Nonparametric tests answer a question every SPSS user asks sooner or later: your data do not meet the assumptions of a t-test or ANOVA, so what do you run instead? This guide gives you the decision, not just a list of test names. Work through the questions below, find your situation, and you will land on the right test — plus the exact SPSS menu path to run it and how to report it.
Why “nonparametric” is not a lesser choice
Parametric tests such as the independent-samples t-test and one-way ANOVA compare means, and they assume your outcome is roughly normally distributed within each group, measured on an interval or ratio scale, and — for the classic versions — that group variances are similar. When those assumptions break badly, the p-value from a parametric test can mislead you.
Nonparametric tests take a different approach. Most of them work with ranks rather than raw scores: they order all your values from smallest to largest and compare the rank patterns between groups. That makes them far less sensitive to outliers and skew, and it means they work with ordinal data such as Likert-scale responses, where a “mean” is often not meaningful in the first place.
That flexibility comes with a trade-off. Rank-based tests usually have slightly less statistical power than their parametric equivalents when the parametric assumptions actually hold. The rule of thumb is simple: if your data meet the assumptions, use the parametric test. If they clearly do not, a nonparametric test is the honest choice — not a second-class one.
When do you need a nonparametric test?
Reach for a nonparametric test when one or more of the following is true:
- Your outcome is ordinal (for example, a 1-to-5 satisfaction rating, a rank, or a grade) rather than truly continuous.
- Your outcome is continuous but strongly skewed or heavy with outliers, and transforming it does not fix the problem.
- Your sample is small, so you cannot meaningfully check normality at all.
- Your data are counts or categories rather than scores.
Before switching, do two quick checks in SPSS. First, look at your data: Analyze > Descriptive Statistics > Explore, and inspect the histogram and the Shapiro-Wilk test. Second, ask whether a transformation (such as a log transformation for right-skewed data) or a robust parametric alternative would solve the problem. If neither helps, move on to the decision guide below.
The decision flowchart: five questions
Start at the top and follow your answers:
Question 1: What kind of outcome do you have?
- Categories or frequencies (for example, pass/fail counts) → chi-square test (see the note at the end of this section).
- Scores, ranks, or ratings → continue to Question 2.
Question 2: Are you testing a relationship or a difference?
- Relationship between two variables → Spearman’s rank correlation.
- Difference between groups or conditions → continue to Question 3.
Question 3: Are the groups independent or related?
- Independent: different people in each group → continue to Question 4a.
- Related: the same people measured twice, or matched pairs → continue to Question 4b.
Question 4a (independent groups): How many groups?
- Two groups → Mann-Whitney U test.
- Three or more groups → Kruskal-Wallis H test.
Question 4b (related groups): How many conditions?
- Two conditions → Wilcoxon signed-rank test.
- Three or more conditions → Friedman test.
Question 5: If Kruskal-Wallis or Friedman is significant, which groups differ?
- Run pairwise follow-up comparisons with a correction for multiple testing (Dunn’s test with a Bonferroni adjustment is the standard route in SPSS) rather than eyeballing the medians.
That is the whole flowchart. The rest of this guide explains each test, its SPSS steps, and its output.
Two independent groups: the Mann-Whitney U test
Use it when: you have two unrelated groups and an ordinal or non-normal outcome. It is the nonparametric equivalent of the independent-samples t-test. Example: comparing anxiety ratings between a treatment group and a control group.
How to run it in SPSS:
- Go to Analyze > Nonparametric Tests > Legacy Dialogs > 2 Independent Samples.
- Move your outcome variable into the Test Variable List and your grouping variable into the Grouping Variable box.
- Click Define Groups and enter the two group codes (for example, 0 and 1).
- Keep Mann-Whitney U ticked and click OK.
How to interpret the output: SPSS reports a U statistic, a Z value, and an asymptotic p-value. Report the median and interquartile range (IQR) for each group, not just the means — the test compares rank distributions, so medians describe what was actually compared. A significant result means the distributions differ; check which group has the higher median to state the direction.
