StatEvery statistics student hits the same wall. You've learned the z-test, the t-test, the paired t-test, the chi-square test, and ANOVA — each one individually makes sense — and then a word problem lands in front of you and you have no idea which tool to pick up. The good news: test selection follows from just three questions about your situation. Learn to ask them in order and the choice usually makes itself.
The three questions that pick your test
- What kind of data do you have? Numerical measurements (heights, times, scores) or categories (yes/no, red/blue/green, pass/fail)?
- How many groups are you comparing? One group against a known value, two groups against each other, or three or more?
- Are the observations independent or paired? Two separate sets of people, or the same people measured twice (before/after)?
Almost every intro-stats test is an answer to a specific combination of these three questions.
Comparing means: z-tests and t-tests
When your data are numerical and your question is about an average, you're in mean-comparison territory.
One-sample z-test
Use it when you compare one sample mean to a known value and you know the population standard deviation. That second condition is rare in real life — you usually meet the z-test in textbooks, or with proportions (where the standard deviation follows from the hypothesized proportion itself).
One-sample t-test
Same question — does this group's mean differ from a claimed value? — but you estimate the standard deviation from the sample, which is the realistic case. The t-distribution's slightly heavier tails account for that extra uncertainty. With large samples, t and z give nearly identical answers.
Two-sample t-test
Two independent groups, one numerical outcome: do students taught with method A score differently from students taught with method B? The key word is independent — different people in each group.
Paired t-test
Same subjects measured twice: blood pressure before and after a medication, quiz scores before and after a study technique. Pairing removes person-to-person variation, which usually makes the test far more sensitive. Choosing "two-sample" when the data are actually paired is one of the most common test-selection errors — always ask whether each observation in one group has a natural partner in the other.
Comparing counts and categories: chi-square tests
When your data are categories rather than measurements, means don't exist — you compare counts instead. That's chi-square territory.
- Goodness of fit: does one categorical variable match a claimed distribution? (Is this die fair? Do customer complaints spread evenly across weekdays?)
- Independence / association: are two categorical variables related? (Does preferred study method differ by grade level?)
The logic is always the same: compare the counts you observed to the counts you'd expect if the null hypothesis were true, and measure the total mismatch.
Comparing three or more means: ANOVA
Three teaching methods, four fertilizers, five app designs — when you have a numerical outcome and more than two independent groups, use analysis of variance (ANOVA). It asks one question: is at least one group mean different from the others?
Why not just run t-tests on every pair? Because each test carries a false-positive risk, and running many tests multiplies your chances of a fluke "significant" result. ANOVA controls that risk with a single overall test.
A cheat sheet
| Your situation | Test |
|---|---|
| One numerical sample vs. known value, σ known | One-sample z-test |
| One numerical sample vs. claimed value, σ unknown | One-sample t-test |
| Two independent groups, numerical outcome | Two-sample t-test |
| Same subjects measured twice | Paired t-test |
| Three or more independent groups, numerical outcome | ANOVA |
| One categorical variable vs. claimed distribution | Chi-square goodness of fit |
| Two categorical variables, testing association | Chi-square test of independence |
| Relationship between two numerical variables | Correlation / linear regression |
Don't skip the conditions
Picking the right test is half the job; verifying it's legitimate to use is the other half. The recurring requirements are random sampling or random assignment, independence of observations, adequate sample sizes (or roughly normal populations for small samples), and — for chi-square — large enough expected counts in each cell. On the AP exam, checking conditions earns explicit rubric points; see our AP Statistics study guide for how graders score inference questions.
Remember what the result means. A small p-value says your data would be surprising if the null hypothesis were true — it doesn't measure how large or how important the effect is. Understanding this (and the related ideas of Type I and Type II errors) matters more than memorizing any formula. The normal-distribution machinery behind most of these tests is covered in our guide to understanding the normal distribution.
How the choice becomes automatic
Nobody internalizes test selection by staring at a flowchart. It becomes automatic through mixed practice: encountering scenario after scenario where you must decide, getting it wrong, seeing why, and trying again. Interleaving different problem types — rather than drilling one test at a time — is exactly the kind of practice that research on learning shows builds durable judgment (more on that in our guide to spaced repetition).
How StatRise helps
StatRise includes calculators for z-tests, t-tests, paired t-tests, chi-square tests, and ANOVA among its 39 statistics and probability tools — and StatRise Pro adds step-by-step worked solutions so you can follow the reasoning, not just the result. The lesson modules cover confidence intervals and hypothesis testing in depth, mixed practice serves scenario questions across topics so test selection becomes reflex, and the Type I and Type II error visualizer lets you see the trade-offs behind every hypothesis test. It all runs on-device, offline, with no ads or account.