Interaction Effect Mapping¶
Essence¶
Interaction Effect Mapping is the practice of making combinations visible. It applies when factors that look understandable in isolation may behave differently when they coexist. The archetype asks a simple structural question: what changes when A and B happen together?
The answer may be synergy, antagonism, interference, independence, threshold dependence, saturation, or context-specific reversal. The point is not merely to produce an analysis table. The map should change design decisions: what to combine, what to separate, what to sequence, what to monitor, and what to stop assuming is additive.
Compression statement¶
When multiple factors are present or planned together and their combined effect may be non-additive, identify the factors, test or infer their combinations, classify the interaction effects, and redesign the intervention around the mapped pattern of synergy, antagonism, interference, independence, or conditional dependence.
Canonical formula: effect(A + B) is not assumed to equal effect(A) + effect(B); map delta_interaction = observed_combined_effect - expected_additive_effect.
When to Use This Archetype¶
Use this archetype when multiple interventions, components, policies, actors, incentives, or conditions will operate together and isolated evidence is not enough. It is especially useful before scaling a bundle, deploying a multi-part program, combining system features, or interpreting inconsistent results across contexts.
It also fits situations where a component is praised or blamed even though its effect may depend on what it was paired with. For example, a coaching program may look weak unless paired with timely feedback, while a retry mechanism in infrastructure may become harmful when paired with aggressive autoscaling under load.
Do not use this archetype merely because there are many parts. The parts must plausibly modify one another's effects. A simple inventory, dependency list, compatibility checklist, or one-factor test is not enough.
Structural Problem¶
The structural problem is hidden non-additivity. The system is being reasoned about as if each factor contributes its own separable effect, but the real outcome depends on the relation among factors.
This creates several recurring errors. Teams overestimate bundles because each part worked alone. They miss synergies because the right pair was never tested. They blame one factor for failure when the problem was the pairing. They generalize evidence from one context to another without noticing that a co-factor has changed.
Interaction Effect Mapping addresses that problem by shifting the unit of evaluation from isolated factors to factor combinations.
Intervention Logic¶
The intervention begins by naming the factors clearly. Those factors may be treatments, design features, resources, incentives, environmental conditions, organizational roles, or technical components. The map then establishes a baseline expectation: what would we expect if effects were merely additive or independent?
Next, the relevant combinations are selected. In small systems, full combination testing may be possible. In larger systems, the map may focus on pairwise combinations, high-risk combinations, high-frequency combinations, or combinations suggested by theory and prior failures.
The combined effects are then measured, simulated, inferred, or reviewed. The result is classified into actionable categories: reinforcing, antagonistic, interfering, independent, conditional, saturating, or uncertain. Finally, each category is connected to a design update: combine, separate, sequence, buffer, monitor, reduce intensity, or abandon.
Key Components¶
Interaction Effect Mapping shifts the unit of evaluation from isolated factors to factor combinations, exposing the non-additivity that single-variable thinking hides. The Factor List names the specific candidates for interaction — treatments, design features, incentives, or technical components precise enough to test or reason about — so the analysis does not dissolve into impressionistic talk about everything in the system. The Interaction Matrix is the structured representation of relations among those factors, recording which combinations have been considered and what is known about each. Effect Measurement defines the outcome being compared and grounds the map in evidence rather than impression, ideally tagging confidence, source, and time horizon since the same combination may improve one measure while worsening another.
Three further components convert observation into design. Interaction Type Classification sorts each pairing into actionable categories — reinforcing, antagonistic, interfering, independent, conditional, saturating, or uncertain — because raw observations without classification remain a collection rather than a guide. A Compatibility Check screens whether a given combination can be safely and meaningfully tested or deployed, acting as a guardrail that runs before interaction mapping rather than substituting for it. The Design Update Rule connects the map to action by specifying what to do when synergy, antagonism, interference, neutrality, or uncertainty is found; without this linkage the archetype collapses into analysis without intervention. Finally, the Review or Iteration Rule keeps the map honest over time, since scale, population, background conditions, and incentives can all change the interaction pattern after the initial study.
