Hypothesis vs Theory vs Law in Science: Definitions and Examples

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-10-02

A hypothesis, a theory, and a law are three different kinds of scientific explanation, and they differ in scope, not in how well they are supported. A hypothesis is a narrow, testable statement about the natural world. A theory is a broad, well-substantiated explanation that ties together facts, laws, inferences, and tested hypotheses. A law is a compact description of how something behaves under stated conditions.

You will meet this vocabulary constantly: in a genetics problem set that asks you to state a null hypothesis, in a journal club where someone dismisses evolution as "just a theory," in a lab meeting where a senior researcher calls a pattern a law and then lists its exceptions. Getting the distinctions right changes how you write your own papers, how you read claims in the news, and how you respond when a reviewer asks what your hypothesis actually predicts.

Quick Answer

A hypothesis is a testable statement about the natural world that can be used to build more complex inferences and explanations [5]. It is not a guess, educated or otherwise [2].
A theory is a comprehensive explanation of some aspect of nature supported by a vast body of evidence [1]. Theories make predictions about events not yet observed [1].
A law is a descriptive generalization about how some aspect of the natural world behaves under stated circumstances [5]. Laws may have exceptions and can be modified or rejected [3].
Hypotheses do not become theories, and theories do not become laws. All three are explanations that differ in breadth [2][3].
A prediction is what you expect to observe if a hypothesis is true. Asking students to "hypothesize" an outcome usually means asking for a prediction [3].
In statistics, the null hypothesis ($H_0$) states that there is no difference between groups, and the alternative hypothesis ($H_1$) is the researcher's stance based on experience or the literature [6].

Definitions and Intuition

Start with the smallest unit. A hypothesis is a proposed explanation for a fairly narrow set of phenomena, built from prior experience, background knowledge, preliminary observations, and logic [2]. The National Academies describe it as a testable statement about the natural world that can be used to build more complex inferences and explanations [5]. The word "testable" carries weight here. Any scientific explanation has to be framed so that observational evidence could support it and other evidence could refute it. If no possible observation could count against an idea, that idea cannot be subjected to scientific testing [1]. AAAS puts the same point practically: a hypothesis should suggest what evidence would support it and what evidence would refute it, and one that cannot in principle be put to the test may be interesting but is not likely to be scientifically useful [4].

A theory sits at a much wider scale. UC Berkeley describes theories as broad explanations for a wide range of phenomena that are concise, coherent, systematic, predictive, and broadly applicable, often integrating and generalizing many hypotheses [2]. The National Academies define a scientific theory as a comprehensive explanation of some aspect of nature supported by a vast body of evidence, and note that many theories are so well established that no new evidence is likely to alter them substantially [1]. One of the most useful properties of a theory is that it generates predictions about natural events or phenomena that have not yet been observed [1].

A law is a different animal again. In science, the term usually refers to a generalization about data, a compact way of describing what we would expect to happen in a particular situation [3]. Some laws are non-mechanistic statements of how observable quantities are related. The ideal gas law is the standard example: it relates pressure, volume, and temperature without explaining why gases behave that way, and real gases do not precisely conform to it [3]. Other laws carry more mechanistic content. Mendel's first law describes how genes are distributed to gametes and offspring, and whether an idea gets called a law depends largely on its discipline and the period when it was developed [3].

One more term belongs in this set. A fact, in the National Academies' usage, is an observation, measurement, or other form of evidence that can be expected to occur the same way under similar circumstances [1]. The NAS-adapted definition is simpler: an observation that has been repeatedly confirmed, such as the finding that human cells have 23 pairs of chromosomes [5].

Term	Scope	Example	What it does
Hypothesis	Narrow, testable statement	Surface area affects dissolution rate [3]	Generates predictions that can be tested
Theory	Broad explanation integrating many findings	Cell theory [5]	Explains a wide range of phenomena and predicts new ones
Law	Descriptive generalization under stated conditions	Ideal gas law, $PV = nRT$ [10]	Describes what to expect in a defined situation
Fact	Repeatedly confirmed observation	Human cells have 23 pairs of chromosomes [5]	Provides the evidence base

How to Test a Hypothesis: The Statistical Machinery

An explanatory hypothesis does not get tested directly. It generates a prediction, and that prediction gets formalized as a statistical hypothesis. StatPearls defines the null hypothesis as the statement of no statistical difference between groups based on the stated research hypothesis, while the alternative hypothesis lets the researcher take a stance based on experience or the literature [6]. Some textbooks call $H_1$ the research hypothesis.

