Wilcoxon Signed-Rank Test Explained

What Is the Wilcoxon Signed-Rank Test?

Frank Wilcoxon introduced the test in a landmark 1945 paper, Individual Comparisons by Ranking Methods, published in Biometrics Bulletin. His goal was pragmatic: many real datasets in biology and agriculture were too small or too skewed to satisfy the normality assumption of the paired t-test. By replacing raw values with their ranks, Wilcoxon created a procedure that depends on the relative ordering of observations, not on their exact scale. This makes the test robust to outliers and applicable to ordinal data.

The test applies in two closely related situations. The most common is the two-sample paired case: you measure the same variable on the same subjects before and after an intervention. The second is the one-sample case: you have a single group of measurements and want to test whether the population median equals some specific value. Both reduce to the same rank-based calculation.

When to Use the Wilcoxon Signed-Rank Test

The most practical question researchers face is whether to run a Wilcoxon Signed-Rank Test or a paired t-test. Both handle paired data, but they suit different situations.

The test is widely used in medicine (before-and-after pain scores, blood pressure, lab values), psychology (pre-test and post-test attitude scales, Likert responses), education (exam scores before and after tutoring), and any setting where the same measurement is repeated on matched subjects. Because it uses rank order rather than raw arithmetic, a single extreme outlier cannot distort the result the way it could in a t-test.

Assumptions of the Wilcoxon Signed-Rank Test

The test has fewer assumptions than the paired t-test — notably, it does not require normally distributed data. There are three conditions that must hold, and one that is sometimes confused with a requirement but is not.

The Wilcoxon Signed-Rank Test Formula

The test statistic W is calculated by ranking the absolute differences between paired observations and then summing the ranks belonging to either the positive differences or the negative differences.

For small samples (n' ≤ 25), compare W directly to the critical value in the Wilcoxon Signed-Rank critical values table. Reject H₀ when W is less than or equal to the tabled critical value (the logic is reversed from most tests — a small W indicates the positive and negative ranks are unequal, suggesting a treatment effect).

Worked Example — Manual Calculation

The following example uses real-style clinical data to demonstrate every calculation step by hand. This is the type of example commonly used in biostatistics courses and is directly applicable to research in medicine, psychology, and education.

Wilcoxon Signed-Rank Test Calculator

Enter your paired data below to compute the W statistic, signed ranks, p-value (normal approximation for n' > 10), and a ready-to-paste APA result.

Effect Size for the Wilcoxon Signed-Rank Test

A statistically significant result tells you the effect is unlikely due to chance — it does not tell you whether the effect is large enough to matter. Effect size quantifies the practical magnitude of the difference. The standard effect size measure for the Wilcoxon Signed-Rank Test is the rank-biserial correlation r.

These thresholds follow Cohen's (1988) benchmarks for r as adapted for nonparametric tests. In our blood pressure example, if the Z approximation yields Z = −2.67 with N = 18 (9 pairs × 2), then r = 2.67/√18 = 2.67/4.24 = 0.63, indicating a large effect. Some software packages instead report the matched-pairs rank-biserial correlation directly from W⁺ and W⁻: r = (W⁺ − W⁻) / [n'(n'+1)/2]. Both approaches are acceptable — state which formula you used in your methods section.

How to Report Results in APA Format

APA 7th edition does not have a specific Wilcoxon reporting template, but the following format is standard in published research and follows the logic of APA p-value reporting guidelines.

Running the Test in R, Python, SPSS, and Excel

R

The wilcox.test() function in base R handles both the paired and one-sample cases. Set paired = TRUE and exact = TRUE for small samples to get the exact p-value rather than the normal approximation.

# Example: blood pressure before and after intervention
before <- c(145, 152, 138, 160, 149, 155, 142, 147, 158, 136)
after  <- c(138, 144, 140, 148, 149, 141, 135, 143, 150, 132)

# Run paired Wilcoxon test (exact p-value for small n)
result <- wilcox.test(before, after,
                   paired = TRUE,
                   alternative = "two.sided",
                   exact = TRUE)

print(result)
# Output: V = 44, p-value = 0.007813
# Note: R reports V = W+ (sum of positive ranks). Here V = 44 means W+ = 44, W- = 1.

# Effect size: r = Z / sqrt(N)
library(rstatix)
wilcox_effsize(data.frame(before, after),
               before ~ after,
               paired = TRUE)  # returns r (effsize)

# Or manually:
z_approx <- qnorm(result$p.value / 2)
r_effect <- abs(z_approx) / sqrt(length(before) * 2)
cat("Effect size r =", round(r_effect, 3))

Python (SciPy)

The scipy.stats.wilcoxon() function computes the test for paired differences. Pass the differences directly or two arrays; the function accepts both formats.

from scipy import stats
import numpy as np

before = [145, 152, 138, 160, 149, 155, 142, 147, 158, 136]
after  = [138, 144, 140, 148, 149, 141, 135, 143, 150, 132]

# Run test — method='exact' for small n, 'approx' for large n
statistic, p_value = stats.wilcoxon(
    before, after,
    alternative='two-sided',
    method='exact'
)

print(f"W statistic: {statistic}")     # smaller of W+ / W-
print(f"p-value:     {p_value:.4f}")

# Effect size
differences = np.array(before) - np.array(after)
nonzero = differences[differences != 0]
n_prime = len(nonzero)
z_score = stats.norm.ppf(1 - p_value/2)
r_effect = z_score / np.sqrt(n_prime * 2)
print(f"Effect size r: {r_effect:.3f}")

SPSS

/* Menu path: Analyze → Nonparametric Tests → Legacy Dialogs →
   2 Related Samples → Select both variables → Choose Wilcoxon */

/* Or via syntax: */
NPAR TESTS
  /WILCOXON=before WITH after (PAIRED)
  /STATISTICS=DESCRIPTIVES
  /MISSING=ANALYSIS.

SPSS output includes the Negative Ranks, Positive Ranks, and Ties counts, the Z statistic, and the asymptotic significance (p-value). For exact significance with small samples, navigate to Exact and select "Exact" rather than "Asymptotic."

Wilcoxon Signed-Rank Test vs Other Tests

Choosing among nonparametric and parametric options for paired data is one of the most common decision points in applied statistics. The table below distills the key differences.

The RANKS Framework for Applying the Test

Here is an original five-step decision and execution framework that makes the Wilcoxon Signed-Rank Test procedure easy to remember and apply across any discipline.

Common Mistakes and How to Avoid Them

Entity and Formula Glossary

Frequently Asked Questions

Related Statistical Tests and Concepts

The Wilcoxon Signed-Rank Test sits within a broader family of rank-based and nonparametric methods. Understanding the connections helps you choose the right tool for each study design.

For a broader overview of which test fits which design, see the Statistical Test Selector and the guide to statistical test assumptions.