Null distribution¶

Represent the sampling distribution of a declared test statistic under the null hypothesis and sampling scheme used to calibrate tail probabilities, critical values, and type-I error.

Version: v1 · 2026-08-30 · History
Domain-specific #: 2399
Origin domain: statistics
Subdomain: statistical hypothesis testing
Aliases: Null reference distribution, Distribution under the null

Core Idea¶

A null distribution is the probability distribution of a specified test statistic under a specified null hypothesis and sampling model. If \(T=T(X)\) is computed from random data \(X\), the null distribution is the law of \(T\) when the data-generating parameter or family satisfies \(H_0\). It supplies the reference against which an observed \(t_{obs}\) is ranked. Critical regions and p-values are therefore properties of the statistic-plus-null model, not generic properties of the raw data or of a named distribution detached from assumptions.^[1]

The distribution can be derived analytically, approximated asymptotically, generated by exact randomization under an exchangeability null, simulated from a fitted null model, or estimated empirically in large-scale testing. A p-value integrates the null probability of outcomes at least as incompatible with \(H_0\) as the observed statistic under a declared ordering. For a composite null, valid calibration may require a nuisance-parameter adjustment, conditioning argument, pivotal statistic, or supremum over null parameter values rather than substitution of one convenient point.^[2]

The null distribution differs from the observed empirical distribution of raw measurements, from the sampling distribution under an alternative, and from the prior or posterior distribution of a parameter. It is not always Gaussian: \(t\), \(F\), chi-square, exact discrete, permutation, and bespoke simulation distributions arise depending on statistic and design. A permutation distribution is a null distribution only when the transformation group is justified by the null. An empirical null estimates a reference law from many test statistics under additional mixture and central-mass assumptions; it does not simply mean any histogram.^[3]

Structural Signature¶

Null hypothesis. A precise parameter value, family, invariance, or exchangeability statement defines the reference world.
Sampling scheme. Randomization, dependence, censoring, and conditioning determine repeated-sample behavior.
Test statistic. A declared function of the data orders evidence against the null.
Reference law. The statistic's probability distribution is derived or generated under the null.
Tail ordering. One- or two-sided extremeness is defined before observing the result.
Critical value. A null quantile establishes a rejection region with intended type-I error.
Observed statistic. The realized value is located within the reference law.
Calibration method. Analytic, asymptotic, randomization, simulation, or empirical construction is documented.
Validity conditions. Pivotality, exchangeability, nuisance handling, and dependence assumptions bound interpretation.

What It Is Not¶

Not the null hypothesis itself. The hypothesis is a claim; the null distribution is a statistic's law under that claim.
Not the distribution of raw data. The reference concerns a chosen test statistic, which may compress the sample.
Not always a normal distribution. Its family depends on statistic, design, parameters, and approximation.
Not the alternative distribution. Power requires the statistic's law under alternatives, a separate object.
Not any permutation histogram. Exchangeability under the null is required for randomization validity.
Not a posterior probability that the null is true. Frequentist tail calibration is not posterior model probability.

Scope of Application¶

The abstraction is literal wherever practitioners can identify the same constitutive roles, apply the same boundary tests, and obtain the same kind of output. The following habitats are uses of Null distribution itself, not metaphors based only on resemblance.

Classical parametric tests. Using exact or asymptotic \(t\), \(F\), chi-square, and related reference laws.
Randomization inference. Enumerating or sampling treatment assignments justified by the design.
Permutation testing. Generating transformed statistics under an exchangeability null.
Monte Carlo tests. Simulating the statistic from a fully specified or fitted null model.
Large-scale multiple testing. Estimating an empirical null when theoretical calibration is systematically distorted.
Diagnostic simulation. Checking whether nominal critical values achieve intended type-I error under realistic assumptions.

