A polished Results section can still misrepresent your research if one number or claim changes during drafting. AI can help with reporting statistical results, but your verified analysis must control every value and conclusion.
The safest approach is to give AI approved statistical output, request a constrained draft, and compare each sentence with the original. Start by assembling the evidence the tool needs, then separate numerical reporting from interpretation.
Key Takeaways
Report the test statistic, p-value, and effect size, supported by sample sizes and confidence intervals.
Use AI to explain and organize verified results. Don't accept generated numbers without comparing them with your analysis.
Keep confidential data out of AI tools, preserve source records, and match your conclusions to the study design.
Prepare Your Statistical Output Before Using AI
Build a verified evidence packet
Run your analysis in statistical software such as R, SPSS, or jamovi before asking AI to write. Supply the test name, dependent and independent variables, sample sizes, descriptive statistics, estimates, and relevant output.
Describe the model and assumption checks in Methods before presenting findings. Depending on the analysis, these include independence, residual normality, and homoscedasticity, meaning similar residual variance across fitted values or groups. Record violations and any robust methods used.
Our PDF research assistant lets you question uploaded documents and receive cited answers. When extracting published findings, inspect the original Results section and table notes, not just the abstract. AI can mistake a cited background study for the current study.
Protect data and constrain the draft
Never share confidential, identifiable, sponsor-restricted, or embargoed data with AI tools without explicit authorization. Prefer approved aggregate output; removing names alone doesn't guarantee anonymity.
Use a bounded prompt:
Use only the supplied output. Draft an APA-style Results paragraph identifying the analysis, variables, sample size, test statistic, degrees of freedom, p-value, effect size, and confidence interval. Preserve all values. Write "Not reported" for missing information. Separate numerical findings from interpretation, and avoid causal language unless the design supports it.
Keep the prompt, output, and original analysis together so revisions remain traceable.
APA Formatting for Reporting Statistical Results
Choose numbers and decimal places consistently
When reporting statistics in APA style, generally write numbers one through nine as words and use numerals for 10 and above. Use numerals for statistical values, sample sizes, and exact measurements, including 9 kg. Leave a space between the number and unit abbreviation.
Use one decimal place for means and standard deviations as a practical default when the measurement scale supports it. Report most other statistics to two decimal places, and keep decimal places consistent across comparable statistics. Preserve additional decimal places when rounding would conceal meaningful information.
Report exact p-values, usually to three decimal places, except values below .001: write p < .001, never p = .000. APA's numbers and statistics guide also addresses exceptions and consistent presentation.
Format symbols, zeros, and intervals
Italicize Latin statistical symbols, including M, SD, t, F, p, r, and d. Greek symbols such as χ² remain upright.
Use a leading zero for statistics that can exceed one, such as d = 0.40, t = 0.78, or z = 0.62. The leading zero is omitted for p, Pearson's r, and Cronbach's alpha.
State the confidence level and enclose interval limits in square brackets, separated by a comma: 95% CI [0.10, 0.90].
Common statistics such as means and t-tests don't need formulas or methodological references unless an unusual calculation requires explanation.
Interpret P-Values, Effect Sizes, and Confidence Intervals
Explain what a p-value establishes
A p-value describes how probable results at least as extreme as those observed would be under the null model and its assumptions. It isn't the probability that your hypothesis is false or that chance caused the finding.
The American Statistical Association's statement warns against treating a threshold as sufficient evidence for a scientific conclusion.
If you prespecified α = .05, a p-value below that threshold meets your significance criterion, but statistical significance alone isn't a complete scientific conclusion.
A nonsignificant result doesn't establish that no effect exists. Its interval may still include effects that matter.
Explain magnitude and uncertainty together
Effect sizes describe magnitude. Examples include Cohen's d, Pearson's r, partial eta squared, and regression coefficients. Interpret effect sizes in their measurement-scale and disciplinary context.
Practical significance concerns whether an effect matters in use. A small p-value cannot establish that a difference is educationally or clinically meaningful.
Confidence intervals describe estimation uncertainty. Under repeated sampling, a valid 95% interval procedure covers the population parameter 95% of the time. It doesn't assign a 95% probability to a fixed parameter lying inside this particular interval.
Because wide intervals indicate limited precision, retain that uncertainty when AI rewrites your findings.
What to Report for Common Statistical Tests
The numerical examples below are illustrative, not findings from an actual study. When reporting statistics in APA, include effect sizes and intervals alongside tests, consistent with APA 7 reporting guidance. In a real report, each p-value must match verified output.
