Future developments interact; a useful forecast makes those interactions inspectable
Cross-impact analysis asks how the occurrence or state of one development changes the plausibility, timing or consequences of another, revealing reinforcing chains, tensions and scenario structure.
Many forecasts treat developments one at a time: energy prices rise, a technology matures, regulation changes, a workforce shortage grows. In practice, each development can alter the conditions for the others. Cross-impact analysis places those conditional influences in a matrix so hidden dependencies become discussable.
The matrix is not a causal oracle. Its entries are judgements or modelled relationships whose direction, scale, timing and evidence must be stated. The value comes from disciplined comparison, consistency checking and sensitivity—not from turning uncertain futures into decorative decimal points.
Chapter contents
- 1. Define the decision, horizon and system boundary
- 2. Define events or states so they can interact
- 3. Separate baseline plausibility from conditional influence
- 4. Score direction, magnitude, timing and evidence
- 5. Build the directional matrix without double counting
- 6. Worked example: cooling, power and public health
- 7. Trace cascades, feedback and inhibiting loops
- 8. Generate scenarios from consistent combinations
- 9. Check probability and logical consistency
- 10. Test sensitivity, disagreement and structural alternatives
- 11. Combine evidence and expert judgement transparently
- 12. Connect findings to action, monitoring and revision
Choose a reading route
- Route 1: Frame the system — decision, horizon and event set
- Route 2: Score influence — direction, condition and scale
- Route 3: Work a matrix — cascades and scenario paths
- Route 4: Challenge results — consistency, sensitivity and dissent
- Route 5: Use and revise — monitoring, action and governance
How to use this guide
Start with the chapters that match your immediate decision, then return to the worked example and audit chapters. Each chapter separates a model from the evidence needed to use it responsibly. A claim is not strengthened merely by being technical; it is strengthened when assumptions, observations, alternatives and limits can be inspected.
| Layer | Question | Minimum record |
|---|---|---|
| Concept | What exactly is being represented? | Definitions and boundary |
| Method | How will the reasoning be carried out? | Steps, inputs and assumptions |
| Transfer | Would it work in a changed setting? | A deliberately altered case |
| Audit | What could prove it wrong? | Evidence, limits and revision trigger |
Chapter 1. Define the decision, horizon and system boundary
The idea
Cross-impact analysis is useful when a decision depends on interacting developments over a defined future horizon. The boundary should be wide enough to capture key dependencies but small enough to analyse.
A disciplined method
Write the decision, time horizon, geography, stakeholders and external conditions; list what sits inside the matrix and what remains an exogenous driver.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which interactions could materially change the decision. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A city explores cooling, electricity and public-health developments through 2035 because the investment plan spans those systems.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Move the horizon to two years and remove slow-moving developments that cannot plausibly change in time.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Boundary choices can hide external shocks or create an unmanageable matrix.
Evidence to retain: decision brief, horizon, boundary map, exclusions and review triggers
Common misconception: A larger matrix is always more comprehensive.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 2. Define events or states so they can interact
The idea
Rows and columns need unambiguous developments: binary events, threshold crossings or mutually exclusive states. Vague trends make conditional judgements impossible to interpret.
A disciplined method
Give each item an observable definition, baseline, time window and occurrence rule; split compound items and avoid overlapping events.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is what counts as occurrence or a state change. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
“Grid storage becomes affordable” is defined by a cost threshold, deployment scale and date rather than by general optimism.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Change the cost threshold and see whether influence judgements still refer to the same phenomenon.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Binary events can oversimplify gradual change and mutually dependent definitions.
Evidence to retain: event dictionary, thresholds, data sources, exclusions and version history
Common misconception: Short event labels are sufficient definitions.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 3. Separate baseline plausibility from conditional influence
The idea
A baseline estimate describes an event before specified matrix events are considered. Cross-impact entries describe how another event changes that estimate or state, not how likely either event is on its own.
A disciplined method
Elicit or model baselines first, then ask P(B|A) or a qualitative equivalent while keeping direction explicit and testing the reverse separately.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is how A changes B compared with B’s baseline. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Baseline probability of heat-pump adoption is 0.45; cheap storage raises it to 0.60, while adoption may have a different effect on storage deployment.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Lower the baseline without changing the mechanism and observe how absolute and relative impacts differ.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Conditional probabilities can be difficult to elicit and may violate joint-probability constraints.
Evidence to retain: baseline estimate, conditional estimate, direction, evidence and elicitor rationale
Common misconception: If A strongly affects B, B equally affects A.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 4. Score direction, magnitude, timing and evidence
The idea
Influence can be reinforcing or inhibiting, immediate or delayed, direct or mediated. A single signed number hides these distinctions unless the scale is defined.
