Neyman-Pearson Lemma

The Neyman-Pearson lemma is the foundational result of testing: for two simple hypotheses, the test that thresholds the likelihood ratio is the most powerful at any given size.

Definition

The Neyman-Pearson lemma states that, for testing a simple null against a simple alternative at a fixed size, the most powerful test rejects the null when the ratio of the alternative to the null likelihood exceeds a constant, with randomization on the boundary.

Scope

This topic covers simple null and simple alternative hypotheses, the likelihood-ratio statistic, the construction of the most powerful test by thresholding that ratio, the use of randomization to achieve an exact size in discrete problems, the existence and uniqueness of the most powerful test, and the lemma's role as the building block for uniformly most powerful and unbiased tests.

Core questions

• Why is the likelihood ratio the optimal test statistic for two simple hypotheses?
• How is the rejection threshold chosen to achieve a prescribed size?
• When is randomization needed to attain an exact size, and how does it work?
• How does the lemma generalize to composite hypotheses?

Key theories

Most powerful likelihood-ratio test

Among all tests of a given size, the one rejecting when the likelihood ratio exceeds a constant maximizes power; any other test of the same size has no greater power against the alternative.

Randomized tests and exact size

In discrete problems an exact size may require a randomized decision on the boundary of the rejection region, which the lemma incorporates to keep the most-powerful property exact.

Clinical relevance

The likelihood-ratio threshold is the optimal decision rule in signal detection, radar, and diagnostic classification, where it defines the receiver operating characteristic and sets the achievable trade-off between detection rate and false-alarm rate.

History

Neyman and Pearson published the lemma in their 1933 paper that introduced the framework of two hypotheses, error probabilities, and power, displacing purely Fisherian significance testing as the optimality foundation of the subject.

Key figures

• Jerzy Neyman
• Egon Pearson
• Erich L. Lehmann
• Joseph P. Romano

Seminal works

• neymanPearson1933

Frequently asked questions

What does the Neyman-Pearson lemma require of the hypotheses?

In its basic form both the null and the alternative must be simple, meaning each fully specifies the distribution; extensions handle composite hypotheses through monotone likelihood ratios or unbiasedness.

Why is randomization sometimes part of the optimal test?

In discrete settings no fixed rejection region may have exactly the desired size, so the optimal test randomizes its decision on the boundary to hit the target size precisely.