Choosing the right statistical test for your thesis

Start from your research question

Before opening SPSS or R, decide what you are trying to show: a difference between groups, a relationship between variables, or a prediction. The question determines the family of tests; the data determines which member of that family you use.

Know your variable types

Categorical variables (gender, department) and continuous variables (scores, income) call for different tests. Likert-scale items are ordinal, although summed scale scores are often treated as continuous. Be explicit about how you treat them and why.

Check the assumptions

Parametric tests assume approximately normal distributions and similar variances across groups. Check normality with Shapiro–Wilk or visual plots, and homogeneity of variance with Levene’s test. If assumptions fail, use a non-parametric alternative.

A quick decision guide

Comparing two groups: independent t-test, or Mann–Whitney U if assumptions fail. Comparing three or more groups: one-way ANOVA, or Kruskal–Wallis. Relationship between two continuous variables: Pearson correlation, or Spearman. Association between two categorical variables: chi-square test. Predicting an outcome from several variables: multiple regression.

When to use SEM or PLS-SEM

If your model includes latent constructs measured by several questionnaire items, or tests mediation and moderation together, structural equation modelling is often appropriate. Covariance-based SEM (AMOS) suits theory testing with larger samples; PLS-SEM (SmartPLS) is common for prediction-oriented and exploratory models.

Report effect sizes, not just p-values

Examiners increasingly expect effect sizes and confidence intervals. A statistically significant result with a tiny effect may have little practical meaning. Reporting both shows you understand what your results actually say.