Quick answer: You reject the chosen hypothesis when your statistical test shows the p-value is less than your significance level (typically 0.05), indicating strong evidence against it; you fail to reject (not “accept”) the hypothesis when the p-value is greater than the significance level, meaning insufficient evidence exists to disprove it.
Researchers and students constantly stumble over this critical decision point in statistical analysis. The confusion stems from misunderstanding what hypothesis testing actually proves and the precise language statisticians use. Getting this wrong can invalidate your entire study or lead to embarrassing corrections in published work.
| Term | Meaning / When to use | Example sentence |
|---|---|---|
| Reject the hypothesis | Use when p-value < significance level (α); evidence strongly contradicts the null hypothesis | “Since p = 0.023 < 0.05, we reject the null hypothesis that the drug has no effect.” |
| Fail to reject the hypothesis | Use when p-value ≥ significance level; evidence is insufficient to disprove the null hypothesis | “With p = 0.12 > 0.05, we fail to reject the null hypothesis.” |
| Accept the hypothesis | Technically incorrect in frequentist statistics; avoid this phrase entirely | ❌ “We accept the null hypothesis” (incorrect) |
When to reject the chosen hypothesis
You reject the null hypothesis when your data provides strong statistical evidence that the null hypothesis is unlikely to be true. This happens when your test statistic falls into the critical region, or equivalently, when your p-value drops below your predetermined significance level (alpha).
the decision rule is straightforward: if the probability of observing your data (or more extreme data) assuming the null hypothesis is true falls below your threshold—usually 0.05—you reject the null hypothesis.
Real-world examples:
Research paper: “The t-test yielded t(48) = 2.87, p = 0.006. We reject the null hypothesis that there is no difference in test scores between the control and experimental groups.”
Email to supervisor: “The A/B test results came back with p = 0.031. We should reject the hypothesis that the new checkout page performs the same as the old one—the new version is statistically better.”
Thesis defense: “Our chi-square analysis produced χ² = 12.45, p = 0.002. We reject the null hypothesis of independence between smoking status and lung disease incidence.”
Key indicators that you should reject:
- P-value is less than 0.05 (or your chosen alpha level)
- Confidence interval does not include the null value
- Test statistic exceeds the critical value
- The observed effect is statistically significant
When to fail to reject the chosen hypothesis
You fail to reject the null hypothesis when your statistical test does not provide sufficient evidence to conclude that the null hypothesis is false. This is NOT the same as accepting or proving the null hypothesis is true—it simply means your data didn’t show a statistically significant effect.
This distinction matters enormously. As explained in hypothesis testing resources, failing to reject doesn’t confirm the null; it just means you lack evidence to disprove it. Your study might have low power, small sample size, or high variability masking a real effect. For more, see Hypothesis Testing.
Real-world examples:
Lab report: “The ANOVA returned F(2, 87) = 1.34, p = 0.267. We fail to reject the null hypothesis that all three fertilizer treatments produce equal crop yields.”
Business presentation: “With p = 0.089, we fail to reject the hypothesis that customer satisfaction is the same across all three regions. We need more data before making regional policy changes.”
Medical study abstract: “The difference in recovery rates was not statistically significant (p = 0.14). We fail to reject the null hypothesis that the new treatment performs no better than the standard protocol.”
Key indicators that you should fail to reject:
- P-value is greater than or equal to 0.05 (or your alpha)
- Confidence interval includes the null value
- Test statistic falls within the acceptance region
- The observed effect is not statistically significant
Critical examples of proper language:
❌ Wrong: “We accept the null hypothesis that the means are equal.”
✅ Correct: “We fail to reject the null hypothesis; there is insufficient evidence to conclude the means differ.”
❌ Wrong: “The hypothesis was proven true.”
✅ Correct: “The data did not provide sufficient evidence to reject the hypothesis.”
How to remember the difference
Use this memory trick that I’ve taught to hundreds of students: “Absence of evidence is not evidence of absence.”
When you fail to reject, you’re saying “I didn’t find enough evidence to disprove it”—not “I proved it’s true.” Think of it like a courtroom: “Not guilty” doesn’t mean “innocent”; it means the prosecution didn’t prove guilt beyond reasonable doubt.
