Covariance Selection-Term Calculation¶
Statistical estimator — instantiates Selection–Transmission Change Attribution
Isolates the selection channel by computing the covariance between a unit's value and its change in relative weight — a single statistic whose sign says whether high-value units gained share.
Sometimes you only need to know whether the mix moved, and you want it as one interpretable number rather than a full ledger. Covariance Selection-Term Calculation computes exactly the selection channel and nothing else, and it computes it as a covariance: across units, it correlates each unit's value with its change in relative weight against a declared baseline weighting. The defining idea is that the selection term is that covariance — a positive value means high-value units systematically gained share, a negative one means they lost it, and the magnitude is how much of the aggregate move that reweighting drove. It never touches within-unit change, never recomposes to the whole Δmean, and never produces a residual line; it is a single statistic answering a single question.
Example¶
A broad equity index's average dividend yield rose from 2.10% to 2.40% across a year, and an analyst needs to know whether companies raised payouts or whether high-yield names simply grew in the index. The covariance calculation takes each continuing constituent's dividend yield (value) and its change in market-cap weight relative to a proportional baseline, then computes the weighted covariance between them. The result is positive and sizeable: high-yield stocks disproportionately gained index weight over the year. Expressed on the index's scale, the selection channel accounts for roughly +0.22 of the +0.30-point move. That single covariance is enough to say "the index got yieldier mostly because its composition tilted toward payers," without ever asking whether any individual company changed its dividend.
How it works¶
For each unit in the continuing set, take its value and its relative weight in each state; express the weight change against a baseline weighting (equal-share, or the prior period's weights); then compute the weighted covariance of value with relative-weight-change. That covariance is the selection term. The interpretation rides on its sign and size, and it can be re-expressed as a regression slope of weight-change on value when a standardized reading is wanted. The move is deliberately partial: unlike a full ledger it isolates one channel, so it is cheapest when the transmission question is either separately handled or genuinely secondary. The identity that licenses reading this covariance as "selection" is the Robertson–Price form, in which the covariance of value with relative fitness is exactly the selection differential.[n1]
Tuning parameters¶
- Baseline weighting — whether relative-weight-change is measured against equal shares or prior-period weights. This is the counterfactual the covariance is read from; changing it changes what counts as "gaining share."
- Standardization — raw covariance (on the value's scale) versus a correlation or regression slope (unit-free). Standardizing eases cross-case comparison but discards the direct contribution magnitude.
- Entrant handling — whether brand-new units are excluded to keep the covariance defined over a stable set, or admitted with an imputed prior weight. Exclusion keeps the statistic clean but leaves entry effects unmeasured.
- Window — the length of the interval over which weight change is measured. Longer windows smooth noise but blur which sub-period drove the reweighting.
When it helps, and when it misleads¶
Its strength is compression and interpretability: one signed number that directly answers "did high-value units gain weight?", cheap to compute and easy to communicate. When selection is the whole question, it beats standing up a full table.
Its failure mode is exactly its narrowness. A near-zero covariance does not mean "nothing happened" — it can hide large, offsetting within-unit changes the statistic is blind to, so reading it as the whole story is the misuse. And like any decomposition term, a large covariance shows arithmetic attribution, not causation: high-value units gaining share may be an effect of the same force that raised the aggregate, not its cause. The guarding discipline is to always pair it with a transmission measurement before drawing a population-level conclusion, and to report the baseline it was computed against, since the number is meaningless without it.
How it implements the components¶
unit_value_measure— the value variable entering the covariance, recorded comparably across states.unit_weight_measure— supplies the weights whose relative change is the covariance's second variable.selection_term_definition— the covariance of value with relative-weight-change is the selection term, delivered as a single statistic.counterfactual_weight_or_value_baseline— the baseline weighting against which "change in relative weight" is defined; the covariance is measured as departure from it.
It computes no within-unit change (transmission_term_definition — that is the Within-Unit Change Assay) and it builds no recomposing identity (decomposition_identity_and_residual_check — that is the Price Equation Decomposition Table). Its nearest twin is that table: this calculation yields only the selection term as one covariance statistic against a baseline, while the table produces both terms and recomposes them exactly.
Related¶
- Instantiates: Selection–Transmission Change Attribution — it supplies the selection channel in its most compressed, statistical form.
- Consumes: Lineage or Panel Correspondence Matrix for the continuing set over which the covariance is taken.
- Sibling mechanisms: Price Equation Decomposition Table · Within-Unit Change Assay · Lineage or Panel Correspondence Matrix · Composition-vs-Transformation Dashboard · Entry/Exit Normalization Protocol · Decomposition Residual Reconciliation Workflow · Selection–Transmission Sensitivity Analysis
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Covariance Selection-Term Calculation operates as a computation, comparison, model, or analytic representation used to infer, estimate, or choose because it isolates the selection channel by computing the covariance between a unit's value and its change in relative weight — a single statistic whose sign says whether high-value units gained share.
Independent corroboration: The frozen evidence defines Covariance Selection-Term Calculation as 'Isolates the selection channel by computing the covariance between a unit's value and its change in relative weight — a single statistic whose sign says whether high-value units gained share', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Biology & Ecology
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Specialized
Rationale: Evolutionary theory cohered the Price-equation selection term as covariance between a unit's trait value and its change in relative reproductive weight.
Related originating lineages:
- Statistics & Experimental Design — Covariance algebra and estimation supply the statistic, weighting baseline, sign, and uncertainty calculation.
Review resolution: Both reviewers agree on evolutionary biology and statistics. The Price-equation selection term is specifically biological, but it is formally constituted by covariance estimation, supporting cross-disciplinary origin and specialized reach.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] The Robertson–Price identity (Alan Robertson's secondary theorem of natural selection) states that the change in a mean trait due to selection equals the covariance between the trait and relative fitness. It is what licenses reading a covariance of value with relative-weight-change as the selection channel specifically. ↩