Parametric vs Nonparametric Tests: Pick the Right One

Key takeaways

• Choose the test by your design first (2 groups? 3+ groups? paired?).
• Use parametric when assumptions (like normal shape) are reasonable.
• Report effect size (how big the difference is), not just p-value.

Parametric vs nonparametric tests is about this: Do your data and design meet the assumptions for mean-based tests? If yes, parametric tests (like t-tests and ANOVA) are often a good fit. If not, nonparametric tests (often rank-based) can be safer.

Quick Answer: Parametric vs Nonparametric Tests

Use a parametric test when your outcome is numeric and the data are roughly normal with no extreme outliers. Use a nonparametric test when your outcome is ordinal (ranked), very skewed, or has strong outliers.

Match the test to your design: t-test ↔ Mann–Whitney (2 independent groups) and ANOVA ↔ Kruskal–Wallis (3+ independent groups).

Key terms you must know (quick definitions)

• Parametric test (uses a model with parameters, often assumes a normal distribution).
• Nonparametric test (uses fewer shape assumptions; often uses ranks).
• Assumption (a condition a method expects; if it fails badly, results can mislead).
• Normal distribution (a “bell-shaped” pattern: most values near the center).
• Outlier (a value far from the rest; it can pull the mean).
• Independent groups (different people in each group; no pairing).
• Paired data (same people measured twice, or matched pairs).
• Ordinal data (ranked levels like 1–5; gaps are not guaranteed equal).
• p-value (how surprising your data would be if the null hypothesis were true).
• Effect size (how big the difference/relationship is).
• Null hypothesis (the “no difference” or “no relationship” claim you test).

👉 Try our Normal Distribution Calculator to quickly check probabilities and z-scores, so you can see if your data looks close to normal before choosing a test.

Comparison table (most common choices)

Fast memory tip:

• 2 groups → t-test or Mann–Whitney
• 3+ groups → ANOVA or Kruskal–Wallis

Assumptions checklist

Check these before choosing:

1. Independence (one person’s score does not affect another’s).
2. Scale (numeric vs ordinal).
3. Outliers (any extreme values?).
4. Normal (is the shape roughly bell-like?).
5. Equal spread (do groups have similar variability?).
6. Sample size (very small samples make shape issues bigger).

Why this matters: Many nonparametric tests use ranks, so they are less sensitive to outliers and non-normal shapes.

Step-by-step method

Step 1: Write your study question in one line

Example: “Do students in Group A score higher than Group B?”

Step 2: Label variables

• Outcome (what you measure).
• Grouping variable (how you split people).

Step 3: Identify the design

• 2 groups or 3+ groups?
• Independent or paired?

Step 4: Do a quick data check

Make a quick plot (dotplot/histogram) and look for:

• non-normal shape (heavy skew)
• big outliers
• very different spreads between groups

Step 5: Choose the test from the table

Use the matching row (2 groups vs 3+ groups, independent vs paired).

Step 6: Plan your reporting (don’t stop at p)

Report:

• test name + statistic + p-value
• effect size (recommended)
  APA’s quick guide supports clear reporting of statistics.

What to write on your paper (mini-checklist)

Copy this:

• Design: (2 groups / 3+ groups; independent / paired)
• Outcome type: (numeric / ordinal / counts)
• Assumptions: (normal? outliers? equal spread?)
• Test chosen: (t-test / Welch / Mann–Whitney / ANOVA / Kruskal–Wallis)
• Result: statistic + p-value
• Effect size: (Cohen’s d, η², or rank-based r)
• Conclusion: one sentence in plain English

Worked examples

Example 1 (very easy): normal data → t-test

Question: Do two classes differ in quiz scores?

• Class A: 78, 80, 82, 79, 81
• Class B: 72, 74, 73, 75, 71

Step 1 (design): 2 independent groups.
Step 2 (check): looks normal; no extreme outliers.
Pick: Independent t-test.

Why: t-test compares means when assumptions are reasonable.

Example 2 (homework level): outlier → Mann–Whitney

Question: Do two groups differ in daily study minutes?

• Group A: 10, 12, 15, 18, 120
• Group B: 20, 22, 25, 28, 30

Design: 2 independent groups.
Check: big outlier (120). Not normal.
Pick: Mann–Whitney U (rank-based).

Why: ranks reduce the “pull” of extreme values.

Example 3 (exam style): 3+ groups → ANOVA

Question: Do 3 study methods lead to different test scores?

• A: 78, 80, 79, 81
• B: 74, 75, 73, 76
• C: 82, 83, 84, 81

Design: 3 independent groups.
Check: scores look normal; spreads similar.
Pick: One-way ANOVA.

