How to Write a Null and Alternative Hypothesis

A methods section template says 'state your null and alternative hypotheses' and gives no example, so a lot of first drafts default to a single sentence of hope: 'I think the breathing exercise will help.' That sentence is not a hypothesis pair, and a supervisor's red pen will find it fast. Learning how to write a null and alternative hypothesis is less about memorizing a formula and more about being precise on paper about a claim you were probably already precise about in your head.

A hypothesis pair is two sentences, not one. The null hypothesis states that nothing is going on: no effect, no difference, no relationship. The alternative hypothesis states the specific effect a researcher predicts instead. Most first-time writers get the two parts backward; they're trying to prove their hypotheses when what they should be doing is testing their null hypothesis against the data. The alternative hypothesis is really just an educated guess.

How to Write a Null and Alternative Hypothesis: The Basic Shape

Take one running example: a ten-minute breathing exercise before an exam, tested against its effect on self-reported test anxiety. The null hypothesis says the exercise changes nothing: 'A pre-exam breathing exercise has no effect on students' self-reported test anxiety.' The alternative hypothesis says it does: 'A pre-exam breathing exercise is associated with a difference in students' self-reported test anxiety, compared with no exercise.' Every hypothesis pair in a methods section follows that same shape, a flat statement of no effect paired with a flat statement of a specific one.

Both halves have to exist on the page, even though only one of them usually turns out to be interesting. Skipping the null and writing only what a researcher expects to find is one of the fastest ways to lose marks on an early-stage proposal. This step sits inside the wider sequence covered in how to plan a research paper, right after a research question has been narrowed down and just before the outline takes shape.

Writing the Null Hypothesis

The null hypothesis, written H0, is the boring default: no effect, no difference, no relationship between the variables in the study. It describes what the data would look like if nothing were going on, not a claim about the real world directly, and that distinction is the baseline every statistical test is actually built to challenge.

Write it as an equality. For the breathing exercise example: the mean anxiety score in the group that did the exercise equals the mean anxiety score in the group that did not. In plain language rather than notation, that becomes: there is no difference in self-reported test anxiety between students who complete a pre-exam breathing exercise and students who do not. Naming both groups and the exact outcome measured, not just the general topic, is what separates a real null hypothesis from a vague restatement of the research question. A hypothesis generator forces that same specificity, since it will not produce either half of the pair until the outcome variable and the comparison group are named.

Directional or Non-Directional: Writing the Alternative Hypothesis

The alternative hypothesis comes in two shapes, and the choice is not just stylistic. A non-directional, or two-tailed, alternative predicts a difference without saying which way it runs: the breathing exercise changes anxiety scores, higher or lower, unspecified. A directional, or one-tailed, alternative commits to a direction: the breathing exercise specifically lowers anxiety scores.

What's being compared	Two-tailed (non-directional)	One-tailed (directional)
Alternative hypothesis says	Anxiety scores differ between groups, direction unstated	Anxiety scores are lower in the breathing-exercise group
Null hypothesis covers	No difference between groups	No difference, or a difference opposite the predicted direction
Significance threshold	Split across both tails, about 2.5% in each at the 0.05 level	Concentrated in one tail, the full 5% at the 0.05 level
What you lose	Some power to detect a strong effect in a specific direction	The ability to claim significance if the real effect runs the other way, however large

The practical cost sits in that bottom row. A one-tailed test is easier to find significance for, because you get the whole 5% sitting in one tail, rather than split between two tails. This also means that if you found out that your breathing exercise made people feel more anxious (i.e., the effect went opposite to what was expected), then a one-tailed test, which can only be used to detect 'lower', would've no way to pick up that there's a real effect going the other way. Choosing directional means betting the entire test on already being right about the direction.

Why the Direction Has to Be Decided Before You Look at the Data

A directional hypothesis is a stronger claim than a non-directional one, and a stronger claim needs grounds. If five earlier trials already found breathing exercises lower anxiety and never once found the opposite, a directional alternative has a real basis. If this is the first study asking the question, a non-directional alternative is the honest default, since there is no prior pattern yet to justify betting on a direction.

Most supervisors and statistical reviewers expect a non-directional test unless the write-up can point to that kind of prior evidence. Choosing 'directional' only after seeing which way the numbers fell makes a result look more significant than a two-tailed test would have allowed, without changing anything about how convincing the data actually are. Decide the shape of the alternative hypothesis while writing the proposal, before a single participant is tested, and which one was picked stops being a question anyone can raise later.

What Rejecting the Null Does Not Let You Claim

Rejecting the null hypothesis means the data collected would be unlikely if the null were actually true, at whatever significance level was set in advance, usually 0.05. That's a narrower claim than it sounds, and most of the overclaiming in a results or discussion section comes from writing something the test never actually supported.

Rejecting the null does not prove the alternative hypothesis is true. On its own it does not establish that the breathing exercise causes lower anxiety, unless the study was a controlled experiment with random assignment rather than a comparison of two existing groups. It says nothing about how large or practically meaningful the effect is: with a big enough sample, a trivial difference in anxiety scores can still clear the 0.05 threshold. Failing to reject the null carries its own limit too, and it does not prove there is no effect, only that this particular study, at this sample size, did not find strong enough evidence against 'no effect' to rule it out.

Weak: what the results do not show	Strong: what the test actually supports
The breathing exercise reduces exam anxiety.	Students who completed the breathing exercise reported lower anxiety scores than those who did not, a gap unlikely to appear if the null hypothesis of no difference were true.
This proves the exercise works.	The result supports rejecting the null hypothesis for this sample. It does not establish that the exercise lowers anxiety for students in general.
The result was significant, so the effect is large.	Statistical significance shows the difference is unlikely to be chance. It says nothing about whether the difference is big enough to matter in practice.
The exercise had no effect, since the result was not significant.	The study did not find enough evidence to reject the null hypothesis at this sample size, which is different from showing the exercise truly has no effect.

Every sentence in that left column would look at home in an early draft, and every one claims something the statistics do not actually support. The fix is rarely a full rewrite. Usually it is one clause, swapping 'proves' for 'supports,' or naming the significance level instead of asserting certainty.

Clinical work uses a structured format for the same purpose, and PICOT question examples are collected separately. In an essay the equivalent move is a thesis statement, which has its own page.

Once the hypotheses are written and the analysis is done, the paragraphs describing the result are the next place quality slips, usually because they were rushed at the end. An AI humanizer built for academic writing is a tool for that later stage, worth knowing about now so it is not a scramble the night before a deadline.