Null hypothesis
A null hypothesis is the baseline statistical claim tested against observed data.
What a null hypothesis is
A null hypothesis is the baseline claim used in a statistical test. It often says there is no difference between groups, no association between variables, or no effect of an intervention. The test asks whether the observed data are unusual enough, under that baseline model, to reject it.
Why it is called null
The word null points to the idea of no change, no effect, or no relationship for the quantity being tested. In a clinical trial, the null might say two treatments have the same average outcome. In a survey, it might say two groups have the same population proportion.
Alternative hypothesis
The alternative hypothesis is the claim considered when the data are inconsistent with the null. It might say a difference exists, an effect is positive, or an association is not zero. The null and alternative should be stated before analysis so that the test has a clear target.
How testing works
A statistical test calculates a test statistic and a p-value using a model in which the null hypothesis is assumed true. If the p-value is smaller than the chosen significance level, the result is often described as rejecting the null. If it is not small enough, the result is described as failing to reject the null.
Rejecting is not proving
Rejecting the null is not proof that the alternative is true. It means the data are sufficiently incompatible with the null under the test's assumptions and threshold. Measurement problems, bias, multiple testing, p-hacking, or a poorly chosen model can still produce misleading rejection.
Failing to reject
Failing to reject the null is not proof of no effect. A study may have too little power, too much noise, too small a sample, or a threshold that is too strict for the question. To argue that effects are meaningfully absent, researchers usually need designs such as equivalence or noninferiority testing.
Choosing a good null
A useful null hypothesis matches the study question, measurement scale, and decision problem. Sometimes a simple no-difference null is appropriate. In other cases, the more relevant baseline is a minimum meaningful effect, a noninferiority margin, a randomization model, or a model-based assumption that needs checking.
Why it matters
The null hypothesis frames what evidence is being tested. If the null is poorly chosen or misread, the final conclusion can sound stronger than the evidence allows. Clear null and alternative hypotheses help readers understand what the data did and did not rule out.