Null Hypothesis vs Alternative Hypothesis

Q: Can the null hypothesis be true?

Yes. In reality, H₀ may be the true state of nature. A hypothesis test cannot tell you whether H₀ is true; it only tells you whether your sample gives sufficient evidence to reject it. If you fail to reject H₀, the null may be true, or you may simply lack the power (sample size, effect size, or measurement precision) to detect a real effect.

Q: What changes between a one-tailed and two-tailed alternative hypothesis?

A two-tailed test (H₁: μ ≠ μ₀) splits the significance level α between both tails of the sampling distribution (e.g., 2.5% in each tail at α = .05). A one-tailed test (H₁: μ > μ₀ or μ < μ₀) puts all of α in one tail (e.g., all 5% in the upper tail). This means a one-tailed test has a smaller critical value and is more likely to reject H₀ if the effect goes in the predicted direction—but if it goes the opposite way, you cannot reject H₀, even if the effect is large. One-tailed tests require strong prior justification.

Q: Do I state H₀ and H₁ before or after collecting data?

Always before. You must pre-register your hypotheses (or at minimum, specify them before you analyze your data) to avoid p-hacking. Changing your hypothesis after peeking at the results inflates your false-positive rate and undermines the validity of the test. If you want to explore unexpected patterns in your data, that is fine—but call it exploratory analysis, not hypothesis testing, and replicate your findings in a new sample.

Q: What if my research question does not naturally fit H₀ and H₁?

Hypothesis testing is designed for research that makes a prediction about a population parameter. If your research is purely exploratory (e.g., "What patterns exist in this dataset?"), hypothesis testing may not be the right framework—descriptive statistics and visualization may be more appropriate. Conversely, if you want to compare groups or estimate a parameter, hypothesis testing is standard. Consult your field's methodological guidelines or a research mentor for guidance.

Q: Why is the phrase "fail to reject H₀" used instead of "accept H₀"?

Because they mean different things. "Failing to reject" means the evidence was insufficient to reject; it does not affirm that H₀ is true. An absence of evidence (finding the test not significant) is not the same as evidence of absence (proving H₀ is correct). This distinction matters: you might fail to reject H₀ because your sample is too small, your measurement is too noisy, or the true effect is too small to detect—not because H₀ is true.

Q: Can I use a one-tailed test to increase my power?

One-tailed tests do have a technical power advantage: they concentrate α in one tail, requiring a smaller critical value. However, this is only valid if your hypothesis was specified before you saw the data and strong prior evidence supports the predicted direction. Using a one-tailed test after observing the direction of your effect is p-hacking and is not accepted. If in doubt, use a two-tailed test—it is the conservative, pre-registered default in most fields.

Learn to define and state your hypotheses before you collect data, distinguish directional from non-directional tests, and avoid the mistake of thinking a test can "prove" anything.

Key Takeaway

You state two competing hypotheses before you gather data. The null hypothesis (H₀) is your default assumption—usually "no difference" or "no effect." The alternative hypothesis (H₁) is what you are investigating. If your test is significant, you reject H₀ in favour of H₁. If it is not significant, you fail to reject H₀—you never "accept" H₀, and the test never "proves" either hypothesis true.

What Is the Null Hypothesis?

The null hypothesis (H₀) is your starting assumption. It usually states that there is no effect, no difference, or no relationship between variables. In classical hypothesis testing, it is the claim you will try to gather evidence against.

Examples:

Research question: "Does coffee improve memory?" → H₀: μ_coffee = μ_{no coffee} (coffee has no effect on memory).
Research question: "Is our new drug different from the placebo?" → H₀: The new drug and placebo produce identical outcomes.
Research question: "Are men and women equally likely to pursue engineering?" → H₀: Proportion of men = Proportion of women in engineering.

H₀ is testable, mathematically precise, and falsifiable—the test collects evidence that may or may not contradict it.

What Is the Alternative Hypothesis?

The alternative hypothesis (H₁), also called the research hypothesis, is the claim you are gathering evidence for. It is what you suspect is true instead of H₀. The alternative hypothesis can be directional (one-tailed) or non-directional (two-tailed).

