A null hypothesis is a formal statement that there is no effect, no relationship, or no difference between the variables or groups being studied. In any statistical test, understanding what is a null hypothesis is the starting point: it is the default claim you try to disprove — not prove — using data. Researchers label it H₀ (read “H-naught” or “H-zero”) and pair it with an alternative hypothesis (H₁) that captures the effect they expect to find.
If you searched “what does null mean” in a statistics context, note that the word is unrelated to null values in programming. In statistics, null means “no effect” — the baseline position you hold until evidence forces you to change it.
What Is a Null Hypothesis?
The null hypothesis meaning in research is the default assumption of no change. When a scientist asks “Does this drug lower blood pressure?”, the null hypothesis says “No — any observed difference is due to chance.” The test then asks: how likely is the observed data if the null hypothesis were actually true?
Formally, the null hypothesis always contains an equality or no-difference statement:
- “There is no difference between group A and group B.”
- “The correlation between X and Y is zero.”
- “The population mean equals a specified value.”
The notation H₀ comes from early twentieth-century work by Karl Pearson and Ronald Fisher on statistical inference. The subscript zero signals zero effect: start by assuming nothing interesting is happening, then let data push you away from that assumption if they are strong enough.
The Paired Hypotheses: H₀ and H₁
Every statistical test involves two competing claims. Here are examples of hypothesis pairs across common test types:
| Test | Null hypothesis (H₀) | Alternative hypothesis (H₁) |
|---|---|---|
| Two-sample t-test | The two means are equal (μ₁ = μ₂) | The means differ (μ₁ ≠ μ₂) |
| Correlation test | No linear relationship exists (ρ = 0) | Some linear relationship exists (ρ ≠ 0) |
| Proportion test | The proportion equals a target (p = p₀) | The proportion differs (p ≠ p₀) |
| Chi-square test | The variables are independent | The variables are associated |
H₁ is the claim you are looking for evidence of. H₀ is what you accept by default if the evidence is not strong enough to overturn it. The logic mirrors a courtroom: the defendant (H₀) is presumed innocent until proven guilty beyond reasonable doubt.
According to Penn State STAT 200 Open Learning Initiative, “6.1 — Introduction to Hypothesis Testing”, “the burden of proof lies with the researcher who believes there is an effect” — the null hypothesis is the position that must be overcome.
How to State a Null Hypothesis
Stating H₀ correctly is a mechanical skill once you know the pattern. Three steps cover every case.
Step 1: Identify the parameter. Decide what quantity you are testing — a mean (μ), a proportion (p), a correlation (ρ), or a variance (σ²).
Step 2: Write the equality claim.
A null hypothesis always uses = (or ≤/≥ for one-tailed boundaries, but equality is the default). If you are testing whether a mean equals 100:
H₀: μ = 100
Step 3: Specify the direction in H₁.
The alternative hypothesis uses ≠ (two-tailed), > (right-tailed), or < (left-tailed):
Two-tailed: H₁: μ ≠ 100
Right-tailed: H₁: μ > 100
Left-tailed: H₁: μ < 100
Choose the direction before collecting data, based on the research question — not after looking at the results.
Worked Example: Stating H₀ for a Drug Trial
A pharmaceutical company claims a new drug reduces systolic blood pressure. Researchers want to test whether it changes blood pressure at all.
- Parameter: Mean change in blood pressure, μ (in mmHg)
- Null hypothesis: No change — the mean difference is zero
H₀: μ = 0 (no change in blood pressure)
H₁: μ ≠ 0 (some change, either direction)
If researchers specifically expected the drug to lower blood pressure rather than simply change it, they would use a one-tailed test:
H₀: μ = 0
H₁: μ < 0 (blood pressure decreases)
The direction in H₁ must match the scientific question before data collection begins.
Null Hypothesis Examples Across Research Areas
A null hypothesis example from one field looks structurally identical to one from another. The parameter name changes; the no-effect logic stays the same.
Example 1 — Education Research
Question: Does a new teaching method improve exam scores compared to the traditional method?
H₀: μ_new = μ_traditional (mean scores are equal)
H₁: μ_new ≠ μ_traditional (mean scores differ)
Researchers collect scores from two groups, run a two-sample t-test, and compare the p-value to α = 0.05. If p ≤ 0.05, they reject H₀ and conclude the methods produce different outcomes on average.
Example 2 — Medical Research (Proportion)
Question: Does a vaccine cut infection rate below the baseline rate of 30%?
H₀: p = 0.30 (infection rate equals baseline)
H₁: p < 0.30 (infection rate is lower)
This is a one-tailed test because researchers care only about a decrease. A two-tailed version (H₁: p ≠ 0.30) would detect any change in either direction.
Example 3 — Business Analytics (Correlation)
Question: Is there a linear relationship between advertising spend and weekly sales?
