ACADEMIC WRITING GUIDE

How to Report Statistical Results in a Dissertation (With Worked Examples)

By Acadelyra Editorial TeamPublished

A statistical test can be correctly calculated and poorly reported. A dissertation results section should let readers identify what was analysed, the estimated magnitude and direction of the finding, the uncertainty around it, and the limits of what the evidence establishes. You do not need to reproduce every line of software output. You do need to make every claim traceable to the analysis.

This guide demonstrates five realistic reporting situations using explicitly hypothetical teaching data. The numbers are illustrative, not findings from real participants. Follow your university handbook, the statistical model you actually fitted, and the reporting conventions in your discipline.

What belongs in a statistical results paragraph?

Begin with the relevant research question and analysis. Give the analytic sample size and describe the groups or variables. Report the estimate or test statistic with degrees of freedom where applicable, then its p-value and an effect size or interval estimate when meaningful. Close with a narrow interpretation that does not overstate what the design can support.

ReportWhy it mattersExample
Analytic sampleClarifies how much data the result reflectsn = 100
EstimateStates the observed magnitude and directionMean difference = 6.0 points
UncertaintyShows a plausible range under the model95% CI [2.03, 9.97]
Test resultProvides the formal statistical testt(98) = 3.00, p = .003
MeaningInterprets the result in contextHigher observed scores; cause not established

Do not confuse statistical significance with practical importance. A small effect can be statistically significant in a large sample; a meaningful effect can be estimated imprecisely in a small sample. A p-value is not the probability that your null hypothesis is true.

How to write statistical results: eight practical steps

1. Match every paragraph to a research question

Make a small analysis map linking each question to its variables, study design and intended test. The data analysis plan guide can help connect research questions with defensible planned tests. Avoid presenting every calculation the software produced.

2. Confirm the analytic dataset

State how many observations were included and explain exclusions or missing values that change the denominator. If different models use different n values, report them rather than implying every test used the full sample. Our research-data cleaning guide explains how to document exclusions and missing responses before analysis.

3. Describe variables and groups precisely

For each variable, give the units, coding and direction. A larger score should have an intelligible meaning. If you reverse-scored an item or transformed a variable, document it. Our operational definitions guide explains how to define each measurement before interpretation.

4. Report descriptive results first

Give counts and percentages for categories, and suitable location and spread statistics for numerical variables. Mean and standard deviation may fit approximately symmetric data; skewed data may call for median and interquartile range. Do not test a difference you have not actually tested.

5. Match statistics to the exact test used

A Student equal-variance t-test and a Welch t-test do not necessarily have the same degrees of freedom. Correlation coefficients, model coefficients, and omnibus test statistics answer different questions. Never take a convenient statistic from one output and pair it with a p-value from another.

6. Include magnitude and uncertainty

Where appropriate, report mean differences, standardized effects, correlations, regression estimates and confidence intervals. An interval helps readers assess precision; it is not a promise that all future results will fall inside its endpoints.

7. Interpret without making unwarranted claims

For observational or cross-sectional data, prefer “was associated with” or “differed between” rather than “caused.” A non-significant test is not proof of equivalence. Separate results reporting from extended causal or theoretical discussion. For chapter-level organization, see our research findings chapter guide.

8. Check the prose against tables and software output

Match rounding, denominators, signs, confidence limits, p-values and variable names across the narrative and tables. Our tables-and-figures guide shows how to present results visually without repeating the complete table in prose.

Worked example 1: Descriptive statistics

Hypothetical study: A survey asks 120 undergraduates how useful they find written feedback. Of these, 114 answer the feedback item; 76 say useful and 38 say not useful. On a separate engagement scale, 110 valid observations yield a mean of 3.42 and a standard deviation of 0.71 on a 1–5 scale.

MeasureExample resultCorrect denominator
Perceived useful feedback76 (66.7%)114 valid responses
Feedback not perceived useful38 (33.3%)114 valid responses
Engagement scoreM = 3.42, SD = 0.71110 valid scores

Weak write-up

“Most students liked feedback and engagement was high.”

Stronger dissertation paragraph

“Of the 114 students who completed the feedback-usefulness item, 76 (66.7%) described the written feedback as useful and 38 (33.3%) did not. Among 110 students with valid engagement-scale scores, mean self-reported engagement was 3.42 (SD = 0.71) on a 1–5 scale. The denominators differ because six students did not answer the feedback item and ten did not have a valid engagement score.”

Why it works

The paragraph states precisely which responses were available, avoids treating 120 as every percentage denominator, and does not declare 3.42 “high” without a justified benchmark. No inferential comparison is implied.

