Which Statistical Test Should I Use?
Choosing the right statistical test is the decision everything else depends on. Pick the wrong one and even flawless calculations lead to invalid conclusions, and to difficult questions from your instructor, committee or reviewers.
This free guide walks you through the same questions I ask students in a consultation. Answer four questions about your research and your data, and you will get a recommended test, the assumptions to check and a non-parametric alternative if those assumptions fail.
How the Guide Works
The interactive guide asks four questions in order:
- What is your research goal?
- Compare groups
- Examine a relationship or association
- Predict an outcome
- Test or validate a measurement model
- What type is your outcome (dependent) variable?
- Continuous (interval/ratio)
- Ordinal (ranked)
- Categorical: two categories
- Categorical: three or more categories
- Count
- How many groups or predictors are involved?
- One sample vs. a known value · Two groups · Three or more groups · One or more predictors
- Are the observations independent or related?
- Independent (different participants in each group)
- Related (the same participants measured more than once, or matched pairs)
Your result includes the recommended test, a plain-language explanation of why it fits, its key assumptions, the non-parametric or robust alternative and links to step-by-step help in your software.
Quick Reference: Common Tests at a Glance
| Research goal | Outcome | Groups / design | Parametric test | Non-parametric alternative |
|---|---|---|---|---|
| Compare to a known value | Continuous | One sample | One-sample t-test | Wilcoxon signed-rank |
| Compare two groups | Continuous | Independent | Independent-samples t-test (Welch) | Mann-Whitney U |
| Compare two conditions | Continuous | Related | Paired-samples t-test | Wilcoxon signed-rank |
| Compare 3+ groups | Continuous | Independent | One-way ANOVA | Kruskal-Wallis |
| Compare 3+ conditions | Continuous | Related | Repeated-measures ANOVA | Friedman |
| Two or more factors | Continuous | Factorial | Factorial / mixed ANOVA | Aligned rank transform |
| Association | Both continuous | — | Pearson correlation | Spearman / Kendall |
| Association | Both categorical | Independent | Chi-square test of independence | Fisher’s exact test |
| Association | Two categories | Related | McNemar’s test | — |
| Predict an outcome | Continuous | 1+ predictors | Multiple regression | Robust / quantile regression |
| Predict an outcome | Binary | 1+ predictors | Logistic regression | — |
| Predict an outcome | Count | 1+ predictors | Poisson / negative binomial regression | — |
| Test a theory with latent constructs | Multiple indicators | — | CFA / SEM | PLS-SEM (with caveats) |
Before You Decide: Three Common Pitfalls
- Choosing a test from the output, not the question. Start from your research question and hypotheses, then match the test to them. See Hypothesis Testing.
- Treating Likert items as automatically continuous. Single Likert items are ordinal; multi-item scale scores are often treated as continuous, but that decision should be justified.
- Ignoring dependence. Repeated measurements or clustered data (students within classes) violate independence and need paired, repeated-measures or multilevel methods.
Guide Limitations
This guide covers the most common designs taught in undergraduate and graduate courses. It does not replace a methodological review of complex designs, such as multilevel, longitudinal, survey-weighted or quasi-experimental studies, where the right choice depends on details a flowchart cannot capture.
Get Your Test Choice Confirmed by a Professor
If your design does not fit neatly into the table, or you need to justify your choice to a committee, book a free 15-minute methods check. Describe your research question, your variables and your sample, and I will confirm the right analysis, usually within 24 hours.
Prefer email or chat? Write to [email protected] or message me on WhatsApp at +1 (425) 999-3993.