How to Write a Statistical Analysis Section

Why the Statistical Analysis Section Matters

The statistical analysis section is one of the most closely evaluated parts of a scientific manuscript. It tells readers, reviewers, and editors how the data were analyzed, why specific methods were selected, and whether the conclusions are supported by an appropriate analytical strategy.

A strong statistical analysis section should be clear enough for another researcher to understand the main steps of the analysis. It should not only name statistical tests, but also explain how variables were summarized, how comparisons were performed, how models were built, how missing data were handled, and how statistical uncertainty was reported.

In many manuscripts, statistical reporting is too brief. Authors may write only that “data were analyzed using SPSS” or that “p < 0.05 was considered significant.” This is usually not enough. Reviewers need to understand the logic of the analysis, not just the software used.

Begin with Descriptive Statistics

The section should first explain how baseline characteristics and study variables were summarized. This helps the reader understand how the dataset was described before any comparisons or models were applied.

Continuous variables should be reported according to their distribution. Normally distributed data are often presented as mean ± standard deviation, while skewed data are better reported as median and interquartile range. Categorical variables should usually be presented as frequencies and percentages.

For example:

“Continuous variables were summarized as mean ± standard deviation or median with interquartile range, depending on data distribution. Categorical variables were presented as counts and percentages.”

This simple statement shows that the authors considered the type and distribution of the data before reporting results.

Explain How Data Distribution Was Assessed

If the manuscript compares continuous variables, the statistical analysis section should explain how normality or distributional assumptions were evaluated. This is important because the choice between parametric and non-parametric tests depends partly on data distribution.

Common approaches include:

• Visual inspection of histograms or Q-Q plots
• Shapiro-Wilk test
• Kolmogorov-Smirnov test
• Assessment of skewness and kurtosis

The manuscript does not need to overexplain these methods, but it should show that test selection was not arbitrary.

Example wording:

“Normality of continuous variables was assessed using visual inspection of histograms and the Shapiro-Wilk test.”

Match Statistical Tests to the Data Type

A strong statistical analysis section should clearly connect each statistical test to the type of variable and comparison being performed.

This does not mean that every test must be explained in detail, but the reader should understand why each method was appropriate.

Define the Primary Outcome and Main Variables

The statistical analysis section should clearly identify the primary outcome. This is especially important in clinical and biomedical research, where manuscripts may include many variables and secondary analyses.

The reader should be able to understand:

• What the main outcome was
• Which exposure, intervention, or predictor was evaluated
• Which secondary outcomes were analyzed
• Whether subgroup or exploratory analyses were predefined

A vague analysis section may make the study look unfocused. A clear definition of the primary outcome helps reviewers judge whether the statistical strategy matches the research question.

Example:

“The primary outcome was postoperative functional improvement at 12 months. The main exposure variable was treatment group. Secondary outcomes included complication rate, reoperation, and radiological recurrence.”

Describe the Main Statistical Model

If the study uses regression or multivariable analysis, the model should be described clearly. The manuscript should specify what type of model was used and why it was appropriate for the outcome.

For example:

• Linear regression for continuous outcomes
• Logistic regression for binary outcomes
• Cox proportional hazards regression for time-to-event outcomes
• Poisson or negative binomial regression for count outcomes
• Mixed-effects models for repeated measures or clustered data

A strong statistical analysis section should also explain how variables were selected for the model. Were they selected based on clinical relevance, prior literature, univariable screening, or a predefined analysis plan?

Example wording:

“Variables included in the multivariable model were selected based on clinical relevance and previously reported associations with the outcome.”

Report Adjustment Variables Clearly

Adjustment variables are important because they show how the analysis attempted to account for confounding. These may include demographic, clinical, radiological, laboratory, or treatment-related variables.

Common adjustment variables include:

• Age
• Sex
• Baseline disease severity
• Comorbidities
• Treatment group
• Tumor size or lesion volume
• Follow-up duration
• Center or surgeon when relevant

Avoid writing only that “multivariable analysis was performed.” Instead, state which variables were included and why.

Example:

“The multivariable logistic regression model was adjusted for age, sex, baseline severity, lesion size, and treatment modality.”

This gives reviewers a clearer understanding of the model.

Address Missing Data

Missing data can affect the reliability of study findings. Therefore, the statistical analysis section should explain how incomplete data were handled.

A weak statement would be:

“Patients with missing data were excluded.”

A stronger statement would be:

“Patients with missing primary outcome data were excluded from the primary analysis. A sensitivity analysis was performed to evaluate whether exclusion of incomplete cases affected the main results.”