For a categorical prediction, the chi-square goodness-of-fit test is the usual tool:

$$ \chi^2 = \sum \frac{(O_i - E_i)^2}{E_i} $$

Here $O_i$ is the observed count in category $i$, $E_i$ is the expected count under the null hypothesis, and the sum runs over all categories. Each term measures squared deviation from expectation, scaled by the size of the expectation, so a deviation of 20 counts matters far more when you expected 30 than when you expected 5,000. The degrees of freedom are the number of categories minus one.

The p-value that comes out of this test is the probability that the observed effect would have occurred by chance if, in reality, there was no true effect [6]. By convention, $p < 0.05$ or $p < 0.01$ is considered statistically significant [6]. For a chi-square test with one degree of freedom, the critical value at $\alpha = 0.05$ is 3.841, so a statistic above that threshold leads to rejecting the null.

Worked Example

Mendel's law of segregation makes a sharp prediction, and we can test it with a null hypothesis. The explanatory hypothesis is that each pea plant carries two copies of a heritable factor for seed shape, and the two copies separate equally into gametes [8]. The prediction follows: selfing $Rr$ F1 plants should give an F2 ratio of 3 round to 1 wrinkled.

The statistical setup:

$H_0$: the F2 population follows a 3:1 ratio, so $p_{\text{round}} = 0.75$
$H_1$: the ratio differs from 3:1, so $p_{\text{round}} \neq 0.75$

The data come from Mendel's seed-shape counts: 5,474 round and 1,850 wrinkled, for $n = 7{,}324$ and an observed ratio of 2.959:1 [9].

Expected counts under $H_0$:

Round: $0.75 \times 7{,}324 = 5{,}493$
Wrinkled: $0.25 \times 7{,}324 = 1{,}831$

Chi-square:

$$ \chi^2 = \frac{(5474 - 5493)^2}{5493} + \frac{(1850 - 1831)^2}{1831} = 0.0657 + 0.1972 = 0.263 $$

With 2 categories minus 1, $df = 1$, and $p = 0.608$. The critical value is $\chi^2(0.95, df=1) = 3.841$, so we fail to reject $H_0$: the data are consistent with a 3:1 ratio. An exact binomial test gives the same $p = 0.608$.

The same test on flower color (705 violet, 224 white, $n = 929$) gives expected counts of 696.75 and 232.25, $\chi^2 = 0.391$, and $p = 0.532$. Again, fail to reject.

In Python:

from scipy.stats import chisquare
chisquare([5474, 1850], [5493, 1831])  # statistic 0.2629, pvalue 0.6081

Read the result carefully. Failing to reject $H_0$ does not prove the law of segregation. It means the observations did not contradict the prediction the law generated. That is a weaker and more honest statement, and it is the correct one.

Reading the Result Without Overreading It

The p-value answers one narrow question: how often would data like these appear if the null hypothesis were true? It does not tell you the probability that the null is true, and it does not measure the size or importance of an effect. A tiny p-value in a study with a trivial effect size is still a tiny effect.

Failing to reject $H_0$ is not evidence that $H_0$ is true. It only means the data are compatible with it at the chosen alpha. A small sample can fail to reject a null for the simple reason that it lacks the power to detect a real deviation. This is why Mendel's large F2 counts matter: with 7,324 seeds, the test has enough resolution to detect meaningful departures from 3:1.

There is also a conceptual gap between the explanatory hypothesis and the statistical one. The law of segregation is a claim about how alleles move through meiosis. $H_0$ is a claim about a proportion in a population. The explanatory hypothesis generates the prediction that $H_0$ formalizes, but they are not the same object. Confusing them leads to writing discussion sections that claim to have proven a biological mechanism when the test only failed to reject a ratio.

Comparisons With Commonly Confused Terms

Theory vs law. A theory explains; a law describes. Theories are deep explanations that apply to a broad range of phenomena and may integrate many hypotheses and laws, while laws are often compact statements of how observable quantities relate [3]. Neither is superior to the other. They do different jobs.

