October 7, 2026

What Is a Hypothesis? How to Write and Test One in High School

TL;DR A hypothesis is a statement about what you expect to find, written so precisely that your own results could prove you wrong. Below, you'll find how a hypothesis differs from a theory, a prediction; the five parts every testable hypothesis contains; what null and alternative hypotheses actually mean; a six-step method for writing one;…

TL;DR

A hypothesis is a statement about what you expect to find, written so precisely that your own results could prove you wrong. Below, you’ll find how a hypothesis differs from a theory, a prediction; the five parts every testable hypothesis contains; what null and alternative hypotheses actually mean; a six-step method for writing one; and how to test it in a school project. Getting this right early is the difference between a project that produces a result and one that produces an opinion, and it is usually the first thing a research mentor fixes. Horizon Academic Research Program pairs you 1:1 with a PhD scholar or professor for a trimester of guided research, ending in a 20-page paper. Apply here, or come to the next info session first.

You have a research question and a teacher asking for your hypothesis by Friday. You write something like ‘I think social media affects teenagers’ sleep.’ It sounds reasonable. The problem is that no result you could collect would show it to be false, and that makes it useless as a hypothesis. Here is how to write one that works, and what to do with it once you have.

What is a hypothesis, in plain terms?

A hypothesis is a testable statement about what you expect to happen, written before you collect your data. It is a tentative answer to your research question, and the word tentative is doing real work. You are not claiming to know.

Two features separate a hypothesis from a guess.

You have read around your topic, you know roughly what other people have found, and your expectation follows from that. A hypothesis pulled out of thin air is a hunch with better grammar.
It is specific enough to be wrong. If you cannot describe a result that would make you abandon it, you have not written a hypothesis yet.

A good hypothesis is vulnerable. It sticks its neck out. ‘Social media affects sleep’ is safe, which is exactly the problem. ‘Among 15- to 17-year-olds, more than two hours of phone use after 9pm is associated with falling asleep later than peers who use their phones for under 30 minutes’ can be checked, and if the data comes back flat, you know.

How is a hypothesis different from a theory, a prediction, and a thesis statement?

These four get used interchangeably in everyday speech and mean quite different things in research.

Term	What it is	Example
Hypothesis	A testable statement about a relationship you expect to find, written before you collect data	Instrumental music while studying improves recall test scores compared with silence
Theory	A well-established explanation supported by a large body of evidence, which generates many hypotheses	Cognitive load theory, which explains how working memory limits learning
Prediction	What specifically should happen in one particular study if the hypothesis is right	In my study, the music group will average at least 2 points higher out of 20
Thesis statement	The central argument of a written paper, which you have already concluded and will now defend	Music’s effect on studying depends more on the task than on the music itself

A hypothesis is written before you know the answer. A thesis statement is written after. If you are writing a literature review or an argumentative essay rather than running a study, what you need is a thesis, and our guide to writing a thesis statement in 10 steps covers that instead.

What makes a hypothesis testable?

A testable hypothesis has five parts. Missing any one of them is what makes a hypothesis feel vague.

The population. Who or what you are studying. ‘People’ is not a population. ’15 to 17-year-olds at my school’ is.
The independent variable. The thing you change, or the thing that differs between groups. Describe it precisely enough that someone else could set it up the same way.
The predicted direction. Higher, lower, faster, more frequent. Saying a variable ‘affects’ another is not a direction.
The dependent variable, and how you will measure it. Not ‘memory’ but ‘score on a 20-item recall test’. Not ‘wellbeing’ but ‘score on a named, published scale’.
The comparison. Compared with what? Silence, no treatment, a different group, last year’s figures. A prediction with nothing to compare against cannot come out either way.

‘Stress’ is a concept. ‘Self-reported stress on a 1 to 10 scale, collected at the same time each day’ is a variable. You cannot measure a concept. You can only measure the specific thing you decide will stand in for it, and being explicit about that choice is what makes your project defensible.

A quick test before you commit. Ask yourself: what result would make me say my hypothesis was wrong? If you can describe that result in one sentence, you are ready. If you cannot, go back to the five parts and find the one that is missing.

What is the difference between a null hypothesis and an alternative hypothesis?

Any time you plan to run a statistical test, you are working with two hypotheses at once.

	Null hypothesis (H0)	Alternative hypothesis (H1)
What it says	There is no effect, no difference, no relationship	There is an effect, difference, or relationship
Example	Music while studying makes no difference to recall scores	Music while studying changes recall scores
Its role	The default position, assumed true until the data gives you reason to doubt it	What you actually expect, and what your study is designed to look for
What a test does	You test whether your data is surprising if the null were true	You never test this one directly
Possible outcomes	Reject it, or fail to reject it	Supported, or not supported

The phrase to learn is ‘fail to reject’. Statistical testing does not let you accept the null hypothesis or prove that nothing is going on. It only ever tells you whether your data would be surprising in a world where the null was true. If the data is not surprising, you have not shown there is no effect. You have shown that your study did not detect one, which might be because there is nothing there, or because your sample was too small to see it.