Clarity¶

A clear account of Null distribution must preserve the recognition invariant stated in the Core Idea rather than rely on the title alone. State the null, statistic, sampling unit, dependence structure, conditioning, and tail ordering together. Name whether the law is exact, asymptotic, simulated, permutation-based, or empirically estimated. Explain nuisance-parameter handling and whether the statistic is pivotal. Distinguish a point-null reference law from a composite-null calibration valid over a family. These declarations are not editorial extras: each changes what observations count, which transformations are licensed, and what conclusion can be drawn. A reader should be able to reconstruct the input, the operative rule, the output, and at least one defeater from the account without consulting an implementation or guessing an unstated convention.

Manages Complexity¶

Null distribution manages complexity by replacing a diffuse field of observations or possible operations with a bounded role structure: null hypothesis supplies a precise parameter value, family, invariance, or exchangeability statement defines the reference world.; sampling scheme supplies randomization, dependence, censoring, and conditioning determine repeated-sample behavior.; test statistic supplies a declared function of the data orders evidence against the null.; reference law supplies the statistic's probability distribution is derived or generated under the null.; tail ordering supplies one- or two-sided extremeness is defined before observing the result.. The compression is useful because it localizes disagreement. One can ask whether the input was properly formed, whether a constitutive relation held, whether an alternative explanation defeats the inference, or whether the output was overinterpreted. The same compression can mislead when its discarded detail is exactly what the decision requires. A reference-grade use therefore reports both the invariant retained and the information intentionally lost.

Abstract Reasoning¶

Write the null hypothesis at the parameter, model, invariance, or assignment level.
Choose a statistic whose ordering corresponds to the intended departure.
Derive or generate its distribution under every material null condition.
Handle nuisance parameters by conditioning, estimation, invariance, or conservative maximization.
Select one- or two-sided extremeness and compute critical values or tail probabilities.
Verify type-I-error calibration analytically or by simulation under realistic dependence and sample size.
Analyze an alternative distribution separately when discussing power.
Test the candidate interpretation against the nearest named confusable rather than accepting a shared surface feature.
State the conclusion at the same scope as the source conditions, and retain uncertainty or nonuniqueness where the construct does not remove it.

Knowledge Transfer¶

The strict upward abstraction is Hypothesis Testing Null Vs Alternative. Null Distribution instantiates Hypothesis Testing (Null vs. Alternative) because it is the reference component that converts a statistic into calibrated evidence and controls false rejection under the null. Within statistical hypothesis testing, the full mechanism transfers literally when the same roles and boundary tests recur. Beyond that domain, only the parent-level skeleton should travel. Reusing the label Null distribution after removing its constitutive vocabulary would hide a change of mechanism behind an analogy. The honest transfer rule is therefore two-stage: recognize the domain-specific pattern first, then lift only the parent relation that remains invariant under a substrate change.

Examples¶

Canonical¶

For a one-sample Gaussian mean test with unknown variance, \(T=(\bar X-\mu_0)/(S/\sqrt n)\) has a Student \(t\) null distribution with \(n-1\) degrees of freedom under the model assumptions. The reference is not standard normal because estimating the scale changes finite-sample uncertainty. A two-sided p-value locates \(|t_{obs}|\) in both tails of that declared \(t\) law.

Mapped back: input and conventions → constitutive role test → bounded output → explicit interpretation and defeater check.

Applied / In Practice¶

In a randomized experiment, analysts compute a difference in means and repeatedly reassign observed outcomes according to the actual randomization scheme. Under the sharp null of no unit-level treatment effect, assignments are exchangeable by design and generate an exact randomization distribution. If clusters rather than individuals were randomized, permuting individuals would create the wrong null law and anti-conservative inference; the sampling unit is constitutive.

Mapped back: field observation or problem → candidate recognition → confusable and limit checks → appropriately scoped conclusion.