| Component | Description |
|---|---|
| Factor List ↗ | The factor list defines what may interact. A good factor list is specific enough to test or reason about. “Support” is usually too vague; “weekly coaching call,” “automated reminder,” or “peer accountability group” is more useful. The factor list prevents the map from becoming an impressionistic discussion about everything in the system. |
| Interaction Matrix ↗ | The interaction matrix is the structured relation map. It records which combinations have been considered and what is known about their combined effect. It may be visualized as a table, graph, dependency map, or experiment design, but the component is the structured representation of relations, not the display artifact alone. |
| Effect Measurement ↗ | Effect measurement defines what outcome is being compared. The same combination may improve one measure while worsening another, so measurement must match the purpose of the intervention. In uncertain settings, effect measurement should also show confidence, evidence source, and time horizon. |
| Interaction Type Classification ↗ | Classification turns observations into design meaning. A combination that reinforces value invites bundling. A combination that cancels value invites separation or redesign. A combination that works only under a threshold invites tuning. Without classification, the map is just a collection of observations. |
| Compatibility Check ↗ | A compatibility check screens whether a combination can safely and meaningfully be tested or deployed. It is a guardrail, not the archetype itself. Compatibility asks whether elements can coexist; interaction mapping asks how coexistence changes their effects. |
| Design Update Rule ↗ | The design update rule connects the map to action. It defines what to do when the map shows synergy, antagonism, interference, neutrality, or uncertainty. This component is what keeps the archetype from becoming analysis without intervention. |
| Review or Iteration Rule ↗ | Interaction maps can expire. Scale, population, background conditions, technology, incentives, and load can all change the interaction pattern. A review rule defines when to revisit the map and what signals should trigger revision. |
Common Mechanisms¶
| Mechanism | Description |
|---|---|
| Factorial Experiment ↗ | A factorial experiment is a formal way to test factors and combinations. It can estimate main effects and interaction effects, making it a powerful mechanism when factors can be varied systematically and ethically. |
| Pairwise Combination Testing ↗ | Pairwise testing examines two-factor combinations when full mapping is too expensive. It is a pragmatic implementation of the archetype, but it should not be mistaken for complete interaction knowledge. Pairwise-safe does not always mean bundle-safe. |
| Interaction Matrix Table ↗ | An interaction matrix table is a visible artifact for recording combinations, expected effects, observed effects, confidence levels, and recommended actions. It helps teams reason together, but the table is only the representation; the archetype is the mapping-and-redesign logic. |
| Compatibility Screening ↗ | Compatibility screening removes unsafe, invalid, infeasible, or uninterpretable combinations before testing. It is often necessary in safety-sensitive settings, but it is narrower than interaction mapping because it does not explain the full range of positive, negative, neutral, or conditional combined effects. |
| Dependency Interaction Map ↗ | A dependency interaction map shows how services, processes, teams, or resources modify one another's behavior through dependency pathways. It is useful in infrastructure, operations, supply chains, and organizational workflows. |
| Treatment Interaction Analysis ↗ | Treatment interaction analysis evaluates whether an intervention's effect changes under different co-treatments, populations, moderators, or conditions. It is common in clinical, behavioral, social, and policy contexts. |
| Design of Experiments Protocol ↗ | A design of experiments protocol plans factor levels, combination coverage, randomization, blocking, and measurement windows. It is useful when the mapping effort needs statistical discipline and efficient coverage of a large design space. |
Parameter / Tuning Dimensions¶
The main tuning dimension is scope: pairwise, selected bundle, full factorial, or higher-order mapping. A narrow scope is faster and cheaper; a broad scope is more complete but can become infeasible.
A second dimension is evidence strength. Some maps are experimental, some are observational, some are simulation-based, and some are expert-elicited. The draft should not pretend those have equal confidence.
A third dimension is factor granularity. Coarse factors make mapping manageable but vague. Fine-grained factors make mapping precise but may create combinatorial explosion.
Other important tuning dimensions include outcome metric, time horizon, context stratification, risk threshold, review cadence, and the action threshold for redesign.
Invariants to Preserve¶
Preserve clear factor identities. If the factors are vague, the map cannot explain what is interacting.
Preserve an explicit baseline. Interaction means deviation from an expected isolated or additive effect; without a baseline, every change can look like an interaction.
Preserve evidence quality. A suspected interaction is not the same as a tested interaction, and a local interaction is not necessarily universal.
Preserve action linkage. The map should inform combination, separation, sequencing, buffering, monitoring, or retirement. Otherwise it is analysis without intervention.
