What Is a P-Value?
The p-value (probability value) is a number between 0 and 1 that tells you how surprising your data is, given a specific assumption about the world. That assumption is called the null hypothesis.
Here's the plain-English version: the p-value answers this question — "If there were really no effect, no difference, or no relationship in the population, how likely would I be to see data like mine — or even more extreme data — just by chance?"
Think of it like a coin-flip test. You flip a coin 20 times and get 16 heads. You want to know whether the coin is biased. The null hypothesis says it's fair (50% heads). The p-value tells you the probability of getting 16 or more heads just by chance if the coin really were fair. If that probability is very low — say, 0.006 — you have reason to suspect the coin isn't fair.
Drug trials work the same way. Researchers give a treatment to one group and a placebo to another, then compare outcomes. The p-value measures how likely the observed difference in outcomes would be if the treatment actually had no effect. A small p-value means the data would be very unlikely under that "no effect" scenario, which is evidence against it.
The formal definition of the p-value comes from frequentist statistics — the framework developed by Ronald Fisher in the 1920s and later refined by Neyman and Pearson. For more on the machinery behind it, see the p-values overview.
A researcher compares exam scores for two teaching methods. She gets p = 0.04. This means: if both methods produced identical results in the population, there would be only a 4% chance of seeing a difference this large (or larger) in her sample. That's unlikely enough that she rejects the "no difference" assumption.
What Does p < 0.05 Mean?
A p-value less than 0.05 means there is less than a 5% probability of observing your results if the null hypothesis were true. This is the standard threshold for statistical significance — when p < 0.05, researchers typically reject the null hypothesis and conclude the result is statistically significant.
The 0.05 threshold is called the significance level, written as α (alpha). When you set α = 0.05 before running your test, you're saying: "I'll accept a 5% chance of wrongly rejecting the null hypothesis." If your p-value comes back below that threshold, you reject H₀ and call your result statistically significant.
So if p = 0.03, there's only a 3% chance of seeing data this extreme by random chance alone, assuming no real effect exists. That's low enough to reject the null hypothesis at α = 0.05. If p = 0.08, the chance is 8% — above the threshold, so you don't reject H₀.
Where does 0.05 come from? Ronald Fisher proposed it in the 1920s as a convenient rule of thumb. He wrote that a 1-in-20 chance was a reasonable threshold for deciding whether an effect was real. It wasn't derived mathematically — it was a practical suggestion that stuck. This matters because 0.05 is a convention, not a law of nature. Medicine and physics often use stricter thresholds like 0.01 or even 0.0000003. Exploratory research sometimes uses 0.10.
| P-Value | Interpretation | Typical Decision |
|---|---|---|
| p < 0.001 | Very strong evidence against H₀ | Reject H₀ |
| p < 0.01 | Strong evidence against H₀ | Reject H₀ |
| p < 0.05 | Moderate evidence — common threshold | Reject H₀ (at α = 0.05) |
| p ≥ 0.05 | Insufficient evidence against H₀ | Fail to reject H₀ |
The significance level is always chosen before the test runs, not after you see the p-value. Choosing α after the fact — sometimes called p-hacking — is a serious problem in research because it makes a false positive look like a genuine finding.
How to Interpret Your P-Value
The p-value always needs context: what was your significance level, and what does the effect actually mean in the real world? Here are three scenarios with concrete numbers to show how interpretation works.
Only a 0.2% chance of data this extreme under H₀. At any standard α (0.10, 0.05, or 0.01), you reject the null hypothesis. This is strong evidence that a real effect exists in the population.
Barely under the α = 0.05 line. Technically, you reject H₀ — but results this close to the boundary warrant caution and ideally replication. The effect could be real but modest.
A 12% chance of data like this under H₀ is not unusual enough to warrant rejection. You fail to reject the null hypothesis. This does not prove H₀ is true — it means your data don't provide sufficient evidence against it.
One phrase matters more than almost anything else here: "fail to reject" is not the same as "accept the null hypothesis." Failing to reject H₀ just means you didn't find enough evidence — your sample size may have been too small, your measurement too noisy, or the effect genuinely absent. You can't tell which from the p-value alone.
"We failed to reject H₀" and "We proved the null hypothesis is true" are completely different claims. The first is a valid statistical conclusion. The second is not — you can never prove H₀ true with a single p-value test.
For step-by-step worked examples of p-value calculations in real tests, see the p-value examples page. And if you want to go deeper on how H₀ and H₁ are constructed, the null and alternative hypothesis guide covers that in full.
What a P-Value Does NOT Mean
The p-value is one of the most misread numbers in statistics. These three myths show up constantly — in textbooks, in research papers, and in news reporting.
Rejecting the null hypothesis is not the same as proving your hypothesis true. A p-value below 0.05 only tells you that your data would be unusual if H₀ were correct. There could be other explanations: a confounded design, a flawed measurement tool, or simple chance. Remember, at α = 0.05 you accept a 5% false-positive rate. Statistical significance is evidence, not proof.