Worked example 2: Independent-samples t-test

Hypothetical study: Group A and Group B each contain 50 students. Their assessment means are 76 and 70, respectively, with SD = 10 in both groups. Assuming independent observations and a defensible equal-variance model, the estimated difference is 6 points. The standard error is 2, giving t(98) = 3.00. Two-sided p is approximately .0034; a 95% interval for the mean difference is approximately [2.03, 9.97]. Cohen’s d = 0.60 using the pooled standard deviation.

StatisticGroup AGroup B
Sample size5050
Mean assessment score76.070.0
Standard deviation10.010.0

Weak write-up

“The t-test proved the teaching method improved achievement because p < .05.”

Stronger dissertation paragraph

“Mean assessment scores were higher in Group A (M = 76.0, SD = 10.0, n = 50) than in Group B (M = 70.0, SD = 10.0, n = 50). An independent-samples Student t-test indicated a difference between the observed groups, t(98) = 3.00, p = .003 (two-sided). The estimated mean difference was 6.0 points, 95% CI [2.03, 9.97], corresponding to Cohen’s d = 0.60. Because this hypothetical comparison does not establish random allocation or control of potential confounders, the result alone cannot establish that the teaching approach caused the difference.”

Why it works

Readers see the group summaries, inferential result, magnitude and precision together. The assumption and design qualification are explicit. If the real analysis uses Welch’s test, report the actual Welch statistic, degrees of freedom, interval and p-value—not these Student-test values.

Worked example 3: Pearson correlation

Hypothetical study: An observational sample of 80 students reports a correlation of r = .35 between perceived feedback usefulness and engagement. The test has df = 78. Its two-sided p-value is approximately .0015. A Fisher z-based 95% CI for the population correlation is approximately [.14, .53].

Weak write-up

“Feedback improves engagement by 35%, and the relationship is proven.”

Stronger dissertation paragraph

“Among 80 students, perceived feedback usefulness was positively associated with self-reported academic engagement, Pearson’s r(78) = .35, p = .001 (two-sided), with a Fisher z-based 95% CI of approximately [.14, .53]. Students who reported more useful feedback tended to report greater engagement. The result describes an association and does not show that feedback usefulness causes engagement.”

Why it works

A correlation of .35 is not a 35% increase. Its square (.1225) describes roughly 12.3% shared variance in a simple bivariate description, but that should not be casually interpreted as the proportion of engagement caused by feedback. Check distributional assumptions, outliers and nonlinearity before choosing Pearson correlation.

Worked example 4: Multiple linear regression

Hypothetical model: Engagement (1–5 scale) is predicted using feedback usefulness (1–5), study hours per week and year of study (coded numerically as an illustrative simplifying choice). The analytic sample has n = 120. The overall fitted model has R² = .24 and adjusted R² = .22. The chosen example coefficient estimates and intervals are internally consistent as illustrative reporting values, not reconstructed real software output.

PredictorB95% CIp
Feedback usefulness0.28[0.13, 0.43].0003
Weekly study hours0.04[0.01, 0.07].010
Year of study0.06[-0.05, 0.17].281

Weak write-up

“All the variables influenced engagement, and the regression explained 24% of students.”

Stronger dissertation paragraph

“In the multiple linear regression fitted to 120 students, the predictors collectively accounted for 24% of the observed variation in engagement scores (R² = .24; adjusted R² = .22). After adjustment for the other included predictors, perceived feedback usefulness was positively associated with engagement (B = 0.28, 95% CI [0.13, 0.43], p < .001). Weekly study hours were also positively associated with engagement (B = 0.04, 95% CI [0.01, 0.07], p = .010), whereas the estimate for year of study was imprecise and its interval crossed zero (B = 0.06, 95% CI [-0.05, 0.17], p = .281). These are conditional associations in a hypothetical observational model, not demonstrated causal effects.”

Why it works

B describes expected change in the outcome’s units per one-unit change in a predictor, holding the other included predictors constant; it is not a percentage unless the outcome/model scale warrants that interpretation. R² concerns variance in observed outcomes, not a percentage of individual students “explained.” In an actual report, also give the precise model specification, diagnostics, any categorical reference groups and omnibus model result where required.

Worked example 5: Non-significant findings and uncertainty

Hypothetical comparison: Two groups of 40 students each have mean ratings of 3.1 and 3.3, with within-group standard deviations of 1.0. The difference (A minus B) is -0.20 points. Under a Student equal-variance model, SE ≈ 0.224, t(78) ≈ -0.89, p ≈ .375, and the 95% CI for the difference is approximately [-0.65, 0.25].

Weak write-up

“There was no difference between the groups.”

Stronger dissertation paragraph

“Group A (M = 3.10, SD = 1.00, n = 40) and Group B (M = 3.30, SD = 1.00, n = 40) had an estimated mean difference of -0.20 points (A minus B). The two-sided independent-samples Student t-test did not provide sufficient evidence of a difference at the prespecified .05 level, t(78) = -0.89, p = .374, 95% CI [-0.65, 0.25]. Because the interval includes both potentially meaningful negative values and smaller positive values, the finding does not establish that the groups are equivalent.”

