Skip to content

Distributional Effects

Prime #
809
Origin domain
Economics And Finance
Subdomain
welfare economics → Economics And Finance

Core Idea

Distributional effects name the pattern in which a system-wide change produces an aggregate outcome that conceals systematically heterogeneous changes at the unit level — different subpopulations, components, or instances are affected to different degrees and often in different directions. The structural commitment is that the same intervention or shock yields a vector of unit-level effects whose distribution across the population matters separately from its summary statistic. A policy that raises average welfare by one percent may raise the top decile by five percent and lower the bottom by four; the aggregate signal hides this. The signature is therefore an intervention, a population of units, a per-unit effect function keyed to unit-specific properties, and a distribution of effects whose shape changes the appropriate evaluation in ways no scalar summary can recover.

Three structural details distinguish distributional effects from sibling patterns. First, the heterogeneity is structural, keyed to identifiable unit properties — income, location, age, genotype, network position — not random; the same unit type re-experiences the same effect direction under repetition. Second, the aggregate measure is not informative about the distribution: a positive average is fully compatible with majority loss when the gains are concentrated. Third, the choice of aggregation rule — mean, median, weighted sum, a social-welfare function with curvature, a Pareto criterion — is itself a normative commitment, and different rules pick different "right answers" from the same unit-level effect vector. The prime thus carries an explicit value-laden surface at exactly the point where the vector is collapsed to a scalar, which is part of why it sits toward the framed end of the spectrum even though its underlying object — vector-versus-summary — is structural.

How would you explain it like I'm…

The Average Hides It

Imagine your class gets candy and on average everyone got more. But that average hides that some kids got a big pile and some kids actually got less than before. Just knowing the average doesn't tell you who won and who lost. You have to look at how the candy was split up.

What The Average Hides

Distributional effects are when a single change affects different people in different amounts, and even in opposite directions, but the overall number hides all of that. Suppose a new rule makes the town "one percent richer on average." That average could mean the richest families gained a lot while the poorest actually lost money. The single summary number can't tell you that, you have to look at how the effect is spread across everyone. And how you choose to summarize it, average, middle person, or something that cares more about the worst-off, is really a choice about what you think is fair.

Spread Behind The Summary

Distributional Effects name the pattern where a system-wide change produces an aggregate outcome that hides systematically different changes at the unit level, different people or groups affected by different amounts and often in opposite directions. The same shock yields a whole vector of unit-level effects, and the shape of that distribution matters separately from any single summary number. A policy that raises average welfare one percent might raise the top tenth by five percent and lower the bottom by four, and the aggregate signal conceals this. Three things make it specific: the differences are structural, keyed to identifiable properties like income, location, or age rather than random; the aggregate is genuinely uninformative about the distribution, since a positive average is fully compatible with most people losing if the gains are concentrated; and the choice of how to aggregate, mean, median, weighted sum, a fairness-weighted measure, is itself a value judgment that picks different "right answers" from the same effects. That's why the prime carries a value-laden surface exactly at the point where the vector gets collapsed into a single number.

 

Distributional Effects name the pattern in which a system-wide change produces an aggregate outcome that conceals systematically heterogeneous changes at the unit level, different subpopulations, components, or instances are affected to different degrees and often in different directions. The structural commitment is that the same intervention or shock yields a vector of unit-level effects whose distribution across the population matters separately from its summary statistic. A policy that raises average welfare by one percent may raise the top decile by five percent and lower the bottom by four; the aggregate signal hides this. The signature is therefore an intervention, a population of units, a per-unit effect function keyed to unit-specific properties, and a distribution of effects whose shape changes the appropriate evaluation in ways no scalar summary can recover. Three details distinguish it from siblings. First, the heterogeneity is structural, keyed to identifiable unit properties, income, location, age, genotype, network position, not random; the same unit type re-experiences the same effect direction under repetition. Second, the aggregate measure is not informative about the distribution: a positive average is fully compatible with majority loss when gains are concentrated. Third, the choice of aggregation rule, mean, median, weighted sum, a social-welfare function with curvature, a Pareto criterion, is itself a normative commitment, and different rules pick different right answers from the same unit-level effect vector. The prime thus carries an explicit value-laden surface exactly where the vector is collapsed to a scalar.

