You have a research question and a data file, and SPSS offers more than forty tests. How do you choose one that you can defend in a single sentence at your defense? Deciding which statistical test to use is not about memorizing names. It is about answering four questions in order: What does your research question ask? What is the measurement level of the dependent variable? How many groups or measurements are you comparing, and are they independent? Are the assumptions of the parametric test met?
This article is part of our guide How to Write a Master’s Thesis by the Mastermind PhD team and relates to our Statistical Analysis service. Sources last checked: November 2026.
This guide turns those four questions into a complete decision tree, applies it to five examples from different disciplines, clears up the most common confusions, and gives you the whole table as a printable PDF to keep beside you while you analyze.
What does choosing the right statistical test mean? It means matching the test to three things at once: the type of question (description, difference, relationship, effect, structure), the measurement level of the variables (nominal, ordinal, interval or ratio) and the design of the comparison (how many groups, independent or related), and then checking the assumptions of the chosen test and switching to its non-parametric alternative if they fail. The right test is the one you can justify with these three elements in one sentence.
Question 1: What does your research question ask?
The verb in the question is the first clue to which statistical test you need, because it identifies the family of tests:
- “What is the level of …” or “To what extent …” leads to description: means, standard deviations and an interpretation key, or frequencies and percentages.
- “Is there a difference …” or “Does … differ …” leads to a comparison of groups or measurements.
- “What is the relationship between …” leads to correlation.
- “What is the effect of … on …” or “How well does … predict …” leads to regression.
- “Does … mediate …” or “Does … moderate …” leads to path models or structural equation modeling.
- “Do the items measure …” leads to factor analysis.
A question with two verbs (“What is the level of satisfaction, and does it differ by gender?”) is two questions, each with its own table.
Question 2: What is the measurement level of the dependent variable?
| Level | Examples | Suitable statistics |
|---|---|---|
| Nominal | Gender, faculty, pass or fail | Frequencies, chi-square, logistic regression |
| Ordinal | Grade band, rank, a single Likert item | Median, non-parametric tests, Spearman’s rho |
| Interval or ratio | Age, test score, time, a score averaged from several Likert items | Mean, parametric tests, Pearson’s r, regression |
The convention in theses, and the one that decides which statistical test applies: a score averaged from several items is treated as interval; a single item is ordinal.
📱 Ask a statistician about your data — Send a short description of your instrument and sample size on WhatsApp and get a suitable analysis plan. Every analysis comes with the raw SPSS file and a results report.
Question 3: How many groups or measurements, and are they independent?
The design of the comparison narrows which statistical test fits:
- Two independent groups (women and men; experimental and control at post-test): independent samples t-test or Mann-Whitney U.
- Three or more independent groups (three qualification levels): one-way ANOVA or Kruskal-Wallis.
- Two measurements of the same people (pre-test and post-test): paired samples t-test or Wilcoxon signed-rank.
- Three or more measurements of the same people (pre-test, post-test, follow-up): repeated measures ANOVA or Friedman.
- Two independent variables at once (gender and qualification): two-way ANOVA with main effects and the interaction.
- A comparison that controls for a covariate (the pre-test score): ANCOVA.
Question 4: Are the assumptions met?
Parametric tests assume an approximately normal dependent variable within each group (or normal residuals in regression), homogeneous variances (Levene’s test), no influential outliers and roughly 30 or more cases per group. If these fail with small groups or ordinal data, use the non-parametric alternative. If the departure is moderate and the groups are large, you may rely on the robustness of the parametric test, and say so with a citation. With unequal variances, use Welch’s t-test, or Welch’s ANOVA with Games-Howell post hoc comparisons.
Which statistical test? The complete decision tree
- A description question. Interval variable: mean, standard deviation and a class-interval key where your program uses one. Nominal or ordinal variable: frequencies, percentages and the median.
- A difference question with an interval dependent variable and assumptions met:
- two independent groups: independent samples t-test (Welch’s version if variances differ);
- three or more independent groups: one-way ANOVA with Tukey or Games-Howell;
- two related measurements: paired samples t-test;
- three or more related measurements: repeated measures ANOVA;
- two independent variables: two-way ANOVA.
- With an ordinal dependent variable or failed assumptions: Mann-Whitney U, Kruskal-Wallis, Wilcoxon signed-rank or Friedman, respectively.
- With a nominal dependent variable: chi-square test of independence (Fisher’s exact test for small 2 × 2 tables), or McNemar’s test for two related measurements.
- A relationship question. Two interval variables with acceptable distributions: Pearson’s r. Ordinal or non-normal data: Spearman’s rho. Two nominal variables: chi-square with Cramér’s V. One interval and one dichotomous variable: point-biserial correlation, which is equivalent to a t-test.
- An effect or prediction question. Interval dependent variable: simple regression (one predictor), multiple regression (several predictors) or hierarchical regression (blocks ordered by theory). Binary dependent variable: binary logistic regression. Ordinal dependent variable: ordinal regression.
- A mediation, moderation or full-model question. The PROCESS macro for simple mediation; AMOS for confirmatory models with large samples; SmartPLS for exploratory models or smaller samples.
- An instrument-structure question. New instrument: exploratory factor analysis. Theory-based instrument: confirmatory factor analysis. Reliability: Cronbach’s alpha or McDonald’s omega.
