One- and Two-Tailed Tests¶
Hypothesis-test designs that allocate rejection probability to one prespecified direction or to extreme departures in both directions according to the scientific alternative.
Core Idea¶
Tail choice is a property of the tested alternative and decision rule. A one-sided test asks whether a parameter departs in one prespecified direction and puts alpha there; a two-sided test treats sufficiently large departures on either side as evidence.
Direction must be chosen from the scientific question before seeing results. Switching afterward or ignoring meaningful opposite effects breaks error calibration. The statistic, null distribution, estimate, effect size, and interval remain necessary for interpretation.
Structural Signature¶
Sig role-phrases:
- Null hypothesis — Defines the reference parameter or distribution. It is baseline. Counterfactual: Without it tail probability has no tested claim.
- Alternative hypothesis — States directional or bidirectional departures. It is decision target. Counterfactual: Tail choice cannot be made independently of the alternative.
- Test statistic — Orders sample outcomes relative to the null. It is evidence mapping. Counterfactual: Raw effect direction may not match statistic direction without convention.
- Null distribution — Assigns probabilities to statistic values under the null. It is calibration. Counterfactual: Tails are distribution-relative, not graphical decoration.
- Rejection region and alpha — Allocate type-I error across one or two directions. It is decision rule. Counterfactual: Using a one-tail cutoff after seeing sign inflates error.
- Observed outcome — Produces p-value and decision under the preregistered rule. It is application. Counterfactual: Nonsignificance does not prove the null.
What It Is Not¶
- It is not a device for shrinking p-values after inspection.
- One-sided does not mean every distribution has only one geometric tail.
- Nonsignificance does not establish equivalence.
- Statistical significance is not practical importance.
- Closest near-miss. A chi-square statistic can have a right-tail rejection region while testing a nondirectional lack-of-fit alternative; 'one-tailed' geometry and scientific direction are not always synonymous.
Scope of Application¶
- Experimental design. Aligns alternatives and power.
- Statistical inference. Defines rejection regions and p-values.
- Quality control. Tests directional exceedance.
- Research reporting. Documents prespecified analysis choices.
Clarity¶
State parameter, null and alternative with inequalities, direction rationale and timing, statistic, null distribution, alpha, multiplicity, assumptions, effect estimate, confidence interval, and treatment of opposite effects.
Manages Complexity¶
A small change in alternative geometry reallocates false-positive risk and changes which evidence the procedure agrees to recognize.
Abstract Reasoning¶
- Formulate the scientific alternative before data.
- Choose a statistic with known null behavior.
- Allocate alpha consistently to one or both directions.
- Compute estimate, uncertainty, statistic, and p-value.
- Interpret direction, magnitude, and errors without post-hoc switching.
Knowledge Transfer¶
A tail decision transfers only with the same parameter, directionally meaningful question, statistic convention, null model, alpha, and preregistration; a previous one-sided choice is not reusable by default.
Examples¶
Canonical¶
Before data collection, a reliability study tests H1: defect rate exceeds 1%; alpha is placed in the upper tail of the declared statistic.
Mapped back: null → rate≤1% boundary; alternative → greater; statistic → rate-based; distribution → null; region → upper tail.
Applied / In Practice¶
Researchers observe an effect's sign and then report the matching one-sided p-value despite planning a two-sided question; type-I error is no longer controlled as claimed.
Mapped back: direction timing → post hoc; alpha rule → changed; verdict → invalid.
Structural Tensions¶
T1 — Directional Power versus Opposite-Direction Detection. One-sided allocation improves sensitivity in one direction by declining symmetric evidence in the other.
Diagnostic: Could an opposite effect change the scientific conclusion?
T2 — Decision Threshold versus Effect Interpretation. A small p-value reflects null incompatibility, not magnitude, utility, or replication.
Diagnostic: Are estimate and uncertainty reported alongside the test?
Structural–Framed Character¶
One- and Two-Tailed Tests are structural as alternative-dependent rejection geometry and framed by scientific design and error control.
Structural Core vs. Domain Accent¶
The core is null, alternative, statistic order, tail region, and alpha. Statistics supplies distributions, power, multiplicity, and interpretation.
Instantiates / Related Primes¶
This entry is a kind of Hypothesis Testing (Null vs. Alternative).
-
Approved root. No reviewed parent entails this directional testing distinction.
-
Related — null hypothesis, alternative hypothesis, p-value, confidence interval, type-I error, and equivalence test. They provide components and contrasts.
Relationships to Other Abstractions¶
Current abstraction One- and Two-Tailed Tests Domain-specific
Parents (1) — more general patterns this builds on
-
One- and Two-Tailed Tests is a kind of Hypothesis Testing (Null vs. Alternative) Prime
One- and Two-Tailed Tests is a strict kind of Hypothesis Testing (Null vs. Alternative): they allocate null-rejection probability according to one- or two-direction alternatives.Every reviewed One- and Two-Tailed Tests instance satisfies Hypothesis Testing (Null vs. Alternative) because they allocate null-rejection probability according to one- or two-direction alternatives. The child adds the domain-specific restrictions stated in its frozen identity. Hypothesis Testing (Null vs. Alternative) is broader and can occur without the restrictions that define One- and Two-Tailed Tests.
Hierarchy paths (5) — routes to 5 parentless roots
- One- and Two-Tailed Tests → Hypothesis Testing (Null vs. Alternative) → Statistical Inference → Inductive Reasoning
- One- and Two-Tailed Tests → Hypothesis Testing (Null vs. Alternative) → Statistical Inference → Uncertainty
- One- and Two-Tailed Tests → Hypothesis Testing (Null vs. Alternative) → Verification → Evaluation → Comparison → Self Checking
- One- and Two-Tailed Tests → Hypothesis Testing (Null vs. Alternative) → Statistical Inference → Probability → Measure → Set and Membership
- One- and Two-Tailed Tests → Hypothesis Testing (Null vs. Alternative) → Statistical Inference → Probability → Measure → Aggregation → Micro Macro Linkage
Neighborhood in Abstraction Space¶
One- and Two-Tailed Tests sits in a moderately populated region (49th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.
Family — Statistical Hypothesis Tests & Diagnostics (9 abstractions)
Nearest neighbors
- Misuse of p-values — 0.88
- CUSUM — 0.87
- Approximate Bayesian Computation — 0.86
- Grey Relational Analysis — 0.86
- ÉLECTRE — 0.86
Computed from structural-signature embeddings · 2026-10-08
Not to Be Confused With¶
- Directional estimate. Tell: Can have a sign without a one-sided test.
- Equivalence test. Tell: Often uses two one-sided tests for a different composite claim.
- Chi-square goodness-of-fit. Tell: Uses one numerical tail for a nondirectional discrepancy.
- Post-hoc subgroup test. Tell: Raises selection and multiplicity issues beyond tail choice.
References¶
- Frozen Wikipedia discovery revision: https://en.wikipedia.org/wiki/One-_and_two-tailed_tests (revision 1305306160).
- Preserved source candidate: https://archive.org/details/modernintroducti00dekk_431
- Preserved source candidate: https://archive.org/details/modernintroducti00dekk_431/page/n392
- Preserved source candidate: http://www.economics.soton.ac.uk/staff/aldrich/1900.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.