Structural Signature

the intervention or shock — the heterogeneous population of units — the per-unit effect function keyed to unit properties — the resulting vector of effects — the aggregation rule that collapses it to a scalar — the structural (non-random) heterogeneity — the non-informativeness of the summary about the distribution

Distributional effects are present when each of the following holds:

  • An intervention or shock (the system-wide change). A single change applied across a population — a policy, treatment, update, or shock — whose effects are at issue.
  • A heterogeneous population (the units). A set of subpopulations, components, or instances with identifiable properties — income, location, age, genotype, network position — over which the change acts.
  • A per-unit effect function (the relating rule). A mapping from unit properties to the effect that unit experiences; the same change yields different effects, often in different directions, keyed to those properties.
  • An effect vector (the distributed outcome). The full distribution of unit-level effects, a higher-dimensional object whose shape matters separately from any summary.
  • An aggregation rule (the collapsing operation). A mean, median, weighted sum, social-welfare function, or Pareto criterion that projects the vector onto a scalar — itself a normative commitment, since different rules pick different "right answers" from the same vector.
  • Structural heterogeneity (the non-randomness invariant). The variation is keyed to unit properties, not random; the same unit type re-experiences the same effect direction under repetition.
  • Summary non-informativeness (the concealment invariant). The aggregate is not informative about the distribution — a positive average is fully compatible with majority loss when gains are concentrated.

The components compose so that any scalar summary is one normative projection among many, and the disciplined move is to disaggregate to recover the vector and name the aggregation rule that collapsed it.

What It Is Not

  • Not effect size. effect_size is a scalar magnitude of an aggregate effect; distributional effects are precisely the vector behind that scalar — the per-unit distribution the effect size collapses. A large effect size can hide a majority harmed.
  • Not aggregation. aggregation is the operation of collapsing many units to a summary; distributional effects are the recognition that the collapse discards a value-laden distribution and that the rule chosen is a normative commitment. Aggregation is the move; distributional effects critiques what it conceals.
  • Not heterogeneity of treatment effect alone. The per-unit variation matters, but distributional effects add the concealment-by-summary and normative-aggregation claims. Mere unit variation without a collapsing summary is not the full pattern.
  • Not heavy-tailed distributions. heavy_tailed_distributions concern the shape of one variable's distribution; distributional effects concern how a single intervention's effects distribute across structurally identified units, often bidirectionally.
  • Not selection bias. selection_bias distorts an estimate through non-representative sampling; distributional effects assume the effect on each unit is real and ask how the true per-unit effects are distributed and summarized. One corrupts the measurement; the other inspects an honest measurement's hidden structure.
  • Not equity as a value. equity is a normative standard for fair distribution; distributional effects are the structural-and-analytical recognition that a vector of unit effects exists and that any summary embeds an aggregation rule. Equity is one aggregation rule one might choose; the prime names the choice.
  • Common misclassification. Acting on an average effect that conceals concentrated harm — treating "raises the average" as "is good." Catch it by asking whether the reported number could coexist with a majority of units worse off; if a concentrated gain among a few could produce the same average, the scalar is hiding the distribution.

Broad Use

Distributional effects recur wherever a single change acts on a heterogeneous population and is then summarized. In welfare and policy economics, a tax change, trade agreement, or minimum-wage adjustment has an average GDP effect that is uninformative without a decile-by-decile decomposition — the original setting for the concept. In epidemiology and public health, an average treatment effect masks heterogeneity by age, comorbidity, and baseline severity, so the same drug may help the severely affected and harm the mildly affected. In climate science, a global average temperature rise hides regionally concentrated extremes, and the distribution of impacts across latitude, coastline exposure, and economic resilience is the policy-relevant signal. In AI fairness and ML evaluation, a model's average accuracy hides per-group performance gaps that group-conditional metrics make legible. In engineering reliability, average part-life hides early-failure subpopulations that carry most of the warranty cost. In drug development, average dose-response hides responder/non-responder distinctions, and personalized medicine is the explicit move to act on the distribution rather than the average. And on networks and platforms, an algorithm update with positive average engagement may concentrate gains on high-volume creators while cratering the long tail.

Clarity

Naming distributional effects exposes that aggregate measures are projections of a higher-dimensional unit-level object onto a single scalar, and that the projection discards information no scalar can re-encode. It separates two evaluative claims that are routinely conflated — "the intervention raises the average" and "the intervention is good" — and surfaces the load-bearing role of the chosen aggregation rule. It also clarifies why disagreements about policy can persist despite agreement on the average effect: the disputants may be applying different aggregation rules to the same distribution, so the dispute is not empirical but normative, and locating it correctly is the first step to resolving it. The clarifying force is to convert a single headline number back into the vector it summarizes and to make explicit the rule that collapsed the vector, so that both the empirical and the value-laden parts of an evaluation can be inspected separately.

