Free tool

Which statistical test should I use?

Answer a few quick questions about your study and this tool tells you which test to run, where to find it in SPSS and how to report it in APA style. It covers 33 tests and procedures, from t-tests and ANOVA to chi-square, regression, moderation and mediation.

How to choose a statistical test

Three questions decide most choices. What is your outcome? Scale scores (such as exam percentages) usually call for t-tests, ANOVA or regression; ordinal ratings or clearly non-normal scores call for non-parametric tests; categories call for chi-square tests. How many groups or conditions? Two means a t-test or its non-parametric partner; three or more means ANOVA or its partner. Same people or different people? Measuring the same people more than once needs a paired or repeated-measures test.

Then check the assumptions for the test you picked: normality (Explore), equal variances (Levene’s test) and, for chi-square, expected counts.

Every test at a glance

Every route through the decision tree, with the SPSS menu path for each test.

Compare groups or conditions

Your situationTestIn SPSS
Scores that are roughly normal → One group against a known valueOne-sample t-test
Scores that are roughly normal → Two → Different peopleIndependent-samples t-test
Scores that are roughly normal → Two → The same peoplePaired-samples t-test
Scores that are roughly normal → Three or more (one grouping variable) → Different people → Yes, the variances are similarOne-way ANOVA
Scores that are roughly normal → Three or more (one grouping variable) → Different people → No, or I’m not sureWelch’s ANOVA with Games-Howell
Scores that are roughly normal → Three or more (one grouping variable) → The same peopleRepeated-measures ANOVA
Scores that are roughly normal → Two or more grouping variables, or a covariate → Two grouping variables, different people in every cellTwo-way ANOVA
Scores that are roughly normal → Two or more grouping variables, or a covariate → One repeated factor and one between-groups factorMixed ANOVA
Scores that are roughly normal → Two or more grouping variables, or a covariate → Groups compared while controlling for a covariateANCOVA
Ratings, ranks or non-normal scores → Two groups, different peopleMann-Whitney U test
Ratings, ranks or non-normal scores → Two conditions, the same peopleWilcoxon signed-rank test
Ratings, ranks or non-normal scores → Three or more groups, different peopleKruskal-Wallis H test
Ratings, ranks or non-normal scores → Three or more conditions, the same peopleFriedman test
Categories → One categorical variable against expected proportionsChi-square goodness of fit
Categories → Two categorical variables, different people → NoChi-square test of independence
Categories → Two categorical variables, different people → Yes, and it’s a 2 × 2 tableFisher’s exact test
Categories → The same people measured twice on a yes/no outcomeMcNemar’s test

See how variables are related

Your situationTestIn SPSS
Two Scale variables, roughly normal and linearPearson correlation
Ordinal, skewed or with outliers → NoSpearman’s rho
Ordinal, skewed or with outliers → YesKendall’s tau-b
Two categorical variablesChi-square test of independence
Two Scale variables, controlling for a thirdPartial correlation

Predict an outcome

Your situationTestIn SPSS
A Scale score → One predictorSimple linear regression
A Scale score → Several predictors, all togetherMultiple regression
A Scale score → In steps, e.g. control variables firstHierarchical regression
Two categoriesBinary logistic regression

Test a third variable’s role

Your situationTestIn SPSS
Does the effect depend on a third variable?Moderation analysis (PROCESS Model 1)
Does the effect happen through a third variable?Mediation analysis (PROCESS Model 4)

Check a questionnaire scale

Your situationTestIn SPSS
Do the items hang together?Cronbach’s alpha
Which items group together?Exploratory factor analysis

Describe or check my data

Your situationTestIn SPSS
Counts and percentages for categoriesFrequencies
Means and standard deviationsDescriptives
Whether data are normalExplore (normality checks)
A chartChart Builder

Questions people ask

How do I choose the right statistical test?

Answer three questions: what type of outcome you have (Scale scores, ordinal ratings or categories), how many groups or conditions you are comparing, and whether the same people or different people are in each group. Then check the assumptions: if Scale scores are clearly not normal, use the non-parametric alternative.

What is the difference between a t-test and an ANOVA?

A t-test compares the means of two groups or conditions. A one-way ANOVA compares three or more. Running several t-tests instead of an ANOVA inflates the chance of a false positive.

When should I use a non-parametric test?

Use one when your outcome is ordinal (such as a rating scale) or when Scale scores are clearly not normal, particularly in small samples. Mann-Whitney U, Wilcoxon, Kruskal-Wallis and Friedman replace the independent t-test, paired t-test, one-way ANOVA and repeated-measures ANOVA respectively.

Should I use chi-square or Fisher’s exact test?

Use chi-square when no more than 20% of cells have expected counts below 5. For a 2 × 2 table with small expected counts, report Fisher’s exact test, which SPSS prints automatically in Crosstabs.

What is the difference between correlation and regression?

Correlation measures how strongly two variables move together. Regression goes further: it gives an equation that predicts one variable from one or more others and shows how much variance they explain.

What is the difference between moderation and mediation?

Moderation asks whether a third variable changes the strength of an effect (when or for whom). Mediation asks whether an effect works through a third variable (how or why). Both are usually run in SPSS with the free PROCESS macro.

Know your test? Learn to run it.

Start learning free