Selection Vs Transmission Decomposition

Prime #

1166

Origin domain

Evolutionary Dynamics

Subdomain

population change accounting → Evolutionary Dynamics

Core Idea

A change in the weighted mean of any trait splits exactly into a selection term — the covariance of the trait with each unit's growth weight, the change from differential weighting — and a transmission term — the share-weighted within-unit change, the change from units changing themselves. The Price equation is its canonical statement.

How would you explain it like I'm…

Grew vs. Swapped

Imagine a class's average height goes up. Maybe the same kids all grew taller, OR maybe some short kids left and tall kids joined — nobody grew, the mix just changed. The average can move for two totally different reasons, and you can split the change into 'who's here' and 'how each one changed.'

Who Counts More Vs Who Changed

When the average of something in a group changes, this prime says there are exactly two separate reasons mixed together, and it splits them apart cleanly. Reason one, called selection: which members count for more changed — the bigger or faster-growing ones got more weight, even if no member changed. Reason two, called transmission: the members themselves changed, each one shifting a bit. A team's average score can rise because the high scorers played more games (selection) or because each player got better (transmission), and those call for different fixes. The split is exact, not a guess, and you can even break each part down again into smaller groups.

Reweighting Versus Changing

Selection Vs Transmission Decomposition cuts any change in a population's weighted average — a trait, score, productivity, or return — into two structurally distinct parts. The first part, selection, is how much the average moved because units with higher trait values gained more weight (got reproduced, copied, or scaled more); formally it's the covariance between each unit's weight and its trait. The second part, transmission, is how much moved because the units themselves changed, averaged by their share. The split is an exact identity whenever you can measure units, weights, and trait values, and it's recursive — each term re-splits by sub-grouping. This matters because a bare average hides which mechanism is running, and the two demand different interventions: changing who gets weighted is not changing what each unit does.

 

Selection vs transmission decomposition splits a change in the population mean (or weighted average) of any trait into two structurally distinct contributions. The first is the covariance of the trait with each unit's growth or replication weight, divided by mean weight — the change produced by differentially reweighting existing units (selection). The second is the share-weighted expectation of each unit's within-unit transformation — the change produced by the units themselves changing (transmission). Formally, the change in the weighted mean equals the covariance of weight with trait over mean weight, plus the weighted expectation of within-unit change; the Price equation is the canonical statement. The decomposition is exact whenever the population is partitioned into measurable units with measurable weights and trait values, and it is recursive, since each term can itself be re-decomposed by sub-grouping. The defining commitment is that an apparent average change in a heterogeneous population can arise from two distinct mechanisms whose intervention implications diverge, and the decomposition is the diagnostic that tells them apart. Where a bare aggregate hides which mechanism is operating, this identity-backed cut recovers it.

Broad Use

• Evolutionary biology: the Price equation splitting mean phenotype change into selection (covariance with fitness) and transmission.
• Firm productivity: the Foster-Haltiwanger-Krizan decomposition — within-firm growth plus between-firm reallocation plus entry/exit.
• Portfolio attribution: Brinson attribution — allocation effect plus within-class security selection.
• Organizational learning: average skill change splitting into hiring and promotion (selection) plus within-person learning.
• Demography: life-expectancy and mortality decompositions — composition shift plus within-region change.
• Machine learning: fleet effects — high-accuracy models retained plus within-model fine-tuning.

Clarity

Makes visible a question the bare aggregate hides: did the units change, or did the weighting of units change? — and the two have completely different intervention implications.

Manages Complexity

Compresses "did the mean change because of X or Y" into one structural cut, recursively re-decomposable by sub-grouping, with no residual — so selection and transmission are guaranteed exhaustive.

Abstract Reasoning

Supports exact counterfactuals — if selection had been zero..., if within-unit change had been zero... — that are not estimates but the two halves of an identity, blocking the error of reading an aggregate rise as evidence the units improved.

Knowledge Transfer

• Biology to economics: FHK is the Price equation for firms — identical algebra, identical within-versus-between logic.
• Biology to finance: Brinson is the Price equation for portfolios, with the same "which margin do I work?" conclusion.
• Any substrate: the role-mapping (unit, weight, trait, selection, transmission) is fixed, so a practitioner applies the cut immediately in a new field.

Example

A 5% rise in aggregate productivity could be entirely within-firm improvement, entirely reallocation toward already-best firms, or any mix — and funding within-firm training when reallocation dominates wastes the money the decomposition would have saved.

Relationships to Other Primes

Parents (1) — more general patterns this builds on

• Selection Vs Transmission Decomposition is a kind of Decomposition — An EXACT, residual-free specialization of decomposition: a weighted-mean change split identically into covariance-of-weight-with-trait (selection) + share-weighted within-unit change (transmission). The Price equation is its generator; FHK and Brinson are the same identity for firms/portfolios. Dossier-confirmed specialization edge.

Path to root: Selection Vs Transmission Decomposition → Decomposition

Not to Be Confused With

• Selection Vs Transmission Decomposition is not Selection Bias because the former's selection term is an exact, intended accounting contribution, whereas selection bias is a distortion to correct.
• Selection Vs Transmission Decomposition is not generic Decomposition because the former is the specific, residual-free, identity-backed split, whereas decomposition is the general method.
• Selection Vs Transmission Decomposition is not Competition because the former accounts for net reweighting arithmetically, whereas competition is a substantive process of contention.