Writing a Hypothesis: Null, Alternative and Testable Statements

A hypothesis is a prediction you can test. It sits at the heart of experimental and many survey-based projects, because it tells you what to measure, how to analyze and what a result would mean. Students often find it confusing, partly because it overlaps with research questions and partly because of the null and alternative language used in statistics.

This guide explains what a hypothesis is, how to write one, how null and alternative forms work and how to avoid common errors. The examples are illustrations only.

What a hypothesis does

A hypothesis turns a question into something falsifiable. It names the variables, predicts how they relate and can be contradicted by evidence. It is not a hunch or a statement of belief. It should follow from theory or from earlier studies, and it should be written before you collect data.

Our guide to research questions and objectives explains how to write the question that your hypothesis answers.

Qualities of a good hypothesis

• Testable. You can gather data that could support or contradict it.
• Specific. It names the variables and the groups.
• Clear. A reader understands what is predicted.
• Grounded. It is based on theory or evidence.
• Falsifiable. There is a possible result that would show it is wrong.
• Simple. It makes one prediction at a time.

A statement such as "Social media is bad for students" cannot be tested as it stands, because "bad" is vague and no result would clearly contradict it.

Variables

A hypothesis relates two or more variables.

• Independent variable: the factor you think has an effect, or the one you manipulate.
• Dependent variable: the outcome you measure.
• Control or confounding variables: other factors that could affect the outcome, which you hold constant or account for.

In "Students who practice in spaced sessions will score higher than students who practice in one block," the independent variable is practice schedule and the dependent variable is exam score.

Operational definitions

You must say how each variable will be measured. This is an operational definition. "Study time" could mean hours reported in a diary, minutes logged in an app or hours in the library. "Well-being" could mean a score on a particular questionnaire. Without an operational definition, a hypothesis cannot be tested, and readers cannot judge your work.

Spaced practice is defined as three 20-minute sessions on separate days within one week. Exam performance is measured by the score out of 100 on a unit test.

Null and alternative hypotheses

Statistical testing works through a pair of hypotheses.

• Null hypothesis (H0): there is no difference or no relationship in the population.
• Alternative hypothesis (H1 or Ha): there is a difference or a relationship.

For the spaced practice example:

H0: There is no difference in mean exam scores between students who practice in spaced sessions and students who practice in one block.

H1: Students who practice in spaced sessions have higher mean exam scores than students who practice in one block.

You collect data and ask how likely your results would be if the null were true. If they would be very unlikely, you reject the null in favor of the alternative. If not, you fail to reject it. Notice the wording: you do not "prove" the alternative, and you do not "accept" the null.

Directional and non-directional hypotheses

Use a directional hypothesis when theory or earlier evidence gives you a clear reason to predict the direction. Otherwise use a non-directional one. Whichever you choose, decide before you see the data, since choosing afterward is a form of cherry-picking.

Types of hypotheses by purpose

• Difference: groups differ on an outcome.
• Relationship: two variables are associated.
• Effect: a change in one variable produces a change in another.
• Prediction: one variable predicts another.

A relationship hypothesis, such as "Higher weekly work hours will be associated with lower course grades," does not claim causation, which is an important distinction to preserve in your write-up.

How to write one step by step

1. Start with your research question.
2. Identify the variables, independent and dependent.
3. Review the literature for theory or findings that support a prediction.
4. Choose the type: difference, relationship or effect.
5. Decide on direction, if you have grounds.
6. Write the alternative hypothesis in one clear sentence.
7. Write the null hypothesis as its opposite.
8. Add operational definitions in your methods.
9. Check testability: what data would support it, and what would contradict it?

More examples

Multiple hypotheses

Many studies test more than one hypothesis, numbered H1, H2 and so on. Keep each clear and separate, and link them to your sub-questions. A large number of tests increases the chance of a false positive, so avoid testing everything you can think of. State your hypotheses in advance, and report all the tests you ran, including the ones that did not support your predictions. Our guide to data analysis basics for students explains how tests relate to hypotheses.

Interpreting results

When you write up results, link back to each hypothesis.

• Supported: the data are consistent with the alternative hypothesis, within the limits of your design.
• Not supported: the data do not provide enough evidence to reject the null. This does not mean there is no effect, only that you did not find one with these data.
• Mixed or unexpected: report honestly and discuss possible explanations.

Avoid the word "proved". Say "the results supported" or "the results did not support".

Hypotheses in qualitative research

Most qualitative projects do not use hypotheses, because they aim to explore, not to test. They use open research questions, and may produce propositions or theory as findings. If your program asks for hypotheses in a qualitative project, check what is meant. Sometimes the term refers to working assumptions or expected themes, which you should hold loosely.

Common mistakes

• Writing a question instead of a statement.
• Vague variables with no operational definition.
• Untestable claims such as "Technology is harmful."
• Several predictions in one hypothesis.
• Choosing direction after seeing the data.
• Claiming causation from a correlational design.
• Saying you proved the hypothesis.
• Writing hypotheses after the analysis.
• Confusing null and alternative.

A quick checklist

• Is the hypothesis a clear, testable statement?
• Are the independent and dependent variables named?
• Are the variables operationally defined?
• Is there a null and an alternative?
• Is the direction justified?
• Does it follow from theory or evidence?
• Did I write it before collecting data?

Final thoughts

A good hypothesis says exactly what you expect to find and how you would know if you were wrong. Write it as a clear statement, define your variables, pair it with a null and keep your language careful when you report the results. The discipline of stating a prediction in advance is one of the best habits in quantitative research.