Example 4 (trick): Likert ratings → Kruskal–Wallis

Question: Do 3 sections differ in stress rating (1–5)?

• Sec A: 2, 2, 3, 3, 4
• Sec B: 4, 4, 5, 3, 4
• Sec C: 1, 2, 2, 1, 3

Key detail: 1–5 is ordinal.
Pick: Kruskal–Wallis.

Sale

Scientific Calculators for Students 991ES Office...

$32.98

Buy on Amazon

Construction Calculator with Writing Tablet...

$42.99

Buy on Amazon

Sale

Texas Instruments TI-84 Plus CE Color Graphing...

$111.99

Buy on Amazon

Practice set + answer key

Directions

For each: (1) design, (2) outcome type, (3) normal/outliers, (4) choose a test.

1. Two independent groups, numeric scores, normal → ?
2. Two independent groups, numeric, big outlier → ?
3. Before vs after scores (same students), normal differences → ?
4. Three independent groups, numeric, very skewed → ?
5. Trap: Two independent groups, numeric, spreads very different → ?
6. Three groups, Likert 1–5 → ?

Answer key

1. Independent t-test
2. Mann–Whitney U
3. Paired t-test
4. Kruskal–Wallis
5. Welch t-test (common fix for unequal spread)
6. Kruskal–Wallis

Two short solutions (only)

• #2: Outlier breaks “normal” → use Mann–Whitney U (rank-based).
• #6: 3 groups + ordinal outcome → use Kruskal–Wallis.

Common mistakes

If you get stuck…

• “I don’t know what test to use.”
  Start with: 2 groups or 3+ groups? independent or paired?
• “My data is not normal.”
  If it’s skewed or has outliers, switch to the rank-based option.
• “My software says Welch t-test.”
  That often means spreads differ. Welch is a common choice.
• “I mix up Mann–Whitney and Kruskal–Wallis.”
  Mann–Whitney = 2 groups. Kruskal–Wallis = 3+ groups.
• When to ask for help (no shame):
  If your teacher has a required rule (like always using a certain test), follow the class rule.

Next steps

• how to choose the right statistical test — Use this when you’re unsure which test matches your design.
• what to do if data is not normal — Use this when your plot looks skewed or has outliers.
• effect size vs p-value — Use this when writing your Results and Discussion.

How we know

• NIST’s handbook explains Mann–Whitney as a nonparametric alternative for 2-group comparisons and notes it uses ranks.
• NIST’s handbook describes Kruskal–Wallis for comparing 3+ groups with unknown distributions.
• The ASA warns that p-values are often misused and do not measure effect size.
• A widely cited tutorial explains why effect sizes matter alongside p-values.
• APA’s Numbers and Statistics guide supports clear, standard reporting.

Use this the right way

Use this guide to understand your data and make honest choices. For a research project, write your method clearly, report what you did, and cite your sources. Don’t copy answers without learning.

Evidence rule for claims

This guide avoids “always” rules. When it says “often,” it is based on widely used references (NIST, ASA, and peer-reviewed teaching resources). Where choices depend on class rules or software options, it says so.

References

American Psychological Association. (2024). Numbers and statistics guide (7th ed.). https://apastyle.apa.org/instructional-aids/numbers-statistics-guide.pdf

National Institute of Standards and Technology. (n.d.). Do two arbitrary processes have the same central tendency? (Mann–Whitney U test). NIST/SEMATECH e-Handbook of Statistical Methods. https://www.itl.nist.gov/div898/handbook/prc/section3/prc35.htm

National Institute of Standards and Technology. (n.d.). How can we compare several populations with unknown distributions? (Kruskal–Wallis test). NIST/SEMATECH e-Handbook of Statistical Methods. https://www.itl.nist.gov/div898/handbook/prc/section4/prc41.htm

Sullivan, G. M., & Feinn, R. (2012). Using effect size—or why the p value is not enough. Journal of Graduate Medical Education, 4(3), 279–282. https://pmc.ncbi.nlm.nih.gov/articles/PMC3444174/

Wasserstein, R. L., & Lazar, N. A. (2016). The ASA’s statement on p-values: Context, process, and purpose. The American Statistician, 70(2), 129–133. https://www.stat.berkeley.edu/~aldous/Real_World/ASA_statement.pdf

Study tools that can help

• If you get overwhelmed: a simple research planner (keeps your variables + test choice clear).
• If your steps get messy: a graph paper notebook (tables stay neat).
• If you compute by hand: a scientific calculator.
• If you get distracted: a Pomodoro timer or basic noise blockers (for focus blocks).