Choosing the right statistical test is the step most people get stuck on, and picking the wrong one is a mistake no amount of careful formatting can fix. The good news: the choice comes down to a few clear questions about your data. This guide walks you through them, then points you to a step-by-step APA 7 reporting guide for whichever test you land on.
Start with three questions
Almost every test choice is settled by three questions:
- What are you trying to do? Compare groups, or measure a relationship between variables?
- How many groups or conditions, and are they related or independent? Related means the same participants measured more than once (before and after); independent means different participants in each group.
- What kind of data is your outcome? Continuous and roughly normal, continuous but skewed or ordinal (ranks, Likert-type ratings), or categorical (counts in categories)?
Hold those three answers in mind and the table below points to your test.
Comparing groups: the decision table
| Your situation | Outcome is continuous and roughly normal | Outcome is ordinal or non-normal |
|---|---|---|
| One group vs a fixed value | One-sample t-test | Wilcoxon signed-rank against a value |
| Two independent groups | Independent t-test | Mann-Whitney U test |
| Two related measurements | Paired t-test | Wilcoxon signed-rank test |
| Three or more independent groups | One-way ANOVA | Kruskal-Wallis test |
| Three or more related conditions | Repeated-measures ANOVA | Friedman test |
| Two categorical factors on one outcome | Two-way ANOVA | (model the ranks or use a generalized model) |
Measuring relationships: the decision table
| Your situation | Use this test |
|---|---|
| Strength of a linear relationship between two continuous variables | Pearson correlation |
| Relationship when data are ordinal or the pattern is monotonic but not linear | Spearman correlation |
| Predict a continuous outcome from one predictor | Simple linear regression |
| Predict a continuous outcome from several predictors | Multiple regression |
| Association between two categorical variables | Chi-square test of independence |
Parametric or non-parametric?
The right-hand column of the comparison table lists non-parametric tests. Reach for them when:
- Your outcome is ordinal (ranks, or Likert-type ratings you do not want to treat as interval), or
- Your outcome is continuous but badly non-normal, especially in a small sample, or
- There are extreme outliers that would distort a mean.
When the assumptions of the parametric test hold, it is more powerful, so prefer it. Each reporting guide above lists the specific assumptions to check and names the non-parametric fallback.
A worked decision
Suppose you measured a depression score (continuous) for the same 30 people before and after therapy, and the difference scores look roughly normal.
- What are you doing? Comparing groups (not measuring a relationship).
- How many, related or independent? Two measurements, related (same people).
- Data type? Continuous and roughly normal.
That leads to a paired t-test. If the difference scores were badly skewed, you would use the Wilcoxon signed-rank test instead.
Report an effect size, whatever you choose
Every modern journal, and APA 7, expects an effect size alongside the test, not just a p value. Each guide above shows the effect size that belongs with its test (Cohen's d, eta squared, Cramer's V, r, and so on) and how to format it.
Let KyroStat choose and run it for you
If you are still unsure, that is exactly the problem KyroStat was built for. Upload your spreadsheet, describe your question in plain language, and it recommends the appropriate test for your data and design, runs it on real Python or R, checks the assumptions, and hands you the APA 7 write-up, the plot, and the code. You get the right test and a defensible result without second-guessing the choice.
Frequently asked questions
How do I know if my data are normal enough for a parametric test? Look at a histogram or a Q-Q plot of the variable (or the difference scores for a paired design), and consider a formal test such as Shapiro-Wilk. In larger samples the tests tolerate mild non-normality.
What if I have more than one outcome variable? That moves you toward multivariate methods (such as MANOVA) or running and correcting separate tests. Start by choosing the right test for each outcome, then plan for multiple comparisons.
Is a Likert item continuous or ordinal? A single Likert item is ordinal. A scale that sums or averages many items is often treated as continuous. When in doubt, the non-parametric option is the safer choice.
Which test gives an effect size automatically? All of them have a matching effect size; the reporting guides above show which one and how to format it.