Manages Complexity

Distributional effects compress heterogeneous-unit analysis into three primitives that scale across substrates: the intervention or shock, the population of units with their relevant properties, and the per-unit effect function. The full unit-level effect vector is reducible to those three plus the chosen aggregation, so an analyst need not track every unit individually to reason rigorously about who is affected and how. The same diagnostic questions then apply in economics, epidemiology, and machine learning: who is the population, what are the relevant unit-level properties, what is the per-unit effect by property, and what aggregation are we applying and why? This reduction is what makes the pattern portable: rather than re-deriving distributional reasoning from scratch in each field, the analyst instantiates the same four-part model and inherits its full apparatus of disaggregation, conditional aggregation, and aggregation-rule selection. The complexity that is managed is not the population's size but its heterogeneity — the three-object model gives a fixed handle on an arbitrarily varied set of units.

Abstract Reasoning

Recognizing distributional effects supports several lines of inference. It predicts that any intervention evaluated only by aggregate measures will produce systematic surprise whenever the distribution differs from prior expectations — the structural source of "policy that worked in theory failed in practice." It supports the move of disaggregating, re-running the analysis at the unit level to recover the distribution, and the converse move of conditional aggregation, re-aggregating by subgroup to reveal per-subgroup effects. And it connects to the broader principle that the choice of aggregation function encodes value judgments: a Rawlsian rule weights the worst-off, a utilitarian rule sums, a Pareto rule forbids any loss, and each selects a different optimal intervention from the same effect vector. The reasoner who has internalized the prime therefore treats every headline effect as provisional until the underlying vector and the aggregation rule are both on the table, and reads disagreement about a "good" intervention as a clue that different aggregation rules are quietly in play.

Knowledge Transfer

The portability of distributional effects rests on its substrate-neutral skeleton — intervention, heterogeneous population, per-unit effect function, aggregation rule — which lets a method built in one field transfer to another by re-identifying the units and their properties. The welfare-economics move of decomposing aggregate effects by subgroup transfers directly to ML fairness, where group-conditional accuracy is a distributional-effects analysis: the income decile becomes the demographic group, the per-unit effect becomes per-group model performance, and the same disaggregation that exposes a regressive tax exposes a discriminatory classifier. Responder-stratification in drug trials transfers to targeted policy design, where an intervention is applied conditionally on unit properties rather than uniformly, so that the pharmacological logic of "treat the subgroup that benefits" becomes the policy logic of "target the population the instrument helps." Vulnerability mapping in epidemiology — identifying which subgroups bear the burden — transfers to climate adaptation planning, where the relevant subgroups are geographies and infrastructure types and the same mapping discipline locates where adaptation investment should concentrate. The engineering-reliability intuition that the average describes the typical part but not the failure-driving subpopulation transfers to user-experience design, where the median user is healthy and the long tail of struggling users drives churn, so a Weibull-tail mindset borrowed from warranty analysis sharpens product retention work. In each transfer the practitioner performs the same four-step diagnosis — name the population, name the unit-level properties, estimate the per-unit effect, and choose and own an aggregation rule — and the same warning travels with it: any scalar summary is one normative choice among many, and acting on it without naming that choice is where interventions that "worked on average" quietly fail the units that mattered.

Examples

Formal/abstract

A machine-learning classifier's average accuracy is a clean formal instance of the prime's vector-versus-scalar structure. The intervention is the deployment of a single model; the heterogeneous population is the set of test cases, partitioned by an identifiable property — say demographic group \(g\). The per-unit effect function maps group properties to per-group error rates, and it is structurally heterogeneous: a model trained on imbalanced data systematically underperforms on under-represented groups, the same group re-experiencing the same deficit on every evaluation, not by chance. The effect vector is the full set of group-conditional accuracies \(\{a_g\}\); the aggregation rule is the overall accuracy, a weighted mean \(\sum_g w_g a_g\) with weights equal to group sizes. The concealment invariant is decisive and quantifiable: a model with 95% overall accuracy can post 99% on a majority group comprising most of the data and 60% on a small minority group, and the headline number conceals the gap entirely because the majority's weight dominates the average. This is the prime's central claim made arithmetic — the scalar is one projection, and a different aggregation rule (the minimax/Rawlsian rule, \(\min_g a_g\)) selects a different "best" model from the same vector, exposing the disparity the mean hides. The disciplined moves follow directly: disaggregate (compute group-conditional metrics) to recover the vector, and name the aggregation rule (equal-weight versus size-weight versus worst-group) as the normative choice it is.

