Stratified Rollup Analysis¶
Aggregation method — instantiates Multi-Scale Signal Monitoring
Summarizes upward while keeping strata intact and each stratum's own baseline attached, so an aggregate cannot hide a vulnerable subgroup or a fattening tail.
Stratified Rollup Analysis is a bottom-up aggregation method with a strict rule: never collapse the distribution. Where a plain rollup replaces many local values with a single mean and throws the spread away, stratified rollup produces the summary while preserving the strata — cohorts, sites, shifts, regions — each carrying its own baseline so that a subgroup drifting into trouble stays visible even when the headline number looks fine. Its defining move is refusing to let averaging erase variation: the aggregate and its constituent strata are computed and reported together, and the method deliberately keeps every stratum decomposable so a tail or a vulnerable cluster can be read straight off the same summary. It is a computation, not a review and not a display — it makes the rollup itself honest about what it contains.
Example¶
A manufacturer tracks yield across three plants. The enterprise yield sits at a comfortable 97% and, on a naïve rollup, nothing is wrong. Stratified rollup analysis produces that same 97% but never as a bare number: it carries the yield distribution across plant, line, and shift, each stratum compared against its own historical baseline rather than the global one.
Read that way, the summary tells a different story. Plant B's night shift on line 3 has slipped to 88% — well below the ~95% baseline that stratum has held for a year — while the day shifts and the other plants run rich enough to hold the enterprise figure at 97%. The masked subgroup is right there in the rollup, flagged because it broke its own baseline, not because it dented the global one. The method does not diagnose why line 3's night shift is failing, and it does not page anyone; it simply refuses to let the healthy aggregate bury the failing stratum, and keeps that stratum decomposable so the next step can start exactly there.
How it works¶
- Aggregate without collapsing. The rollup carries the distribution — strata, tails, and spread — alongside the summary statistic, rather than replacing them with a single mean.
- Keep every stratum decomposable. The aggregate stays traceable to its constituent subgroups by construction, so a flagged stratum can be read directly off the rollup.
- Compare each stratum to its own baseline. A subgroup is judged against the reference band it normally holds, so a chronically-lower-but-stable stratum does not false-alarm and a stratum breaking its own norm does.
- Surface the tail, not just the center. The method reports the worst-performing strata and distributional tails as first-class outputs of the summary, not as an optional add-on.
Tuning parameters¶
- Stratification depth — how finely the population is sliced (plant vs plant-line-shift). Finer strata catch smaller vulnerable groups but shrink cell sizes toward statistical noise.
- Baseline scope — whether each stratum is judged against its own history or a shared global reference. Per-stratum baselines catch relative deterioration; a global baseline catches absolute underperformance; they flag different things.
- Flagging threshold — how far below its baseline a stratum must fall to surface, trading missed vulnerable subgroups against a flood of minor flags.
- Summary statistic — mean, weighted rate, or percentile for the headline, deciding how much the center reflects or masks the tails.
- Minimum cell size — the floor below which a stratum is too small to trust, trading resolution against spurious precision.
When it helps, and when it misleads¶
Its strength is that it makes the rollup itself trustworthy: a stratified summary cannot commit aggregate masking, because the vulnerable subgroup and the fattening tail are computed into the same output as the headline. It is the direct antidote to "the average looks healthy while clusters of failure accumulate underneath."
Its failure mode is the ecological fallacy[n1] run in reverse — the temptation to read the aggregate as if it described every stratum, which is exactly what a non-stratified rollup invites and what a careless reader of even a stratified one can still do by glancing only at the headline. The classic misuse is slicing so finely that every cell becomes noise, so real subgroup signals drown in a hundred spurious flags from tiny strata. The guarding discipline is to pin stratification depth to cell sizes large enough to be stable, to judge each stratum against its own baseline so relative and absolute problems are told apart, and to treat the flagged strata as the point of the summary rather than a footnote beneath the reassuring average.
How it implements the components¶
aggregation_rule— its core: a rollup rule that produces the summary while preserving strata, spread, and tails rather than collapsing them to a mean.disaggregation_rule— the aggregate is kept decomposable by construction, so any flagged stratum can be read straight off the rollup without a separate query.baseline_by_scale— each stratum carries its own reference band, so deterioration is judged relative to what that subgroup normally holds.
It does NOT implement scale_bridge_owner or cross_scale_pattern interpretation — the owned, top-down diagnosis of an aggregate shift — that's Drill-Down Root Signal Review; stratified rollup is the bottom-up computation that keeps strata intact so such a diagnosis is possible, not the review that performs it.
Related¶
- Instantiates: Multi-Scale Signal Monitoring — supplies the distribution-preserving aggregation the architecture rolls up with.
- Consumes: Local / Regional / Global Indicator Set supplies the per-level indicators and strata the rollup summarizes.
- Sibling mechanisms: Cross-Scale Anomaly Heatmap · Drill-Down Root Signal Review · Ecological Monitoring Network · Local / Regional / Global Indicator Set · Multi-Level Dashboard · Nested Early-Warning System · Organizational Health by Unit Monitoring · Public-Health Sentinel / Aggregate Surveillance · Supply-Chain Tier Monitoring
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Stratified Rollup Analysis operates as an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution because it summarizes upward while keeping strata intact and each stratum's own baseline attached, so an aggregate cannot hide a vulnerable subgroup or a fattening tail.
Independent corroboration: The frozen evidence defines Stratified Rollup Analysis as 'Summarizes upward while keeping strata intact and each stratum's own baseline attached, so an aggregate cannot hide a vulnerable subgroup or a fattening tail', so its operative form is Analysis, Modeling & Optimization.
Nearest alternative: Representation, Specification & Plan — Stratified Rollup Analysis includes features of a static representation, map, specification, schema, or prospective plan that externalizes information, but its defining operation is an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Convergent development
Present-day reach: Multi-domain
Rationale: Aggregation retaining stratum baselines is hierarchical statistical reporting.
Related originating lineages:
- Data Science & Analytics — Pipelines preserve subgroup fields.
- Mathematics — Mathematical modeling, proof, and abstract-structure practice supplies a parallel or contributing lineage for the mechanism's defining operation: summarizes upward while keeping strata intact and each stratum's own baseline attached, so an aggregate cannot hide a vulnerable subgroup or a fattening tail.
- Public Administration & Policy — Vulnerable groups must remain visible.
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 convergent because the combined evidence shows independent disciplinary development. 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] The ecological fallacy — inferring facts about individuals or subgroups from an aggregate statistic that need not hold for any of them. It is the exact error a stratified rollup exists to prevent, by keeping each stratum's own value and baseline attached to the summary. ↩