Structural Tensions¶

T1: Exact law versus asymptotic convenience. Closed-form limits are easy to use but may miscalibrate small samples. Diagnostic: Has finite-sample error been checked?
T2: Simple null versus nuisance uncertainty. A composite null lacks one automatic reference law. Diagnostic: How are nuisance parameters handled?
T3: Pivotal statistic versus raw interpretability. Studentization can stabilize the law while making the statistic less intuitive. Diagnostic: What invariance does the transformation buy?
T4: Permutation flexibility versus exchangeability. Resampling looks assumption-light but requires a valid transformation group. Diagnostic: What null makes the permutation legal?
T5: Theoretical null versus empirical null. Estimated references can correct systematic distortion but add mixture assumptions. Diagnostic: Why is empirical estimation identifiable here?
T6: Autonomous reference object versus generic hypothesis testing. Testing travels; the statistic's null law is the specific calibration object. Diagnostic: Can the critical value be defined without this distribution?

Structural–Framed Character¶

Null Distribution is structural-leaning: once null, statistic, and sampling design are fixed the law is mathematical, while choosing those ingredients and an extremeness ordering is framed by the inquiry. The five framing criteria point in a consistent direction. Evaluative weight is limited to whether the defining conditions are met, not whether the outcome is desirable. Human practice matters to the extent that experts choose conventions, instruments, or reporting thresholds, but those choices do not make every verdict arbitrary. Institutional history explains the name and standard use; it does not replace the recognition rule. The operative vocabulary travels within the home field and closely adjacent subfields, while transfer farther away requires translation to the parent prime. Thus recognition remains disciplined even where interpretation is defeasible.

Structural Core vs. Domain Accent¶

What is skeletal. Null Distribution instantiates Hypothesis Testing (Null vs. Alternative) because it is the reference component that converts a statistic into calibrated evidence and controls false rejection under the null. This is the part that can be expressed without the candidate's specialist nouns.

What is domain-bound. The accent is a statistical null, sampling distribution of a test statistic, tail ordering, critical values, nuisance parameters, and exact/asymptotic/resampling calibration. Remove those elements and the result is no longer Null distribution; it is only the parent relation or a loose analogy.

Why this does not clear the prime bar. The name does not recur with unchanged diagnostics across three independent domains. What transfers is already represented by prime:hypothesis_testing_null_vs_alternative. The candidate remains autonomous because its in-domain recognition rule, failure modes, and consequences are stable, but its vocabulary and interventions do not float free of the home substrate.

Null Distribution instantiates Hypothesis Testing (Null vs. Alternative) because it is the reference component that converts a statistic into calibrated evidence and controls false rejection under the null.

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 Null distribution Domain-specific

Parents (1) — more general patterns this builds on

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

Null Distribution instantiates Hypothesis Testing (Null vs.

Neighborhood in Abstraction Space¶

Null distribution sits in a sparse region of the domain-specific corpus (91^st percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Statistical Tests & Distribution Calibration (7 abstractions)

Nearest neighbors

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

Not to Be Confused With¶

Null hypothesis. The model claim under which the reference distribution is formed.
Sampling distribution. A broader family including laws under both null and alternative parameter values.
Permutation distribution. A generated reference law valid only under an appropriate transformation null.
Empirical null. A null law estimated from many statistics under additional mixture assumptions.
Alternative distribution. The statistic's law when a specified alternative is true, used for power.
Posterior distribution. A Bayesian distribution over parameters or models after observing data.

References¶

[1] Lehmann, E. L. and Romano, J. P. (2022). Testing Statistical Hypotheses, 4^th ed. Springer. https://doi.org/10.1007/978-3-030-70578-7 registry ↩

[2] Fisher, R. A. (1935). The Design of Experiments. Oliver and Boyd, foundational treatment of randomization and significance testing. registry ↩

[3] Efron, B. (2004). “Large-Scale Simultaneous Hypothesis Testing: The Choice of a Null Hypothesis.” Journal of the American Statistical Association 99(465), 96–104. https://doi.org/10.1198/016214504000000089 registry ↩