Count the groups. Check the data. Name the test.
Research items on the NCE usually come down to one decision: which statistical test fits the design. Answer a few questions about the study and see the test, then practice on scenarios until the choice is quick.
{{ question }}
{{ resultName }}
{{ resultText }}
{{ resultAlt }}
The rule of thumb
Count the groups and check the data. Two means: t-test. Three or more means: ANOVA. Categories and counts: chi-square. Two continuous variables, relationship: correlation. Predicting one from others: regression.
Whenever an item describes counts of people in categories, the answer is chi-square.
Practice: name the test
{{ practiceIntro }}
{{ pText }}
{{ pWhy }}
{{ pResult }}
{{ pNote }}
Scenarios were written for this site from the lesson's rules. None is from the actual exam.
Reading a research item, one question at a time
A statistics item on the NCE usually describes a study in two or three sentences and asks which test fits. You do not need to compute anything. You need to answer three questions about the design, in order, and the test follows from the answers. The decision helper above asks them for you; with practice you will ask them yourself in a few seconds.
- Is the question about a difference or a relationship? Comparing groups points to the t-test or ANOVA. Asking how two variables move together points to correlation. Predicting one variable from others points to regression.
- How many groups, and are they the same people? Two different groups take an independent-samples t-test. The same people measured twice, or matched pairs, take a dependent (paired) t-test. Three or more groups take an ANOVA, which tests them with one F-test instead of a string of t-tests that would inflate the Type I error rate.
- What kind of data is the outcome? Means need interval or ratio scores. Counts of people in categories are nominal data, and nominal data take chi-square. Ranked data take the nonparametric versions: Mann-Whitney U, Wilcoxon signed-rank, Kruskal-Wallis or Spearman's rho.
Worked examples
- A counselor compares depression scores for clients in three conditions: cognitive therapy, solution-focused therapy and a waitlist. Difference, three groups, interval scores: one-way ANOVA. If the F is significant, a post hoc test such as Tukey's HSD shows which groups differ.
- The same group members complete an anxiety scale before and after ten sessions. Difference, two measurements of the same people: dependent (paired) t-test.
- A career center records how many first-year and fourth-year students choose each of four majors. Counts in categories: chi-square test of independence.
- A researcher predicts graduate GPA from an admissions test score and undergraduate GPA. Prediction from two predictors: multiple regression.
Watch for the add-ons the exam likes: a second independent variable makes it a factorial ANOVA, a covariate such as pretest scores makes it an ANCOVA, and more than one dependent variable makes it a MANOVA. The full explanation of each test, with the terms the items use, is in the Which statistical test lesson. When the choice feels automatic, convert scores on the score converter or take a research set on the practice page.