Stuck choosing between t-test, ANOVA, chi-square, or correlation?
This statistical test selector helps you narrow your choice based on what you are trying to do, what kind of data you have, and whether your groups are independent or related. It is built for students who want a fast, sensible starting point without guessing. Use it before you open SPSS, Excel, Jamovi, R, or your class notes.
What this means
This recommendation is based on your goal, variable type, number of groups, and whether the observations are independent or related. Before finalizing your test, check the assumptions required by your teacher, adviser, or software output.
You can use this as a starting sentence in your methodology: “The [test name] was used because the study aimed to [compare/test/predict] using [type of data] across [groups/variables].”
Show logic / decision path
Common mistakes
- Choosing a test based only on the title of the study
- Confusing categorical variables with numerical variables
- Using an independent-groups test for paired or repeated data
- Using Pearson correlation for ordinal or clearly non-normal data without checking assumptions
- Using chi-square when expected counts are too small
- Reporting a test without checking assumptions first
Need to check one of the most common assumptions next? Use our Normality Test Calculator after you get your recommended test.
What this calculator does
This tool helps you choose a statistical test by asking the same core questions teachers and textbooks use: What is your goal, what type of variable is your outcome, how many groups do you have, and are those groups independent or related? Those are the main decision points behind common beginner and early-college tests.
In practice, the selector points you toward tests such as the independent samples t-test, paired t-test, one-way ANOVA, chi-square test of independence, Pearson correlation, Spearman rank correlation, and simple regression. These choices depend less on the title of your research and more on the structure of your variables and study design.
It is a decision-support tool, not a substitute for your instructor’s rules or your software output. A good test choice still depends on assumptions such as independence, variable type, and, for many parametric tests, whether the data are roughly normal.
When to use it (and when not to)
Use this tool when you already know your research question but are unsure which test matches it. It is most helpful for homework, practical research classes, undergraduate projects, and thesis drafts where you need a strong first answer.
Use it when your question sounds like one of these:
- Do two groups differ?
- Are these two variables related?
- Are these categorical variables associated?
- Which test fits pretest-posttest data?
- Should I use a parametric or nonparametric test?
Do not use it as your only basis when your design is advanced, such as ANCOVA, MANOVA, mixed models, survival analysis, or multilevel analysis. Those cases usually need more than a simple selector. Also do not use it to ignore your course conventions. Some teachers simplify Likert data as interval, while others require ordinal methods. That can change the recommended test.
How it works
The selector follows a decision path.
First, it asks your goal. Are you describing data, comparing groups, testing a relationship, testing association, or predicting an outcome? That choice immediately narrows the family of tests. For example, comparing group means points toward t-tests or ANOVA, while testing association between categorical variables points toward chi-square or Fisher’s exact test.
Second, it asks about your variable type. A numerical variable behaves differently from a categorical variable. That is why a numerical score can lead to a t-test or correlation, while pass/fail or yes/no outcomes often lead to proportion tests, chi-square, or logistic regression.
Third, it asks about group structure. Two independent groups are different from paired scores from the same students. Three or more groups are different from two groups. This is where the path splits into tests such as independent samples t-test, paired t-test, one-way ANOVA, repeated measures ANOVA, Mann–Whitney U, or Friedman test.
Finally, it asks about assumptions. If a numerical outcome is roughly normal and the design fits, a parametric test may be appropriate. If assumptions fail, the better answer may be a nonparametric alternative such as Mann–Whitney U, Wilcoxon signed-rank, Kruskal–Wallis, or Spearman correlation.
Step-by-step example
Imagine your research question is:
“Do students taught with Method A, Method B, and Method C have different mean quiz scores?”
Here is the decision process:
- Your goal is to compare groups.
- Your outcome variable is quiz score, which is numerical.
- You have three groups.
- The groups are independent because each student belongs to only one method.
- If the scores are reasonably normal and the variances are not wildly different, the best test is one-way ANOVA.
A simple write-up sentence would be:
“A one-way ANOVA was used to compare mean quiz scores across the three teaching methods.”