Download the full table as a PDF: Statistical test selection table (two pages, printable) lists, for every question type, the variable types, the parametric test, the non-parametric alternative, the effect size and what to report, together with the assumption checklist.
Five worked examples from different disciplines
Each row applies the four questions to decide which statistical test fits a typical thesis question.
| Discipline and question | Analysis | Test | One-sentence justification |
|---|---|---|---|
| Education: Does teacher burnout differ by years of experience (three bands)? | Difference; interval DV (scale score); three independent groups | One-way ANOVA with Tukey | “The DV is interval, the three groups are independent, the distribution is acceptable and Levene’s test is non-significant (p = .23).” |
| Management: What is the effect of the five service-quality dimensions on customer satisfaction? | Effect; interval DV; five interval predictors | Multiple regression (Enter method) | “An effect question with several interval predictors and a sample of 300, which exceeds 50 + 8 × 5 = 90.” |
| Medicine: Is the type of treatment (three types) associated with complications (yes or no)? | Association between two nominal variables | Chi-square test of independence with Cramér’s V | “Both variables are nominal and all expected counts exceed 5.” |
| Psychology: Did anxiety scores improve in 24 participants after the program? | Difference between two related measurements; small sample | Wilcoxon signed-rank (or the paired t-test if the differences are normal) | “The same people were measured twice, the sample is under 30 and the differences are skewed.” |
| Engineering or computer science: What is the relationship between experience band (ordered) and acceptance of the new system? | Relationship; one ordinal variable | Spearman’s rho | “One variable is ordinal, so Pearson’s r is not appropriate.” |
Common confusions about which statistical test applies, and how to settle them
- Independent or paired t-test? If the same person is measured twice, the samples are paired; if each group contains different people, they are independent. Experimental and control groups at post-test are independent.
- Chi-square or t-test? Chi-square when the dependent variable is a count of nominal categories; t-test when it is the mean of a continuous variable.
- Correlation or regression? Correlation for the strength and direction of a relationship with no dependent variable specified; regression for effect, prediction and several predictors.
- One-way or two-way ANOVA? One-way for a single independent variable with several categories; two-way when two independent variables enter together and you want their interaction.
- Mediation or moderation? A mediator explains how an effect happens and lies between the independent and dependent variables; a moderator explains when the effect is stronger or weaker by changing the strength of the relationship.
- Kruskal-Wallis or ANOVA for Likert scores? Averaged scale scores with large groups and an acceptable distribution call for ANOVA; small groups, marked skew or a single item call for Kruskal-Wallis.
How do you justify which statistical test you chose in the methodology chapter?
A template: “To answer the third research question, an independent samples t-test was used because the dependent variable (job satisfaction score) is interval, the independent variable (gender) is nominal with two independent categories, and the distribution within each group was approximately normal (skewness and kurtosis within ±1). Because the variances were not homogeneous (Levene’s test, p = .041), Welch’s correction was applied.”
Applied to the data, the result reads: “Women (M = 3.49, SD = 0.64, n = 158) reported higher job satisfaction than men (M = 3.21, SD = 0.78, n = 142), t(274.8) = 3.35, p = .001, d = 0.39, a small-to-medium effect.” One justification sentence per test, written before the result, saves pages of defense in the discussion chapter.
📱 Ask on WhatsApp now — 5 years of experience across all disciplines and 14 academic services. Two-stage payment (80/20); revisions are free until supervisor approval, within 30% of the chapter.
Frequently asked questions
What do I do if one question needs two tests?
Split it, and decide which statistical test fits each part. “What is the level of satisfaction, and does it differ by gender?” is a descriptive question followed by a difference question. One question, one table, one interpretation.
Can I use both a parametric and a non-parametric test for the same question?
Do not present two results and keep the one that suits your hypothesis; that is selective reporting. You may put the alternative in an appendix to show that the result is stable when the assumptions are borderline, stating which result the conclusions rest on and why.
How do I choose between multiple regression and structural equation modeling?
Use regression when your variables are measured directly or as scale scores and you ask about the effect of predictors on one dependent variable. Use SEM when the variables are latent constructs with several dimensions, or when several dependent variables and mediators sit in one model and your sample is large enough.
Which test do I use for open-ended “suggestions” questions in a questionnaire?
None. They call for content analysis: classify the answers into categories and present frequencies and percentages with illustrative quotations.
Is the test that software or a website suggested automatically correct?
Tools that propose a test from variable types are useful for checking, but they do not know your question, your design or your assumption results. Use the tree to decide which statistical test fits first, then verify with the tool if you wish.
Sources
- Field, A. (2018). Discovering Statistics Using IBM SPSS Statistics (5th ed.). SAGE.
- Pallant, J. (2020). SPSS Survival Manual (7th ed.). Open University Press.
- Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences (2nd ed.). Lawrence Erlbaum.
- Laerd Statistics. Statistical test selector and SPSS tutorials. https://statistics.laerd.com/statistical-guides/
- McHugh, M. L. (2013). The chi-square test of independence. Biochemia Medica, 23(2), 143–149. https://doi.org/10.11613/BM.2013.018
- IBM. IBM SPSS Statistics documentation. https://www.ibm.com/docs/en/spss-statistics
Sources last checked: November 2026. This article is general educational guidance and does not replace your university regulations or your supervisor’s instructions. How we prepare our content.