Target Outcomes¶
A successful interaction map reveals beneficial combinations that should be reinforced, harmful combinations that should be avoided or controlled, and neutral combinations that need not receive special attention.
It also improves attribution. Teams stop saying “this intervention worked” or “this feature failed” when the real explanation is that it worked with one co-factor and failed with another.
The strongest outcome is combination-aware design: the system is deliberately shaped around known or suspected interaction patterns rather than assembled from individually attractive parts.
Tradeoffs¶
Interaction mapping trades simplicity for realism. It gives a more truthful view of combined behavior, but it adds complexity, evidence burden, and decision overhead.
It also trades completeness against feasibility. Full combination testing may be impossible. Reduced mapping may be necessary, but it can miss higher-order interactions.
Finally, it can trade optimization against robustness. Designing around a synergy can improve performance but also create dependency on a fragile combination.
Failure Modes¶
The most common failure mode is spurious interaction mapping: noise or confounding is mistaken for a real relation. This can be mitigated through replication, confidence ratings, causal guardrails, and clear evidence labels.
Another failure mode is combinatorial explosion. Too many factors can make the map unusable. Prioritize high-impact, high-risk, high-frequency, theory-relevant, or irreversible combinations.
A third failure mode is pairwise blindness. Pairwise tests are useful, but they can miss three-way or higher-order interactions. Use targeted bundle testing where higher-order effects are plausible.
A fourth failure mode is static-map drift. A map made at one scale or in one context may fail later. Build in review triggers.
The final failure mode is analysis without action. Every interaction category should point to a design response.
Neighbor Distinctions¶
Interaction Effect Mapping is distinct from Catalytic Pairing. Catalytic Pairing is a downstream design move that deliberately uses amplification; Interaction Effect Mapping first discovers whether amplification exists.
It is distinct from Synergistic Bundle Design. Bundle design constructs a package around reinforcing effects; interaction mapping diagnoses whether proposed combinations reinforce, cancel, interfere, or merely coexist.
It is distinct from Antagonism Screening and Separation. Antagonism screening specializes in harmful negative interactions; interaction mapping covers the full interaction space, including positive, neutral, conditional, and uncertain relations.
It is distinct from Compositional Assembly. Assembly asks how to build a whole from parts; interaction mapping asks how the parts change one another's effects when combined.
It is distinct from a compatibility matrix. Compatibility tells whether coexistence is allowed or feasible. Interaction mapping tells what coexistence does.
Variants and Near Names¶
Factorial Interaction Mapping is the formal experimental variant. It uses factorial or fractional-factorial designs to estimate interaction terms.
Pairwise Interaction Mapping is the tractable approximation variant. It maps two-factor interactions when full combination testing is infeasible.
Dependency Interaction Mapping is a systems variant. It is useful when processes, services, resources, or teams change one another's effects through dependency pathways.
Near names include combined-effect mapping, factor interaction mapping, treatment interaction analysis, dependency interaction maps, and combination effect screening. These should generally point back to this archetype unless they become formal methods in a separate experimental-design family.
Cross-Domain Examples¶
In public policy, a housing package may include subsidies, zoning reform, construction incentives, and tenant protections. The value of any one policy depends on what else is present, so the package should be mapped for reinforcement and interference before scale.
In infrastructure, caching, retries, rate limiting, and autoscaling may each improve reliability alone. Under load, however, their combined behavior can either stabilize the system or amplify failure. Interaction mapping identifies which combinations require tuning or separation.
In education, tutoring, adaptive software, parent communication, and assessment cadence may reinforce one another for some students while creating overload for others. Mapping interactions by context helps avoid one-size-fits-all conclusions.
In organizational design, incentives, autonomy, reporting lines, and review cadence often modify one another's effects. A performance system that works under one governance arrangement may backfire under another.
Non-Examples¶
A single A/B test of one feature against a control is not Interaction Effect Mapping unless it examines how the feature's effect changes with another factor.
A dependency inventory is not Interaction Effect Mapping unless it evaluates how dependencies modify outcomes when active together.
A compatibility checklist is not Interaction Effect Mapping by itself. It may prevent invalid combinations, but it does not explain synergy, antagonism, interference, or conditional effects.
A copied best-practice bundle is not Interaction Effect Mapping if the local combination effects are not examined.