Category-Membership Attribution Audit¶
Workflow — instantiates Associative Transfer Warrant Audit
Tests whether a property has been attributed to an individual or item merely because it belongs to a category, class, cluster, or population.
Category-Membership Attribution Audit targets one specific associative move: reasoning down from a group to a member — this item is in a class with property P, therefore this item has P. Its defining concern is the legitimacy of that inference direction. A category base rate is a perfectly good prior; it becomes a fallacy when it is treated as a verdict on an individual for whom individuating evidence exists or could be gathered. The workflow measures how strongly the category actually predicts the property, checks whether the category is merely a proxy for something already measured, insists that individuating evidence be sought and allowed to dominate, and — because category attribution is where stereotyping and essentialism live — checks for the distinctive harm of a wrong member-level call.
Example¶
A loan-underwriting model flags an applicant as high default-risk, and when the analyst traces why, the weight turns out to rest largely on the applicant's residential cluster, a neighborhood whose historical default rate is elevated. The audit runs. Property attributed: elevated default probability. Attributed via: membership in a geographic cluster. Category-level strength: the cluster correlates with default only moderately. Independence probe: the cluster is largely a proxy for income and existing debt — variables the model already measures directly — so the geographic signal is double-counting a confounder, not adding information. Individuating evidence: this applicant's own two-year payment history is clean and dominates the base rate. Harm check: the cluster is a well-known proxy for a protected class, so a wrong call here is both stigmatizing and legally fraught.
The verdict: drop the category signal as redundant and hazardous, and decide on the individuating payment history. The base rate remains available as background, but it is no longer standing in for a person.
How it works¶
- Identify the category channel — name the class, cluster, or population and the property being read off it.
- Measure category-level strength, honestly, rather than assuming the group rate is decisive.
- Run the ecological-fallacy check: does the group-level rate actually license an individual-level claim, and is the category a redundant proxy for a confounder already in hand?
- Demand individuating evidence and let it override the prior when present.
- Apply the harm check for stereotyping, essentialism, and protected-attribute proxying before any category-based attribution is allowed to stand.
Tuning parameters¶
- Category granularity — how coarse the class is (broad population vs. narrow cluster). Coarser categories carry more stereotype risk; narrower ones may thin the base rate to noise.
- Individuating-override weight — how strongly a member's own evidence displaces the base rate. High override respects individuals; too high discards genuine priors.
- Confound stringency — how readily a category is judged a mere proxy for a measured variable. High stringency strips redundant signals.
- Protected-attribute sensitivity — how aggressively proxy-discrimination risk is flagged. High sensitivity is fairer but rejects more otherwise-predictive categories.
When it helps, and when it misleads¶
Its strength is blocking stereotyping and essentialism — and their operational cousin, proxy discrimination — while keeping legitimate base rates in their proper role as priors. It is the concrete guard against the ecological fallacy, inferring an individual's traits from group aggregates.[1]
Its failure mode is the opposite excess: discarding base rates entirely is itself an error (base-rate neglect), and calibrating exactly how much a category may legitimately inform an individual judgment is genuinely hard. The classic misuse is letting a category rate function as a verdict on a member when individuating evidence is available and simply was not consulted. The guarding discipline is the rule that a category is a prior, never a conclusion: individuating evidence must be sought, and where it exists it dominates.
How it implements the components¶
property_transfer_claim— forces the property attributed to the member (risk, quality, guilt, aptitude) to be named explicitly rather than assumed from the label.independence_and_confounding_probe— runs the ecological-fallacy and proxy checks, testing whether the category adds information beyond measured confounders.association_strength_indicator— quantifies how strongly the category actually predicts the property, rather than treating membership as decisive.harm_and_stigma_check— flags stereotyping, essentialism, and protected-class proxying that a wrong member-level attribution would inflict.
It does not maintain the multi-link ledger (the Association-to-Evidence Matrix), test a transmission channel (the Contact/Contagion Warrant Test), or scope a positive endorsement (the Endorsement Scope Checklist).
Related¶
- Instantiates: Associative Transfer Warrant Audit — this workflow guards the group-to-member inference.
- Sibling mechanisms: Association-to-Evidence Matrix · Guilt-by-Association Review · Halo and Taint Decomposition Table · Contact/Contagion Warrant Test · Endorsement Scope Checklist · Trust Transitivity Breakpoint Review · Associative Claim Red Team
Editorial Notes¶
Form Classification¶
Form family: Assessment, Review & Assurance
Rationale: Tests whether a property has been attributed to an individual or item merely because it belongs to a category, class, cluster, or population, making its operative form a bounded evaluation of existing evidence or work that produces a finding or disposition.
Independent corroboration: The frozen evidence defines Category-Membership Attribution Audit as 'Tests whether a property has been attributed to an individual or item merely because it belongs to a category, class, cluster, or population', so its operative form is Assessment, Review & Assurance.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Convergent development
Present-day reach: Universal
Rationale: Statistical methodology named the ecological fallacy: inferring an individual's property from an aggregate pattern in the group containing it.
Related originating lineages:
- Psychology — Stereotyping and illusory-correlation research explains attribution of perceived group traits to individuals.
- Sociology & Anthropology — Sociological research exposed how aggregate group categories and institutional statistics are misapplied to members.
Review resolution: Statistics is primary because the audit's central error is ecological inference: transferring a group-level association to an individual without individual evidence. Sociology contributes analysis of institutional group categories, and psychology contributes stereotyping and illusory-correlation research, so the broader audit has convergent universal provenance.
Review outcome: Researched adjudication after independent review; high confidence.
Sources consulted:
- Robinson: Ecological Correlations and the Behavior of Individuals
- Hamilton and Gifford: Illusory Correlation in Interpersonal Perception
References¶
[1] The ecological fallacy (W. S. Robinson, 1950) is the error of inferring an individual's characteristics from statistics aggregated over the group to which the individual belongs. A category base rate is a legitimate prior but not a verdict on any given member when individuating evidence exists. withdrawn registry ↩