null hypothesis

Null Hypothesis

The default assumption in hypothesis testing that there is no effect — the two groups are the same, the treatment has no impact, the drug works no better than placebo.

When you'd see it: Every statistical test implicitly or explicitly specifies a null hypothesis. In A/B testing: 'Version A and Version B have the same conversion rate' is the null. Rejecting the null doesn't prove the alternative — it only says the data is inconsistent with the null at your chosen threshold.

Why it matters: The null hypothesis defines what you're testing against. The null is also important for understanding false positives: if every null hypothesis you test is true, 5% of your tests will still show P < 0.05 by chance — which is p-hacking at scale.

Common mistakes: Treating failure to reject the null as proof that the null is true. 'Not significant' means insufficient evidence against the null — it doesn't mean the effect is zero. Absence of evidence isn't evidence of absence, especially in underpowered studies.

Related hypothesis testing, P-value, statistical significance, T-test, p-hacking