What is a hypothesis? How it differs from a theory

A hypothesis is a proposed explanation, or a proposed answer to a specific question, stated so that an observation or experiment could show it to be wrong. "Plants grow taller when music is played to them" is a hypothesis: it says what should happen, and a test can catch it out. A theory is something much bigger, a broad explanation that has survived many such tests and ties many facts together.

In short

• A hypothesis is a testable, specific claim about how something works, written before the test that will check it.
• It turns into a prediction ("if we do this, then we should see that"), and statistics compares the result with a null hypothesis of no effect.
• A hypothesis is never proven by one test; it is supported or rejected, and a theory is the large structure that well-tested hypotheses build.

What makes a statement a hypothesis?

A hypothesis has to do three things. It proposes an answer to a real question. It is specific enough that a test can give a clear verdict. And it can fail: some possible result would count against it. "Something in the environment affects plant growth" is too vague to fail. "Music affects plants in mysterious ways" can't fail at all. "Seedlings exposed to music for three hours a day grow taller in three weeks than seedlings kept in silence" can.

The everyday definition, "an educated guess", misses the point. A guess can be untestable; a hypothesis must state, at least implicitly, what would prove it wrong, which is the idea behind falsifiability. It also usually rests on what is already known. As the OpenStax biology textbook puts it, a hypothesis is "a suggested explanation for an event, which one can test", and a valid one must be testable and falsifiable.

A hypothesis is not the same as a prediction. The hypothesis is the general claim: music helps plants grow. The prediction is what that claim says will happen in one particular test, usually written "if… then…": if 20 seedlings get music and 20 don't, then the music group will be taller on day 21. One hypothesis can produce many predictions, and checking several of them is what makes the test strong.

How do you turn a question into a hypothesis?

Take the question many school science fairs ask: do plants grow faster with music? Here is how it becomes something testable, step by step. The numbers below are an illustration, not a real study.

1. Narrow the question. "Faster" becomes a measurement: height in centimeters after 21 days. "Plants" becomes one kind: bean seedlings of the same variety, planted the same day.
2. State the hypothesis. Bean seedlings that hear three hours of music a day grow taller over 21 days than seedlings kept in silence.
3. Name the variables. The independent variable is the one the experimenter changes: music or silence. The dependent variable is the one measured: height. The controlled variables are everything held the same: light, water, soil, temperature, pot size.
4. Write the null and alternative hypotheses. The null hypothesis, written H₀, is that music makes no difference: the two groups' average heights are equal. The alternative, H₁, is that the music group is taller on average.
5. Make the prediction. If H₁ is right, the music group's average height will be clearly greater than the silent group's, by more than the ordinary plant-to-plant variation would explain.
6. Design the test so it can fail fairly. Put 40 seedlings in identical pots, assign 20 to each group at random so that no stronger seedlings end up in one group, and have someone who doesn't know which plant is which do the measuring.

Suppose the music group averages 14.1 cm and the silent group 13.8 cm, with individual plants in both groups ranging from about 11 cm to 17 cm. A difference of 0.3 cm is small next to that spread, and a statistical test would most likely say it could easily arise by chance. The researcher would then fail to reject the null hypothesis. That is not the same as proving music does nothing: a bigger experiment might detect a small effect. How surprising a result would be if H₀ were true is exactly what a p-value measures, and the NIST/SEMATECH handbook sets out how the null and alternative are framed.

What does a hypothesis look like in real science?

In 1982, stress and lifestyle were thought to be the main causes of stomach ulcers. The Australian doctors Barry Marshall and Robin Warren proposed a different hypothesis: a bacterium, later named Helicobacter pylori, causes them. After studying biopsies from 100 patients, they found the organism in almost all of those with stomach inflammation or ulcers, as the Nobel Assembly's 2005 press release describes.

Their hypothesis made predictions that could have failed. If the bacterium causes ulcers, then killing it with antibiotics should cure them for good, where acid-reducing drugs only healed them for a while before they returned. Treatment studies showed exactly that, and Marshall famously swallowed a culture of the bacterium in 1984 and developed stomach inflammation. A hypothesis earns trust not by sounding right but by passing tests it could have failed. Marshall and Warren won the 2005 Nobel Prize in Physiology or Medicine. The bacterium is now known to cause more than 90% of duodenal ulcers and up to 80% of stomach ulcers, which a short course of antibiotics and acid inhibitors can cure.

How is a hypothesis different from a theory and a law?

In everyday speech, "theory" means a hunch, which is why "it's just a theory" sounds dismissive. In science the words sit on a different ladder. What is science? sets out the basic distinction; the table below adds the prediction and the null hypothesis, and what happens to each when a test goes against it.

Term	What it is	Scope	What a failed test does	Example
Hypothesis	A proposed, testable explanation for something specific	One question	Rejects it or forces a revision	H. pylori causes stomach ulcers
Prediction	What the hypothesis says will happen in one test	One experiment	Counts against the hypothesis behind it	Antibiotics will stop ulcers coming back
Null hypothesis	The default claim of no effect or no difference	One analysis	Is "rejected" when results would be very surprising under it	Antibiotics make no difference to relapse
Law	A description of a regular pattern, often an equation, without saying why	A wide class of events	Shows the conditions where it stops holding	Boyle's law: at constant temperature, a gas's pressure times its volume stays the same
Theory	A well-tested explanation of why a wide range of things happen	A whole field	Rarely overturns it; usually refines or limits it	The germ theory of disease

Two misreadings are common. A hypothesis doesn't become a law once it passes enough tests, and a law doesn't grow into a theory: they do different jobs, one describing and the other explaining. And a theory isn't a hypothesis waiting for promotion. It is built from many hypotheses that survived testing, such as Marshall and Warren's, which became one more strand of the germ theory.

Can a hypothesis ever be proven?

Not in the way a theorem in mathematics is proven. A hypothesis about the world is supported each time one of its predictions comes true, but no number of passed tests rules out the next one failing. Reasoning from many observations to a general rule is inductive, and it never reaches certainty; reasoning from a hypothesis to a prediction is deductive, and it is what makes the test sharp. Deductive vs inductive reasoning explains the difference.

What a single failed prediction shows is less clear-cut than it looks. The failure might lie in the hypothesis, or in an assumption about the equipment, the sample or the measurement. That is why scientists repeat experiments and test several predictions from the same hypothesis before rejecting it, and why a hypothesis that keeps passing tests becomes something people act on. It is also why scientists often write models, simplified pictures of how a system works, to turn a hypothesis into precise numbers: what a scientific model is picks up there. And all of it rests on what counts as evidence in the first place.

Check yourself

• Without looking back, turn "does coffee help people concentrate?" into a hypothesis, with its independent, dependent and controlled variables.
• What is the difference between a hypothesis and a prediction, and why does one hypothesis need several predictions?
• Why does failing to reject the null hypothesis not prove there is no effect?

Sources

• The Science of Biology, in Biology 2e — OpenStax, Rice University
• Null and Alternative Hypotheses, in Introductory Statistics 2e — OpenStax, Rice University
• What are statistical tests? — NIST/SEMATECH e-Handbook of Statistical Methods
• The Nobel Prize in Physiology or Medicine 2005, press release — Nobel Assembly at Karolinska Institutet, 2005
• How science works — Understanding Science, University of California Museum of Paleontology
• Hypothesis — Wikipedia