Cross-Impact Balance Analysis
- Introduced by
- Wolfgang Weimer-Jehle
- Year
- 2006
- Type
- Semi-quantitative scenario-construction pipeline
- Family
- Cross-impact / systematic scenario analysis
What does it do?
Cross-Impact Balance (CIB) analysis is a semi-quantitative foresight method that turns a panel of qualitative expert judgments into a small set of internally consistent scenarios.
Introduced by Wolfgang Weimer-Jehle in 2006, CIB describes a system as a set of descriptors, each of which can take one of several discrete future states, and asks experts to judge, pairwise, how strongly each state promotes or restricts every other state. These judgments form a cross-impact matrix; a balance algorithm then searches the combinatorial space of state combinations for configurations in which every descriptor's chosen state is the one most strongly supported by all the others. These self-consistent combinations are the scenarios. CIB has become a standard tool for building qualitative socio-technical scenarios, including the shared socio-economic pathways used in climate research.
The intuition behind the method
Imagine you are trying to picture coherent futures for a region: will the economy grow or stagnate, will governance be centralized or fragmented, will energy be cheap or expensive, and so on. Each of these questions has a few plausible answers, and the answers pull on one another — cheap energy makes growth more likely, fragmented governance makes coordinated energy policy harder. If you simply list every combination of answers you get an unmanageable and largely nonsensical pile, full of futures that contradict themselves. CIB gives you a disciplined way to throw out the contradictions. You write down how each possible answer pushes for or against every other answer, and then you look for combinations that hang together: futures where every assumption is reinforced rather than undermined by the rest. Those stable, self-supporting bundles are the scenarios worth telling stories about.
History
Cross-impact analysis originated in the late 1960s with Gordon and Hayward's probabilistic cross-impact matrices, but those early variants struggled with mathematical coherence and dependence on raw probabilities. Wolfgang Weimer-Jehle's 2006 paper in Technological Forecasting and Social Change reframed the problem in system-theoretical terms, replacing probabilities with ordinal promoting/restricting judgments and introducing the balance algorithm that defines consistency as a fixed point. This reformulation gave cross-impact analysis a rigorous and reproducible footing. CIB was subsequently adopted for constructing qualitative socio-technical and environmental scenarios; Schweizer and Kriegler and colleagues used it to systematize the narratives behind the shared socio-economic pathways in climate-change research, cementing CIB as a mainstream foresight technique.
When to use it
Use CIB when you need a manageable set of internally consistent qualitative scenarios from a system with many interacting, hard-to-quantify factors and you have access to subject-matter experts who can make pairwise impact judgments. It is especially valuable when the dimensions of uncertainty are categorical rather than numeric, when stakeholders demand transparency about why certain combinations of assumptions belong together, and when the raw combinatorial space of possibilities is too large to inspect by hand. CIB is less appropriate when the system is well captured by continuous dynamic equations (where simulation may be better), when no credible expert judgments about cross-impacts are available, or when only a single, most-likely projection is wanted rather than a portfolio of alternatives.
Limitations and considerations
- Results depend heavily on the chosen descriptors, states, and the quality of the pairwise impact judgments, which are inherently subjective.
- Eliciting a full cross-impact matrix is demanding, and the number of judgments grows rapidly with descriptors and states.
- The additive balance rule assumes impacts combine linearly, which may understate threshold effects or strong interaction among three or more descriptors.
- Consistency is internal coherence, not empirical likelihood; a consistent scenario is not guaranteed to be probable or to actually occur.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Reduces an intractable combinatorial space of possible futures to a small set of internally consistent, self-supporting scenarios.
- Handles categorical, qualitative descriptors that resist numeric modeling while still applying a rigorous, reproducible algorithm.
- Makes the reasoning fully transparent and auditable: every scenario can be traced back to the underlying cross-impact judgments.
- Integrates heterogeneous expert knowledge into a single coherent system representation usable for socio-technical and climate scenarios.
Applications
- Constructing internally consistent socio-economic scenarios such as the shared socio-economic pathways in climate research.
- Energy-transition and infrastructure foresight where many qualitative drivers interact.
- Regional and urban development scenarios combining economic, governance, and social descriptors.
- Technology assessment and policy analysis requiring transparent, expert-grounded scenario portfolios.
- Combining qualitative scenario logic with downstream quantitative models by selecting consistent assumption sets.
How it works
How it works
The analysis begins by decomposing the system into descriptors — the key dimensions of uncertainty, such as economic growth, governance structure, or technology adoption — and assigning each a small set of mutually exclusive, discrete future states. A full configuration of the system is a choice of one state per descriptor. Weimer-Jehle stresses that descriptors should be chosen so that their states are clearly distinguishable and that the set is reasonably comprehensive, because the scenarios that emerge are nothing more than particular configurations of these descriptor states.
