A manufacturer deciding whether to adjust a production process needs more than a measurement: it needs a rule for deciding when the evidence warrants action. The Neyman–Pearson framework turns statistical evidence into decisions under uncertainty by specifying hypotheses, error risks, and a decision rule before analysis. section 4.0 section 4.3.1
Introduced by Jerzy Neyman and Egon Pearson in 1933, the framework is not a particular statistical test. It is a general procedure for choosing and applying tests consistently across different questions and data. section 4.0 section 4.3.3
The procedure separates design, evaluation, and interpretation:
- Before analyzing the data, formulate the hypotheses, set a significance level, select an appropriate test, and specify the decision rule. section 4.3.2 section 4.3.3
- Evaluate the evidence using the chosen test and apply the prespecified rule to reach a statistical conclusion. section 4.3.3
- Interpret the conclusion in context and consider the action it supports, recognizing that a statistical decision remains uncertain. section 4.5 section 4.2.4
A significance level sets a standard for the risk of a Type I error: rejecting a true null hypothesis. A Type II error is failing to reject a false null hypothesis. The relative consequences of these errors depend on the problem, so setting the rules requires subject-matter judgment as well as statistical reasoning. section 4.0 section 4.2.4 section 4.3.1
Manufacturing decision
A manufacturer may use a hypothesis test to distinguish ordinary process variation from evidence of a production problem. The hypotheses, test, significance level, and decision rule are set in advance; the resulting conclusion can inform whether adjustment is warranted. The practical consequences of an unnecessary adjustment and a missed problem matter when designing the test. section 4.4 section 4.3.2 section 4.2.4
Prespecification makes conclusions reproducible: researchers applying the same rules to the same data should reach the same statistical conclusion. It also helps guard against p-hacking, in which analyses are modified or results selectively reported to obtain statistical significance. A test does not prove or disprove a hypothesis; it provides a disciplined decision under uncertainty. section 4.3.2 section 4.2.4
