Local / Global Analysis¶
Method — instantiates Scale Reframing
Contrasts local cases, subgroups, or sites with aggregate system behavior so local variation and global trends can be interpreted together rather than confused.
Local / Global Analysis is a two-pole reconciliation that fires specifically when the local evidence contradicts the aggregate — a subgroup, site, or case says one thing while the pooled number says the opposite. The contradiction is not noise to be smoothed over; it is the method's whole reason to exist and its central finding. Rather than picking a favorite pole, it holds the two together, locates the confound or aggregation that produces the disagreement, and marks the boundary past which the global figure stops being trustworthy for a given subgroup. Its defining question is not "which level is right?" but "why do these two levels disagree, and where does that disagreement bite?"
Example¶
Two hospitals' outcome data are pooled, and the aggregate says Treatment A has a higher survival rate than Treatment B. But when the same patients are split by case severity, B wins among mild cases and B wins among severe cases. A only looked better because it was given disproportionately to mild patients, whose survival is high regardless. The local (within-severity) evidence flatly contradicts the global (pooled) evidence.
Local / Global Analysis names that contradiction as the signal, then finds the confound — case-mix — that manufactures the reversal. It marks the boundary explicitly: the pooled survival rate is invalid as a comparison because treatment assignment is entangled with severity, so the severity-stratified rate is the one that governs. The verdict reconciles both poles rather than crowning one: B is the better treatment; the aggregate was an artifact of who received it.
How it works¶
- Set the two poles against one outcome. Local cases or subgroups on one side, the pooled aggregate on the other, measured on the same metric.
- Treat disagreement as the finding. When the poles reverse, the reversal is the result, not a data-quality problem to be averaged away.
- Locate the confound. Identify the composition, selection, or weighting that makes the pooled figure diverge from the within-group figures.
- Mark the validity boundary. State where the aggregate can still be trusted and where a subgroup breaks it.
Tuning parameters¶
- Subgroup granularity — how finely the local pole is cut. Finer cuts catch more reversals but shrink each group toward noise.
- Reversal threshold — how large a local/global disagreement must be before it counts as a real contradiction rather than sampling wobble.
- Default-trust pole — which pole is presumed correct until the other overturns it, guarding against reflexive faith in either the anecdote or the average.
- Boundary strictness — how narrowly to fence the aggregate's remaining valid domain.
When it helps, and when it misleads¶
Its strength is catching aggregation paradoxes that a single-level read renders invisible — the reversals where a pooled number is not just imprecise but pointing the wrong way. The canonical case is Simpson's paradox, where a trend present in every subgroup disappears or flips when the groups are combined.[n1] Its failure mode is anecdotal overcorrection: treating one vivid local case as if it overturns a large, stable aggregate. The matching misuse is cherry-picking — hunting subgroups until one flips the headline the way you wanted. The guarding discipline is to require the local pattern to be structural (a genuine confound that persists across cuts), not a lone anecdote, and to keep the aggregate on the page as context rather than discarding it.
How it implements the components¶
current_scale— names which pole the reasoning currently trusts, so the contradiction can be located rather than assumed.scale_mismatch_signal— its star: the local-versus-aggregate contradiction that both triggers the method and is its finding.scale_boundary_note— marks where the pooled figure stops being valid for a subgroup, fencing the aggregate's real domain.
It does not lay a reveals/hides ledger across a fixed micro/meso/macro triad and return the pattern-bearing rung (alternate_scale_set, scale_comparison, revealed_pattern — that's Micro / Meso / Macro Analysis), nor pick a governing layer and translate it into action (decision_scale_selection, translation_back_path — that's Organizational Level Analysis).
Related¶
- Instantiates: Scale Reframing — the two-pole, contradiction-driven specialization of the scale shift.
- Sibling mechanisms: Micro / Meso / Macro Analysis · Zoom-In / Zoom-Out Diagnosis · Scale-Specific Policy Analysis · Organizational Level Analysis · Ecological Scale Review · User-Level / System-Level Analytics Comparison
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Local / Global Analysis operates as a computation, comparison, model, or analytic representation used to infer, estimate, or choose because it contrasts local cases, subgroups, or sites with aggregate system behavior so local variation and global trends can be interpreted together rather than confused.
Independent corroboration: The frozen evidence defines Local / Global Analysis as 'Contrasts local cases, subgroups, or sites with aggregate system behavior so local variation and global trends can be interpreted together rather than confused', 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 multilevel and hierarchical analysis formalized the joint interpretation of local variation and aggregate behavior.
Related originating lineages:
- Sociology & Anthropology — Multilevel social research materially shaped comparison of individual, subgroup, site, and aggregate behavior.
- Systems Thinking & Cybernetics — Systems analysis independently distinguishes component dynamics from whole-system patterns.
Review resolution: Both independent reviews assign primary provenance to statistics_experimental_design. The queued secondary differences (alternate_origin_disagreement, origin_mode_disagreement) are reconciled by retaining sociology_anthropology, systems_cybernetics only as formative or independently established lineage(s), not merely as application domains. origin_mode=convergent records the provenance relationship, while domain_reach=universal separately records applicability breadth. confidence=high preserves the more cautious assessment, and encyclopedia_synthesis=false records whether either reviewer identified a corpus-specific synthesis.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
Its nearest twin is Micro / Meso / Macro Analysis, the other level-comparing method. The one-sentence difference: this method fires on a contradiction between two poles — local versus aggregate — and its deliverable is reconciling that disagreement and fencing where the aggregate fails (it owns scale_mismatch_signal), whereas Micro / Meso / Macro compares three fixed rungs whether or not they conflict and returns the rung the pattern lives on.
[n1] Simpson's paradox — a statistical reversal in which an association that holds within every subgroup vanishes or inverts once the subgroups are pooled, because a lurking variable is unevenly distributed across them. It is the archetypal reason a global figure can be true and yet misleading for every local case beneath it. ↩