Why it works

A non-significant result does not prove the null hypothesis or a zero effect. A formal equivalence claim requires a justified equivalence margin and an appropriate design and test. Report the estimate and interval even when p exceeds .05.

APA-style reporting conventions: a quick practical checklist

ItemUseful practice
Statistical symbolsItalicize symbols such as p, t, F, r, M and SD where required by the chosen style
p-valuesReport exact values to appropriate precision; use p < .001 when smaller than .001
IntervalsIdentify the confidence level and the parameter being estimated
RoundingKeep consistent, sensible decimals without changing conclusions
Sample sizesState the n used in each analysis and explain missing-data exclusions
Tables and narrativeDo not narrate every cell; highlight the answers to the research questions

Conventions differ among statistical approaches and university departments. Do not force a test statistic into a formula it does not use. Follow the reporting standard relevant to your method and preserve the exact output of the verified analysis.

Common mistakes—and how to fix them

  • Copying raw SPSS, R or Excel output without explaining the substantive result: select the statistics that answer the specific question.

  • Writing “significant” without magnitude, direction or uncertainty: add an estimate, appropriate interval and real-world unit.

  • Misstating p as the probability the hypothesis is true: explain p as a model-based probability for data at least as extreme under the null and assumptions.

  • Calling a non-significant result proof of no effect: report the interval and acknowledge limited precision.

  • Interpreting correlation as a percentage increase or a causal effect: describe direction and strength of association.

  • Failing to disclose changes in analytic n: report the denominator used for each result and material missing-data decisions.

  • Reporting contradictory numbers between a paragraph and a table: reconcile the single analysis output and apply a consistent rounding rule.

  • Ignoring a prespecified analysis because its result was disappointing: distinguish planned tests from exploratory follow-ups.

An adaptable dissertation results paragraph template

“To address [research question], [test/model] was applied to [n] eligible observations. [Groups or variables] were summarized as [appropriate statistics]. The estimated [difference/association/coefficient] was [estimate and units], with [95% CI or other uncertainty]. The test result was [test statistic, degrees of freedom if applicable, and p-value]. This indicates [bounded finding] in the observed data. Because [design/measurement limitation], the analysis does not by itself establish [unsupported stronger claim].”

Use the template selectively: a descriptive paragraph does not need an inferential p-value; a multivariable model may require several predictors and diagnostic details. The aim is an accurate, readable report, not a rigid formula.

Frequently asked questions

Should I report p = .000?

No. Statistical software may display .000 after rounding; report p < .001 when the value is sufficiently small, rather than implying a probability of exactly zero.

Must every result include an effect size?

Use the most informative estimate and uncertainty for the question and the statistical approach. Many inferential results benefit from an effect size, but some reports emphasize raw-unit coefficients or rate ratios instead of standardized effects.

Where do I discuss why a result occurred?

Usually in the Discussion chapter. Results sections should report what the analyses found; brief interpretation is appropriate, but lengthy theoretical explanation belongs elsewhere.

Can I include a non-significant result?

Yes. Report planned analyses regardless of direction or statistical significance, with estimates and uncertainty.

What if my supervisor requires APA 7?

Use your program’s APA-based guidance for statistics, tables, italicization and rounding, and check examples for your exact model.

Final quality-control checklist

  • [ ] Every paragraph answers a research question or analysis objective.

  • [ ] Sample sizes and denominators are correct.

  • [ ] Variable names, units and coding are unambiguous.

  • [ ] Means, percentages and spread statistics match the source output.

  • [ ] Test statistics, degrees of freedom and p-values agree.

  • [ ] Effect direction, magnitude and confidence intervals agree.

  • [ ] Non-significant results are reported without equivalence claims.

  • [ ] Correlations or regression coefficients are not labeled causal without justification.

  • [ ] Reported values agree across narrative, tables and figures.

  • [ ] Planned and exploratory tests are distinguished.

  • [ ] Methods and assumptions are accurately described.

  • [ ] All worked examples in this guide are clearly recognized as hypothetical.

Further reading and credible sources

American Psychological Association — Journal Article Reporting Standards (JARS)

Purdue OWL — Reporting Statistics in APA Style

American Statistical Association — Statement on Statistical Significance and P-Values

EQUATOR Network — SAMPL Guidelines

Final takeaway

Strong statistical reporting tells readers what you tested, how much the evidence suggests, how uncertain that estimate is, and what the research design can legitimately support. A useful results section is transparent, selective and faithful to the analysis—not a collection of impressive-looking numbers.

If you need help improving the organization, clarity or reporting of a research chapter, Acadelyra offers academic writing guidance and editorial support. Keep your original data, calculations and interpretation accountable to your approved methods and institutional rules.