Crossover Interaction¶
Core Idea¶
A crossover interaction occurs when the effect of one input reverses direction across levels of a second condition. The same treatment, signal, event, or perturbation raises an outcome in one state and lowers it in another. Consequently, neither input has a context-free sign, and reporting only an averaged main effect can conceal or even cancel the operative mechanism.
Broad Use¶
The pattern appears in factorial experiments, genotype-by-environment response, receptor pharmacology, state-dependent memory, social evaluation, and engineered systems whose response to an input depends on operating mode.
Clarity¶
The prime separates ordinary effect modification — larger here, smaller there — from true directional reversal. It also distinguishes a real conditional effect from an aggregate reversal caused only by changing subgroup weights.
Manages Complexity¶
Instead of maintaining unrelated rules for every case, the analyst records two inputs, their conditional-effect table, and the zero-crossing. This prevents a near-zero average from being mistaken for “no effect.”
Abstract Reasoning¶
Estimate the effect of X separately at each level of Z. If the conditional effects have opposite signs, preserve the interaction and stop interpreting the marginal main effect as a portable intervention rule. Interventions must specify the state into which the input will land.
Knowledge Transfer¶
A minor blunder helping a highly competent agent but hurting a mediocre one, a partial agonist raising low receptor tone but lowering a full-agonist-driven state, and a retrieval state helping memory only when it matches the encoding state share the same conditional-sign table.
Example¶
If an intervention changes an outcome by +3 in state A and -2 in state B, a balanced
sample reports a main effect of only +0.5, and a different mixture can report zero or a
negative average. The stable fact is the crossover: effect(X | A) > 0 and
effect(X | B) < 0.
Relationships to Other Abstractions¶
Current abstraction Crossover Interaction Prime
Parents (1) — more general patterns this builds on
-
Crossover Interaction is a kind of Synergy and Antagonism Prime
Crossover Interaction is the sign-reversing species of interaction effect in which one factor's conditional effect is positive at one level of a second factor and negative at another.
Children (3) — more specific cases that build on this
-
Partial Agonist Domain-specific is a decomposition of Crossover Interaction
Partial Agonism is the receptor-pharmacology form of a crossover interaction because the same ligand raises response when acting alone and lowers it when displacing a stronger full agonist.
-
Pratfall Effect Domain-specific is a decomposition of Crossover Interaction
The Pratfall Effect is the social-evaluation form of a crossover interaction because the same minor blunder raises liking under a high competence prior and lowers it under a mediocre prior.
-
State-Dependent Learning Domain-specific is a decomposition of Crossover Interaction
State-Dependent Learning is the memory form of a crossover interaction because each retrieval state helps material encoded in the matching state and hurts relative recall of material encoded in the other state.
Hierarchy path (1) — routes to 1 parentless root
- Crossover Interaction → Synergy and Antagonism → Nonlinearity
Not to Be Confused With¶
- Synergy and Antagonism: the broader interaction-effect family; crossover is its sign-reversing species.
- Simpson's Paradox: an aggregate association can reverse because subgroup weights change even when no conditional effect reverses.
- Bayesian Updating: a fixed likelihood ratio changes posterior magnitude but does not switch from supporting to opposing a hypothesis merely because the prior changes.
- Nonlinearity: the broad failure of additivity or homogeneity; most nonlinearities do not cross zero across a second condition.
Notes¶
(Pending Claude style and citation pass; mathematical identity, aliases, and DAG placement should be preserved.)