Positive and novel results are likelier to be published, so the literature overstates effects.
Sterling documented it in 1959, and the file-drawer problem is the standard name. Registered reports and pre-registration exist because of it, and they measurably reduce the positive-result rate.
In practice
- A drug's four published trials all show benefit and its three unpublished ones do not, and the meta-analysis reads the four.
- A company that only writes up successful experiments has an internal literature in which every initiative worked.
- Further reading
- Wikipedia
- Subject
- Science & Epistemology
- Essay
- The ways evidence misleads