The Courtroom Analogy:
- Null hypothesis = Defendant is innocent
- Reject null = Guilty verdict (enough evidence to convict)
- Fail to reject null = Not guilty verdict (not enough evidence to convict, but not necessarily innocent)
The Alpha Threshold Rule:
Remember: α = 0.05 is your line in the sand
- P < 0.05 → Reject (significant evidence against null)
- P ≥ 0.05 → Fail to reject (insufficient evidence against null)
Editor’s insight: I always tell my graduate students to physically write “FAIL TO REJECT” in their drafts whenever p > 0.05. After doing this 20-30 times, the correct phrasing becomes automatic. The moment you write “accept,” your statistical training should trigger an alarm bell.
Common mistakes and exceptions
Mistake #1: Saying “accept the null hypothesis”
This is the most pervasive error I see in undergraduate theses and even some published papers. You never “accept” the null in frequentist statistics. You can only reject it or fail to reject it. Accepting implies proof of truth, which hypothesis testing doesn’t provide.
Mistake #2: Confusing statistical significance with practical importance
Just because you reject the null doesn’t mean the effect matters. With a huge sample size, you can detect trivial differences that are statistically significant but practically meaningless.
Example: “We rejected the null hypothesis (p < 0.001) that the weight loss programs have equal effects. Program A produced 0.3 pounds more weight loss than Program B over 12 weeks.” Statistically significant? Yes. Clinically relevant? Debatable.
Mistake #3: Misinterpreting “fail to reject” as “no effect”
A non-significant result doesn’t prove there’s no effect—it might mean your study was underpowered. I’ve reviewed manuscripts where authors claimed “there is no relationship” based on p = 0.07 with n = 15. That’s not evidence of absence; that’s an underpowered study. For more, see Hypothesis.
Exception: Bayesian hypothesis testing
In Bayesian statistics, you CAN talk about accepting or supporting a hypothesis because you’re calculating the probability that the hypothesis is true given the data, not the probability of the data given the hypothesis. However, this article focuses on traditional frequentist methods, which dominate most scientific fields.
Exception: Equivalence testing
When you specifically want to demonstrate that two treatments are equivalent (not different), you use equivalence testing with two one-sided tests (TOST). Here, you can reject the hypothesis that the treatments differ by more than a specified margin. This flips the traditional framework on its head.
US vs. UK terminology:
The statistical concepts are identical, but you’ll occasionally see “hypothesis” spelled the same way in both variants—no difference here. However, British researchers sometimes use “significance test” where Americans say “hypothesis test,” though both terms are understood globally.
Common error in reporting:
I constantly see this in cover letters and research statements:
❌ “My research accepted the hypothesis that social media use correlates with anxiety.”
✅ “My research failed to reject the hypothesis” or better yet, “My research found statistically significant evidence supporting the alternative hypothesis that social media use correlates with anxiety (p = 0.012).”
Frequently Asked Questions
Why don’t we ever “accept” the null hypothesis?
Because hypothesis testing is designed to look for evidence against the null, not for it. Failing to find evidence against something isn’t the same as proving it’s true. You might have missed a real effect due to small sample size, measurement error, or low statistical power.
What if my p-value is exactly 0.05?
By convention, p = 0.05 is the boundary where you reject the null (p ≤ 0.05 means reject). However, this is an arbitrary cutoff. A p-value of 0.049 and 0.051 represent essentially the same evidence, so report the exact p-value and interpret it thoughtfully rather than treating 0.05 as a magical threshold.
Can I say “the hypothesis was supported” instead of “failed to reject”?
Yes, when referring to your alternative hypothesis (the research hypothesis), you can say “the data support the alternative hypothesis” if you rejected the null. But never say you “supported” or “accepted” the null hypothesis.
What’s the difference between Type I and Type II errors in this context?
A Type I error occurs when you reject a true null hypothesis (false positive). A Type II error happens when you fail to reject a false null hypothesis (false negative). Your significance level (α) controls Type I error risk, while statistical power (1-β) relates to Type II error risk.
Susan Robinson is a seasoned language expert with over 15 years of experience specializing in English word comparisons. With a Ph.D. in English Literature from Oxford University, Susan possesses an in-depth understanding of language nuances and usage. Her academic journey revealed the profound impact of word choice on effective communication, fueling her passion for this niche field. Prior to joining WhichWordHub, Susan taught advanced ESL courses, where she developed innovative methods to clarify word distinctions for non-native speakers. At WhichWordHub, she is dedicated to crafting comprehensive guides that deconstruct complex language concepts into easily digestible information. Her articles often delve into common language pitfalls and provide practical advice on vocabulary enhancement. Susan’s goal is to foster a deeper appreciation for word precision and aid readers in mastering the art of eloquent expression.