datasnooping

Data snooping refers to the practice of examining data to extract patterns, correlations, or conclusions, often beyond the original purpose for which the data were collected. In statistics and data analysis, data snooping describes the misuse of data to generate hypotheses after looking at the data, which can produce overly optimistic results and inflate the risk of false positives. It is closely related to terms such as data dredging and p-hacking, and it typically involves trying multiple analyses or models on the same dataset until something appears significant, then presenting that result as evidence.

Beyond statistics, the term is used more broadly to describe the unauthorized or excessive examination of data to infer sensitive information or track behavior. This form raises privacy, ethical, and legal concerns and may violate data protection laws or consent terms. Organizations may address these concerns through data governance, access controls, audit trails, anonymization, and the use of privacy-preserving techniques.

Mitigation and best practices include preregistering analyses, using independent holdout samples or cross-validation, and correcting for multiple testing. In privacy contexts, approaches such as data minimization, transparent data-use notices, and user consent are recommended. When used responsibly, data analysis aims to balance insight generation with safeguards against biased conclusions and privacy violations.

See also: data mining, p-hacking, preregistration, cross-validation, data governance, privacy-preserving data analysis.