Non-directional (two-tailed) alternative: You predict a difference, but not which direction.

H₁: μ_coffee ≠ μ_{no coffee} ("Coffee changes memory—either up or down.")
H₁: σ₁ ≠ σ₂ ("The variances are different.")

Directional (one-tailed) alternative: You predict the direction of the effect.

H₁: μ_coffee > μ_{no coffee} ("Coffee improves memory.")
H₁: r > 0 ("Study hours and exam scores are positively correlated.")

Directional hypotheses are less common and only justified when prior theory or evidence strongly predicts the direction. Changing your hypothesis after peeking at the data is a form of p-hacking and inflates your false-positive rate.

How to State Hypotheses for Two-Tailed and One-Tailed Tests

The format for H₀ and H₁ differs slightly depending on whether your test is directional or not.

Two-Tailed Test (Non-Directional)

Use this when: You have no strong prior reason to predict the direction, or you want to detect a difference in either direction.

H₀: μ = μ₀ H₁: μ ≠ μ₀

Plain English: "There is no difference in the mean" vs. "There is a difference in the mean." The rejection region is split between both tails of the sampling distribution.

One-Tailed Test (Directional)

Use this when: Prior evidence or theory predicts the direction, and you only care about detecting a difference in that one direction.

H₀: μ ≤ μ₀ H₁: μ > μ₀

H₀: μ ≥ μ₀ H₁: μ < μ₀

Plain English (upper tail): "The mean is less than or equal to the null value" vs. "The mean is greater than the null value." The entire rejection region is in the upper (right) tail.

Plain English (lower tail): "The mean is greater than or equal to the null value" vs. "The mean is less than the null value." The entire rejection region is in the lower (left) tail.

Critical difference: A one-tailed test at α = .05 puts all 5% of the rejection region in one tail. A two-tailed test at α = .05 splits it: 2.5% in each tail. This is why one-tailed tests have less power (they use a larger critical value) but are only valid if your prediction matches the observed direction.

Visualizing H₀ and H₁ on the Sampling Distribution

Below is a normal sampling distribution with H₀ centred at the null value and the rejection region (red) marking where we reject H₀ in favour of H₁.

Fig. 1 — Sampling distribution under H₀ (null hypothesis), centred at the null value. For a two-tailed test with α = 0.05, the rejection region occupies both tails (shaded red) at z ≈ ±1.96. Observed test statistics in these red regions lead to rejection of H₀ in favour of H₁.

Cross-Discipline Examples: New H₀ and H₁ Pairs

Below are H₀ and H₁ formulations from different fields. Notice how H₀ always expresses "no effect" or "no difference," and H₁ expresses what you are investigating.

Discipline	Research Question	H₀ (Null)	H₁ (Alternative, Two-Tailed)
Psychology	Does cognitive therapy reduce anxiety scores?	μ_before = μ_after	μ_before ≠ μ_after
Biology	Does temperature affect enzyme reaction rate?	r = 0 (no correlation)	r ≠ 0 (correlation exists)
Business	Is customer satisfaction different between two store locations?	p₁ = p₂ (proportions equal)	p₁ ≠ p₂ (proportions differ)
Education	Do students taught with multimedia show different learning gains than those taught traditionally?	μ_multimedia = μ_traditional	μ_multimedia ≠ μ_traditional

Common Student Mistakes

Three errors appear repeatedly when students write hypotheses or interpret test results.

Mistake 1: Confusing "fail to reject H₀" with "accept H₀"

When a test is not significant, you have insufficient evidence to reject H₀. This does not mean H₀ is true; it means the data do not give you enough reason to abandon it. You may have low statistical power, a small effect that is real but undetected, or sampling error. Always say "fail to reject H₀," never "accept H₀."

Mistake 2: Stating H₁ as your preferred outcome, without justification

If you write H₁: μ > μ₀ (a one-tailed test), you must have stated this hypothesis before you looked at your data. If you run the test, observe the mean is lower, and then switch to H₁: μ < μ₀, you are engaging in p-hacking. Changing your hypothesis after seeing the data inflates your false-positive rate and is considered scientific misconduct in most fields.

