What a research hypothesis is
A hypothesis is a specific, testable prediction about how variables relate. It is not a vague hunch or a research question — it is a precise statement, derived from theory or previous research, that your data can support or contradict. Where a research question asks (‘does sleep affect exam performance?’), a hypothesis predicts (‘more sleep is associated with higher exam scores’).
Hypotheses belong to quantitative, deductive research: you start with a theory, derive a prediction, collect numerical data, and test whether the data support the prediction. Qualitative research, which explores meaning rather than testing predictions, uses open research questions instead. So the first question to ask is whether your study is the kind that should have hypotheses at all — if you are exploring experiences or perceptions, it probably should not.
How a hypothesis is developed and tested
Hypotheses are not invented from nowhere; they are derived from existing theory and literature, then tested through a defined statistical procedure. The flow below shows the journey from an idea to a decision.
The null and alternative hypotheses
Every hypothesis test involves a pair of hypotheses. The alternative hypothesis (H1 or Ha) is the statement you actually believe and want to find evidence for — that there is a relationship, difference or effect. The null hypothesis (H0) is its logical opposite — that there is no relationship, difference or effect. Counter-intuitively, the statistical test is built around the null: you assume the null is true, then ask how likely your data would be under that assumption.
Research question: Does sleep duration affect exam performance?
Independent variable: sleep duration (hours). Dependent variable: exam score.
Alternative hypothesis (H1): Students who sleep longer the night before an exam achieve higher exam scores.
Null hypothesis (H0): There is no relationship between sleep duration and exam score.
Directional (one-tailed) version of H1: Longer sleep is associated with higher scores (you predict the direction).
Non-directional (two-tailed) version: Sleep duration is associated with exam score (direction unspecified).
If the data would be very unlikely under the null (conventionally, a probability below 5%, i.e. p < .05), you reject the null and conclude there is support for the alternative. If not, you fail to reject the null. Note the careful wording in the next section — it matters.
Why you never ‘prove’ a hypothesis
Students routinely write ‘this proves my hypothesis’, and lose marks for it. Statistics does not prove things; it weighs evidence. You reject or fail to reject the null hypothesis — you never ‘accept’ or ‘prove’ either hypothesis outright. A non-significant result does not prove the null is true; it may simply mean your sample was too small to detect a real effect.
“You never ‘prove’ a hypothesis. You either reject the null hypothesis or fail to reject it — the language of statistics is deliberately cautious for good reason.”
— A rule worth memorising for your results chapter
This cautious language reflects the logic of hypothesis testing and the ever-present risk of error: a Type I error is wrongly rejecting a true null (a false positive), and a Type II error is failing to reject a false null (a false negative). Writing your conclusions in the correct, tentative language — ‘the results support H1’, ‘we reject the null’ — signals that you understand what a statistical test can and cannot tell you.
Identifying and operationalising your variables
A testable hypothesis depends on clearly defined variables. The independent variable (IV) is the presumed cause or predictor (the one you manipulate or that varies); the dependent variable (DV) is the presumed effect or outcome (the one you measure). In ‘more sleep improves exam scores’, sleep is the IV and exam score is the DV.
The crucial skill is operationalisation — defining each variable in terms of how you will actually measure it. ‘Sleep’ becomes ‘hours slept the night before the exam, self-reported’; ‘exam performance’ becomes ‘percentage score on the end of module exam’; ‘authenticity’ becomes ‘score on a validated 5-item authenticity scale’. A hypothesis whose variables are not operationalised cannot be tested, because you have not said what counts as evidence. Vague constructs (‘wellbeing’, ‘success’, ‘engagement’) must be tied to a concrete, measurable indicator before they can appear in a hypothesis.
Directional vs non-directional (one- vs two-tailed)
A directional (one-tailed) hypothesis predicts the direction of the effect: ‘more sleep leads to higher scores’. A non-directional (two-tailed) hypothesis predicts that there is an effect but not its direction: ‘sleep duration affects scores’. The choice should follow from your theory and prior evidence: if previous research strongly suggests a direction, a directional hypothesis is justified; if the evidence is mixed or you genuinely do not know, use a non-directional one.
