Aggregation Rules¶
Method — instantiates Degrees-of-Freedom Reduction
Combines multiple variables into a composite value, category, score, or state so decisions are made over fewer dimensions.
Aggregation Rules collapse several independently-varying inputs into a single derived quantity — a score, grade, band, or state — so the decision then ranges over that one summary instead of the many raw variables. Its defining move is that it manufactures a new decision-ready value by a stated combining function (a weighted sum, a category roll-up, a worst-of, a vote) and hands the decision that value, not the parts. That is what separates it from tying variables together so they share one setting, and from projecting a table into several new axes: aggregation produces one number or label that a person or rule can act on directly, and the whole point is that the act of combining is where the weighting judgment lives.
Example¶
A consumer lender receives roughly twenty independently-moving signals on each applicant: months of on-time payments, how much of the available credit is used, the age of the oldest account, how many hard inquiries landed last quarter, and so on. Asking an underwriter to weigh twenty dials per file is slow and wildly inconsistent between reviewers. An aggregation rule replaces the twenty dials with one: a credit score on a fixed 300–850 scale. The real FICO scoring family is exactly this — a published combining rule that assigns each category a share of the total (payment history the largest slice, utilization the next, and so on) and returns a single number.
The lender now underwrites over that one dimension: applicants above a cutoff are approved, those below are declined or priced up. The twenty variables still exist, but they no longer each demand attention — they have been folded into a composite whose weighting is stated once and applied uniformly. The payoff is consistency and speed; the price, which the rest of the mechanism exists to watch, is that the single number can hide which input actually moved it.
How it works¶
What distinguishes aggregation from the reductions around it is that the reduction happens inside a combining function:
- Pick the combining form. Additive weighted sum (compensatory — a strong input can offset a weak one), categorical binning (twenty ages of account become "thin / established / seasoned"), logical roll-up (a status is "blocked" if any precondition fails), or an ordering rule (worst-of, best-of, median).
- Fix the weights and the scale. Decide how much each input counts and what the output ranges over, and — critically — write those weights down so the judgment they encode is inspectable rather than buried in a black box.
- Emit one value per case, then monitor. The output is a single score/label, checked against cases where the composite and the underlying parts disagree.
Tuning parameters¶
- Combining function — sum vs. roll-up vs. worst-of; compensatory forms are smooth but let a decisive weakness be masked, non-compensatory forms are brittle but refuse to average away a red flag.
- Weighting scheme — how much each input counts; heavier weights sharpen the intended signal but amplify any error in that input.
- Output granularity — a continuous score vs. a few bands; coarser bands are easier to act on but throw away separation the decision may need.
- Compensation stance — whether a high input may fully offset a low one, or whether some inputs are hard gates.
- Re-weighting cadence — static weights vs. periodic refit as outcomes accrue.
When it helps, and when it misleads¶
Its strength is that it turns an unrankable bundle into one comparable, consistent quantity that a threshold or an ordering can act on — and, when the weights are published, it makes the weighting argument visible instead of leaving it to each reviewer's gut.
Its central failure is that a composite averages, and averaging can erase the one variable that should have decided the case: a spotless-but-thin file and a long-but-shaky file can land on the same score for opposite reasons. Once a composite becomes a target, it also invites gaming — people optimize the score rather than the thing it was meant to summarize.[n1] The classic misuse is to keep tuning the weights until the composite ranks the cases the author already favored. The discipline that guards against this is to monitor the retained-variation signal — track the cases where the score and the raw inputs point different ways, and treat a rising count as evidence the rule has compressed away something the decision needs.
How it implements the components¶
Aggregation Rules fill the judgment-and-monitoring slice of the archetype's machinery, not the mapping or the encapsulation:
relevance_to_task_criterion— the weighting scheme is an explicit statement of which inputs matter for the decision and how much, made inspectable rather than left implicit.retained_variation_signal— the disagreement monitor tracks whether the composite still separates the cases the decision cares about, catching over-compression before it does damage.
It does not enumerate an independent_variable_map or keep a reduction_reversibility_log for a lower-dimensional transform — that's Dimensionality Reduction; nor does it apply a variable_coupling_rule that makes separate parameters share one value or update rule — that's Parameter Tying.
Related¶
- Instantiates: Degrees-of-Freedom Reduction — supplies the composite that shrinks the decision to a single dimension.
- Sibling mechanisms: Dimensionality Reduction · Parameter Tying · Option-Set Simplification · Configuration Profiles · Controlled Vocabularies · Default Presets · Design Constraint Templates · Modular Interfaces
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Combines multiple variables into a composite value, category, score, or state so decisions are made over fewer dimensions, making its operative form a computation or analytic transformation that produces an inference, comparison, or optimized result.
Independent corroboration: The frozen evidence defines Aggregation Rules as 'Combines multiple variables into a composite value, category, score, or state so decisions are made over fewer dimensions', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Convergent development
Present-day reach: Universal
Rationale: Statistical index construction and measurement theory characteristically combine multiple observed variables into a composite score while making scale, compensation, and weighting explicit.
Related originating lineages:
- Data Science & Analytics — Production scoring pipelines operationalize composite values at scale and monitor drift and gaming.
- Economics & Finance — Index numbers, credit scores, utility aggregation, and compensatory weighting supplied influential decision-facing composite measures.
- Mathematics — Algebraic aggregation operators, logical conjunctions, order statistics, and weighted sums provide the formal combining structures.
Review resolution: The mechanism is not an abstract algebraic operator alone: it documents the weighting and aggregation procedure used to construct a composite score for empirical decision use. OECD and JRC treat weighting, aggregation, uncertainty, and sensitivity as a statistical composite-indicator methodology, supporting statistics as primary, with mathematics, data science, and economics as material convergent lineages.
Attribution caveat: Combining parts into a summary is ancient and universal; statistics is the nearest formalizing tradition for the mechanism's explicit weighting and diagnostic concerns.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Researched adjudication after independent review; medium confidence.
Sources consulted:
Notes¶
[n1] Goodhart's law — "when a measure becomes a target, it ceases to be a good measure." A composite score summarizing many inputs is a prime target: actors optimize the aggregate rather than the underlying variables it was meant to compress, which is why the retained-variation monitor matters. ↩