Mathwords logoMathwords

Nonparametric Statistics — Definition, Formula & Example

Nonparametric statistics is a family of methods that test hypotheses without assuming your data come from a specific distribution, such as the normal distribution. These methods usually work with signs, ranks, or counts instead of the raw values.

A statistical procedure is nonparametric if its validity does not depend on the population belonging to a family of distributions indexed by finitely many parameters (such as normal with mean μ\mu and variance σ2\sigma^2). Many nonparametric tests are distribution-free. This means the null distribution of the test statistic is the same for every continuous population, so exact p-values can be found by counting. Named examples include the sign test, Wilcoxon signed-rank test, Mann-Whitney U (Wilcoxon rank-sum) test, Kruskal-Wallis test, and Spearman rank correlation.

Key Formula

P(S≥s)=∑k=sn(nk)(12)nP(S \ge s) = \sum_{k=s}^{n} \binom{n}{k}\left(\tfrac{1}{2}\right)^{n}
Where:
  • SS = Number of positive differences, the sign test statistic
  • ss = Observed number of positive differences
  • nn = Number of nonzero paired differences
  • kk = Summation index running from s to n

Worked Example

Problem: Ten drivers measure their gas mileage before and after using a fuel additive. The changes in miles per gallon (after minus before) are: +2, +1, +3, −1, +4, +2, +1, +15, +2, +3. The +15 is an extreme outlier, so a t-test that relies on normality is risky. Use a one-sided sign test at the 0.05 level to decide whether the median change is greater than 0.
Step 1: Replace each difference with its sign and ignore its size. The outlier +15 now counts the same as +1. There are 9 positive differences and 1 negative difference, so n=10n = 10 and s=9s = 9.
H0:median change=0Ha:median change>0H_0: \text{median change} = 0 \qquad H_a: \text{median change} > 0
Step 2: If H0H_0 is true, each change is equally likely to be positive or negative, so SS follows a binomial distribution with n=10n = 10 and p=0.5p = 0.5. No normality assumption is needed. Find the probability of 9 or more positives.
P(S≥9)=(109)+(1010)210=10+11024=111024≈0.0107P(S \ge 9) = \frac{\binom{10}{9} + \binom{10}{10}}{2^{10}} = \frac{10 + 1}{1024} = \frac{11}{1024} \approx 0.0107
Step 3: Compare the p-value with the significance level. Since 0.0107<0.050.0107 < 0.05, reject H0H_0.
0.0107<0.050.0107 < 0.05
Answer: The p-value is 111024≈0.0107\frac{11}{1024} \approx 0.0107, which is below 0.05. You reject the null hypothesis. The data give convincing evidence that the median change in gas mileage after using the additive is greater than 0, and the outlier did not distort the conclusion.

Why It Matters

In psychology, biology, and medical research, samples are often small, skewed, or measured on ordinal scales such as 1-to-5 survey ratings. These features break the assumptions behind t-tests and ANOVA, and nonparametric tests such as Mann-Whitney U and Kruskal-Wallis are the standard replacements. Most college statistics courses and statistical software packages pair each parametric test with a nonparametric counterpart.

Common Mistakes

Mistake: Believing nonparametric tests make no assumptions at all.
Correction: They drop the assumption of a specific distribution shape, but they still require random sampling and independent observations. Some tests add their own conditions. For example, the Wilcoxon signed-rank test assumes the differences are symmetric about the median. Check these conditions before you use the test.

Related Terms