Three or more independent groups: the Kruskal-Wallis H test
Use it when: you have three or more unrelated groups and an ordinal or non-normal outcome. It is the nonparametric equivalent of one-way ANOVA. Example: comparing satisfaction scores across three teaching methods.
How to run it in SPSS:
- Go to Analyze > Nonparametric Tests > Legacy Dialogs > K Independent Samples.
- Move your outcome into the Test Variable List and your grouping variable into the Grouping Variable box.
- Click Define Range and enter the lowest and highest group codes.
- Keep Kruskal-Wallis H ticked and click OK.
How to interpret the output: SPSS reports a chi-square value (labelled “Chi-Square” — this is the H statistic) with degrees of freedom equal to the number of groups minus one. A significant result tells you only that at least one group differs from another. Follow it with pairwise comparisons (in the newer Nonparametric Tests dialog, double-click the test output to open Model Viewer, then use the Pairwise Comparisons view, which applies a Bonferroni adjustment).
Two related groups: the Wilcoxon signed-rank test
Use it when: the same people are measured twice, or you have matched pairs, and the difference scores are not normally distributed. It is the nonparametric equivalent of the paired-samples t-test. Example: stress scores before and after a workshop for the same students.
How to run it in SPSS:
- Go to Analyze > Nonparametric Tests > Legacy Dialogs > 2 Related Samples.
- Move the two measurements into the Test Pairs list.
- Keep Wilcoxon ticked and click OK.
How to interpret the output: SPSS reports a Z value and a p-value, plus a ranks table showing how many cases increased, decreased, or tied. Report the median for each time point and the median difference if it is meaningful. Ties (no change) are informative — a large number of ties can shrink the effective sample size, so mention them if they are common.
Three or more related conditions: the Friedman test
Use it when: the same people are measured under three or more conditions, and the scores are ordinal or non-normal. It is the nonparametric equivalent of repeated-measures ANOVA. Example: pain ratings at baseline, week 2, and week 4 for the same patients.
How to run it in SPSS:
- Go to Analyze > Nonparametric Tests > Legacy Dialogs > K Related Samples.
- Move all the measurement variables into the Test Variables box.
- Keep Friedman ticked and click OK.
How to interpret the output: SPSS reports a chi-square statistic with degrees of freedom equal to the number of conditions minus one, plus mean ranks for each condition. As with Kruskal-Wallis, a significant result needs pairwise follow-up comparisons to say which conditions differ.
Testing a relationship: Spearman’s rank correlation
Use it when: you want the association between two ordinal variables, or two continuous variables where the relationship is monotonic (consistently increasing or decreasing) but not linear, or where outliers distort Pearson’s correlation. Example: the relationship between hours of revision and exam rank.
How to run it in SPSS:
- Go to Analyze > Correlate > Bivariate.
- Move both variables into the Variables box.
- Tick Spearman, untick Pearson, and click OK.
How to interpret the output: Spearman’s rho (rs) ranges from -1 to +1, interpreted like Pearson’s r: values near 0 mean little monotonic association, values near 0.30 are a moderate association, and values near 0.50 or above are strong (these are rough guides, not cut-offs). Always plot the data first — a correlation, parametric or not, can hide a non-monotonic pattern.
A note on categorical outcomes: the chi-square test
If your outcome is a count in categories (for example, how many students in each group passed or failed), you are not choosing between the tests above at all — you need a chi-square test of independence (Analyze > Descriptive Statistics > Crosstabs, then tick Chi-square under Statistics). It is often listed alongside nonparametric tests because it also makes no normality assumption, but it answers a different question: whether two categorical variables are associated, not whether scores differ.
How to report effect sizes for nonparametric tests
A p-value alone is not enough. For Mann-Whitney and Wilcoxon, compute r from the Z value: r = Z divided by the square root of the total sample size (r = Z / sqrt(N)). Rough guides: 0.10 is small, 0.30 is medium, and 0.50 is large. For Kruskal-Wallis and Friedman, report the test statistic, degrees of freedom, p-value, and group medians with IQRs, and report pairwise effect sizes or adjusted p-values for any follow-ups. SPSS does not print r automatically for these tests, so computing it by hand from Z and N is a normal, expected step — show the calculation in your methods note if your supervisor asks.