Mistake 3: Thinking a test can "prove" H₁

Rejecting H₀ provides evidence against H₀ in favour of H₁, but it does not "prove" H₁. A low p-value means "the probability of seeing data this extreme, if H₀ were true, is small." It does not prove your hypothesis is correct. Your result could be due to a real effect, or you could be in the unlucky 5% when α = .05. This is why replication is vital: one significant test is suggestive, but repeated independent confirmation is much more convincing.

Related Resources

Build your understanding by reading these companion articles:

How to Write a Research Question — Develop your research question before you form your hypotheses. This guide covers the "FINER" criteria and shows how a clear question guides your hypotheses.
Which Statistical Test Should I Use? — Once you have your hypotheses, this decision tree helps you pick the right test for your data.
t-Test Calculator — Run a one-sample or independent-samples t-test online and see your test statistic, degrees of freedom, and p-value calculated step by step.

Frequently Asked Questions

Can the null hypothesis be true?

Yes. In reality, H₀ may be the true state of nature. A hypothesis test cannot tell you whether H₀ is true; it only tells you whether your sample gives sufficient evidence to reject it. If you fail to reject H₀, the null may be true, or you may simply lack the power (sample size, effect size, or measurement precision) to detect a real effect.

What changes between a one-tailed and two-tailed alternative hypothesis?

A two-tailed test (H₁: μ ≠ μ₀) splits the significance level α between both tails of the sampling distribution (e.g., 2.5% in each tail at α = .05). A one-tailed test (H₁: μ > μ₀ or μ < μ₀) puts all of α in one tail (e.g., all 5% in the upper tail). This means a one-tailed test has a smaller critical value and is more likely to reject H₀ if the effect goes in the predicted direction—but if it goes the opposite way, you cannot reject H₀, even if the effect is large. One-tailed tests require strong prior justification.

Do I state H₀ and H₁ before or after collecting data?

Always before. You must pre-register your hypotheses (or at minimum, specify them before you analyze your data) to avoid p-hacking. Changing your hypothesis after peeking at the results inflates your false-positive rate and undermines the validity of the test. If you want to explore unexpected patterns in your data, that is fine—but call it exploratory analysis, not hypothesis testing, and replicate your findings in a new sample.

What if my research question does not naturally fit H₀ and H₁?

Hypothesis testing is designed for research that makes a prediction about a population parameter. If your research is purely exploratory (e.g., "What patterns exist in this dataset?"), hypothesis testing may not be the right framework—descriptive statistics and visualization may be more appropriate. Conversely, if you want to compare groups or estimate a parameter, hypothesis testing is standard. Consult your field's methodological guidelines or a research mentor for guidance.

Why is the phrase "fail to reject H₀" used instead of "accept H₀"?

Because they mean different things. "Failing to reject" means the evidence was insufficient to reject; it does not affirm that H₀ is true. An absence of evidence (finding the test not significant) is not the same as evidence of absence (proving H₀ is correct). This distinction matters: you might fail to reject H₀ because your sample is too small, your measurement is too noisy, or the true effect is too small to detect—not because H₀ is true.

Can I use a one-tailed test to increase my power?

One-tailed tests do have a technical power advantage: they concentrate α in one tail, requiring a smaller critical value. However, this is only valid if your hypothesis was specified before you saw the data and strong prior evidence supports the predicted direction. Using a one-tailed test after observing the direction of your effect is p-hacking and is not accepted. If in doubt, use a two-tailed test—it is the conservative, pre-registered default in most fields.

Sources

OpenStax. (2024). Introductory Statistics 2e. — Standard textbook on hypothesis testing and statistical inference used in universities across the United States.
Duke University Medical Center Library. (2024). Research guides and statistics resources. — Institutional guidance on hypothesis development and research design.
Bland, J. M., & Altman, D. G. (1994). "One and two sided tests of significance." BMJ, 309(6958), 248. — Classic reference clarifying one-tailed vs. two-tailed tests and when each is justified.
Goodman, S. N. (2008). "A Comment on Replication, P-values and Evidence." Statistics in Medicine, 11(7), 875–879. — Explains why p-values do not prove hypotheses and the importance of replication.