How to write a hypothesis
Short answer: A hypothesis is a specific, testable prediction about the relationship between two or more variables, derived from theory or earlier findings. A good hypothesis names the variables and the population, states the expected direction of the relationship, and could be shown to be wrong by data. In statistical testing it is paired with a null hypothesis (H0: no effect or no difference) and an alternative hypothesis (H1: the predicted effect); results either reject H0 or fail to reject it, and a single study never proves a hypothesis true.
On this page
- What makes a hypothesis testable
- Research hypothesis vs statistical hypotheses
- Directional and non-directional hypotheses
- Step 1: Start from your research question
- Step 2: Identify the variables
- Step 3: State the expected relationship and its reason
- Step 4: Write it as one clear sentence
- Step 5: Check that the design can test it
- Examples by design
- Numbering and placement
- Common mistakes
- Quick checklist
If the research question asks what you want to know, the hypothesis states what you expect to find and why. It turns a question into something data can contradict. That is what makes it useful: a hypothesis that no possible result could disprove tells readers nothing.
What makes a hypothesis testable
A good hypothesis is:
- Specific: it names the variables and the population ("first-year undergraduates"), not just a topic.
- Directional where theory allows: it says which way the relationship should go ("higher… lower").
- Measurable: each variable can be observed or measured with the data you plan to collect.
- Falsifiable: some realistic result would count against it.
- Grounded: it follows from theory or earlier findings, which you explain in the introduction or literature review.
Not every study needs hypotheses. Descriptive and exploratory studies, and most qualitative research, are usually guided by research questions instead. Hypotheses are standard when you test a relationship or an effect.
Research hypothesis vs statistical hypotheses
| Research hypothesis | Null hypothesis (H0) | Alternative hypothesis (H1) | |
|---|---|---|---|
| What it states | Your substantive prediction in words | No effect, no difference or no association | The effect, difference or association you predict |
| Example | Students who receive weekly feedback will write better final essays than students who do not | Mean essay scores are equal in the feedback and control groups | Mean essay scores are higher in the feedback group |
| Where it appears | Introduction or end of the literature review | Usually implicit in the analysis; stated explicitly in some fields and courses | Usually the same as the research hypothesis |
A statistical test assesses how compatible the data are with H0. You either reject H0 (the data are unlikely if there is no effect) or fail to reject H0. Failing to reject H0 does not show that there is no effect, and rejecting it does not prove H1; it means the data support H1 in this study.
Directional and non-directional hypotheses
| Type | Wording | When to use |
|---|---|---|
| Directional | "X will be positively associated with Y"; "Group A will score higher than Group B" | Theory or prior evidence predicts the direction |
| Non-directional | "X will be associated with Y"; "Groups A and B will differ" | You expect an effect but have no firm basis for its direction |
A directional hypothesis does not oblige you to run a one-tailed test; many fields and journals expect two-tailed tests either way. Whatever you choose, decide the wording and the test before you see the data. Switching to a one-tailed test after looking at the results is a well-known way to produce misleading p values.
Step 1: Start from your research question
Each hypothesis should answer one research question, or part of one. "Is weekly paid work associated with first-semester grades?" leads naturally to a prediction about that association. See how to write a research question.
Step 2: Identify the variables
Name the independent (predictor) and dependent (outcome) variables, and any mediator or moderator. Decide how each will be measured: grade point average from records, hours from a weekly diary, exhaustion from a validated scale.
Step 3: State the expected relationship and its reason
Use theory or earlier studies to justify the direction. The justification usually comes in the sentences just before the hypothesis: "Because paid work competes with study time (Author, Year), we expect…"
Step 4: Write it as one clear sentence
Templates that work in most fields:
Step 5: Check that the design can test it
Make sure your data and analysis can actually test the prediction. A cross-sectional survey can test an association but not that X causes Y. A mediation hypothesis needs measures of the mediator. If the design cannot test the hypothesis, change one or the other before you collect data. Many researchers now pre-register their hypotheses and analysis plan (for example on the Open Science Framework) so readers can see they were set in advance.
Examples by design
| Design | Hypothesis |
|---|---|
| Experiment | Participants who receive weekly written feedback will have higher final essay scores than participants in a no-feedback control group. |
| Correlational survey | Weekly hours of paid work will be negatively associated with first-semester GPA among full-time undergraduates. |
| Moderation | The negative association between paid work hours and GPA will be weaker for students whose jobs are related to their degree. |
| Mediation | The association between remote-work intensity and emotional exhaustion will be partly explained by weaker work–home boundaries. |
| Panel data | Within firms, years with higher R&D intensity will be followed by more patent applications in the next year. |
Numbering and placement
State hypotheses at the end of the introduction or literature review, after the argument that justifies them. Number them (H1, H2, H3, or H1a and H1b for related parts) so you can refer to them in the methods and results. In the results, report each hypothesis in order and say plainly whether it was supported; our results section guide shows how.
Common mistakes
- Writing a hypothesis that cannot be wrong ("X may or may not affect Y")
- Using vague concepts ("social media is bad for teenagers")
- Stating a causal hypothesis that the design cannot test
- Packing several predictions into one sentence; split them into H1a, H1b
- Adding or rewording hypotheses after seeing the data without saying so
- Saying a hypothesis was "proved"; use "supported" or "not supported"
Quick checklist
- Does each hypothesis follow from a research question?
- Are the variables, population and direction named?
- Could a realistic result contradict it?
- Is the reasoning behind it given in the text just before?
- Can your design and analysis test it?
- Are hypotheses numbered and reported in the same order in the results?
- Were they set before the data were analyzed?
For how hypotheses shape the rest of the study, see how to write a methodology section.