Null Hypothesis

The null hypothesis (H₀) is the default proposition in statistical hypothesis testing that states there is no effect, no difference, or no association between variables. It serves as the baseline assumption against which the observed data are evaluated, and statistical tests calculate the probability of obtaining the observed results (or more extreme results) if the null hypothesis were true.^[1][3]

The null hypothesis is the cornerstone of the frequentist statistical framework that dominates quantitative research. It provides a concrete, falsifiable statement that can be tested with data: if the data are sufficiently unlikely under H₀, the researcher rejects it in favor of the alternative hypothesis.^[1] This framework structures experimental design, sample size calculation, and the interpretation of results across clinical trials, epidemiology, psychology, economics, and many other disciplines that use statistical inference. Understanding what the null hypothesis means, and what rejecting or failing to reject it does and does not imply, is essential for the correct interpretation of research findings.

In the standard hypothesis testing framework, the researcher begins by stating the null hypothesis (no effect) and the alternative hypothesis (an effect exists). For a clinical trial comparing a drug to placebo, H₀ might state that the mean blood pressure reduction is the same in both groups, while the alternative hypothesis (H₁) states that the means differ.

A test statistic is calculated from the data and compared against its sampling distribution under H₀. The p value, the probability of observing the test statistic (or a more extreme value) if H₀ were true, quantifies how incompatible the data are with the null hypothesis.^[3] If the p value falls below the prespecified significance level (alpha, usually 0.05), H₀ is rejected, and the result is declared statistically significant.

Critically, failing to reject H₀ does not prove that H₀ is true.^[2] It means only that the data do not provide sufficient evidence to reject it, which may reflect a true absence of effect, an effect too small to detect with the available sample size, or measurement imprecision. This distinction between "absence of evidence" and "evidence of absence" is fundamental but frequently misunderstood.^[2]

The choice of null hypothesis influences the entire analysis. A two sided test (H₀: no difference, H₁: a difference in either direction) is the default for most research. A one sided test (H₀: no difference or a difference in the direction not predicted, H₁: a difference in the predicted direction) is appropriate only when the researcher has strong a priori reason to expect the effect in one direction only and when an effect in the other direction would be treated the same as no effect.

State a null hypothesis whenever conducting a formal statistical test. It should be formulated before data collection, based on the research question and the theoretical framework. In most studies the null hypothesis is a statement of no effect, no difference, or no association. It should be paired with a clearly stated alternative hypothesis and a prespecified alpha level. In equivalence and noninferiority trials, the null hypothesis is reversed: in an equivalence trial it states that the treatments differ by at least a specified margin in either direction, and in a noninferiority trial that the new treatment is worse than the comparator by at least that margin; the goal is to reject this null to demonstrate equivalence or noninferiority.

A randomized trial tests whether a new rehabilitation program reduces 30 day readmission rates compared to standard discharge planning. The null hypothesis states that the readmission rate is the same in both groups. The observed readmission rate is 18% in the intervention group and 22% in the control group (risk difference: 4 percentage points, 95% CI: negative 1 to 9 percentage points, p = 0.12). Because p exceeds 0.05, the researchers fail to reject the null hypothesis. They note that the confidence interval includes both clinically meaningful reductions and no difference, suggesting that the study may have been underpowered, and they recommend a larger trial.