Statistical Analysis for a Seminar Paper or Thesis: The Complete Guide

Every step of the statistics in an academic paper - from hypotheses and data preparation to choosing the test and writing the results chapter in APA style. With examples, common mistakes and when to get help.

The statistical analysis is the part of a paper students fear most - and a wrong test or a sloppy report can sink a whole results chapter. The good news: the process is always the same. Almost every quantitative seminar paper or thesis goes through the same seven steps, in this order. This guide walks through each one and links to the detailed guide for every step.

1. Start from your hypotheses, not your data

The analysis is decided before you open SPSS. For every hypothesis, know three things:

• The dependent variable - what you measure (e.g. an anxiety score).
• The independent variable - what predicts or separates (gender, treatment group, years of experience).
• The kind of relationship: a difference between groups, an association, or a prediction.

"Students who work more than 20 hours a week will report more burnout" is a two-group difference. "The more hours students work, the higher their burnout" is a correlation - and is analysed differently. The wording of the hypothesis picks the test.

2. Know your measurement levels

Each variable is nominal (gender, department), ordinal (rank), or interval/ratio (age, score, a questionnaire total). The level of the dependent variable decides between parametric tests (t, ANOVA, Pearson) and non-parametric tests or chi-square. A Likert scale averaged over items is usually treated as interval; a single Likert item is ordinal.

3. Prepare and clean the data

The most underestimated step, and often longer than the analysis itself:

• Coding: clear variable names, value labels (1 = male, 2 = female), missing values defined.
• Reverse-scored items: if you don't reverse them before summing, the scale score is wrong.
• Impossible values: a 55 on a 1-5 item is a typo, not a finding.
• Scale scores and reliability (Cronbach's alpha). If alpha is low, there are options - see Low Cronbach's alpha: what to do.

4. Choose the right test

The full decision guide, including when to switch to a non-parametric test: How to choose a statistical test. When Mann-Whitney is the right call: Mann-Whitney: when to use it. Detailed guides to the most common tests: which t-test fits and one-way ANOVA and post-hoc tests.

5. Check assumptions before you run

Parametric tests assume a roughly normal dependent variable (or a large enough sample), equal variances between groups (Levene's test), and for correlation and regression, a linear relationship. A violated assumption is not the end: there is a correction (e.g. Welch's t) or an alternative test. What matters is checking and reporting it. Think about sample size up front, too - see Sample size and power analysis.

6. Run the analysis and read the output

Most of SPSS's output is noise. Each test has three or four numbers that matter: the statistic (t, F, r, χ²), degrees of freedom, the p-value and an effect size. A line-by-line walkthrough: Reading logistic regression output in SPSS. Significance is not importance: APA 7 asks for effect sizes and confidence intervals, not only p.

7. Write the results chapter

Organise it by hypothesis, not by the order of the output: descriptive statistics first (one table), then each hypothesis - what was tested, the test, the APA-formatted result, and whether it was supported.

Example: "Burnout was significantly higher among students working more than 20 hours a week (M = 3.82, SD = 0.61) than among those working less (M = 3.41, SD = 0.72), t(118) = 3.34, p = .001, d = 0.61."

Interpretation belongs in the discussion. Formats: Reporting regression in APA; before you submit: APA checklist.

Common mistakes

• Choosing the test by which one came out significant.
• Running many t-tests instead of one ANOVA.
• Reporting p = .000 (write p < .001).
• Forgetting reverse-scored items.
• Reading a correlation as causation.
• Skipping the sample description.

Every guide in this series

• How to choose a statistical test
• Missing data in a questionnaire
• Low Cronbach's alpha
• Testing normality
• Which t-test fits
• One-way ANOVA and post-hoc tests
• Two-way ANOVA and interactions
• The chi-square test
• Correlation or regression
• Logistic regression in SPSS
• Mediation and moderation with PROCESS
• Non-parametric tests
• SEM, path analysis and latent variables
• How to write the results chapter - formula and templates
• ChatGPT for statistical analysis - what works and what is risky
• Mann-Whitney
• Effect sizes
• Sample size and power
• Reporting regression in APA
• APA checklist before submission

When to get expert help

When the hypotheses aren't clear, when your supervisor asks for a model you never learned (mediation with PROCESS, multilevel models), when the data is messy, or when time is short. Good help means guidance and explanation, so you can defend every decision yourself. Tell us where you're stuck.

FAQ

Do I have to use SPSS? No - JASP (free) and R give the same results. Check what your supervisor expects.

None of my hypotheses were supported. Now what? Report it exactly like a significant result; a null result is a finding, and the discussion can explain it.

Is it allowed to get help with the analysis? Guidance, explanation and checking your own work - yes, statistical consulting is standard practice. Submitting work someone else wrote in your name - no. Check your institution's rules.

Last updated: 2026-10-01