A disciplined method
Use an anchored scale or conditional probabilities, record lag and mechanism, cite evidence and include an unknown option rather than forcing a score.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which pathway carries the influence and when. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Water restrictions accelerate efficiency investment within one year but inhibit water-intensive industry over five years.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Change the assumed policy enforcement and rescore the pathway rather than the label alone.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Scores from different mechanisms or horizons may not be comparable.
Evidence to retain: scale anchors, mechanism note, lag, evidence, confidence and dissent
Common misconception: A precise score is stronger evidence than a well-documented range.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 5. Build the directional matrix without double counting
The idea
The matrix places influencing developments on one axis and affected developments on the other. Diagonal cells usually have no direct interpretation, and mediated effects should not be entered again as direct impacts without evidence.
A disciplined method
Complete rows systematically, compare reciprocal cells, flag common causes and distinguish direct influence from downstream cascade.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is whether an entry is direct, indirect or confounded. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A skills shortage directly slows retrofits; slower retrofits later affect energy demand, but the second pathway is not copied into the first cell.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Add a common funding shock and reassess which apparent pairwise links were confounded.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Pairwise structure cannot fully represent nonlinear combinations or institutional feedback.
Evidence to retain: matrix version, cell rationales, directness flags, confounders and reviewer notes
Common misconception: Every observed association belongs as a direct cross-impact.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 6. Worked example: cooling, power and public health
The idea
A small fictional matrix shows how conditional influence changes scenario structure without pretending to predict one certain future.
A disciplined method
Define six developments, score directional impacts on a -3 to +3 scale, calculate row and column summaries, trace two-step chains and write narrative explanations for the strongest cells.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which developments are active drivers, dependent outcomes or critical links. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Affordable storage supports reliable cooling; widespread cooling raises peak demand; building retrofits reduce demand and heat exposure. The loop creates both reinforcement and constraint.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Introduce a prolonged fuel-price shock and update only pathways whose mechanisms change.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: The example is exploratory; scores are not measured causal effects or calibrated probabilities.
Evidence to retain: definitions, matrix, rationales, calculations, alternative scores and scenario notes
Common misconception: The highest row sum identifies the policy to implement.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
| Development | Strong influence | Important dependency |
|---|---|---|
| Affordable storage | Grid reliability | Supply chains |
| Building retrofits | Lower cooling demand | Workforce capacity |
| Heat alerts | Protective behaviour | Trust and access |
| Peak demand | Grid investment | Weather and adoption |
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 7. Trace cascades, feedback and inhibiting loops
The idea
Two-step and longer paths reveal cascades that pairwise reading misses. Feedback can amplify or stabilise change, while opposing paths can cancel or dominate at different times.
A disciplined method
Trace only substantively plausible paths, multiply or combine scores using a declared rule, inspect signs and delays, and return to the mechanisms before interpreting totals.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which path transmits the strongest conditional change. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Storage enables cooling, cooling raises demand and demand accelerates grid investment; retrofits create a balancing path by reducing demand.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Delay grid investment by five years and see how the reinforcing loop becomes a short-term vulnerability.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Matrix algebra can create spurious paths and cannot prove feedback strength.
Evidence to retain: path list, combination rule, lags, mechanism review and excluded paths
Common misconception: Every long path found mathematically is a meaningful cascade.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 8. Generate scenarios from consistent combinations
The idea
Cross-impact results can help identify combinations of developments that reinforce one another or create tension. Scenarios should remain plural and internally explained.
A disciplined method
Select high-impact uncertainties, form candidate combinations, use consistency scores or probabilistic simulation, reject only with documented logic and write causal narratives.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which combinations are plausible enough to explore. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
One scenario combines strong retrofits and low peak strain; another combines rapid cooling adoption with delayed grid investment and frequent outages.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Add a policy reversal and test which scenario assumptions remain coherent.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Consistency is not likelihood, and rare inconsistent-looking combinations can occur after shocks.
Evidence to retain: scenario rules, combinations, consistency checks, narratives and residual uncertainties
Common misconception: The most internally consistent scenario is the forecast.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 9. Check probability and logical consistency
The idea
Conditional estimates can contradict Bayes rules, event definitions or each other. Qualitative scores can also imply impossible or circular claims.
A disciplined method
Run bounds and coherence checks, compare reciprocal conditions, review mutually exclusive events and revise estimates with elicitors rather than silently normalising them.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which entries cannot belong to one plausible joint account. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Experts judge A makes B almost certain but also judge B makes A much less likely despite no timing or selection explanation; the pair is returned for clarification.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Split one compound event and see whether the contradiction disappears.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Consistency algorithms impose assumptions and cannot select the true judgement automatically.
