Shapiro–Wilk Test¶
A statistical normality test whose statistic measures how closely ordered sample values align with expected order statistics from a normal population.
Core Idea¶
The Shapiro–Wilk test assesses whether an iid univariate sample is compatible with a normal population.
W compares ordered observations with expected normal order-statistic structure and is calibrated for sample size.
A small p-value is evidence against normality at α; nonrejection can reflect normality or low power and should be paired with graphics and consequence analysis.
Structural Signature¶
Sig role-phrases:
- iid sample. Supplies univariate observations. Constitutive input. If altered: Dependence changes calibration.
- ordered values. Expose sample quantile pattern. Constitutive transform. If altered: Raw order labels are irrelevant.
- normal order-statistic weights. Encode expected normal spacing/covariance. Constitutive reference. If altered: Different weights define another test.
- W statistic. Normalizes weighted order alignment by sample spread. Constitutive measure. If altered: W alone needs n-specific calibration.
- null calibration/p-value. Connects W and sample size to evidence. Inferential role. If altered: No universal W cutoff exists.
- alpha/context consequence. Sets rejection and practical relevance. Decision frame. If altered: Failure to reject is not proof.
What It Is Not¶
- Not proof of normality. Nonrejection is limited evidence.
- Not multivariate. The standard test is univariate.
- Not a robustness test. A method may tolerate deviations.
- Not sample-size invariant. Power/calibration change with n.
Scope of Application¶
The concept applies in statistical diagnostics and related work when its scope and evidence are explicit.
- Statistical diagnostics. Checks distributional assumptions.
- Regression. Can assess residuals under model conditions.
- Quality control. Screens process distributions.
- Experimental analysis. Guides transformation/model choice.
- Simulation validation. Checks generated marginals.
Clarity¶
Report data unit, independence basis, n, transformations, W, p, α, ties/censoring, graphical diagnostics, and practical impact. Do not test raw outcomes when the model assumes normal residuals.
Manages Complexity¶
The test compresses an entire order-pattern comparison into W and p, gaining reproducibility while losing detail about which tail, skew, or mixture caused rejection. The Shapiro–Wilk statistic compares ordered sample values with the pattern expected from normal order statistics. Values of W near one generally indicate close alignment, but inference uses sample-size-specific null calibration rather than a universal distance threshold. Rejecting normality at chosen α supplies evidence against the null; failing to reject does not prove normality. Small samples can miss meaningful deviations, while large samples can flag negligible departures. Estimated parameters, ties, discreteness, dependence, censoring, and prior data transformation can invalidate routine calibration or change interpretation. The test addresses marginal normality of an iid sample, not normal residuals, multivariate normality, or whether a downstream method is robust enough. Graphical diagnostics and substantive consequences should accompany the p-value.
Abstract Reasoning¶
- Identify the actual normality assumption.
- Check iid/univariate applicability.
- Compute ordered values and W with validated software.
- Use n-appropriate calibration.
- Interpret with QQ plot, power, and downstream robustness.
Knowledge Transfer¶
Order-statistic goodness-of-fit reasoning transfers to other distributions/tests, but Shapiro–Wilk requires its normal weights and calibration.
Examples¶
Canonical¶
An iid sample is sorted, W is computed from normal-order-statistic weights, and the p-value is compared with a preregistered α without calling nonrejection proof.
Mapped back: iid sample → independent observations; ordered values → x(1)…x(n); normal order-statistic weights → Shapiro–Wilk coefficients; W statistic → computed W; null calibration/p-value → n-specific p; alpha/context consequence → bounded decision.
Applied / In Practice¶
A regression analyst tests model residuals, not raw skewed outcomes, and combines W with a QQ plot and a sensitivity analysis showing whether inference changes under a robust method.
Mapped back: iid sample → model residuals with dependence caveat; ordered values → ordered residuals; normal order-statistic weights → software implementation; W statistic → reported; null calibration/p-value → reported; alpha/context consequence → practical robustness.
Structural Tensions¶
T1: sensitivity vs. practical importance. Large n detects tiny harmless departures. Diagnostic: Does the deviation affect the intended analysis?
T2: power vs. assurance. Small n nonrejection may be weak evidence. Diagnostic: What deviations could the test detect?
