How to report statistical results in APA style

Updated 8 September 2026 · 13 min read

You ran the analysis, the p-value came out significant, and now comes the part nobody teaches: turning numbers into academic prose. This is the stage where statistically sound work still comes back covered in comments — not because of calculation errors, but because of poor wording, off-standard formatting and omitted mandatory information.

The results section has one golden rule that solves half the problems: it presents, it does not interpret. Interpretation, comparison with the literature and causal explanations belong in the discussion. In results you report what the data showed, objectively and completely.

What every inferential result must contain

Whatever the test, five elements are mandatory under contemporary standards:

  1. The name of the test applied
  2. The test statistic with degrees of freedom (t, F, U, χ², H)
  3. The exact p-value, or "< .001" when very small
  4. An effect size (Cohen's d, η², r, Cramér's V)
  5. Descriptive statistics for the groups: mean and standard deviation, or median and interquartile range

Item 4 is the most neglected and the most frequently demanded. APA 7 requires effect sizes, and since the American Statistical Association's statement on p-values (2016), reporting significance alone is considered insufficient practice. Effect size answers the question that actually matters: is the effect large enough to have practical relevance? Adding the confidence interval — technically optional but strongly recommended, and required by many journals — raises quality further.

Numeric conventions in APA 7

Sentence templates by test

Independent-samples t-test

The intervention group (M = 74.32, SD = 8.15, n = 25) performed significantly better than the control group (M = 62.08, SD = 9.42, n = 25), t(48) = 4.91, p < .001, 95% CI [7.23, 17.25], Cohen's d = 1.39, indicating a large effect.

If Welch's correction was applied, state it: "Welch's correction was used because of unequal variances (Levene's test, F = 6.12, p = .017)", and the degrees of freedom will be fractional — t(43.7).

Paired-samples t-test

Scores increased significantly from pretest (M = 58.40, SD = 7.21) to posttest (M = 68.95, SD = 6.88), t(29) = 8.14, p < .001, mean difference = 10.55 points, 95% CI [7.90, 13.20], d = 1.49.

Mann-Whitney U

Because the normality assumption was violated according to the Shapiro-Wilk test, the Mann-Whitney U test was used. The experimental group (Mdn = 8.00, IQR = 7.00–9.00) scored significantly higher than the control group (Mdn = 6.00, IQR = 5.00–7.00), U = 142.5, z = −3.42, p < .001, r = .48, a moderate effect.

For non-parametric tests report median and interquartile range, not mean and SD — using the mean here contradicts your own justification for choosing the test.

Wilcoxon signed-rank

The Wilcoxon signed-rank test indicated a significant change between time points (n = 28), T = 34.0, z = −3.88, p < .001, r = .52.

One-way ANOVA with post hoc

A one-way analysis of variance revealed a significant difference among the three groups, F(2, 87) = 12.47, p < .001, η² = .223, a large effect. Homogeneity of variances was confirmed by Levene's test (F = 1.08, p = .344). Tukey's post hoc test showed that Group A (M = 78.2, SD = 6.1) differed significantly from Group B (M = 68.5, SD = 7.3, p = .002) and from Group C (M = 65.1, SD = 6.9, p < .001), with no difference between B and C (p = .182).

Always name the post hoc test used, and run it only if the global F is significant.

Kruskal-Wallis

The Kruskal-Wallis test indicated a significant difference among groups, H(2) = 15.82, p < .001, ε² = .178. Dunn's post hoc test with Bonferroni correction identified a difference between Groups A and C (p = .003).

Pearson correlation

There was a moderate, significant positive correlation between hours of study and academic performance, r(48) = .624, p < .001, 95% CI [.42, .77], R² = .389, indicating that 38.9% of the variance in performance is shared with hours of study.

Never write that one variable "caused" the other. Use "was associated with", "was related to", "co-varied with".

Spearman correlation

Spearman's correlation indicated a moderate, significant negative association between anxiety level and sleep quality, rs(62) = −.438, p < .001.

APA prefers rs to the Greek rho for Spearman's coefficient.

Simple linear regression

The simple linear regression was significant, F(1, 98) = 41.23, p < .001, explaining 29.6% of the variance in the outcome (R² = .296, adjusted R² = .289). The slope was B = 0.42, SE = 0.07, t = 6.42, p < .001, 95% CI [0.29, 0.55], yielding the equation Y = 42.36 + 0.42X. Each additional unit in the predictor was associated with a mean increase of 0.42 units in the outcome.

Use italic capital B for unstandardised coefficients and β only for standardised ones. Software prints unstandardised values by default, so B is usually correct — labelling them β is a frequent error.

Chi-square

The association between sex and treatment adherence was statistically significant, χ²(1, N = 105) = 8.74, p = .003, Cramér's V = .264. Among women, 72.4% (n = 42) adhered, compared with 48.9% (n = 23) of men.

APA asks for the sample size inside the parentheses alongside the degrees of freedom for chi-square. If any expected frequency was below 5, report Fisher's exact test and justify the substitution.

Non-significant result

No statistically significant difference was observed between groups, t(58) = 1.24, p = .220, d = 0.32. Observed power of .63 suggests that smaller effects may not have been detected; equivalence between the groups cannot be inferred.

Never write "there was no difference" or "the groups were equal". The correct wording is always "no statistically significant difference was observed". Reporting effect size and power for null results signals rigour.

Table formatting

Every major style guide converges on one essential point: scientific tables have no vertical rules and no full gridlines. Only three horizontal rules — above the header, below the header, and at the foot. Pasting a table from Excel with all its borders is the most immediately visible error a reviewer spots.

Figure formatting

Recommended structure for the results section

  1. Sample characteristics. Final N, attrition and exclusions with reasons, demographic data. If participants were lost, state how many and why — omitting this raises suspicion of selection bias.
  2. Assumption checks. Shapiro-Wilk and Levene results, justifying the choice between parametric and non-parametric approaches.
  3. Descriptive statistics for the main variables, usually in a table.
  4. Inferential analyses in the order of the specific objectives stated in the introduction. Reviewers check that correspondence.
  5. Additional or exploratory analyses, clearly labelled as such.

Errors that trigger revisions

How to state the methods

Normality of the data was assessed with the Shapiro-Wilk test and homogeneity of variances with Levene's test. Where assumptions were met, between-group comparisons used independent samples t tests; where they were violated, the Mann-Whitney U test was used instead. Correlations were estimated with Pearson's coefficient. Effect sizes were calculated as Cohen's d and interpreted following Cohen (1988). An alpha level of .05 was adopted for all analyses.

Generate this automatically

The statistics calculator drafts the paragraph in this format, with effect size and confidence interval, plus the table in editorial format and the figure at 300 dpi on a white background. Just review and adapt.

Open calculator

References