Read the result, not just the threshold
How to interpret a p-value: effect size, confidence intervals and five checks
A p-value cannot tell you how large a finding is or how likely your hypothesis is to be true. Read the estimate, its uncertainty and the study conditions together.
AcademicFix Editorial Team · Sources checked · Türkçe
Three ways to read a finding
One number. Three different questions.
Before asking “Is it significant?”, choose the information you actually need.
p=0.03≠97% certainty
You cannot reverse the conditional question. p=0.03 does not mean a 97% chance that the hypothesis is true. It does not measure effect size either. [1]
Three parts of the same teaching example Hypothetical data · follow a card to its calculation
How extreme is this result?
Under the null and test assumptions, the probability of a statistic at least this extreme.
Inspect the test 02 / Estimated magnitude5 pointsHow large is the difference?
The sample mean minus the reference value. Read it in the original measurement unit.
Inspect the difference 03 / Uncertainty[1.08, 8.92]How precise is the estimate?
The 95% confidence interval for the difference adds uncertainty to the point estimate.
Inspect the intervalFor researchers reading and reporting a statistical test. This guide covers classical tests and confidence intervals. It cannot choose a test for your study without examining its design and data.
1. What does a p-value describe?
Assuming the null hypothesis and the test model conditions hold, it is the probability of a test statistic at least as extreme as the observed one. What counts as “extreme” depends on whether the test is one-sided or two-sided. [2]
This asks about results under a model. It does not answer the reverse question, “What is the probability that the null is true?” [1]
Choose the significance level, alpha, before testing. With a prespecified alpha of 0.05, p=0.03 falls below the threshold. That decision alone does not establish a large or practically important effect. [2] [1]
2. Five checks before drawing a conclusion
- Identify the null hypothesis, test, direction and analysis plan.
- Read the estimated effect in its measurement unit. Was a practically important difference justified in advance?
- Inspect the interval: which effect values does it span, and how precise is the estimate?
- Check sampling, independence and the assumptions that make the model appropriate.
- Look for the planned analyses, additional attempts and decisions about missing data.
This reading sequence is our editorial application of the ASA principles on context and transparent reporting. It is not a universal test-selection algorithm. [1]
3. One result, three useful questions
Let the sample mean be 55 and the fixed reference be 50. For the two-sided test H₀: μ=50, the standard error is 10/√25=2 and z=(55−50)/2=2.50.
| Question | Calculation | Interpretation |
|---|---|---|
| Compatibility with the reference | Two-sided p≈0.0124 | Under this model, the probability of |z|≥2.50. |
| Estimated magnitude | 55−50=5 points | The estimated raw difference, separate from the p-value. |
| Precision | Mean 95% CI: [51.08, 58.92] Difference 95% CI: [1.08, 8.92] | 55±1.95996×2, using the known population standard deviation. [3] |
Would a five-point estimate settle the practical question? Suppose a minimum important difference of three points had been justified before analysis. The interval for the difference spans values below and above three. The small p-value therefore does not establish that the difference is at least three points. That conclusion follows from this hypothetical calculation; three points is not a universal cutoff.
If the population standard deviation is unknown and estimated from the sample, do not copy this z calculation. For an appropriate one-sample normal-mean analysis, the t method uses n−1 degrees of freedom. Other study designs require their own methods. [4]
4. Four mistakes worth catching
| Shortcut | Better interpretation |
|---|---|
| p=0.03 → 97% chance the hypothesis is true | The p-value is conditional on the null and test model; it is not a probability assigned to the hypothesis. |
| Small p → large effect | Read the difference or association coefficient separately, in a meaningful unit. |
| p>0.05 → no effect | Crossing that threshold does not prove absence. Examine the estimate and the interval. |
| 95% CI → 95% of observations lie inside it | This interval estimates a population parameter; it does not describe the spread of individual observations. |
The first two distinctions come from the ASA p-value principles; the third from the limits of threshold labels; the fourth from the NIST explanation of confidence intervals. [1] [5] [4]
The 95% describes the coverage of the interval-building method over repeated samples under the stated conditions. For one classical interval, it is not a posterior probability that the fixed population mean lies inside it. [4]
5. A reporting template you can adapt
Keep the method, direction, measurement unit and uncertainty together. This example is a teaching report, not a mandatory journal or APA format.
In the hypothetical example, a two-sided one-sample z-test was used assuming a known population standard deviation of 10 points. The sample mean was 55 points (n=25). Relative to the 50-point reference, the estimated difference was 5 points, with a 95% confidence interval of [1.08, 8.92] and p≈0.0124. The interval spanned values below and above the example's prespecified three-point minimum important difference.
- Disclose deviations from the analysis plan and additional analyses.
- Report the planned findings, rather than selecting only the smallest p-value. [1]
- Name the sample size, effect unit and interval method.
- Follow the current reporting instructions of your journal or institution.
Two questions readers often ask
Are p=0.049 and p=0.051 fundamentally different findings?
They may fall on opposite sides of a decision threshold. The scientific interpretation still needs the estimates, intervals, design and context. [5]
Can a small p-value repair a weak study design?
The design and conduct remain part of the interpretation. A p-value alone does not certify the quality or truth of a study. [1]
Sources and verification
The relevant sections were reviewed on 14 October 2026 and matched to the claims. The numerical example was separately recalculated using the standard normal distribution. These citations do not imply a partnership or endorsement.
- 1ASA statement on statistical significance and p-values (2016)
Page 2: hypothesis probability, effect magnitude, context and transparent reporting.
- 2NIST: Critical values and p values
The conditional definition and a significance level chosen before testing.
- 3NIST: What are confidence intervals?
Repeated sampling and the normal-mean interval with a known population standard deviation.
- 4NIST: Confidence Limits for the Mean
Precision, coverage and the t interval when the standard deviation is estimated.
- 5ASA: Limits of statistical significance labels
Why a threshold label alone cannot establish whether an effect exists or matters.
Put the reading into practice
Check how the analysis answers your research question, then make the result table clear.
- Choose an analysis for your research question
- Plan sample size before collecting data
- Format the result table — the tool does not calculate statistics.