Sign test¶

Test a paired-difference or one-sample median null by reducing non-tied observations to positive and negative signs and evaluating the positive count against its exact binomial distribution under a declared null probability, usually one half.

Version: v2 · 2026-08-30 · History
Domain-specific #: 2779
Origin domain: statistics
Subdomain: nonparametric hypothesis testing

Core Idea¶

The sign test discards magnitude and counts the signs of independent nonzero differences; under the usual continuous null of no directional tendency or median difference zero, the positive count \(W\) has distribution \(\operatorname{Binomial}(m,1/2)\) after ties are omitted.^[1] each non-tied pair contributes one Bernoulli outcome determined only by direction; under the null, exchangeability or the median condition gives equal positive and negative probabilities, so an unusually imbalanced count supplies evidence against the null through calibrated binomial tails.

Its autonomous residual is the deliberate loss of magnitude and exact calibration of a sign count, not generic hypothesis testing, a binomial test with arbitrary outcomes, Wilcoxon signed ranks, McNemar's matched-binary table, or a visual plus-minus summary. The identity fails when paired units are treated as independent observations, ties are silently discarded or recoded, a symmetric distribution is claimed as necessary for the basic sign test, normal approximation is used in a small sample without checking error, or failure to reject is reported as proof of equality.

Recognition requires an analyst to define the paired difference and sign orientation, verify independence across pairs, count positive, negative, and tied cases, set m to the non-tied total when using the classical rule, choose the alternative before seeing results, and compute the exact binomial tail including its discrete two-sided convention. Once established, it supports testing paired directional change with minimal distributional assumptions, testing a population median under continuity conditions, analyzing ordinal paired preferences, and providing a robust alternative when magnitudes are unreliable or heavy-tailed without turning those uses into the definition.

Structural Signature¶

Carrier: independent paired differences or independent one-sample deviations from a declared median value, observed on at least an ordinal scale
Inputs or antecedent state: a directional comparison for each unit, a null probability for a positive sign, a tie rule, effective non-tied sample size, one- or two-sided alternative, significance level, and exact or approximation method
Constitutive operation: each non-tied pair contributes one Bernoulli outcome determined only by direction; under the null, exchangeability or the median condition gives equal positive and negative probabilities, so an unusually imbalanced count supplies evidence against the null through calibrated binomial tails
Invariant: observations are independent across units, within-unit comparisons are meaningfully ordered, the null fixes sign probability, ties follow a declared rule, and the rejection probability is computed from the resulting sign-count distribution
Recognition test: define the paired difference and sign orientation, verify independence across pairs, count positive, negative, and tied cases, set m to the non-tied total when using the classical rule, choose the alternative before seeing results, and compute the exact binomial tail including its discrete two-sided convention
Output or consequence: testing paired directional change with minimal distributional assumptions, testing a population median under continuity conditions, analyzing ordinal paired preferences, and providing a robust alternative when magnitudes are unreliable or heavy-tailed
Failure boundary: paired units are treated as independent observations, ties are silently discarded or recoded, a symmetric distribution is claimed as necessary for the basic sign test, normal approximation is used in a small sample without checking error, or failure to reject is reported as proof of equality

What It Is Not¶

It is not the whole field of statistics; many objects in that field do not satisfy its constitutive rule.
It is not its canonical example. For ten paired measurements with eight positive and two negative differences and no ties, the two-sided exact sign-test p-value under a fair-sign null is twice the smaller binomial tail, with the discrete convention stated. That is an instance, not a definition.
It is not Hypothesis Testing (Null vs. Alternative). Hypothesis Testing supplies the parent decision architecture. Student's t-Test uses magnitudes and an estimated standard error, Wilcoxon signed-rank uses ranks and symmetry, and McNemar's test is a matched-binary discordance test that shares a conditional binomial calculation but has a different carrier.
It is not an unrestricted metaphor. with many ties, the classical omission rule changes the estimand and effective sample size; exact tie-aware, trinomial, permutation, or interval-null procedures may be preferable and must not be presented as the unchanged basic test

Scope of Application¶

Sign test applies when the analyst can specify independent paired differences or independent one-sample deviations from a declared median value, observed on at least an ordinal scale and establish that observations are independent across units, within-unit comparisons are meaningfully ordered, the null fixes sign probability, ties follow a declared rule, and the rejection probability is computed from the resulting sign-count distribution. This is descriptive statistical reference content, not an instruction to select a test without study-design, sampling, missingness, multiplicity, and domain-effect considerations.^[2]

Recognition. define the paired difference and sign orientation, verify independence across pairs, count positive, negative, and tied cases, set m to the non-tied total when using the classical rule, choose the alternative before seeing results, and compute the exact binomial tail including its discrete two-sided convention
Comparison. Compare legitimate instances through paired versus one-sample carrier, sign orientation, null sign probability, independence, continuity, ties, effective sample size, one- or two-sided alternative, exact-tail convention, approximation, confidence interval, power, and effect interpretation.
Boundary. with many ties, the classical omission rule changes the estimand and effective sample size; exact tie-aware, trinomial, permutation, or interval-null procedures may be preferable and must not be presented as the unchanged basic test
Use. Preserve every assumption when using the identity for testing paired directional change with minimal distributional assumptions, testing a population median under continuity conditions, analyzing ordinal paired preferences, and providing a robust alternative when magnitudes are unreliable or heavy-tailed.