A p-value of 0.06 misses the 0.05 threshold by a hair. That doesn't make the finding worthless. In exploratory research, a result near significance is a signal worth following up. The 0.05 line is arbitrary — there's nothing special about it. What matters is the full picture: effect size, sample size, study design, and whether the result replicates. Treating 0.049 as a discovery and 0.051 as a failure is a category error.
The p-value depends on both the effect size and the sample size. In a large enough sample, even a trivially small effect — say, a drug that lowers blood pressure by 0.1 mmHg — can produce p = 0.0001. That result is highly statistically significant but clinically meaningless. Always report effect size (Cohen's d, odds ratio, R²) alongside p-values. The p-value tells you whether an effect exists; effect size tells you whether it matters.
P-Value vs Statistical Significance
The p-value and statistical significance are tightly linked but not identical. Statistical significance is a binary decision: your result either crosses the α threshold or it doesn't. The p-value is the continuous number that drives that decision.
When researchers say a result is statistically significant, they mean p fell below their pre-set α. That's the only thing the phrase means — nothing more.
The bigger distinction is between statistical significance and practical significance. A result can be statistically significant but practically unimportant — a new fertilizer increases crop yield by 0.2 kg per hectare, p = 0.001. The p-value is tiny because a massive study was run. But no farmer would change their practice for 0.2 kg per hectare. Equally, a small pilot study might find a drug reduces symptoms by 30% but reach only p = 0.08. The sample was too small to reach significance, but the effect size is worth studying further.
This distinction becomes especially misleading with large datasets. When you have thousands of observations, even minuscule differences become statistically significant. Always pair your p-value with an effect size measure. For most means comparisons, that's Cohen's d. For regression models, it's R². For proportions, it's odds ratios or relative risk.
The p-value tells you whether an effect likely exists. Effect size tells you how large it is. You need both to draw a meaningful conclusion.
Type I and Type II errors are also closely tied to this relationship. A Type I error is a false positive — you reject H₀ when it's actually true. The probability of a Type I error is exactly α. A Type II error is a false negative — you fail to reject H₀ when an effect really exists. The probability of a Type II error is β, and 1 − β is the power of the test.
P-Value in Different Statistical Tests
The same p-value logic applies across every common statistical test — the calculation differs, but the interpretation is identical: compare p to α and decide whether to reject H₀.
| Test | What H₀ says | Example p-value interpretation |
|---|---|---|
| T-test | The group means are equal | If p = 0.03, the means differ significantly — the groups are probably not drawn from populations with the same mean. See t-test examples. |
| Chi-square test | The two variables are independent | If p = 0.01, the observed frequencies differ enough from what independence predicts to reject that assumption. The variables are associated. See the chi-square test guide. |
| ANOVA | All group means are equal | If p < 0.05, at least one group mean differs from the others. ANOVA doesn't tell you which pair — follow-up tests like Tukey's HSD do. See the ANOVA guide. |
| Regression (F-test) | All coefficients equal zero | If p < 0.05 for the overall model, at least one predictor explains variance in the outcome. Individual coefficient p-values test each predictor separately. |
| Correlation test | Population correlation = 0 | If p = 0.02 for a Pearson correlation, the linear relationship between variables is unlikely to be zero in the population. |
Every output table from R, Python (scipy, statsmodels), SPSS, or Excel's Analysis Toolpak includes a p-value column. In regression output, look for "Pr(>|t|)" or "P-value." In ANOVA tables, it's usually the rightmost column. Once you understand what the p-value means, reading any of these becomes straightforward.
Quick Reference: P-Value Checklist
When you get a p-value from any statistical test, run through this checklist before reporting your result.
- Identify your significance level — usually α = 0.05, but check your field's convention
- Set your threshold in advance — compare your p-value to α before looking at it
- If p < α, reject the null hypothesis and report a statistically significant result
- If p ≥ α, fail to reject the null hypothesis — do not say you "proved" H₀
- Always report effect size alongside your p-value — significance alone isn't enough
- Never say "the null hypothesis is proven true" — failing to reject H₀ is different from accepting it
- Check your sample size — very large samples can make tiny, unimportant effects significant
- Report the exact p-value, not just "p < 0.05" — modern journals prefer precision
Conclusion
The p-value is a single number with a specific, limited meaning: the probability of your data under the null hypothesis. When it's below your significance level, you have grounds to reject the null — that's all. It doesn't prove anything, it doesn't measure the size of an effect, and it doesn't tell you whether your finding matters in the real world.
The most useful habit you can build is reading p-values alongside effect sizes and confidence intervals. A p < 0.05 result with a tiny effect and a borderline confidence interval tells a very different story than p = 0.001 with a large, well-replicated effect. Context is everything. For more on how hypothesis testing decisions fit together, see the hypothesis testing main page, which covers the full framework from null hypothesis to decision rule.
Ready to Calculate Your Own P-Value?
Enter your data directly and get an instant p-value with a plain-English interpretation.
Use the Free P-Value Calculator →