Stratified Scale Sampling¶
Sampling design — instantiates Scale-Invariance Testing
Designs evidence-gathering across deliberate scale bands, and registers non-scale differences, so a conclusion is not overgeneralized from a narrow range of sizes.
Stratified Scale Sampling is a sampling design. Instead of collecting evidence wherever it is convenient — which almost always means a narrow band of similarly-sized cases — it deliberately draws observations across defined scale strata (small, medium, large) and, alongside, registers the non-scale differences that ride along with size so they are not later mistaken for scale effects. Its defining idea is coverage-by-design plus confound bookkeeping: the point is to guarantee that any eventual invariance claim rests on the full span of scales it will be applied to, and that a difference blamed on "scale" is not really a difference in context that merely correlates with scale. It builds the evidence base; it does not run the test or issue the verdict.
Example¶
Ecologists want to know whether a pollinator-abundance-per-area relationship holds across meadow patches. Convenient sampling would grab the patches nearest the field station — all roughly the same middling size — and generalize from them. Instead the team stratifies: small patches under 0.5 hectares, medium patches, and large patches over 5 hectares, sampling equally within each band so the eventual claim spans the whole size range it will be applied to. Crucially, they open a confound register: larger patches in this landscape also tend to be older, so any abundance difference could be driven by patch age rather than area. Age gets its own column, tracked separately at every stratum, so it cannot silently masquerade as a scale effect. The design's contribution is not a conclusion — it is evidence honest enough that a real conclusion becomes possible.
How it works¶
- Define the strata to span the deployment range. Set band boundaries so the smallest and largest scales the claim will face are actually represented, not interpolated.
- Allocate samples across, not proportional to availability. Equal or targeted allocation per band prevents the abundant middle from drowning out the sparse extremes.
- Register the confounds that co-vary with scale. For each stratum, log the non-scale attributes (age, funding, density, staffing) that travel with size, keeping them visible as separate variables.
Tuning parameters¶
- Number and boundaries of strata — finer strata resolve where behavior changes but demand more samples; coarse strata can bury a regime shift inside one band.
- Allocation across strata — equal allocation protects the extremes; availability-weighted allocation is cheaper but re-imports the narrow-range bias.
- Register depth — how many confounds are tracked. Tracking more catches the sneaky ones but adds bookkeeping; tracking too few lets an unregistered context effect pass as scale.
When it helps, and when it misleads¶
Its strength is that it defeats two errors at once: overgeneralizing from a narrow band of sizes — the restriction-of-range problem, where a relationship estimated over a thin slice is projected across a range it was never observed on[n1] — and the conflation of scale with the context that happens to accompany it. Its failure mode is subtle: if the strata are drawn too coarsely, or a confound goes unregistered, the design will quietly launder a context effect as a scale effect, lending false credibility to whatever test consumes its data. The classic misuse is stratifying diligently on size while ignoring the one variable that co-varies with size. The guarding discipline is to pre-register the confounds before sampling and to confirm the strata actually span the range the conclusion will be applied to.
How it implements the components¶
Stratified Scale Sampling realizes the evidence-design slice of the archetype — the part that decides where to look so the later test is not fooled:
scale_sampling_plan— the strata, their boundaries, and the allocation across them constitute its sampling plan.comparison_scale_set— the small / medium / large bands are the concrete set of scales the evidence will compare.context_confound_register— it logs the non-scale differences that ride with size, keeping them separable from scale effects.
It designs where to look and separates scale from context, but it does not decide the outcome or trace the failure: the bounded transfer_limit and the action_update_rule are Breakpoint Review Table's, and running the bands through a model rather than the field — with its scaling_ratio and breakpoint_detection — is Simulation Rescaling Sweep's.
Related¶
- Instantiates: Scale-Invariance Testing — it is the archetype's evidence-design front end, ensuring the data span the scales and separate out context.
- Sibling mechanisms: Pilot-to-Scale Validation · Normalized Metric Check · Dimensional Scaling Test · Per-Unit Invariance Check · Simulation Rescaling Sweep · Breakpoint Review Table · Log-Log Scaling Check
Editorial Notes¶
Form Classification¶
Form family: Representation, Specification & Plan
Rationale: Stratified Scale Sampling operates as a static representation, map, specification, schema, or prospective plan that externalizes information because it designs evidence-gathering across deliberate scale bands, and registers non-scale differences, so a conclusion is not overgeneralized from a narrow range of sizes.
Independent corroboration: The frozen evidence defines Stratified Scale Sampling as 'Designs evidence-gathering across deliberate scale bands, and registers non-scale differences, so a conclusion is not overgeneralized from a narrow range of sizes', so its operative form is Representation, Specification & Plan.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Sampling across size bands prevents unsupported scale extrapolation.
Related originating lineages:
- Biology & Ecology — Ecological processes vary by scale.
- Data Science & Analytics — Data science, analytics, and operational monitoring supplies a parallel or contributing lineage for the mechanism's defining operation: designs evidence-gathering across deliberate scale bands, and registers non-scale differences, so a conclusion is not overgeneralized from a narrow range of sizes.
- Engineering & Design — Scale changes performance.
- Mathematics — Mathematical modeling, proof, and abstract-structure practice supplies a parallel or contributing lineage for the mechanism's defining operation: designs evidence-gathering across deliberate scale bands, and registers non-scale differences, so a conclusion is not overgeneralized from a narrow range of sizes.
Review resolution: The blind reviewers agree that statistics_experimental_design is the primary origin and differ only on alternate origin disagreement, origin mode disagreement, domain reach disagreement, encyclopedia synthesis disagreement. I preserve every independently explained alternate from both records rather than imposing a numeric cap. I retain cross_disciplinary_synthesis because the combined evidence shows material contributions from several lineages. The broader reach of multi_domain records portability separately from historical provenance; encyclopedia_synthesis=true preserves the affirmative synthesis judgment where either reviewer identified one.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] Restriction of range — when a relationship is estimated from a narrow slice of the possible values of a variable, the estimate can badly misrepresent the relationship over the full range. Deliberately sampling across scale strata is the standard corrective. ↩