Decision framework

Hypothesis Testing and P Values

Run hypothesis tests by stating null and alternative claims, calculating a test statistic and p value, and separating statistical significance from practical impact.

How this page is maintained

Written for learners, checked against the sources below, and reviewed every year. Last reviewed July 22, 2026.

Short answer

Hypothesis testing evaluates whether sample evidence is unusual under a null model. The p value is the probability, assuming the null is true, of observing a result at least as extreme as the sample. Small p values indicate incompatibility with the null, not proof of practical importance.

  • State null, alternative, and significance level before seeing results.
  • P value is about data under the null, not probability the null is true.
  • Statistical significance and practical importance are different questions.

Core testing steps

A standard workflow is: define H0 and Ha, choose alpha, compute a test statistic, find p value, and make a decision in context. The test statistic compares observed effect to expected variation under H0.

Type I error rejects a true null with probability alpha. Type II error fails to reject a false null. Power is 1 - beta, the chance of detecting a real effect under a specific alternative.

Interpreting p values responsibly

A p value does not measure effect size or importance. With large samples, tiny effects can produce small p values. With small samples, meaningful effects can miss significance.

Report an estimate and confidence interval with the test result. This keeps focus on direction, size, and uncertainty instead of a single threshold decision.

  • Do not switch one-sided and two-sided tests after looking at data.
  • Avoid treating p=0.049 and p=0.051 as fundamentally different evidence.
  • Predefine decision rules in analysis plans.

One-sample mean test with known sigma

Historical average handling time is 50 minutes. A new process sample has n=36, mean 47, assume sigma=9.

  1. Set hypotheses: H0: mu=50, Ha: mu<50.
  2. Compute z=(xbar-mu0)/(sigma/sqrt(n))=(47-50)/(9/6)=-3/1.5=-2.0.
  3. Find one-sided p value P(Z<=-2.0)=0.0228.
  4. Compare to alpha=0.05 and conclude evidence supports lower mean handling time.
Result: At alpha=0.05, reject H0; the sample provides evidence that mean handling time decreased.

Common mistakes

  • Interpreting p value as probability H0 is true.
  • Using significance alone as proof of practical value.
  • Choosing alpha after seeing the p value.
  • Failing to report effect estimate with uncertainty.

Try one

If p=0.03 and alpha=0.05, what is the formal decision?

Reject H0 at the 0.05 level, while still interpreting effect size and context separately.

Sources

Learn this with a tutor

Tell LearnLive what you already know and what you need to do with hypothesis tests.

Build this course