SMIC Prob-Expert
- Introduced by
- Michel Godet (LIPSOR)
- Year
- 2006
- Type
- Probabilistic cross-impact pipeline for ranking scenario combinations
- Family
- Futures studies / French prospective school
What does it do?
SMIC Prob-Expert — from the French Systeme et Matrices d'Impacts Croises, Systems and Matrices of Cross-Impacts — is the probabilistic cross-impact method in Michel Godet's la prospective toolkit.
It takes a small set of fundamental hypotheses about the future and asks experts for both the simple probability that each hypothesis comes true and the conditional probabilities linking the hypotheses to one another. Because experts' raw estimates are rarely mutually consistent, SMIC's core is a quadratic optimisation that adjusts them minimally into a coherent joint probability distribution over the 2^n possible combinations of the hypotheses. Each combination is an image of the future — a scenario — and the corrected, or net, probabilities rank these images from most to least likely. The method thereby turns scattered expert opinion into a probabilistically weighted set of scenarios, identifying the few core futures that concentrate most of the probability mass.
The intuition behind the method
Suppose the future of a system hinges on a handful of yes-or-no questions: will the regulation pass, will the technology mature, will demand grow. With n such hypotheses there are 2 to the n possible worlds, one for each pattern of yeses and noes, and the natural goal is to know which of those worlds are most probable. The trouble is that experts find it far easier to judge one hypothesis at a time, and to say how one would shift the odds of another, than to assign a probability to every complete combination at once. So SMIC asks the easier questions — the chance of each hypothesis alone, and the chance of each given that another holds or fails. But people's separate guesses never quite fit together: the conditional probabilities they give are mathematically incompatible with the simple ones. SMIC's quadratic program nudges all the estimates as little as possible until they form a single consistent probability distribution over the full set of combinations. Out comes a ranked list of scenarios with coherent probabilities, so the analyst can focus on the two or three images of the future that the experts, taken together, actually consider most likely.
History
SMIC Prob-Expert emerged from the cross-impact analysis tradition that developed in futures studies from the late 1960s, when Theodore Gordon and Olaf Helmer first proposed letting the occurrence of one event adjust the probabilities of others. Michel Godet, building the French school of la prospective at the Conservatoire national des arts et metiers and later LIPSOR, formalised a probabilistic variant — Systeme et Matrices d'Impacts Croises — that uses a quadratic optimisation to reconcile experts' simple and conditional probability estimates into a coherent joint distribution over scenario combinations. Integrated with MICMAC structural analysis, MACTOR actor analysis, and morphological scenario building, SMIC became the quantifying, probability-weighting module of Godet's toolbox, documented in works including Creating Futures: Scenario Planning as a Strategic Management Tool (2006). LIPSOR's free SMIC-PROB-EXPERT software spread the method through European foresight and strategic-planning practice.
When to use it
Use SMIC Prob-Expert when you want to attach coherent, comparable probabilities to a set of scenarios built from a small number of well-defined uncertain hypotheses, and when a panel of experts can credibly judge those hypotheses and their interactions. It suits the later, quantifying stage of a la prospective study, where the key variables and actor stakes have already narrowed the future down to a few fundamental binary questions and decision-makers want to know which combinations are most likely. It is appropriate when stakeholders value an explicit probabilistic ranking over purely qualitative narratives. It is poorly suited to problems with many hypotheses, since the two-to-the-n combinations grow unmanageable beyond five or six; to situations where experts cannot meaningfully estimate conditional probabilities; or to deeply novel futures where any probability elicitation would be spurious and a purely qualitative scenario approach is safer.
Limitations and considerations
- The number of combinations grows as 2^n, so the method is practically limited to about five or six hypotheses.
- Reducing the future to binary hypotheses discards gradations and forces complex uncertainties into yes-or-no form.
- Results depend entirely on subjective expert probabilities, and the correction cannot fix poor or biased underlying judgements.
- Aggregating divergent experts into one distribution can mask genuine disagreement behind a falsely confident consensus.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Attaches coherent, comparable probabilities to whole scenarios, going beyond purely qualitative narrative methods.
- Elicits only simple and pairwise-conditional probabilities, judgements experts can make far more reliably than full-combination estimates.
- The quadratic correction enforces probabilistic coherence while honouring expert input through a minimal adjustment.
- Ranks the 2^n images of the future so analysts can concentrate on the few core scenarios that hold most of the probability.
Applications
- Assigning coherent probabilities to scenarios in a la prospective study after the key hypotheses have been identified.
- Ranking combinations of uncertain events to find the most probable images of the future in strategic foresight.
- Reconciling a panel of experts' inconsistent probability judgements into a single coherent distribution.
- Complementing qualitative morphological scenarios with a quantitative sense of their relative likelihood.
- Supporting sensitivity analysis on how favoured scenarios shift as hypotheses or expert weightings change.
How it works
How it works
The method begins by defining a small set of fundamental hypotheses, each a clear binary proposition about the future state of the system — typically derived from the key variables identified in structural analysis and the stakes mapped in actor analysis. Godet recommends keeping the number of hypotheses small, around five or six, because the number of combinations grows as two to the power of the number of hypotheses, and both the elicitation burden and the interpretive load explode beyond that. Each combination of the hypotheses, with every one either realised or not, constitutes an image of the future, so this set of hypotheses defines the entire space of scenarios the analysis will weigh.
