Derivation of the Test Statistic for the Wilcoxon Signed-Rank Test

Q: Why use ranks instead of the raw differences themselves?

The use of ranks makes the Wilcoxon Signed-Rank test non-parametric. It mitigates the disproportionate impact of outliers and allows the test to be valid even when the underlying distribution of the differences is not normal. It assesses the median difference rather than the mean difference, providing a robust inference for location shifts.

Q: How are tied absolute differences or zero differences handled?

For tied absolute differences, the standard procedure is to assign the average of the ranks they would have received if they were distinct (the 'midrank' method). For zero differences (where $ D_i = 0 $), these observations are typically removed from the analysis before ranking, and the sample size $ n $ is reduced accordingly, as they provide no information about the direction of the difference.

Q: What is the key assumption underpinning the derivation of $ E[W] $ and $ Var[W] $?

The critical assumption under the null hypothesis is that each non-zero difference $ D_i $ has an equal probability ($ 0.5 $) of being positive or negative, independent of its magnitude or the signs of other differences. This implies symmetry of the distribution of differences around zero, allowing us to treat the signs $ S_i $ as independent Bernoulli(0.5) random variables, whose ranks are fixed once the absolute differences are ordered.

Exploring the cinematic intuition of Derivation of the Test Statistic for the Wilcoxon Signed-Rank Test.

Visualizing...

Our institutional research engineers are currently mapping the formal proof for Derivation of the Test Statistic for the Wilcoxon Signed-Rank Test.

Apply for Institutional Early Access →

The Formal Theorem

Let

(D_1, \dots, D_n)

be a random sample of

n

differences, where

D_i = X_i - Y_i

for paired data or

D_i = X_i - M_0

for one-sample data against a hypothesized median

M_0

. Let

|D_i|

denote the absolute differences, and let

R_i

be the rank of

|D_i|

when all

n

non-zero absolute differences are ranked from smallest to largest (observations with

D_i = 0

are typically removed, reducing

n

). Define an indicator variable

S_i

such that

S_i = 1

D_i > 0

and

S_i = 0

D_i < 0

. The Wilcoxon Signed-Rank test statistic,

W

, is defined as the sum of the ranks corresponding to positive differences:

W = \sum_{i=1}^n R_i S_i

Under the null hypothesis

H_0

that the population median of the differences is zero (i.e., the distribution of differences is symmetric about zero), the expected value of

W

is:

E[W] = \frac{n(n+1)}{4}

And the variance of

W

is:

Var[W] = \frac{n(n+1)(2n+1)}{24}

For large

n

W

can be approximated by a normal distribution, leading to the standardized test statistic

Z

Z = \frac{W - E[W]}{\sqrt{Var[W]}}

Analytical Intuition.

Imagine a highly sensitive analytical scale, calibrated to detect even the most subtle imbalances. Our experimental data, consisting of

n

paired observations or

n

measurements compared to a hypothesized median

M_0

, generates a set of differences

D_i

. Each

D_i

is a 'deviation' from neutrality. First, we strip away the initial judgment, focusing solely on the magnitude of these deviations. We rank the absolute differences

|D_i|

, assigning higher ranks to larger deviations. This is akin to gauging the 'weight' of evidence each difference presents. Next, we reintroduce the 'direction' of the deviation – positive or negative. Under the null hypothesis, a perfect symmetry is expected; a positive deviation is just as likely as a negative one of the same magnitude. The test statistic

W

then sums up the ranks only for those differences that are positive. If our sample truly reflects the null hypothesis, we'd expect the sum of positive ranks

W

to be roughly half the total possible sum of ranks, as positive and negative deviations should balance out. A significantly higher or lower

W

signals a departure from this symmetry, suggesting a real effect. The derivation then quantifies this 'expected balance' and its natural variability under pure chance, providing the bedrock for our statistical decision.

CAUTION

Institutional Warning.

A common point of confusion arises from the definition of the test statistic $W$ . Some texts define it as $W^+$ (sum of positive ranks), while others use $W$ (sum of all signed ranks, where $D_i<0$ corresponds to $-R_i$ ). Ensure clarity on which definition is being used, as it affects the expected value.

Academic Inquiries.

Why use ranks instead of the raw differences themselves?

The use of ranks makes the Wilcoxon Signed-Rank test non-parametric. It mitigates the disproportionate impact of outliers and allows the test to be valid even when the underlying distribution of the differences is not normal. It assesses the median difference rather than the mean difference, providing a robust inference for location shifts.

How are tied absolute differences or zero differences handled?

For tied absolute differences, the standard procedure is to assign the average of the ranks they would have received if they were distinct (the 'midrank' method). For zero differences (where $D_i = 0$ ), these observations are typically removed from the analysis before ranking, and the sample size $n$ is reduced accordingly, as they provide no information about the direction of the difference.

What is the key assumption underpinning the derivation of $E[W]$ and $Var[W]$ ?

The critical assumption under the null hypothesis is that each non-zero difference $D_i$ has an equal probability ( $0.5$ ) of being positive or negative, independent of its magnitude or the signs of other differences. This implies symmetry of the distribution of differences around zero, allowing us to treat the signs $S_i$ as independent Bernoulli(0.5) random variables, whose ranks are fixed once the absolute differences are ordered.

Standardized References.

Definitive Institutional SourceConover, W. J. (1999). Practical Nonparametric Statistics (3rd ed.). John Wiley & Sons.

Intermediate

Proof of Chebyshev's Inequality

Exploring the cinematic intuition of Proof of Chebyshev's Inequality.

Intermediate

Derivation of the Mean and Variance of the Binomial Distribution

Exploring the cinematic intuition of Derivation of the Mean and Variance of the Binomial Distribution.

Intermediate

Derivation of the Mean and Variance of the Poisson Distribution

Exploring the cinematic intuition of Derivation of the Mean and Variance of the Poisson Distribution.

Advanced

The Conceptual Proof of the Central Limit Theorem (CLT)

Exploring the cinematic intuition of The Conceptual Proof of the Central Limit Theorem (CLT).

Institutional Citation

Reference this proof in your academic research or publications.

NICEFA Visual Mathematics. (2026). Derivation of the Test Statistic for the Wilcoxon Signed-Rank Test: Visual Proof & Intuition. Retrieved from https://nicefa.org/library/applied-statistics/derivation-of-the-test-statistic-for-the-wilcoxon-signed-rank-test

Dominate the Logic.

"Abstract theory is just a movement we haven't seen yet."

Subscribe for Full Proofs Early Access

Visualizing...

The Formal Theorem

Analytical Intuition.

Institutional Warning.

Academic Inquiries.

Why use ranks instead of the raw differences themselves?

How are tied absolute differences or zero differences handled?

What is the key assumption underpinning the derivation of E[W] E[W] E[W] and Var[W] Var[W] Var[W]?

Standardized References.

Related Proofs Cluster.

Proof of Chebyshev's Inequality

Derivation of the Mean and Variance of the Binomial Distribution

Derivation of the Mean and Variance of the Poisson Distribution

The Conceptual Proof of the Central Limit Theorem (CLT)

Institutional Citation

Dominate the Logic.

What is the key assumption underpinning the derivation of $E[W]$ and $Var[W]$ ?