Formula
D = \{d_1,\dots,d_n\},\quad d_i \in \{s_{i1},\dots,s_{im_i}\},\ \ \text{configuration } \mathbf{x}=(x_1,\dots,x_n)Experts then fill the cross-impact matrix, the heart of the method. For every ordered pair consisting of a state of one descriptor and a state of another, they record a direct impact judgment on an ordinal scale (typically from strongly restricting to strongly promoting). The judgment answers a counterfactual question: holding everything else aside, if this state were realized, how much more or less plausible would that other state become? Crucially these are direct, pairwise judgments, which keeps the elicitation cognitively manageable even when the implied scenario space is enormous.
Formula
c_{(i,a)\to(j,b)} \in \{-3,\dots,+3\},\quad \text{direct promoting}(+)/\text{restricting}(-)\text{ judgment of state }s_{ia}\text{ on }s_{jb}Given a candidate configuration, the impact balance on each descriptor is computed by summing the cross-impacts that all the other descriptors' currently selected states exert on each possible state of that descriptor. The balance is a kind of net plausibility score: it aggregates all the promoting and restricting pressures acting on each state from the rest of the system. A descriptor's states can then be ranked by how strongly the surrounding configuration supports them.
Formula
IB_j(b\mid \mathbf{x}) = \sum_{i\neq j} c_{(i,x_i)\to(j,b)},\quad \text{balance on state }b\text{ of descriptor }jA configuration is judged internally consistent when, for every descriptor, the state that is actually selected is also the state with the highest impact balance — that is, no descriptor would 'prefer' to switch given what every other descriptor is doing. This makes consistency a fixed-point condition: a consistent scenario is one that reproduces itself under the balance rule. Inconsistent configurations contain at least one descriptor whose chosen state is dominated by an alternative, signalling an internal contradiction.
Formula
x_j \text{ consistent} \iff IB_j(x_j\mid\mathbf{x}) = \max_{b} IB_j(b\mid\mathbf{x});\ \ \mathbf{x}\text{ consistent if all }j\text{ consistent}The balance algorithm searches the configuration space for all such fixed points, typically by iterating from many starting points and letting each descriptor relax toward its best-supported state until the configuration stops changing. Because only a tiny fraction of the combinatorially huge space is self-consistent, the method collapses an intractable number of raw combinations into a manageable handful of consistent scenarios, a property that Schweizer and Kriegler exploited to systematize qualitative scenario sets for environmental-change research.
Formula
S^* = \{\mathbf{x} : \mathbf{x}\text{ is a fixed point of the balance algorithm}\},\quad |S^*| \ll \textstyle\prod_i m_iFinally the consistent scenarios are organized and interpreted. Scenarios that differ in only a few descriptor states are near-duplicates and can be grouped, using a simple Hamming-type distance over descriptor states, into a smaller number of distinct scenario families. Each family is then narrated and characterized, and the cross-impact structure can be inspected to explain why certain assumptions cluster together, giving analysts a transparent, traceable account of why the future portfolio looks the way it does.
Formula
S^* \xrightarrow{\text{cluster}} \{\text{scenario families}\},\quad d(\mathbf{x},\mathbf{x}') = \#\{i : x_i \neq x'_i\}Common pitfalls
- Choosing overlapping or redundant descriptors, which double-counts impacts and distorts the balances.
- Treating the number of consistent scenarios as meaningful in itself rather than clustering and interpreting them substantively.
- Filling the matrix inconsistently across experts without reconciling or documenting the divergent judgments.
- Confusing internal consistency with probability and presenting CIB scenarios as forecasts of the most likely future.
Frequently asked questions
How is CIB different from classical probabilistic cross-impact analysis?
Classical cross-impact analysis, going back to Gordon and Hayward, works with event probabilities and conditional adjustments, which proved mathematically awkward and sensitive to estimation error. Weimer-Jehle's CIB instead uses ordinal judgments of how strongly one descriptor state promotes or restricts another, and defines scenarios as fixed points of a deterministic balance algorithm. The output is a set of internally consistent configurations rather than revised probabilities, which makes CIB better suited to qualitative, categorical foresight problems.
What does 'internally consistent' actually mean in CIB?
A scenario is internally consistent when no descriptor would change its state given what all the other descriptors are doing. Formally, for each descriptor the selected state must have the highest impact balance — the largest net sum of promoting minus restricting influences from the rest of the configuration. This is a self-reproducing fixed point: the configuration regenerates itself under the balance rule, so there is no internal contradiction pulling any single assumption in a different direction.
Does CIB tell me which scenario is most likely?
No. CIB filters for internal coherence, not empirical probability. A consistent scenario is one whose assumptions reinforce rather than undermine each other, but coherence and likelihood are different things; an internally consistent future can still be unlikely. Analysts sometimes layer additional information, such as expert plausibility ratings or the number and depth of supporting impacts, to discuss relative plausibility, but the core method deliberately separates consistency from forecasting.
Related methods
Sources
Weimer-Jehle, W. (2006). Cross-impact balances: A system-theoretical approach to cross-impact analysis. Technological Forecasting and Social Change, 73(4), 334-361.
Open DOI referenceCopied.Schweizer, V. J., & Kriegler, E. (2012). Improving environmental change research with systematic techniques for qualitative scenarios. Environmental Research Letters, 7(4), 044011.
Open DOI referenceCopied.