In science, a single study is rarely enough to anything. Other researchers must be to repeat the study and get results. This process is replication. In 2015, a large tried to repeat 100 published psychology . Only about third to one half of the results were , depending how success was .

When a published result cannot be repeated, it does not usually mean that the original researchers were dishonest. More often, the problem lies in ordinary features of how research has been done and published.

One factor is small samples. A study with only a few dozen participants can produce a striking result simply by chance. Another factor is flexibility in analysis: when researchers try several ways of analyzing their data, at least one may produce an impressive-looking result even if no real effect exists.

Publishing practices also play a role. Journals have traditionally preferred new and surprising findings. Studies that found nothing were often left unpublished, a situation sometimes called the file drawer problem. As a result, the published record can make effects appear stronger and more reliable than they really are.

In response, many scientists now preregister their studies, publicly stating their hypotheses and analysis plans before collecting data. Larger samples, shared data, and journals that publish replication attempts are also becoming more common. These reforms may make research slower to produce headlines, but they aim to make it more trustworthy.