H₀: ρ = 0 (no linear relationship)
H₁: ρ ≠ 0 (some linear relationship)
A Pearson correlation test checks whether the observed correlation in the sample is too large to be attributed to random chance alone.
Example 4 — Quality Control (Variance)
Question: Has a manufacturing process change increased the variability of part weights above the target σ² = 0.04?
H₀: σ² = 0.04 (variance at target)
H₁: σ² > 0.04 (variance has increased)
A chi-square test for variance (or an F-test if comparing two machines) would be used here. This is a common null hypothesis example in manufacturing and quality engineering.
What Does “Null” Mean in Statistics?
Several searches for “null what does it mean”, “what does in null mean”, and similar phrases arrive at statistics articles from a programming background, where null means an empty or missing value (NULL pointers, null references, missing database fields). In statistics, the word carries a completely different meaning.
In hypothesis testing, null is short for null effect — the assumption that no meaningful difference, relationship, or change exists. The word traces to the Latin nullus (“none”) and was adopted in statistical science to denote the condition of zero effect. It has nothing to do with empty memory addresses or missing data.
When you encounter the phrase “what does null null mean” in a search, it typically reflects a repeated or emphatic search for “null hypothesis meaning” — there is no standard statistical concept called “null null.” The term you want is the null hypothesis (H₀): the baseline claim of no effect.
| Term | Statistics meaning | Programming meaning |
|---|---|---|
| null | No effect / no relationship | Empty or missing value |
| H₀ | The null hypothesis | (not used) |
| null hypothesis | The claim that there is no effect | (not used) |
| NULL value | Rarely used in statistics | A missing or undefined value |
Rejecting the Null Hypothesis
You reject H₀ when the data provide enough evidence that the null claim is implausible. The decision rule combines the p-value and the pre-specified significance level (α).
- If p-value ≤ α → reject H₀ (the data are inconsistent with the null)
- If p-value > α → fail to reject H₀ (insufficient evidence to overturn it)
A common choice is α = 0.05, meaning you accept a 5% risk of wrongly rejecting a true H₀ (a Type I error). Stricter disciplines use α = 0.01 or α = 0.001.
The p-value is the probability of observing data at least as extreme as what you collected, assuming H₀ is true. A small p-value means that the data would be very unlikely if H₀ were true — so you reject H₀.
The Four Possible Outcomes
| Decision | H₀ is actually true | H₀ is actually false |
|---|---|---|
| Reject H₀ | Type I error (false positive), probability = α | Correct decision (power = 1 − β) |
| Fail to reject H₀ | Correct decision (specificity = 1 − α) | Type II error (false negative), probability = β |
Researchers control Type I error by choosing α. Type II error (β) decreases as sample size, effect size, or α increases. The probability of correctly detecting a true effect (1 − β) is called the statistical power of the test.
According to the NIST/SEMATECH e-Handbook of Statistical Methods, “7.1.3 What are hypothesis tests?”, the outcome of a hypothesis test is always a decision to reject or fail to reject H₀ — it is never a proof that H₀ is true.
Why You Cannot Prove a Null Hypothesis True
This is one of the most misunderstood points in applied statistics. You can accumulate strong evidence against H₀ by rejecting it, but you can never collect enough evidence to prove H₀ is true.
Here is why: every sample has variability. Even if H₀ were perfectly true, a particular sample might look as if it isn’t — purely by chance. The hypothesis test accounts for this by asking “how often would data this extreme occur if H₀ were true?” It does not ask “how probable is H₀ given this data?” (that is a Bayesian question, which requires a prior).
Practical consequence: when a study finds p > 0.05, the correct conclusion is “we found no statistically significant evidence of an effect at α = 0.05” — not “there is no effect.” Absence of evidence is not evidence of absence. A study that is underpowered (too small a sample) may fail to detect a real effect; that failure is a Type II error, not proof H₀ is true.
In medical research this distinction is critical. Declaring a drug “does not work” based solely on a non-significant result — without reporting confidence intervals, effect sizes, and power — is a methodological error that has led to premature abandonment of treatments that later proved effective.
Null Hypothesis vs. Alternative Hypothesis
| Property | Null hypothesis (H₀) | Alternative hypothesis (H₁) |
|---|---|---|
| Default position | Yes — assumed true until disproven | No — must be supported by data |
| Notation | = (equality or boundary) | ≠, >, < (inequality) |
| What a rejection means | Data are inconsistent with H₀ | H₁ becomes the working conclusion |
| What non-rejection means | Insufficient evidence against H₀ | H₁ is not established |
| Controls | Type I error rate (α) | Statistical power (1 − β) |
The key principle: your alternative hypothesis H₁ should reflect a specific, directional prediction when you have one. If existing theory tells you that a new drug lowers blood pressure, your H₁ should say μ < 0 (one-tailed) — using a two-tailed H₁ (μ ≠ 0) when you had a directional prediction dilutes your statistical power.