Clarity¶

A clear claim names the carrier, governing rule, assumptions, and recognition test. This matters because sign test can refer to paired differences or a one-sample median formulation, and software packages can implement different two-sided exact p-value conventions. The disciplined statement is that the object counts as Sign test exactly when observations are independent across units, within-unit comparisons are meaningfully ordered, the null fixes sign probability, ties follow a declared rule, and the rejection probability is computed from the resulting sign-count distribution

Identity and measurement remain separate. Rounding creates artificial ties, dependence reduces effective information, and p-values do not measure effect magnitude or the probability that the null is true. Approximation or noisy evidence may weaken a classification without changing its definition.

Manages Complexity¶

The abstraction compresses paired sign test, one-sample median sign test, one- and two-sided forms, exact and normal-approximate calibration, randomized and conservative discrete tests, and tie-aware extensions into a stable carrier, rule, invariant, and failure boundary. It makes comparison tractable while retaining the variables that control validity.

Compression can hide assumptions. A responsible use therefore declares paired versus one-sample carrier, sign orientation, null sign probability, independence, continuity, ties, effective sample size, one- or two-sided alternative, exact-tail convention, approximation, confidence interval, power, and effect interpretation and returns to the full diagnostic whenever a convention or boundary case changes.

Abstract Reasoning¶

Type the carrier. Establish independent paired differences or independent one-sample deviations from a declared median value, observed on at least an ordinal scale and reject examples from a different problem.
Lock the rule. Express that observations are independent across units, within-unit comparisons are meaningfully ordered, the null fixes sign probability, ties follow a declared rule, and the rejection probability is computed from the resulting sign-count distribution independently of one notation or implementation.
Derive carefully. Infer testing paired directional change with minimal distributional assumptions, testing a population median under continuity conditions, analyzing ordinal paired preferences, and providing a robust alternative when magnitudes are unreliable or heavy-tailed only under the stated assumptions.
Stress-test. Contrast the legitimate boundary case—with many ties, the classical omission rule changes the estimand and effective sample size; exact tie-aware, trinomial, permutation, or interval-null procedures may be preferable and must not be presented as the unchanged basic test—with this counterexample: counting eight positive changes out of ten without a predeclared null, independence assumptions, tie rule, and tail calibration is a descriptive tally rather than a sign test.

Knowledge Transfer¶

Transfer within statistics is strong when new cases preserve the same carrier, mechanism, and diagnostic. The move from For ten paired measurements with eight positive and two negative differences and no ties, the two-sided exact sign-test p-value under a fair-sign null is twice the smaller binomial tail, with the discrete convention stated. to For a one-sample median test, subtract the hypothesized median from each observation, omit exact ties under the classical continuous rule, and test whether positive signs occur with probability one half. demonstrates that continuity.^[3]

Outside the domain, only the skeleton—discard magnitude, retain direction, and compare the resulting binary imbalance with a calibrated chance model—travels automatically. The terms paired difference, sign, tie, Bernoulli trial, binomial distribution, median, null hypothesis, alternative, exact p-value, tail, significance level, and power retain domain-specific meanings, so every role and inference must be revalidated.

Examples¶

Canonical¶

For ten paired measurements with eight positive and two negative differences and no ties, the two-sided exact sign-test p-value under a fair-sign null is twice the smaller binomial tail, with the discrete convention stated. Here \(W=8\) and \(m=10\); summing probabilities for counts at least as far from five as eight gives the exact two-sided result, while the sign orientation changes the named tail but not the symmetric two-sided value. It is canonical because the carrier, rule, invariant, and consequence are all inspectable.^[1]

Mapped back: independent paired differences or independent one-sample deviations from a declared median value, observed on at least an ordinal scale → each non-tied pair contributes one Bernoulli outcome determined only by direction; under the null, exchangeability or the median condition gives equal positive and negative probabilities, so an unusually imbalanced count supplies evidence against the null through calibrated binomial tails → observations are independent across units, within-unit comparisons are meaningfully ordered, the null fixes sign probability, ties follow a declared rule, and the rejection probability is computed from the resulting sign-count distribution → testing paired directional change with minimal distributional assumptions, testing a population median under continuity conditions, analyzing ordinal paired preferences, and providing a robust alternative when magnitudes are unreliable or heavy-tailed

