2026-10-07
Which statistical test should I use? A simple guide for thesis students
Choosing the statistical test scares many students, but it depends on only three questions. Answer them in order and the test chooses itself.
Question 1: What type is your outcome?
- Numbers (pain on a 0–10 scale, range of motion in degrees, a questionnaire score) → continuous outcome.
- Categories (improved / not improved, male / female, yes / no) → categorical outcome.
Question 2: How many groups, and are they independent?
- Independent groups: different people in each group (e.g. a treatment group and a control group).
- Related (paired) measurements: the same people measured more than once (e.g. before and after treatment).
Question 3: Are the data normally distributed?
For continuous outcomes, check normality with the Shapiro–Wilk test (best for samples under about 50) and look at a histogram or Q–Q plot. If p > 0.05, you can treat the data as normal and use a parametric test; if not, use the non-parametric alternative.
The quick table
| Your design | Normal data | Not normal |
|---|---|---|
| 2 independent groups | Independent t-test | Mann–Whitney U |
| 1 group, before vs after | Paired t-test | Wilcoxon signed-rank |
| 3 or more independent groups | One-way ANOVA | Kruskal–Wallis |
| 1 group, 3 or more time points | Repeated-measures ANOVA | Friedman |
| 2 groups, before and after | Mixed (split-plot) ANOVA, or ANCOVA with baseline | Compare change scores with Mann–Whitney |
| Relationship between two numbers | Pearson correlation | Spearman correlation |
| Two categorical variables | Chi-square test | Fisher's exact test (if expected counts < 5) |
A typical physiotherapy RCT
"Two groups (exercise vs. exercise + TENS), pain measured before and after 6 weeks." The classic choice is a mixed ANOVA (group × time) or an ANCOVA on the post-treatment score with the baseline score as a covariate. Many theses also report the within-group change (paired t-test or Wilcoxon) and the between-group difference at the end.
Do not forget
- Report effect sizes, not only p-values (Cohen's d, η², or r). Committees and journals now expect them.
- Post-hoc tests after ANOVA (e.g. Bonferroni or Tukey) to know which groups differ.
- Set α = 0.05 before you see the data, and state the software you used (e.g. SPSS version 26).
Let Neura choose — and run — the test
Neura's Results section tool reads your Excel file and your study design, checks normality, proposes the right test for each outcome with the reason, and waits for your approval. Then it runs the analysis with audited statistical code and writes your Results chapter with APA-style tables. Its statistics are tested against SPSS-equivalent results (48 of 48 checks correct). Your first analysis each day is free.
Summary: outcome type → number of groups and paired or not → normal or not → pick from the table → report effect sizes.