Sign Test (paired, direction-only)

Verified

Nonparametric

Independently verified. Every statistic this test reports has been re-derived against an independent reference — never the library the pipeline itself calls — the rendered output was read back in a browser, and the result is locked with a committed regression suite.

Run this test straight away on a free built-in teaching dataset — no data of your own needed — or bring your own. Either opens the guided workspace: variable setup, assumption diagnostics, results with effect sizes and confidence intervals, figures, and APA-ready reporting.

Loading teaching datasets…

Or use your own dataset

Loading your datasets…

Tests whether the median paired difference is zero by counting the proportion of positive vs negative paired differences and comparing to the binomial(n_nonzero, 0.5) null.

Discards the magnitude of differences entirely — only their sign is used. The most assumption-light paired test available: requires only that paired observations are independent across pairs and the null is symmetric in direction. Apt when the difference distribution is wildly non-normal AND the Wilcoxon signed-rank's symmetry assumption is also untenable.

Worked example

Do more people prefer version B over version A?

24 people tried both versions. The sign test counts only who preferred which (direction), ignoring by how much — useful when magnitudes aren't meaningful.

Result

18 of 24 preferred version B and 6 preferred A; the sign test gives p = .023 — a significant majority for B.

How you'd report it (APA)

A sign test showed a significant preference for version B (18 of 24 favouring B), p = .023.

Try it yourself: Load this ready-made sample and follow the run above.

When to use it

  • Paired ordinal or severely skewed continuous DV
    30 patients rate symptom severity on a 5-point Likert scale before and after a single dose of an experimental analgesic.
  • Matched-pair directional outcome (preference / improvement)
    60 wine-tasters compare two vintages and report which they prefer.

When NOT to — use instead

Hypotheses

H₀: P(v1 > v2 | v1 ≠ v2) = 0.5 — among non-tied paired observations, v1 is as likely to exceed v2 as the other way around. This is the operative parameter the sign test actually evaluates; a median-shift interpretation only follows as a corollary under continuity and a symmetric paired-difference distribution, which is exactly the assumption the sign test is designed not to require.
Hₐ: P(v1 > v2 | v1 ≠ v2) ≠ 0.5 — non-zero paired differences favour one direction more often than would be expected under the null.

Parameter tested: P(v1 > v2) — probability that measurement 1 exceeds measurement 2 on a randomly chosen pair

Assumptions (and what to do if they fail)

Effective sample size (non-zero paired differences)medium

Check: Count pairs with v1 ≠ v2.

If violated: Too few non-zero pairs leave the sign test with almost no power.

Tie burden on the paired differenceshigh

Check: tie_ratio ≤ 10% → negligible; 10–25% → declare the power loss; 25–50% → the test is running on a remnant and inference is weak; > 50% → the paired-difference distribution is zero-inflated and the sign

If violated: Many tied pairs erode the effective sample size and may hide a ceiling/floor effect.

Ready to run a Sign Test (paired, direction-only) on your own data?

Guided setup, automatic assumption checks, effect sizes, figures and an APA write-up.

Run this test →