Evidence to retain: constraint report, flagged cells, discussion, revisions and unresolved inconsistencies
Common misconception: Mathematical adjustment converts inconsistent opinions into truth.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 10. Test sensitivity, disagreement and structural alternatives
The idea
Results can hinge on a few uncertain cells, scale choices or event definitions. Group averages may hide polarised causal stories.
A disciplined method
Perturb influential cells, compare individual or subgroup matrices, change aggregation rules and test an alternative network structure.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which assumptions change driver rankings or scenario membership. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A driver ranking reverses when one disputed policy-to-adoption score moves from +2 to 0, so the report highlights that link instead of the ranking alone.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Remove the most central event and see whether another pathway carries the cascade.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Sensitivity ranges still depend on chosen alternatives and may miss omitted variables.
Evidence to retain: parameter sweeps, subgroup matrices, ranking stability, structural variants and dissent
Common misconception: A stable average matrix means participants agree.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 11. Combine evidence and expert judgement transparently
The idea
Some cells can use empirical models; others rely on expert or stakeholder judgement. Their provenance and uncertainty should remain visible rather than blended into one authority.
A disciplined method
Label evidence type, calibrate scales, use structured elicitation, disclose conflicts and keep unsupported cells as unknown or ranges.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which entries rest on observation, modelling or judgement. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Power-demand elasticity uses operational data, while future regulatory response uses a Delphi panel; the matrix preserves the distinction.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Replace expert judgement with new evidence and update the cell plus downstream scenarios.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Historical data may not transfer to structural change, and panels may share blind spots.
Evidence to retain: source type, data window, model, elicitation protocol, uncertainty and update history
Common misconception: Putting all evidence on one numeric scale makes it equally reliable.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 12. Connect findings to action, monitoring and revision
The idea
Cross-impact analysis supports robust choices, signposts and contingency plans. It should not end with a static matrix.
A disciplined method
Identify actions useful across scenarios, assign indicators to critical links, define thresholds for review and update the matrix when events occur or mechanisms change.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which monitored relationship would alter today’s decision. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A city funds no-regret retrofit training and watches storage costs, heat-related outages and adoption; a threshold triggers capacity-plan revision.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Observe an event occurring earlier than expected and update baselines, conditional paths and decisions.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Monitoring can become confirmation-seeking if disconfirming indicators are excluded.
Evidence to retain: action map, indicators, owners, thresholds, observations, decisions and matrix versions
Common misconception: Once an event occurs, only its row needs updating.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Field workbook
Choose six non-sensitive future developments around one decision. Define each event, set baselines, score directional pairwise effects, document the strongest mechanisms, trace one reinforcing and one balancing path, then build two distinct scenarios and a sensitivity test.
- Write the decision question in one sentence and name the decision owner.
- List observations separately from interpretations, with dates and provenance.
- Record at least two plausible alternatives before choosing a preferred account.
- Run the changed-context test and note which assumptions fail.
- State the limit, confidence and next review trigger in plain language.
| Layer | Evidence | Failure signal |
|---|---|---|
| Definitions | Observable events and horizon | Vague overlapping trends |
| Influence | Directional mechanism and lag | Symmetric guess |
| Consistency | Bounds and logic checks | Impossible conditions |
| Decision | Scenarios, signposts and actions | Matrix as final forecast |
Continue through the verified ecosystem
- Scenario analysis — morphological spaces and cross-consistency
- Delphi method — structured expert elicitation and dissent
- Horizon scanning — weak signals and emerging developments
- Indications and warning — signposts, thresholds and surprise
- Assumption-based planning — shaping actions and hedges
Frequently asked questions
What is cross-impact analysis?
A family of methods for examining how one future event or state changes the plausibility or effect of others.
Must the matrix use probabilities?
No. Some methods use conditional probabilities; others use anchored qualitative influence or consistency scores.
Is influence symmetric?
No. A may strongly affect B while B has little or a different effect on A.
Does a high row score identify the best policy?
No. It indicates modelled influence, not desirability, feasibility, rights or evidence quality.
Why check consistency?
Conditional estimates and event definitions can contradict probability rules or each other.
Is the most consistent scenario the most likely?
No. Consistency addresses internal fit; likelihood and decision value require separate judgement.
Authoritative sources and further reading
- Gordon and Hayward: Initial Experiments with the Cross-Impact Matrix Method — foundational primary paper
- Weimer-Jehle: Cross-Impact Balances — system-theoretical consistency method
- UK Government Office for Science: Futures Toolkit — current official foresight practice
- Millennium Project: Futures Research Methodology — cross-impact and related methods
- Methods of Future and Scenario Analysis — cross-impact and scenario method guidance
This guide explains an exploratory foresight method. Matrix scores and scenarios are conditional judgements, not causal proof, calibrated forecasts or instructions to act without wider evidence and accountability.