Mapped back: The classifier instantiates every component — single intervention (deployment), heterogeneous units (test cases by group), per-unit effect function (group error), the concealing weighted-mean aggregation, structural (non-random) heterogeneity, and summary non-informativeness — and shows the prime's point precisely: average accuracy is one normative projection, and choosing it over \(\min_g a_g\) is a value choice that hides who the model fails.

Applied/industry

A minimum-wage increase shows the same structure in its origin substrate, welfare economics, with the prime's normative surface fully exposed. The intervention is a single policy raising the wage floor; the heterogeneous population is workers and firms, partitioned by skill level, region, and firm size. The per-unit effect function is sharply non-uniform and bidirectional: workers who keep their jobs at the higher wage gain, while marginal workers in low-margin firms may lose hours or jobs, and the direction is keyed to identifiable properties (skill, local labor-market tightness) — the prime's structural-heterogeneity invariant. The effect vector is the full distribution of income changes across workers; the aggregation rule routinely reported is aggregate employment or average earnings. The concealment invariant is exactly why the policy debate persists despite agreement on the average: a near-zero average employment effect is fully compatible with concentrated job losses among the least-skilled in the poorest regions, so disputants applying a utilitarian sum versus a Rawlsian worst-off rule reach opposite verdicts on the same vector — the prime's diagnosis that the disagreement is normative, not empirical, and is located at the aggregation step. The interventions the prime prescribes are the actual tools of good policy analysis: disaggregate by decile and region to recover who wins and loses, apply conditional aggregation to surface per-subgroup effects, and design targeted variants (regional phase-ins, small-firm exemptions) that act on the distribution rather than the average. The same four-step diagnosis transfers to epidemiology (an average treatment effect hiding responders and harmed non-responders) and to climate policy (a global-average temperature target concealing regionally concentrated extremes).

Mapped back: The wage case runs the prime end-to-end — one policy, heterogeneous bidirectional unit effects, a concealing aggregate, and a normative aggregation choice driving the dispute — and demonstrates the transfer: name the population, the unit properties, the per-unit effect, and the owned aggregation rule, and the same discipline that disaggregates a tax disaggregates a drug trial or a climate target.

Structural Tensions

T1 — Aggregate versus Unit Vector (Scalar Concealment). The prime's defining tension: any scalar summary is a projection of a unit-level effect vector, and the projection is non-informative about the distribution — a positive average is compatible with majority loss. The failure mode is headline reasoning: acting on an average effect that conceals concentrated harm, so an intervention that "worked on average" fails the units that mattered. Diagnostic: ask whether the reported number could coexist with a majority of units worse off; if a concentrated gain among a few could produce the same average, the scalar is hiding the distribution and disaggregation is mandatory before any verdict.

T2 — Empirical Effect versus Aggregation Rule (Fact/Value Boundary). Distributional effects fuse an empirical object (the effect vector) with a normative one (the rule collapsing it), and the two are routinely conflated. The failure mode is smuggled normativity: presenting "the intervention raises welfare" as an empirical finding when it is the mean-aggregation rule doing the evaluative work, so a value choice masquerades as a measurement. Diagnostic: ask whether a disagreement about whether an intervention is "good" survives agreement on the full vector; if it does, the dispute is about the aggregation rule (utilitarian sum versus Rawlsian worst-off), not the facts, and must be argued normatively rather than empirically.

T3 — Structural versus Random Heterogeneity (Predictability). The variation is keyed to identifiable unit properties, not chance, so the same unit type re-experiences the same effect direction — which is what makes targeting possible and what makes ignoring it unjust. The failure mode is treating systematic effects as noise: averaging over a structural split as if it were sampling variation, so a reliably harmed subgroup is dismissed as statistical scatter. Diagnostic: ask whether the per-unit effect correlates with a stable property (income, group, genotype, position); if the same units lose every time, the heterogeneity is structural and the average is masking a reproducible distributional pattern, not smoothing random error.