Descriptive statistics, t-tests, and correlations
For descriptive results, identify the observations and report central tendency with variation. Standard deviations describe spread; standard errors describe estimation precision. Use numerals for sample sizes and statistical values.
For an independent-samples t-test, provide group sizes, means, standard deviations, t, degrees of freedom, p, and Cohen's d. Report the mean difference and its interval.
With 50 participants per group, mean scores were 12.0 and 10.0 (SDs = 5.0 and 5.0), respectively. The difference was statistically significant, t(98) = 2.00, p = .048, d = 0.40; mean difference = 2.00, 95% CI [0.02, 3.98].
Use a leading zero for d = 0.40, but omit it for p = .048.
For Pearson correlation, report the variables, sample size, r, degrees of freedom, p, and interval. With 100 observations, an illustrative report is: "The variables were positively correlated, r(98) = .30, p = .002, 95% CI [.11, .47]." Correlation doesn't establish causation.
Chi-square tests and ANOVA
For chi-square, report degrees of freedom in parentheses, χ², sample size, p, and an appropriate effect size:
"Group and response category were associated, χ²(1, N = 100) = 4.00, p = .046, φ = .20."
For ANOVA, report numerator and denominator degrees of freedom, F, p, group descriptions, and effect size:
"Scores differed across three groups (N = 90), F(2, 87) = 4.20, p = .018, ηp² = .09."
A significant omnibus ANOVA doesn't identify which groups differ. Report follow-up comparisons, their multiplicity correction, and relevant confidence intervals separately.
Illustrative Example: Regression Output to Results Paragraph
Consider an observational study relating final exam scores to weekly study hours while adjusting for baseline scores. The following values are simulated solely to demonstrate reporting.
The output packet contains these model and coefficient results.
Output item | Illustrative value |
|---|---|
Observations | 100 |
Model fit | R² = .14 |
Overall model test | F(2, 97) = 8.00, p < .001 |
Study-hours coefficient | B = 0.50, SE = 0.20 |
Coefficient test | t(97) = 2.50, p = .014 |
Coefficient interval | 95% CI [0.10, 0.90] |
A suitable paragraph is below; compare each of its numerals with the supplied output before using it.
"Multiple linear regression examined final exam scores in relation to weekly study hours and baseline scores among 100 participants. The model explained approximately 14% of the observed variance, R² = .14, F(2, 97) = 8.00, p < .001. Each additional weekly study hour was associated with 0.50 additional exam points after adjustment for baseline score, B = 0.50, SE = 0.20, t(97) = 2.50, p = .014, 95% CI [0.10, 0.90]."
The coefficient describes an adjusted association. It doesn't show that increasing study time causes higher scores, because adjustment cannot rule out all confounding.
Residual diagnostics weren't supplied, so AI must label assumption checks "Not reported." Likewise, it cannot invent participant selection details or claim the sample represents all students.
For a full model, include every predictor in a coefficient table rather than presenting only the statistically significant ones.
Choose Between Prose, Tables, and Figures
Use prose for the main finding and its direction. A table works better for several groups, multiple regression coefficients, or repeated analyses. Figures help readers see distributions, trends, and uncertainty.
Don't repeat descriptive statistics already displayed in a table or figure. Instead, refer to the display and explain the important comparison. Define abbreviations and significance markers in table notes; asterisks alone don't replace clearly reported results.
Before submitting, make three final passes:
Check every numeral, sign, degree of freedom, and interval endpoint against the original software output.
Inspect every interpretation for unsupported causal claims, exaggerated practical importance, or omitted limitations.
Preserve the analysis files, drafts, prompts, and verified source passages, and disclose AI assistance when institutional or journal rules require it.
When reporting statistical results from published studies, keep source facts, your own analysis, and unverified AI suggestions separate. Read the original methods, results, and discussion before paraphrasing. Cite the underlying research, not the chatbot's answer.
Frequently Asked Questions
Is a p-value of .05 statistically significant?
That depends on the prespecified decision rule, including whether it uses p < .05 or p ≤ .05. Also, a displayed .050 may hide rounding. Report adequate precision and the stated criterion rather than treating the boundary as a universal verdict.
Can AI fill in missing statistical results?
Don't ask a writing assistant to guess missing values. Request "Not reported," then obtain any required estimate through a documented statistical calculation. If an effect size or interval isn't available, distinguish that omission from a result showing no effect.
Keep the Evidence in Control
AI can make statistical explanations clearer, but traceable evidence makes them defensible. Keep verified estimates, uncertainty, and study-design limits attached to every claim.
A polished paragraph earns its place in your paper only when its numbers match the analysis and its conclusions stay within what the study supports.