bnr#94 =>

Significance Tests

Object: 54
Date of origin: 1925
Author/inventor/context: Ronald Aylmer Fisher
A statistical method to assess the likelihood of the results of an experiment being due to chance. Fisher proposed it as an informal index for use as part of non-quantifiable processes of drawing conclusions from observations. It calculates the probability of obtaining the experimental results under the assumption that the item tested, for example a new treatment, has no effect. Results within the arbitrary but now widely standard 5% cut-off are typically considered statistically significant opening doors for publication. A review found 95% of articles in psychology journals claim statistical significance. Such publication behaviour is statistically improbable and, as meta-analysis confirms, generates significant levels of false positives, i.e. the claim that a new treatment works when in fact it doesn't.

Martin Feuz