Applied / In Practice¶

For a one-sample median test, subtract the hypothesized median from each observation, omit exact ties under the classical continuous rule, and test whether positive signs occur with probability one half. The inference concerns the declared median or sign probability and assumes independent sampling; censoring, clustering, many ties, or informative measurement rounding require another model or explicit modification. It qualifies only after the same diagnostic and failure boundary are checked.^[2]

Mapped back: declared instance → recognition test → boundary check → qualified use

Structural Tensions¶

T1: Exact identity vs. practical recognition. The constitutive condition may be exact while evidence is indirect. Diagnostic: Can the reviewer state both the condition and the warrant?
T2: Canonical form vs. variants. paired sign test, one-sample median sign test, one- and two-sided forms, exact and normal-approximate calibration, randomized and conservative discrete tests, and tie-aware extensions can preserve or change the identity. Diagnostic: Which named role is invariant across the variants?
T3: Compression vs. hidden assumptions. The label is useful only while prerequisites remain visible. Diagnostic: Can each downstream inference be traced to a declared assumption?
T4: Autonomy vs. reduction. The candidate uses broader structures but claims the deliberate loss of magnitude and exact calibration of a sign count, not generic hypothesis testing, a binomial test with arbitrary outcomes, Wilcoxon signed ranks, McNemar's matched-binary table, or a visual plus-minus summary. Diagnostic: Does that residual still support independent recognition after the parent and neighbors are subtracted?

Structural–Framed Character¶

The entry is structurally mixed but domain-framed. Its portable skeleton is discard magnitude, retain direction, and compare the resulting binary imbalance with a calibrated chance model; its identity-bearing terms are paired difference, sign, tie, Bernoulli trial, binomial distribution, median, null hypothesis, alternative, exact p-value, tail, significance level, and power. Those terms determine admissible objects, evidence, and consequences inside statistics.

Structural Core vs. Domain Accent¶

The structural core is a carrier governed by each non-tied pair contributes one Bernoulli outcome determined only by direction; under the null, exchangeability or the median condition gives equal positive and negative probabilities, so an unusually imbalanced count supplies evidence against the null through calibrated binomial tails and tested by define the paired difference and sign orientation, verify independence across pairs, count positive, negative, and tied cases, set m to the non-tied total when using the classical rule, choose the alternative before seeing results, and compute the exact binomial tail including its discrete two-sided convention. The domain accent is constitutive rather than decorative, so an analogy that preserves only the skeleton is not another instance of Sign test.

The proposed strict upward parent is prime:hypothesis_testing_null_vs_alternative. The sign test literally compares a calibrated null with a directional or two-sided alternative and uses a statistic to decide whether evidence crosses a rejection rule; sign reduction and binomial calibration supply the specialization. The edge is proposal-only and points to a frozen prior-baseline Prime.

The entry does not collapse into the parent because the deliberate loss of magnitude and exact calibration of a sign count, not generic hypothesis testing, a binomial test with arbitrary outcomes, Wilcoxon signed ranks, McNemar's matched-binary table, or a visual plus-minus summary A thematic neighbor is declined whenever it does not literally subsume that rule.

The prospective workspace queue contains one strict upward edge to prime:hypothesis_testing_null_vs_alternative. No live DAG mutation is authorized.

Relationships to Other Abstractions¶

Current abstraction Sign test Domain-specific

Parents (1) — more general patterns this builds on

Sign test is a kind of Hypothesis Testing (Null vs. Alternative) Prime

The proposed strict upward parent is prime:hypothesis_testing_null_vs_alternative.

Neighborhood in Abstraction Space¶

Sign test sits in a sparse region of the domain-specific corpus (63^rd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Statistical Dispersion & Testing (44 abstractions)

Nearest neighbors

Kendall rank correlation coefficient — 0.88
Nemenyi test — 0.86
Paired difference test — 0.85
Exchangeable random variables — 0.85
Count data — 0.85

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

Wilcoxon signed-rank test. Uses ranks of absolute differences and typically invokes symmetry, gaining power when magnitudes are meaningful.
Paired t-test. Uses numerical differences and a mean-and-standard-error model rather than only signs.
McNemar's test. Tests marginal homogeneity from discordant matched binary outcomes; its conditional calculation resembles a sign count but the data structure differs.
Binomial test. The computational genus; the sign test derives Bernoulli outcomes from paired or median comparisons.
Median test. Often names an independent-samples contingency-table procedure rather than the one-sample sign test for a median.

References¶

[1] W. J. Conover, Practical Nonparametric Statistics, 3^rd ed., Wiley, 1999, ISBN 978-0-471-16068-7. registry ↩a ↩b

[2] Myles Hollander, Douglas A. Wolfe, and Eric Chicken, Nonparametric Statistical Methods, 3^rd ed., Wiley, 2013, DOI 10.1002/9781119196037. registry ↩a ↩b

[3] E. L. Lehmann and Joseph P. Romano, Testing Statistical Hypotheses, 3^rd ed., Springer, 2005, DOI 10.1007/0-387-27605-X. registry ↩