Common Mistakes When Stating a Null Hypothesis
1. Placing an Inequality in H₀
Write H₀ with an equality, not an inequality. H₀: μ > 100 is incorrect; the null should be the boundary condition: H₀: μ = 100. The direction belongs in H₁.
2. Choosing the Direction of H₁ After Seeing the Data
If you observe that the treatment group scored higher than the control and then write H₁: μ_treatment > μ_control, you have engaged in post-hoc hypothesis construction (sometimes called HARKing — Hypothesizing After Results are Known). This inflates Type I error and invalidates the p-value calculation. The entire pair H₀/H₁ must be written before data collection.
3. Confusing “Reject H₀” With “H₁ Is Proven”
Rejecting H₀ means you have evidence against the null. It does not guarantee H₁ is the correct explanation — there could be a confounding variable, a measurement error, or a sampling bias. “Statistically significant” means unlikely under H₀, not that the effect is large, important, or causally explained by H₁.
4. Setting α After Seeing the p-Value
Choosing α = 0.05 after you see that your p-value is 0.048 is called p-hacking. The significance level must be fixed before the test. Similarly, switching from a two-tailed to a one-tailed test after seeing the data direction is a form of post-hoc manipulation.
5. Using the Wrong Test for Your Hypotheses
A t-test tests differences in means; a chi-square test of independence tests whether two categorical variables are associated; a correlation test checks for a linear relationship. Applying a test that does not match your H₀ produces a meaningless p-value. Confirm that the statistical test you choose directly addresses the parameter stated in your null hypothesis.
Frequently Asked Questions
What is a null hypothesis?
A null hypothesis (H₀) is a formal statistical claim that there is no effect, no difference, or no relationship between the variables under study. It is the default assumption in hypothesis testing: you need data strong enough to reject it before concluding that something interesting is happening.
What is an example of a null hypothesis?
A standard example of a null hypothesis: “The mean exam score of students taught by Method A equals the mean exam score of students taught by Method B (H₀: μ_A = μ_B).” This asserts that the two teaching methods produce identical average outcomes. The paired alternative would be H₁: μ_A ≠ μ_B.
What are examples of hypothesis pairs?
Examples of hypothesis pairs (H₀ and H₁) cover every research question:
- Mean comparison: H₀: μ = 50 vs H₁: μ ≠ 50
- Proportion: H₀: p = 0.20 vs H₁: p > 0.20
- Correlation: H₀: ρ = 0 vs H₁: ρ ≠ 0
- Variance: H₀: σ² = 4 vs H₁: σ² > 4
Each pair shares the same parameter, stated as an equality in H₀ and an inequality in H₁.
What does null mean in statistics?
In statistics, null means “no effect” or “no relationship.” It comes from the Latin nullus (“none”). The null hypothesis is simply the baseline claim that nothing interesting is happening — it is not related to null values or null pointers in programming.
What does “null null” mean?
There is no standard statistical concept called “null null.” Searches for “what does null null mean” typically mean “what does the null hypothesis mean.” The null hypothesis (H₀) is the claim of no effect; the word appears once, not twice.
What is null hypothesis meaning in research?
In research, the null hypothesis meaning is the skeptical default: the assumption that any observed difference or pattern in sample data could be explained by random variation alone. Researchers design experiments to collect data strong enough to overturn this default.
How do you know when to reject the null hypothesis?
Reject H₀ when the p-value from your statistical test is less than or equal to your pre-specified significance level α. At α = 0.05: if p ≤ 0.05, reject H₀ and conclude the data are inconsistent with the null. If p > 0.05, fail to reject H₀ — but do not interpret this as proof that H₀ is true.
What is the difference between H₀ and H₁?
H₀ is the null hypothesis: the default claim of no effect, stated with an equality. H₁ is the alternative hypothesis: the claim you are trying to support, stated with an inequality. The statistical test asks whether the observed data give enough evidence against H₀ to favor H₁ instead.
Summary
A null hypothesis is the default claim of no effect or no difference — the baseline every statistical test starts from. Written as H₀, it always contains an equality and pairs with an alternative hypothesis H₁ that specifies the effect you expect to find. A test produces a p-value: if p ≤ α you reject H₀ and treat H₁ as the working conclusion; if p > α you fail to reject H₀, but you never prove it true.
The core skill is a three-step process: (1) identify the parameter being tested, (2) write the equality claim in H₀, (3) write the matching inequality in H₁ based on your directional prediction — and always do this before collecting data.
For computing the test statistics and p-values you need in practice, the calculators hub includes a t-test calculator, a z-score calculator, and a p-value calculator to work through your own hypotheses.