This shows that the authors considered the possible impact of missing data.

Include Sensitivity and Subgroup Analyses When Appropriate

Sensitivity analyses help show whether the main findings remain stable under alternative assumptions or analytical approaches. They are especially useful when the study includes missing data, borderline significance, retrospective design, or complex modeling.

Subgroup analyses may be useful when the effect of an exposure or treatment may differ across patient groups. However, subgroup analyses should be used carefully. Too many subgroup analyses can create a risk of false-positive findings.

Examples of subgroup analyses include:

• Age groups
• Disease severity groups
• Treatment subtypes
• Tumor size categories
• Follow-up duration groups
• Center-specific analyses in multicenter studies

If subgroup analyses were performed, state whether they were predefined or exploratory.

Example:

“Predefined subgroup analyses were performed according to age group and baseline disease severity. Interaction terms were tested to evaluate whether the association between treatment and outcome differed across subgroups.”

Report Statistical Significance and Confidence Intervals

The statistical analysis section should define the threshold for statistical significance, but it should not rely only on P values. Confidence intervals are important because they show the precision and uncertainty around an estimate.

A typical statement is:

“A two-sided P value of <0.05 was considered statistically significant. Effect estimates were reported with 95% confidence intervals.”

This is better than reporting P values alone. Confidence intervals help the reader judge whether the result is precise, clinically meaningful, or highly uncertain.

Mention Statistical Software

The statistical software used should be reported clearly, including the version when possible. This improves transparency and reproducibility.

Examples include:

• R
• SPSS
• Stata
• SAS
• GraphPad Prism
• Python
• Jamovi

Example:

“All analyses were performed using R version 4.3.0.”

Software reporting should usually come near the end of the statistical analysis section.

Avoid Common Statistical Reporting Problems

Many manuscripts are weakened by incomplete or unclear statistical reporting.

A strong statistical analysis section should make the analytical pathway transparent.

Practical Structure for a Statistical Analysis Section

A clear structure may follow this order:

1. Descriptive statistics — Explain how continuous and categorical variables were summarized.
2. Distribution assessment — State how normality or skewness was evaluated.
3. Group comparisons — Describe the tests used for categorical and continuous variables.
4. Primary outcome analysis — Define the main outcome and main statistical model.
5. Adjustment variables — List covariates included in multivariable models.
6. Missing data — Explain how incomplete data were handled.
7. Sensitivity or subgroup analyses — Describe additional analyses when relevant.
8. Statistical significance and software — Report P value threshold, confidence intervals, and software.

Annotated checklist

A strong statistical analysis section typically covers:

• Data summary — how continuous and categorical variables are described
• Key variables — primary outcome and main exposure defined clearly
• Choice of tests — methods matched to data type and distribution
• Main model — regression approach and key covariates for adjustment
• Assumption checking — how model assumptions were evaluated
• Missing data — how incomplete records were handled
• Sensitivity analysis — robustness under alternative approaches
• Subgroups / interaction — predefined group and interaction testing
• Significance & reproducibility — P values, confidence intervals, software, code review

Click to enlarge

A strong statistical analysis section shows a clear analytical strategy—not just a list of test names.

Statistical Analysis Checklist Before Submission

Before submitting your manuscript, check whether your statistical analysis section answers these questions:

• Are continuous and categorical variables summarized correctly?
• Is the primary outcome clearly defined?
• Are the main exposure or predictor variables identified?
• Are statistical tests matched to variable type and distribution?
• Is the main regression model described clearly?
• Are adjustment variables listed?
• Is missing data handling explained?
• Are sensitivity analyses described when relevant?
• Are subgroup analyses predefined or clearly labeled as exploratory?
• Are confidence intervals reported?
• Is the statistical significance threshold stated?
• Is the software and version reported?

A Simple Formula

A strong statistical analysis section can often be built with this formula:

Data were summarized as…
Normality was assessed using…
Group comparisons were performed with…
The primary outcome was…
The main model was…
The model was adjusted for…
Missing data were handled by…
Sensitivity or subgroup analyses were performed when…
Statistical significance was defined as…
Analyses were performed using…

This formula helps ensure that the section is complete, transparent, and reviewer-friendly.

In short

A strong statistical analysis section does not just say which tests were used.

It shows why those tests were appropriate, how the models were built, and how the reliability of the findings was checked.

For broader manuscript structure, see our how to write a scientific paper guide—or run a pre-submission review that flags statistical reporting gaps.