Now change only one detail: suppose the same students took a pretest, midtest, and posttest. You still have three measurements, but now they are related because they come from the same people. In that case, the better test becomes repeated measures ANOVA, or Friedman test if assumptions are not met.
That is the key idea behind this tool: small design changes can change the correct test.
Common mistakes
One common mistake is choosing a test from the research title alone. A title like “relationship between study habits and performance” does not automatically mean Pearson correlation. You still need to check whether both variables are numerical, whether ranking is involved, and whether the relationship is treated as linear or monotonic.
Another common mistake is confusing independent groups with paired data. Two classes of different students are independent. Pretest and posttest scores from the same students are paired. Mixing these up leads to the wrong test.
Students also often ignore assumptions. For example, a t-test is not just “for two groups.” It also assumes a numerical outcome and a design that fits the test. Likewise, chi-square is not ideal when expected counts are too small, which is why Fisher’s exact test exists.
A final mistake is forgetting that teachers may use different classroom rules. Some courses accept parametric tests for Likert totals. Others prefer nonparametric methods. If your answer differs from your teacher’s, the difference is often in the assumptions or course convention, not because one of you is automatically wrong.
FAQs: Statistical Test Selector
What does this result mean?
It means the tool found the test that best matches your goal, variable type, number of groups, and study design. It is your best starting point, not a blind final answer. You still need to check assumptions.
Which test should I use for two independent groups?
If your outcome is numerical and assumptions are reasonably met, use an independent samples t-test. If the data are ordinal or clearly non-normal, the alternative is often Mann–Whitney U.
Which test should I use for pretest and posttest scores?
Usually paired t-test if the differences are treated as numerical and assumptions are acceptable. If not, Wilcoxon signed-rank test is a common alternative.
When should I use chi-square instead of correlation?
Use chi-square when both variables are categorical, such as strand and pass/fail status. Use correlation when the variables are numerical or rank-based and you want to measure how they move together.
Why is my answer different from my teacher’s?
Your teacher may be using a different rule for assumptions, Likert data, small samples, or class level. Some courses also simplify the choices to match the topics already covered. Check the variable type, independence, and assumptions first.
Can this tool choose between Pearson and Spearman?
Yes. In a beginner-friendly way, it does that by checking whether the variables are numerical or rank-based and whether parametric assumptions are reasonably met. Pearson usually fits numerical data with stronger assumptions; Spearman fits ranked or non-normal data more often.
Is this tool only for hypothesis tests?
Mostly, but not only. It also helps with descriptive choices and can point toward reliability measures such as Cronbach’s alpha or agreement measures such as Cohen’s kappa when the study goal is different from a standard hypothesis test.
Do I still need to check normality?
Yes, for many numerical parametric tests you should still check distribution shape, outliers, and sometimes equal variance. That is why the selector gives a primary answer and a common fallback when assumptions fail.
Related BrainMattersLearning statistics tools
After you choose a test, the next step is usually checking assumptions, summarizing variables, or standardizing scores. On BrainMattersLearning, the most useful companion pages are the Statistics Calculators and Study Guides pillar, the Research Tools hub, and focused tools such as the Normality Test Calculator, Descriptive Statistics Calculator, Z-Score Calculator, and t Distribution Calculator. These pages already match the same student-friendly, answer-first style.
Related tools
- Statistics Calculators and Study Guides
- Normality Test Calculator
- Descriptive Statistics Calculator
- Z-Score Calculator
- t Distribution Calculator
References
OpenStax. (2023). Introductory statistics 2e. OpenStax. https://openstax.org/details/books/introductory-statistics-2e
Penn State Eberly College of Science. (n.d.). STAT 200: Hypothesis testing, Part 2. STAT Online. https://online.stat.psu.edu/stat200/lesson/6
University of California, Los Angeles, Institute for Digital Research and Education. (n.d.). What statistical analysis should I use? UCLA IDRE Statistical Consulting. https://stats.oarc.ucla.edu/spss/whatstat/what-statistical-analysis-should-i-usestatistical-analyses-using-spss/