Your alternative hypothesis can be directional (music improves scores) or non-directional (music changes scores, up or down). Directional is stronger and riskier, so use it when previous research gives you a reason to expect a particular direction. Use non-directional when the literature is genuinely mixed.

If your project does not involve statistics, you do not need a null hypothesis. A well-specified hypothesis is enough. The null and alternative pair belongs to statistical testing specifically, and writing one into a qualitative history project because a template told you to is a common and avoidable mistake.

How do you write a hypothesis, step by step?

Start with a question, not a topic. ‘Sleep’ is a topic. ‘Does phone use after 9pm predict later sleep onset in 15 to 17 year olds?’ is a question. Our guides to choosing a research topic and picking broad versus niche topics cover how to narrow one down.
Read enough to know what is already known. You are looking for what previous studies found, what they measured, and where they disagree. Google Scholar, Semantic Scholar and the free databases in our guide will get you there without a university login.
Write your first answer in plain language. One sentence, no jargon, no hedging. It will be too vague. That is fine, it is a draft.
Add the five parts. Population, independent variable, direction, dependent variable, comparison. This is the step that turns a sentence into a hypothesis.
Check it against your actual methods. Can you really collect that measure, from those people, in the time you have? A hypothesis you cannot test with the resources you have is not a hypothesis, it is a wish. Shrink it until it fits.
Write the null version if you are using statistics. State it plainly as no difference or no relationship, and note which statistical test you will use before you collect anything.

Write the final version down somewhere dated, before you collect a single data point. This matters more than it sounds. It is your evidence that you predicted the result rather than noticed it afterwards, and both research competitions and journals take that seriously.

Can you ever prove a hypothesis true?

No, and understanding why will make your write-up noticeably more sophisticated than most of what you are competing against. The argument is old and simple. A universal claim covers infinitely many cases, so no finite number of confirming observations can establish it. A single genuine counter-example, though, is enough to bring it down. Karl Popper built his account of science on exactly that asymmetry: a claim that nothing could ever contradict is not doing scientific work.

So the right verbs are support, are consistent with, provide evidence for. Not prove, not confirm, not show conclusively. This is not academic fussiness; it is accuracy, and examiners notice.

The same applies to statistics, where the misunderstanding is more specific. The American Statistical Association’s statement on p-values says plainly that p-values ‘do not measure the probability that the studied hypothesis is true, or the probability that the data were produced by random chance alone.’ A p-value tells you how compatible your data is with a particular model, nothing more. The ASA also notes that a p-value does not measure the size of an effect or the importance of a result, which is why a ‘significant’ finding can still be far too small to matter.

Two sentences worth stealing for your own discussion section. ‘These results are consistent with the hypothesis that X.’ And, when they are not, ‘These results do not support the hypothesis that X.’ Both are honest, and both sound like someone who knows what they are doing.

What if your results don’t support your hypothesis?

Then you have a finding, and your project is complete. This is the single biggest source of unnecessary panic in high school research, and it rests on a misunderstanding of what you are being assessed on. Nobody is marking you on whether your guess was right. They are marking you on whether your method could have told the difference.

What to do:

Report it plainly, in the results, without apology. ‘The difference between groups was not statistically significant’ is a complete, respectable sentence.
Say what it might mean. There may be no effect. The effect may be too small for your sample to detect. Your measure may not have captured what you intended. All three are worth discussing and none of them is an admission of failure.
Do not rewrite your hypothesis to match your results. Changing the prediction after seeing the data and presenting it as what you expected all along is a genuine research integrity problem. It also tends to be obvious.
Do not go hunting through your data for something that did come out. Testing twenty things and reporting the one that worked produces findings that will not replicate, which is why pre-committing in step one matters.
Under ISEF rules, if you change your research plan after approval, the change has to be reviewed and approved before you carry on collecting data. See the ISEF International Rules for what that involves.

A null result, honestly reported, is publishable at high school level. Journals such as Journal of Emerging Investigators and Journal of Student Research care about method quality rather than exciting outcomes. Our guides to publishing research in high school and research journals for high school students go through what each one looks for.

Do all research projects need a hypothesis?

No, and forcing one onto a project that does not need it makes the work worse. Hypotheses belong to research that tests a specific expectation. Plenty of good research does something else.