Structural–Framed Character¶
Shapiro–Wilk is structural-measured with inferential framing. Individuation is statistic-specific; analyst agency selects data/alpha; statistical norms govern interpretation; temporality is absent unless data dependence enters; robustness depends on assumptions. Its portable test skeleton is supplied by Normality test. Its character: normal-order alignment summarized into a calibrated rejection test.
Structural Core vs. Domain Accent¶
Skeletal core. Observed values are compared with a calibrated normal-reference pattern to test a normality null.
Domain-bound accent. W's order-statistic coefficients, sample-size calibration, and named implementation distinguish Shapiro–Wilk from other normality tests.
Why not prime. The broader Normality Test parent supplies the genus; Shapiro–Wilk adds one statistic-specific differentia.
Instantiates / Related Primes¶
This entry is a kind of Normality test.
- Strict parent — Normality test. Shapiro–Wilk is a named normality test: it evaluates an iid sample against a normal null, while W's ordered-value alignment supplies the narrower method.
- Related — QQ plot. It shows deviations that W compresses.
Relationships to Other Abstractions¶
Current abstraction Shapiro–Wilk Test Domain-specific
Parents (1) — more general patterns this builds on
-
Shapiro–Wilk Test is a kind of Normality test Domain-specific
Shapiro–Wilk is a strict kind of Normality Test: it tests an iid sample against a normal null using W and expected normal order statistics.Every Shapiro–Wilk application is a normality test: it evaluates sample evidence against a normal-population null under a calibrated decision rule. The child adds the W statistic and its ordered-value alignment with expected normal order statistics; other normality tests can instantiate the parent without that construction.
Hierarchy paths (4) — routes to 4 parentless roots
- Shapiro–Wilk Test → Normality test → Statistical Inference → Inductive Reasoning
- Shapiro–Wilk Test → Normality test → Statistical Inference → Uncertainty
- Shapiro–Wilk Test → Normality test → Statistical Inference → Probability → Measure → Set and Membership
- Shapiro–Wilk Test → Normality test → Statistical Inference → Probability → Measure → Aggregation → Micro Macro Linkage
Neighborhood in Abstraction Space¶
Shapiro–Wilk Test sits in a crowded region of the domain-specific corpus (37th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.
Family — Empirical Measurement & Statistical Inference Methods (50 abstractions)
Nearest neighbors
- D'Agostino's K-squared test — 0.89
- Median Absolute Deviation — 0.89
- M-Estimator — 0.89
- Bootstrapping populations — 0.88
- MAP estimator — 0.88
Computed from structural-signature embeddings · 2026-10-08
Not to Be Confused With¶
- QQ plot. Tell: Graphical diagnosis or formal test?
- Shapiro–Francia. Tell: Approximate/different weights?
- Kolmogorov–Smirnov. Tell: CDF distance or order-correlation statistic?
- Robustness analysis. Tell: Assumption truth or consequence?
References¶
- Frozen Wikipedia discovery revision: https://en.wikipedia.org/wiki/Shapiro%E2%80%93Wilk_test (revision 1362068790).
- Preserved source candidate: https://math.mit.edu/~rmd/465/shapiro.pdf
- Preserved source candidate: https://web.archive.org/web/20240729131413/https://math.mit.edu/~rmd/465/shapiro.pdf
- Preserved source candidate: https://web.archive.org/web/20221004220433/https://apps.dtic.mil/dtic/tr/fulltext/u2/a053857.pdf
- Preserved source candidate: http://www.jmp.com/support/notes/35/406.html
- Preserved source candidate: https://web.archive.org/web/20120525103314/http://www.jmp.com/support/notes/35/406.html
- Preserved source candidate: https://www.researchgate.net/publication/267205556
- Preserved source candidate: https://www.graphpad.com/guides/prism/latest/statistics/stat_choosing_a_normality_test.htm
- Preserved source candidate: https://www.stata.com/manuals15/rswilk.pdf
The frozen Wikipedia revision is discovery provenance. The retained source set was reviewed for identity, formal or operational relation, and scope. The encyclopedia's structural synthesis is bounded to those claims; a thin authority surface is recorded as a nonblocking source-strengthening repair rather than concealed.