T4 — Uniform versus Targeted Intervention (Scopal Design). Recognizing the effect vector opens a design choice the average hides: apply the change uniformly, or condition it on unit properties to act on the distribution. The tension is between the simplicity of uniform policy and the precision of targeting. The failure mode is uniformity by default: deploying one treatment across a heterogeneous population because the average is positive, when a targeted variant (responder-stratified, regionally phased) would avoid the concentrated harm. Diagnostic: ask whether the per-unit effect function is known well enough to condition on; if it is, uniform application leaves welfare on the table and inflicts avoidable losses on the subgroups the instrument hurts.

T5 — Unit Definition versus Effect Distribution (Population Framing). The distribution depends entirely on how units and their properties are defined; re-slicing the population (by a different group variable) can reveal or hide disparities. The tension is that the choice of partition is itself consequential and often unexamined. The failure mode is partition blindness: disaggregating by a convenient axis (income) while the harm concentrates on an unmeasured one (geography, disability), so a distributional analysis declares fairness on the wrong slicing. Diagnostic: ask which unit properties could carry the effect and whether the analysis conditioned on them; a distribution that looks flat under one partition may be sharply unequal under another, and the relevant partition is the one the downstream stakes turn on.

T6 — Static Distribution versus Dynamic Response (Temporal Re-sorting). A measured effect vector is a snapshot, but units (especially agents) re-sort over time — migrating, adapting, exiting — so the distribution at deployment differs from the distribution later. The failure mode is static extrapolation: designing on the initial distribution of effects while units relocate across the partition (firms exit, workers retrain, populations move), so the targeted intervention misses its intended beneficiaries. Diagnostic: ask whether units can change the property that keys their effect; if the population re-sorts in response to the intervention, the distribution is endogenous and a static disaggregation predicts the wrong incidence over time.

Structural–Framed Character

Distributional effects is a hybrid on the structural–framed spectrum, landing on the framed side of the midline with a frontmatter aggregate of 0.5 — and the balanced scores, every criterion at 0.5, register a structurally clean object wearing a welfare-economics frame of equal weight. The underlying object is genuinely structural: a vector of per-unit effects versus a scalar summary, the same many-to-one collapse whether the units are income deciles, demographic groups, engineering parts, or patients. A reliability engineer's early-failure subpopulation hidden behind an average part-life is the prime's cleanest near-value-free case.

But the prime carries an explicit value-laden surface at exactly the point where the vector is collapsed to a scalar, and each criterion sits at the fence for a reason the prime's own content supplies. The vocabulary half-travels (vocab_travels 0.5): the vector-versus-summary structure ports to ML fairness and engineering reliability, but the welfare idiom — who wins, who loses, incidence, equity — comes with it. It carries real but partial evaluative weight (evaluative_weight 0.5): the choice of aggregation rule (utilitarian sum, Rawlsian worst-off, Pareto) is an unavoidable normative commitment, and "who is harmed" is morally charged, yet the per-unit effect vector itself is a neutral empirical object. Its origin is welfare economics (institutional_origin 0.5), a formal field entangled with policy institutions. Its centroid is policy evaluation (human_practice_bound 0.5), though the reliability and ML cases run with no human welfare at stake. And invoking it half-imports (import_vs_recognize 0.5): one partly recognizes a distribution already present, partly imports the evaluative frame of distributive justice.

The honest reading is the one the prime states itself — the underlying object (vector-versus-summary) is structural, but the policy-evaluation centroid and the normative aggregation step pull it onto the framed side. The 0.5 aggregate correctly records a structurally analyzable prime whose defining move, collapsing a vector to a scalar, is precisely where a value judgment enters.

Substrate Independence

Distributional effects is a strongly substrate-independent prime — composite 4 / 5 on the substrate-independence scale. Its signature — a vector of per-unit effects standing behind a scalar summary, the same many-to-one collapse regardless of what the units are — is genuinely relational, which earns structural abstraction a 4. It recurs with the same force across distinct substrates (domain breadth 4): tax, trade, and minimum-wage incidence by decile in welfare economics; treatment-effect heterogeneity by severity in epidemiology; regionally concentrated impacts in climate science; per-group performance gaps in ML fairness; early-failure subpopulations behind average part-life in engineering reliability — the prime's cleanest near-value-free case; responder/non-responder distinctions in drug development; and gains concentrated on high-volume creators in platform algorithms. The transfer is concrete (4): the disaggregate-before-you-summarize discipline ports unchanged, and a reliability engineer's hidden early-failure cohort is recognizably the same object as a hidden income-decile loser. What holds the composite to 4 rather than 5 is that the defining move — collapsing the vector to a scalar — is exactly where an aggregation rule and a value judgment enter, so the welfare-economics idiom of winners, losers, and equity travels with the prime in its policy-centroid applications even though the per-unit effect vector itself is a neutral empirical object.