Scientific theory vs hypothesis. The difference is breadth. A hypothesis addresses a fairly narrow set of phenomena; a theory addresses a wide range and often generalizes many hypotheses [2]. A supported hypothesis is not upgraded to a theory. UC Berkeley explicitly corrects this misconception: hypotheses cannot become theories, and theories cannot become laws [2]. The three are like apples, oranges, and kumquats, which cannot grow into one another [3].

Hypothesis vs prediction. A prediction is what you expect to observe. A hypothesis has explanatory power. When a textbook asks you to "hypothesize" that table salt dissolves faster than rock salt, it is asking for a prediction. The explanatory version would be that the amount of surface area a substance has affects how quickly it can dissolve, which is testable because it generates expectations such as powdered sugar dissolving faster than granular sugar [3].

Is a theory just a guess? No. The everyday meaning of "theory" as a hunch or speculation is quite different from the formal scientific definition [1]. Calling something "just a theory" conflates the two and ignores that a scientific theory is a powerful explanation for a broad set of observations supported by many lines of evidence [2].

Common Mistakes

Treating the terms as a ladder. The idea that a hypothesis becomes a theory and a theory becomes a law is wrong. They differ in breadth, not in level of support [2][3].
Calling a prediction a hypothesis. "I hypothesize that group A will have a higher mean" is a prediction. A hypothesis explains why, and that explanation is what makes the prediction testable [3].
Saying a law is proven and cannot change. Laws may have exceptions and may be modified or rejected based on new evidence [3]. Mendel's law of independent assortment is violated by linked genes on the same chromosome, which shows that laws carry stated conditions [8].
Treating failure to reject $H_0$ as proof of $H_0$. It means the data are compatible with the null at the chosen alpha, nothing more.
Using "theory" in the everyday sense in a scientific context. A hunch and a scientific theory are different things, and the National Academies flag the gap between the two meanings directly [1].
Assuming a significant p-value validates the whole explanatory hypothesis. The test evaluates one prediction under one set of assumptions. Other theories may fit the same observations as well or better, and the testing, improving, and occasional discarding of theories goes on all the time [4].

Limitations

Usage of "law" varies by discipline and era. UC Berkeley stresses that whether an idea is called a law is partly historical, so a rigid hierarchy should not be taught [3]. What chemistry calls a law, biology might call a principle or a postulate.

Philosophers of science debate falsifiability as the criterion that separates science from non-science. The practical standard used by the National Academies and AAAS is testability: there must be possible observational consequences that could support an idea and others that could refute it [1][4]. That standard is workable for research, but it is not the final word in the philosophy literature.

Statistical null and alternative hypotheses are not the same as explanatory scientific hypotheses. Some textbooks use "research hypothesis" for $H_1$ to keep the two apart, and the terminology is not fully standardized.

Frequently Asked Questions

What is the difference between a hypothesis and a theory?

A hypothesis is a narrow, testable statement about the natural world [5]. A theory is a broad, well-substantiated explanation that incorporates facts, laws, inferences, and tested hypotheses [5]. The difference is scope, not strength of evidence.

Is a scientific theory just a guess?

No. The everyday use of "theory" to mean a hunch is different from the scientific definition, which requires a vast body of supporting evidence [1]. A scientific theory also makes predictions about phenomena not yet observed [1].

What is a scientific law, and can it have exceptions?

A law is a descriptive generalization about how some aspect of the natural world behaves under stated circumstances [5]. Laws may have exceptions and can be modified or rejected based on new evidence [3]. Mendel's law of independent assortment, for example, is violated by linked genes [8].

How is a hypothesis different from a prediction?

A prediction states what you expect to observe. A hypothesis explains why you expect it. Asking students to hypothesize an outcome is usually asking for a prediction, while a genuine hypothesis has explanatory power and generates testable expectations [3].

What is a null hypothesis in this context?

The null hypothesis states no statistical difference between groups based on the stated research hypothesis, while the alternative hypothesis reflects the researcher's stance from experience or the literature [6]. The explanatory hypothesis generates a prediction, and the null hypothesis formalizes that prediction for a statistical test.