The choice also affects your statistical test: a directional hypothesis uses a one-tailed test and a non-directional one uses a two-tailed test. Two-tailed tests are more conservative and more common in student work, because they do not assume a direction. Decide before you collect data — choosing a one-tailed test after seeing the results, to scrape under .05, is a serious methodological error.
Testing your hypothesis
You test a hypothesis with a statistical test chosen to match your variables and design — a t-test to compare two group means, ANOVA for three or more groups, correlation or regression for relationships between continuous variables, and chi-square for categorical data. The test produces a p-value, which you compare against your significance level (alpha), conventionally .05.
If p < .05, you reject the null and report support for your alternative hypothesis, alongside the test statistic, degrees of freedom and an effect size. If p ≥ .05, you fail to reject the null. For the detail of choosing and running each test, see our companion guides on t-tests, ANOVA and regression and on interpreting SPSS output, which walk through the output table by table.
What makes a strong hypothesis
Before you commit, test each hypothesis against a few criteria. A strong hypothesis is testable (you can gather data to support or refute it), specific (it names the variables and the expected relationship precisely), falsifiable (it is possible, in principle, for the data to contradict it), and grounded in theory or previous research rather than plucked from the air. It should also be concise — one clear sentence per hypothesis.
Falsifiability is the one students overlook. A ‘hypothesis’ that could never be proven wrong — because it is vague, circular or untestable — is not a scientific hypothesis at all. If you cannot describe what result would contradict your prediction, rewrite it until you can.
The most common hypothesis mistakes
- Writing a hypothesis for qualitative research. Use open research questions instead; hypotheses belong to quantitative work.
- Saying you will ‘prove’ the hypothesis. You reject or fail to reject the null — never prove.
- Not operationalising variables. If you have not said how you will measure a variable, the hypothesis cannot be tested.
- Forgetting the null hypothesis. Every test needs both H1 and H0.
- Choosing one-tailed after seeing the data to get under .05 — a serious error.
- Vague or unfalsifiable predictions that no result could contradict.
- Confusing correlation with causation in a causal-sounding hypothesis tested by a correlational design.
Associational versus causal hypotheses
Hypotheses predict different kinds of relationship, and being clear which you are claiming protects you from over-stating your results. An associational (correlational) hypothesis predicts that two variables are related — that they move together — without claiming one causes the other: ‘screen time is associated with sleep quality’. A causal hypothesis predicts that one variable produces a change in another: ‘increased screen time reduces sleep quality’.
The distinction matters because only certain designs support causal claims. A correlational design, measuring both variables as they naturally occur, can establish association only; to claim causation you generally need an experiment with manipulation and control, or a strong longitudinal design. A frequent and costly error is writing a causal hypothesis (‘X reduces Y’) and then testing it with a correlational survey before concluding causation. Match the strength of your claim to the strength of your design, and where you can show only association, say so explicitly: ‘the data show an association, though causation cannot be inferred from this design’.
Hypothesis examples across disciplines
Strong hypotheses look similar across fields: each names the variables and predicts a specific, testable relationship that a defined test could support or refute.
Psychology: H1 — higher daily social-media use is associated with lower self-reported sleep quality in undergraduates.
Business: H1 — employees with higher perceived autonomy report higher job satisfaction.
Education: H1 — students using retrieval practice achieve higher test scores than students using re-reading.
Marketing: H1 — cause-related advertising produces higher purchase intention than product-only advertising.
Health: H1 — a 12-week exercise programme reduces systolic blood pressure in adults with hypertension.
Each pairs with a null hypothesis of no relationship or difference, and each implies a specific test — a correlation for the psychology example, an independent t-test for the education example, a paired t-test for the before-and-after health example. The hypothesis and the test are chosen together.
