datamangel

Datamangel is a term used in data governance and information science to describe the manipulation or misrepresentation of data in ways that alter its meaning, conclusions, or perceived reliability. It encompasses deliberate acts such as fabrication, selective reporting, or data poisoning, as well as unconscious practices such as biased preprocessing, inappropriate aggregation, or improper handling of missing values that distort results.

The term is not standardized but is used to denote the spectrum of data handling choices and actions that degrade data integrity. In discussions of ethics and governance, datamangel highlights that data quality depends not only on technical accuracy but also on transparency and accountability throughout the data supply chain, from collection to publication.

Common mechanisms include altering records, removing inconvenient observations, cherry-picking statistics, or applying transformations that favor certain outcomes. Examples include omitting nonconforming data points to produce a desired trend, mislabeling timestamps, introducing biased sampling, or versioning data in ways that are not reproducible. Data poisoning in machine learning, where training data are subtly manipulated to affect models, is a related concern.

Impacts of datamangel can range from misleading research conclusions to flawed policy decisions and loss of public trust. It can undermine reproducibility, inflate false positives, and create biased or unfair systems.

Mitigation emphasizes data provenance, transparent preprocessing, and reproducible workflows. Practices include maintaining immutable audit logs, using version-controlled data and code, conducting regular data quality checks, and implementing independent replication or auditing of results. Ethical guidelines and governance frameworks encourage disclosure of data limitations and potential biases to counter datamangel.

See also data integrity, data fabrication, data poisoning, p-hacking, reproducibility.