Variance Partitioning Report¶
Reporting artifact — instantiates Shared-Source Variance Isolation
Splits each dimension's variance into true-signal, shared-source, dimension-specific, and noise shares, carries each share's precision, and rewrites the claim to match.
A Variance Partitioning Report is the communication artifact that splits each dimension's total variance into named shares — true signal, shared-source, dimension-specific, and noise — and then rewrites the claim to match the split. Its defining idea is attribution-and-relabeling: it does not test correlations or design collection; it takes a fitted decomposition and reports how much of what you see is which, changing the language from "four independent indicators agree" to "these four share a measurement source that accounts for most of their common variance, and only two retain independent signal." It carries the precision of each share so that readers see how firmly the attribution is pinned, not just its headline value.
Example¶
A cattle-breeding program reports that several carcass-quality traits are strongly heritable and move together, implying that selecting on one will improve all. Before acting, the geneticist commissions a variance partitioning report. Using a mixed-model decomposition, each trait's phenotypic variance is split into additive-genetic, contemporary-group (pen / batch / season — a shared management source), and residual components, each with a standard error. The report shows that a large slice of the apparent between-trait covariance loads on contemporary group, not genetics: the traits looked jointly heritable partly because animals raised together shared pens and feed. The intraclass correlation[n1] for group is high; the genetic correlation, once group is separated, is modest. The report rewrites the breeding claim — heritability is real but lower than the raw figure, and the cross-trait promise is largely management, not genes.
How it works¶
Take a fitted variance-components model and allocate each dimension's variance — and the between-dimension covariance — to signal / shared-source / specific / noise. Attach a precision to each share (a standard error or credible interval), then rewrite each claim in proportion to its surviving signal share. The distinctive move is that its output is a labeled budget and revised wording, not a correlation test or a hypothetical sweep: it is a transparency-and-attribution artifact. It consumes the model's estimates rather than fitting them.
Tuning parameters¶
- Partition granularity — a few coarse buckets versus many fine source components. Fine partitions localize blame but spread the data thin and widen each share's error.
- Precision display — point shares versus shares with intervals. Showing intervals stops readers over-reading a 58%-versus-42% split the data cannot actually distinguish.
- Covariance versus variance focus — partitioning each dimension's own variance versus the between-dimension covariance. The covariance view is the one that speaks directly to leakage across dimensions.
- Claim-rewrite aggressiveness — how far a low signal share downgrades the headline. Aggressive rewriting is honest but can under-sell a genuinely robust share.
When it helps, and when it misleads¶
Its strength is that it turns a fitted model into a decision-ready, honest sentence and makes the shared-source share impossible to ignore — which is exactly what stops dashboard double-counting from surviving into a published report. Its failure mode is that a partition is only as trustworthy as the model beneath it: garbage components in, confident-looking shares out, and a crisp "most of the covariance is shared source" can rest on an unidentified factor. The classic misuse is quoting a variance share as though it were a validated causal decomposition when the underlying model was merely exploratory. The guarding discipline is to always publish shares with their uncertainty, name the model and its assumptions inside the report, and treat the partition as an attribution given a model — not a fact about the world.
How it implements the components¶
residual_claim_boundary— the report is the interpretive firewall in written form: it states which claims survive, which are shared-source artifacts, and which are too weak to keep.precision_weight_update— it carries each share's precision, so the firmness of the attribution (not just its point value) governs how far a claim is downgraded.
It attributes and relabels but runs no test of its own: unlike its nearest twin [Residual Correlation Diagnostic], it does not inspect which pairwise correlations survive (independence_diagnostic_panel); it does not fit the decomposition it reports (common_variance_adjustment_rule) — that's [Common Factor or Random-Effect Model] — and it does not sweep hypothetical contamination levels to find a breakpoint — that's [Leakage Sensitivity Grid].
Related¶
- Instantiates: Shared-Source Variance Isolation — it is the transparency step that turns a decomposition into revised claim language.
- Consumes: Common Factor or Random-Effect Model supplies the fitted variance components this report attributes and communicates.
- Sibling mechanisms: Source Variance Audit Matrix · Multitrait-Multimethod Matrix · Residual Correlation Diagnostic · Negative-Control Outcome Probe · Batch, Rater, or Instrument Counterbalancing Protocol · Leakage Sensitivity Grid
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Variance Partitioning Report operates as an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution because it splits each dimension's variance into true-signal, shared-source, dimension-specific, and noise shares, carries each share's precision, and rewrites the claim to match.
Independent corroboration: The frozen evidence defines Variance Partitioning Report as 'Splits each dimension's variance into true-signal, shared-source, dimension-specific, and noise shares, carries each share's precision, and rewrites the claim to match', so its operative form is Analysis, Modeling & Optimization.
Nearest alternative: Representation, Specification & Plan — Variance Partitioning Report 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: Single lineage
Present-day reach: Universal
Rationale: Both independent reviews identify statistics experimental design as the historical home of the operation—Splits each dimension's variance into true-signal, shared-source, dimension-specific, and noise shares, carries each share's precision, and rewrites the claim to match.. The retained alternates document formative adjacent traditions; the reach field, not the origin field, carries later applicability.
Related originating lineages:
- Data Science & Analytics — Data science's modeling, validation, and monitoring tradition contributes a separate formative lineage to the mechanism's variance partitioning report logic.
- Mathematics — Mathematical modeling, proof, and abstract-structure practice supplies a parallel or contributing lineage for the mechanism's defining operation: splits each dimension's variance into true-signal, shared-source, dimension-specific, and noise shares, carries each share's precision, and rewrites the claim to match.
Review resolution: Both blind reviewers independently place the defining operation—Splits each dimension's variance into true-signal, shared-source, dimension-specific, and noise shares, carries each share's precision, and rewrites the claim to match.—in statistics experimental design. Their queued differences are secondary: alternate_origin_disagreement, origin_mode_disagreement, domain_reach_disagreement, encyclopedia_synthesis_disagreement. Reviewer A uniquely contributes no additional alternate; reviewer B uniquely contributes ['mathematics']. I preserve the full evidence-supported union of 2 alternate domain(s), without a numeric cap. origin_mode=single_lineage reflects the more specific lineage judgment in reviewer B's evidence, while domain_reach=universal separately records present-day portability. The affirmative encyclopedia-synthesis finding is preserved, and confidence=high uses the more conservative reviewer level.
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 intraclass correlation is the fraction of total variance attributable to a grouping factor — the workhorse quantity of variance partitioning. A high ICC for a shared source (rater, batch, pen) is precisely the sign that a supposedly dimension-level pattern is really riding on the group the measurements shared. ↩