Reporting in APA style (7th ed.)
APA style reports the test statistic, sample size information, and exact p-value, with medians for the groups. Examples you can adapt:
- “Anxiety ratings were higher in the treatment group (Mdn = 6, IQR = 4–7) than in the control group (Mdn = 4, IQR = 3–5), U = 312.00, z = -2.41, p = .016, r = .29.”
- “Satisfaction differed across teaching methods, H(2) = 9.84, p = .007. Pairwise comparisons with a Bonferroni correction showed that Method A (Mdn = 8) scored higher than Method C (Mdn = 5), p = .012.”
- “Stress scores decreased from pre-workshop (Mdn = 22) to post-workshop (Mdn = 17), z = -3.05, p = .002, r = .41.”
- “Revision hours and exam rank were positively associated, rs(48) = .52, p < .001.”
Two formatting notes: APA uses “p = .016” (no leading zero before the decimal for p-values), and italicizes the test symbols (U, H, z, rs) in a manuscript — in SPSS output they appear in plain text, so apply italics when you type them into Word.
Common mistakes to avoid
- Running a nonparametric test just because your sample is small. Small samples make normality hard to check, but they do not automatically require a rank test — think about your measurement scale first.
- Reporting means and standard deviations as the headline result for a rank test. Report medians and IQRs alongside them.
- Stopping at a significant Kruskal-Wallis or Friedman result without follow-up comparisons. “Groups differ” is not a complete finding.
- Running many pairwise tests without a correction. Each extra test inflates your chance of a false positive, which is exactly what the Bonferroni adjustment controls.
- Treating a non-significant result as proof of no difference. It means you did not find evidence of a difference in this sample — especially with small samples, that is not the same thing.
Key Takeaways
- If your data are ordinal, skewed, or full of outliers, a nonparametric test is usually the right call — and it is not a weaker fallback, just a different assumption set.
- Two independent groups → Mann-Whitney U. Three or more independent groups → Kruskal-Wallis H. Two related measurements → Wilcoxon signed-rank. Three or more related measurements → Friedman. A monotonic relationship → Spearman’s rho. Categories and counts → chi-square.
- A significant Kruskal-Wallis or Friedman result always needs pairwise follow-ups with a correction.
- Report medians and IQRs, the test statistic, exact p-values, and an effect size (r = Z / sqrt(N) for Mann-Whitney and Wilcoxon).
- When in doubt, check your measurement scale first, then independence, then the number of groups — those three answers pick the test.
Frequently asked questions
Can I use a nonparametric test if my data are normally distributed?
Yes, but you usually should not. If the parametric assumptions hold, the t-test or ANOVA has more power, so you are more likely to detect a real effect. Use the nonparametric test when the assumptions clearly fail, not as a default.
Do nonparametric tests need a minimum sample size?
There is no strict minimum, but very small samples (for example, fewer than 5 per group) have almost no power to reach significance with a rank test, because there are only so many possible rank arrangements. If your sample is that small, report descriptive statistics carefully and be cautious about any test result.
What if my data have many tied values?
Ties reduce the information in the ranks and SPSS applies a tie correction automatically. With heavily tied data (for example, a 3-point scale), interpret results cautiously and consider whether a chi-square analysis of the response categories fits your question better.
Should I transform my data or switch to a nonparametric test?
Try a sensible transformation first if your data are continuous and skewed — a log transformation often restores normality. If transformation fails, distorts interpretation, or your data are ordinal to begin with, switch to the nonparametric test rather than forcing the data to fit.
Can I report a nonparametric result alongside a parametric one?
You can run both as a sensitivity check, but report one as your primary analysis — the one whose assumptions your data meet — and mention the other only if it changes the conclusion. Reporting whichever p-value looks better is selective reporting, not robustness.
Need this done on your data? Get a free quote → /services/