Likelihood Ratio

How likelihood ratio works

A likelihood ratio compares how well one hypothesis explains the observed data against an alternative hypothesis. It is a relative measure, so it does not tell you whether either hypothesis is true on its own, only which one the data supports more strongly.

In statistical testing, that distinction matters because the same observation can be more compatible with one model than another even when the evidence is still weak overall. The ratio becomes especially useful when you need to compare competing explanations without reducing the result to a simple pass or fail.

Likelihood ratio in sequential testing

In a sequential probability ratio test, the likelihood ratio is updated after each new observation. That makes the test adaptive, because the evidence can accumulate until it reaches a decision boundary rather than waiting for a fixed sample size.

This is why sequential testing is often efficient: strong evidence can end the test early, while ambiguous evidence keeps the process running. The ratio is the control signal that determines whether the current data are already sufficient or whether more observation is needed.

Why the ratio is useful for decision-making

The value of a likelihood ratio lies in turning data into a comparison that can support an explicit decision rule. It helps separate the question of how much the data favour one hypothesis from the question of whether that evidence is enough to act.

That makes it useful in settings where uncertainty is unavoidable and where decisions must be made under changing evidence. It is also why the term appears in both classical statistical testing and applied detection workflows, where repeated updates matter more than a single snapshot.

Common interpretation pitfalls

A likelihood ratio is not the same as a probability that a hypothesis is true, and it should not be read as a standalone confidence score. It only compares support between two hypotheses, so the result depends on which hypotheses are being contrasted and how they are defined.

Misreading the ratio can lead to overconfidence, especially when people treat a large value as proof rather than as relative evidence. The strongest interpretations come from pairing the ratio with the test design, the update rule, and the decision threshold used in context.

Practitioner Guidance

Common misunderstanding: Practitioners often treat a likelihood ratio as if it were an absolute measure of truth. It is better understood as a comparison tool, so the surrounding hypothesis set and stopping rule must be defined clearly before the test begins.

Practitioner takeaway: Use the ratio to compare competing hypotheses consistently, then interpret it through the lens of the decision threshold rather than as an isolated statistic.

Related resources from NHI Mgmt Group

• What does the 144:1 NHI-to-human ratio mean for IAM governance programmes?
• What is the difference between vulnerability severity and exploit likelihood?
• Who is accountable when a merchant’s dispute ratio exceeds network thresholds?
• What makes a data protection ratio hard to interpret without context?