  • Composite substrate independence — 4 / 5
  • Domain breadth — 4 / 5
  • Structural abstraction — 4 / 5
  • Transfer evidence — 4 / 5

Relationships to Other Primes

One-hop neighborhood: parents above, mutual partners to the right, children below.DistributionalEffectscomposition: AggregationAggregation

Parents (1) — more general patterns this builds on

  • Distributional Effects presupposes Aggregation

    distributional_effects is the critical recognition of what the aggregation operation conceals — the vector behind the scalar; it presupposes aggregation as the collapsing step.

Path to root: Distributional Effects → Aggregation → Micro Macro Linkage

Neighborhood in Abstraction Space

Distributional Effects sits in a sparse region of abstraction space (80th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely rather than landing on a neighbor.

Family — Aggregation & Scale Artifacts (16 primes)

Nearest neighbors

Computed from structural-signature embeddings · 2026-06-14

Not to Be Confused With

The nearest neighbor (similarity 0.89) is effect_size, and the relation is exactly the vector-versus-scalar contrast the prime is built on. Effect size is a scalar magnitude — a standardized measure of how large an aggregate effect is (Cohen's d, a regression coefficient, an average treatment effect). Distributional effects are the higher-dimensional object the scalar summarizes: the full vector of per-unit effects whose shape the effect size discards. The two are not rivals at the same level; one is the projection, the other the thing projected. A large effect size is fully compatible with a majority of units made worse off, if the gains are concentrated — which is precisely the concealment the prime exists to flag. Treating an effect size as a complete account of an intervention is the prime's headline failure, headline reasoning. The disciplined relationship is that effect size answers "how big, on average?" while distributional effects answer "distributed how, across whom, and under which aggregation rule?" — and the second question is the one no scalar, however well-standardized, can recover.

A second genuine confusion is with aggregation itself. Aggregation is the operation of combining many unit-level values into a summary — taking a mean, a weighted sum, a median. Distributional effects are not that operation but the critical recognition surrounding it: that the operation discards a value-laden distribution, and that the choice of aggregation rule (utilitarian sum, Rawlsian worst-off, Pareto criterion) is a normative commitment that selects different "right answers" from the same unit vector. Aggregation is morally neutral as a mechanical step; distributional-effects analysis is the insistence that the step is never neutral in evaluation, because the rule embeds a value judgment about whose gains and losses count and how much. Conflating them lets normativity be smuggled in as if it were arithmetic — the prime's smuggled normativity failure, where "the intervention raises welfare" is presented as an empirical finding when the mean-aggregation rule is doing the evaluative work. The discriminating move is to separate the empirical vector (a fact) from the rule that collapses it (a value choice), which aggregation-as-operation never prompts you to do.

A third confusion is with selection_bias. Both involve a population and a worry that a headline number misleads, but the locus of the problem differs fundamentally. Selection bias is a measurement defect: the sample is non-representative, so the estimated effect is wrong — it does not reflect the true effect even on average. Distributional effects assume the per-unit effects are correctly measured and ask how those honest effects distribute and what summary was chosen. One says "your number is biased because of who you sampled"; the other says "your number is accurate but conceals a distribution you must inspect." The cures diverge: selection bias is corrected by fixing the sampling or reweighting to the target population; distributional concealment is addressed by disaggregating the true effects and naming the aggregation rule. Confusing them sends the analyst to fix sampling when the data were fine and the problem was a hidden distribution, or to disaggregate when the underlying estimate was never valid.

For a practitioner the distinctions decide what to do with a headline number. Confusing distributional effects with effect size treats a scalar as a verdict and misses concentrated harm. Confusing them with aggregation lets a value choice pass as arithmetic. Confusing them with selection bias misdiagnoses a concealment problem as a measurement problem. The unifying discipline is the prime's four-step move applied after any summary: name the population and its effect-keying properties, recover the per-unit effect vector, identify the aggregation rule that produced the headline, and own that rule as the normative choice it is.

Solution Archetypes

No catalogued solution archetypes reference this prime yet.