Formula
H = \{H_1, \ldots, H_n\}\ \text{binary hypotheses},\ \ 2^{n} \text{ possible combinations (images)}Experts are then surveyed for three kinds of probability for each hypothesis: its simple probability of occurring on its own, and its conditional probabilities given that each other hypothesis is realised and given that each other hypothesis fails. Godet notes that asking for these pairwise conditionals captures the experts' sense of how the hypotheses reinforce or undermine one another, which is the cross-impact information at the heart of the method. Eliciting probabilities one and two hypotheses at a time is deliberately chosen because experts can reason about such focused judgements far more reliably than about the probability of a full multi-hypothesis combination.
Formula
P(H_i),\quad P(H_i \mid H_j),\quad P(H_i \mid \bar{H_j})\ \text{from each expert}The raw simple and conditional probabilities an expert supplies are almost never mutually coherent — they violate the laws of probability that bind marginals and conditionals together. SMIC resolves this with a quadratic program that searches for a joint probability distribution over the two-to-the-n combinations which respects the axioms of probability (the probabilities are non-negative and sum to one, and the implied marginals and conditionals are consistent) while staying as close as possible to the expert's original estimates. Minimising the squared deviations from the raw inputs ensures the correction is the smallest adjustment that restores coherence, so the expert's judgement is honoured as far as logic permits.
Formula
\min \sum (\pi_k - \pi_k^{0})^2\ \text{s.t. } \sum_k \pi_k = 1,\ \pi_k \geq 0,\ \text{consistency constraints}Solving the program yields the net probabilities: a coherent probability for each of the two-to-the-n images of the future, corrected from the experts' inconsistent raw inputs. These net probabilities are the method's key output, since they assign a defensible likelihood to every complete combination of the hypotheses rather than to the hypotheses in isolation. Godet contrasts these net, post-correction probabilities with the raw elicited ones, treating the gap between them as itself informative about where experts' intuitions were internally contradictory. Aggregating across a panel produces a consensus distribution, often reported alongside the spread of individual expert views.
Formula
\pi = (\pi_1, \ldots, \pi_{2^n}),\quad \pi_k = P(\text{image } k)\ \text{coherent net distribution}Finally the images of the future are ranked by their net probabilities, and the analyst reads off the structure of the scenario space. Typically a few combinations capture most of the probability mass — the core or most probable scenarios — while the long tail of low-probability images can be set aside or treated as contrasting alternatives. Godet uses this ranking to select the handful of scenarios worth developing in detail, complementing the qualitative coherence judgements of morphological analysis with a quantitative sense of relative likelihood. The probabilities also support sensitivity analysis, showing how the favoured scenarios shift if particular hypotheses or expert views are weighted differently.
Formula
\text{rank images by } \pi_k;\quad \text{core scenarios} = \{k : \pi_k \geq \tau\}Common pitfalls
- Choosing too many hypotheses, so the combinations and the elicitation burden become unmanageable and estimates degrade.
- Treating the corrected net probabilities as objective forecasts rather than as a coherent rendering of subjective opinion.
- Defining hypotheses that are not genuinely binary or that overlap, muddying the meaning of the resulting combinations.
- Reporting only the consensus distribution and hiding the spread of expert views that may signal real, decision-relevant disagreement.
Frequently asked questions
Why does SMIC need a quadratic program?
Because the probabilities experts supply are almost never mutually consistent. When someone gives a simple probability for each hypothesis and conditional probabilities linking them, those numbers usually violate the laws of probability that bind marginals and conditionals together, so no genuine joint distribution exactly matches them. SMIC's quadratic program searches for the joint distribution over all combinations that obeys the probability axioms while minimising the squared distance from the expert's raw estimates. The quadratic objective makes the correction the smallest possible adjustment that restores coherence, so the expert's intent is preserved as far as logic allows while the contradictions are removed.
What is an 'image of the future' in SMIC?
An image of the future is one complete combination of the hypotheses, with each hypothesis set to either realised or not. With n binary hypotheses there are two-to-the-n such images, and each corresponds to a distinct scenario. SMIC's purpose is to assign a coherent net probability to every image, so that the analyst can see not just how likely each individual hypothesis is but how likely each full pattern of outcomes is. Ranking the images by their net probabilities identifies the core scenarios — usually a few combinations that hold most of the probability mass and are worth developing in detail.
How does SMIC differ from MICMAC?
Both are matrix-based tools in la prospective but they do different jobs. MICMAC performs structural analysis, classifying the system's variables by influence and dependence to reveal which are the key drivers, and produces no probabilities. SMIC works at a later, quantifying stage: it takes a small set of binary hypotheses, elicits simple and conditional probabilities from experts, and computes a coherent probability distribution over scenario combinations to rank them by likelihood. In short, MICMAC tells you which variables matter, while SMIC tells you which combinations of outcomes are most probable, so they are complementary rather than alternatives.
Related methods
Sources
Godet, M. (2006). Creating Futures: Scenario Planning as a Strategic Management Tool (2nd ed.). Economica.
ISBN: 9782717852448Copied.