelines and varied teams help catch problems early.
Analysts can question or refuse work they believe is harmful, which is part of professional responsibility.
In practice, ethical questions are cheapest to address at the design stage, before data is collected.
The common misconception is that data is neutral, when choices about what to collect and measure shape the results.