How to Report p Values, Confidence Intervals, and Effect Sizes in APA 7
A statistical result is easiest to evaluate when it communicates three different ideas: whether the data are inconsistent with the null model, how large the estimated effect is, and how uncertain that estimate remains. A p value alone cannot answer all three questions.
What should a complete statistical result include?
For most common analyses, report:
- The name of the analysis and the variables or groups involved.
- The relevant descriptive statistics.
- The test statistic and degrees of freedom when applicable.
- The exact p value, except for very small values.
- An effect-size estimate with a clear definition.
- A confidence interval for the main estimate or effect size when available.
- A plain-language interpretation that respects the study design.
The exact set depends on the method. A logistic regression result needs an odds ratio, while a mean comparison may be clearer with a mean difference and Cohen's d.
How should p values be formatted?
Use a lowercase italic p and report exact values to a sensible precision, commonly three decimal places. Write p = .032 rather than p < .05. For a value smaller than .001, write p < .001 rather than p = .000.
Do not describe p = .051 as “almost significant.” State the estimate, confidence interval, and uncertainty without turning the threshold into a graded measure of truth.
How should confidence intervals be reported?
Name the confidence level and show the lower and upper limits in brackets. For example:
The estimated mean difference was 4.20 points, 95% CI [1.35, 7.05].
A confidence interval helps readers see which effect sizes remain compatible with the data under the model. It should be interpreted together with measurement quality, design, assumptions, and subject-matter relevance.
How should effect sizes be selected?
Choose an effect size that matches the research question and analysis:
| Analysis | Useful effect sizes |
|---|---|
| Two-group mean comparison | Mean difference, Cohen's d, Hedges' g |
| ANOVA | Eta squared, partial eta squared, omega squared |
| Correlation | Pearson r or Spearman rho |
| Logistic regression | Odds ratio |
| Cohort comparison | Risk ratio or risk difference |
| Survival analysis | Hazard ratio |
Define the direction and reference group. A positive standardized difference is not interpretable if the reader does not know which group was subtracted from which.
Reporting examples
Independent-samples t test
Group A had a higher mean score (M = 24.60, SD = 5.10) than Group B (M = 20.90, SD = 4.80), t(54) = 2.76, p = .008, mean difference = 3.70, 95% CI [1.02, 6.38], Cohen's d = 0.74.
One-way ANOVA
Mean scores differed across the three groups, F(2, 57) = 6.31, p = .003, η² = .18. Tukey-adjusted comparisons indicated that Group 3 scored higher than Group 1, mean difference = 5.40, 95% CI [1.40, 9.40].
Linear regression
Study time was positively associated with examination score, B = 2.10, SE = 0.58, 95% CI [0.94, 3.26], β = .39, t = 3.62, p = .001.
These values are illustrative. Your report should use the actual output and describe any robust, adjusted, or alternative estimator used.
Common reporting mistakes
- Reporting only “significant” or “not significant.”
- Writing p = .000.
- Omitting units, reference groups, or effect direction.
- Mixing standardized and unstandardized coefficients.
- Claiming clinical or practical importance from the p value.
- Copying a complete software table without editing it for the research question.
Use DataStatPro's publication-ready tools and APA table generator to organize verified results. The effect-size interpretation guide can help you select and explain the magnitude measure.
Frequently asked questions
Should every result include an effect size?
Report an effect size when it meaningfully represents the research question. The most useful measure depends on the design, model, and audience.
Should I report p values in tables and text?
Avoid unnecessary duplication. Report the key result in the narrative and use a table when several related estimates need to be compared.
Is a 95% confidence interval always required?
Not universally, but an interval is usually valuable because it communicates uncertainty. Follow the requirements of the journal, discipline, and analysis.
Can a result be important when p is greater than .05?
Yes. The estimate and interval may still inform feasibility, uncertainty, or future research. Avoid treating a threshold as the only measure of importance.