Significance level, power and sample size
Two further decisions shape whether your test can do its job. The significance level (alpha) is the threshold of evidence you require to reject the null — conventionally .05, occasionally .01 for stricter tests — and it is also your accepted risk of a Type I error. Statistical power is the probability of detecting a real effect if one exists, and it depends heavily on sample size: too small a sample may fail to detect a genuine effect, producing a misleading non-significant result.
For a strong study, ideally conduct a power analysis in advance to estimate the sample size needed to detect the effect you expect. Even if a full power analysis is beyond your scope, recruit as large a sample as is feasible, and interpret a non-significant result cautiously — it may reflect low power rather than a true absence of effect. Showing that you understand the link between alpha, power and sample size signals genuine methodological literacy, and it pre-empts the obvious examiner question about why your sample was the size it was.
Working with several hypotheses
Many studies test more than one hypothesis. Number them clearly (H1, H2, H3…), give each its own null, and make sure each maps to a specific analysis in your results chapter. Keep the set focused: a dissertation testing ten hypotheses usually lacks depth on any of them, and testing many hypotheses also raises the chance of a false positive somewhere — the multiple-comparisons problem.
If you do run many tests, be aware that the risk of at least one Type I error rises with the number of tests, and more advanced work applies a correction (such as Bonferroni) to guard against it. For most student dissertations the better advice is simpler: test a small number of well-justified hypotheses thoroughly, report each with its statistic, p-value and effect size, and tie each back to your research questions so the whole results chapter stays coherent.
Need watertight hypotheses and the right test? Our statistics-literate writers frame testable hypotheses, choose the correct test and interpret the results.
Related guides
- Research proposal writing services (hub)
- Research aims, objectives & questions
- T-tests, ANOVA & regression explained
Frequently asked questions
What is the difference between a null and an alternative hypothesis?
The alternative hypothesis (H1) states the relationship or effect you expect to find. The null hypothesis (H0) states that there is no relationship or effect. The statistical test assumes the null is true and asks how likely your data would be under it; if very unlikely (p < .05), you reject the null in favour of the alternative.
Do qualitative studies have hypotheses?
Generally no. Hypotheses belong to quantitative, deductive research that tests predictions. Qualitative research explores meaning and experience and uses open research questions instead. Writing a hypothesis for a qualitative study is a common mistake.
What does it mean to ‘operationalise’ a variable?
It means defining the variable in terms of how you will actually measure it — for example defining ‘sleep’ as ‘self-reported hours slept the night before the exam’. A hypothesis whose variables are not operationalised cannot be tested.
What is the difference between a one-tailed and a two-tailed hypothesis?
A one-tailed (directional) hypothesis predicts the direction of the effect; a two-tailed (non-directional) hypothesis predicts an effect without specifying direction. The choice, made before collecting data, determines whether you use a one- or two-tailed statistical test.
Can I say my results ‘prove’ my hypothesis?
No. Statistics weighs evidence rather than proving things. You reject the null hypothesis or fail to reject it, and report that the results ‘support’ the alternative. A non-significant result does not prove the null is true; it may reflect a small sample or low power.
What is the difference between an associational and a causal hypothesis?
An associational hypothesis predicts that two variables are related without claiming one causes the other; a causal hypothesis predicts that one variable produces a change in another. Only experimental or strong longitudinal designs support causal claims — a correlational design can establish association only.
How big should my sample be to test a hypothesis?
Large enough to give adequate statistical power to detect the effect you expect. Ideally you run a power analysis in advance to estimate the required sample size; if that is beyond your scope, recruit as large a feasible sample as possible and interpret any non-significant result cautiously, since it may reflect low power.
Can someone help me write and test my hypotheses?
Yes — our statistics-literate writers frame testable hypotheses, operationalise variables, choose the correct test and interpret the results. See our research proposal writing services page or place an order.
Need help framing testable hypotheses and choosing the right statistical test for your study? Place an order or explore our research proposal writing services — rated 4.4/5 across 871 verified Trustpilot and Sitejabber reviews.