Type of project	What it uses instead
Experiment or quantitative study	A hypothesis, usually with a null and alternative pair
Exploratory or qualitative study	Research questions, because the point is to find out what is there rather than to confirm an expectation
Literature review	A thesis or a guiding question, since you are synthesizing existing work rather than generating data
Case study	Research questions, sometimes with propositions that are looser than a formal hypothesis
Historical or archival research	A research question or an argument, supported by sources

If your teacher has asked for a hypothesis and your project is genuinely exploratory, say so and propose research questions instead. That conversation demonstrates more understanding than quietly inventing a hypothesis you cannot test. The key components of a research paper look slightly different for each of these, and getting the match right at the start saves a rewrite later.

Why does a clear hypothesis make your project stronger?

Beyond marks, a well-specified hypothesis does four practical things.

It forces your methods to be decided early. You cannot write the five parts without settling what you will measure and how, which means the hard decisions happen before you waste time collecting the wrong thing.
It makes your project defensible under questioning. Science fair judges and interviewers ask what would have disproved your idea. Students who can answer that immediately sound completely different from students who cannot. See our guides to science competitions and research conferences for what that questioning looks like.
It keeps you honest. A prediction on record before data collection stops you from drifting toward whatever the numbers happen to show.
It makes the work publishable. Every journal that takes high school work expects a clearly stated question or hypothesis and a method that matches it. It is also what makes a paper usable later, whether you are submitting to a competition or conference or writing about the project on an application.

If you want a sense of what this looks like at full strength, our pieces on what you can actually do with a high school research paper and why research experience helps a college application both come back to the same thing: the quality of the question, not the size of the result.

Frequently asked questions

What is a hypothesis in simple terms?

A hypothesis is a testable statement predicting what you expect to find, written before you collect data. It names who you are studying, what you are changing, what you are measuring, which direction you expect, and what you are comparing against.

What is an example of a good hypothesis?

‘Among high school students, listening to instrumental music while studying will produce higher scores on a 20-item recall test than studying in silence.’ It names the population, the independent variable, the direction, the measured outcome and the comparison, so a result could clearly contradict it.

What is the difference between a hypothesis and a theory?

A hypothesis is a single testable statement; a theory is a well-established explanation supported by a large body of evidence. A theory generates hypotheses, not the other way round. Everyday use of ‘theory’ to mean ‘hunch’ is the reverse of the research meaning.

What is a null hypothesis?

The null hypothesis states that there is no effect, no difference or no relationship. It is the default position in statistical testing. You either reject it or fail to reject it, and failing to reject it is not the same as proving it true.

Does a hypothesis have to use the if-then format?

No. If-then is one useful template, not a rule, and it fits experiments better than surveys or observational studies. What matters is that all five parts are present, whatever sentence structure you use.

Can a hypothesis be proven?

No. Hypotheses can be supported or refuted, never proven true, because no finite set of observations can establish a universal claim while a single counter-example can undermine one. Write ‘the results support’ rather than ‘the results prove’.

What happens if my hypothesis is wrong?

Nothing bad. A result that does not support your hypothesis is still a result, and it is assessed the same way as any other. Report it, discuss why it might have come out that way, and do not rewrite the hypothesis afterwards.

How long should a hypothesis be?

One sentence, usually 15 to 35 words. If it runs longer you are probably testing two things at once, which is worth splitting into separate hypotheses so each can succeed or fail on its own.

Do I write the hypothesis before or after my literature review?

After. The reading is what gives you a reason to expect one outcome rather than another, and it stops you proposing something that has already been settled or that nobody can measure.

Does Horizon help students develop research questions and hypotheses?

Yes. Narrowing a topic into a testable question with your mentor is where most Horizon projects start, and it is usually the part students say changed how they think. Apply here or join an info session.

Resources

External

Scribbr on writing a hypothesis, a clear walkthrough of structure and the null and alternative pair
Scribbr research methodology library, plain-language explanations of variables, designs and terms
Scribbr on sampling methods, for deciding who and how many before you test anything
The American Statistical Association’s statement on p-values, the authoritative six principles on what a p-value can and cannot tell you
Stanford Encyclopedia of Philosophy on Karl Popper, for why falsifiability matters and why proof is not available
ISEF International Rules and Guidelines, including what happens if you change your research plan mid-project
ISEF rules for human participants research, required reading before any study involving other people
Google Scholar and Semantic Scholar, the two free tools most student literature searches are built on
Journal of Emerging Investigators and Journal of Student Research, High School Edition, both publish well-designed studies regardless of whether the hypothesis was supported

More from Horizon

Image source: Horizon Academic Research Program

Bridget Carl

Bridget is a Program Director at Horizon Academic. She holds a BS in Public Health and Economics from The George Washington University, where she earned special distinction from her research on the Affordable Care Act individual mandate and open healthcare markets in the US. She also holds a MS in Data Analytics and Policy from Johns Hopkins University where she continued her research focus on public policy and social services, fueled with data science, machine learning and modeling, and cloud computing.

All author posts