Developed by Theodore J. Gordon and Olaf Helmer for the Kaiser Aluminum Future game (1966), and first tested by Gordon and H. Hayward (Initial Experiments with the Cross Impact Matrix Method of Forecasting, 1968).
Builds a matrix of how each possible future event would raise or lower the odds of every other, then simulates the combinations to show which joint scenarios are likely.
A thousand futures for a battery plant, with each event changing the odds of the next
Three events that may or may not happen, each run in turn: a carbon tax, a halving of storage costs, and the regional utility's contract. The sliders set how much each event raises the odds of the next. Play runs 1,000 futures and counts each combination. Set both to zero to see the events treated as independent. Illustrative.
- Carbon tax passes: a 40% chance in any future.
- Storage costs halve: a 25% chance, plus 30 points in futures where the tax passes.
- Utility signs the contract: a 15% chance, plus 40 points in futures where storage costs halve.
What happened in each future
Runs so far: 1,000 of 1,000
- Futures with the contract
- 30%
- Futures with all three
- 11%
- Futures with none of them
- 37%
An example
A battery maker was weighing a second plant that paid off only if the regional utility signed a large storage contract. The strategy team watched three events, each with a probability from its forecasters: a carbon tax passing (40 percent), cell costs halving within three years (25 percent), and the contract itself (15 percent). Multiplied as if independent, the case where all three lined up came to about 1.5 percent, and the board was ready to shelve the plant.
A planner asked one more question for each pair: if this happens, how much likelier does that become? A tax would add about 30 points to cheap storage, and cheap storage would add about 40 points to the contract, since the utility's case rested on cell prices. Run a thousand times with those links, the contract turned up in 30 percent of futures, against 17 without them. All three together came to 11 percent, against 1, and was the likeliest route to the contract. The board approved site work, with the full commitment tied to the tax vote.
Why it works
Forecasts usually arrive as a list of events, each with its own probability, and a list invites the reader to treat them as separate. They rarely are. One event changes the conditions for the next, so some combinations are far likelier than multiplying the separate odds suggests, and others far less likely. A plan built on the separate odds misjudges exactly the scenarios that matter most, the ones where several things happen together.
Asking the pairwise question, if this happens how does that change, puts the dependence on the table where it can be argued over. Simulating many futures then turns a matrix of judgments into counts of joint outcomes, which is what a decision needs: how often the case that pays off occurs, and which chain of events usually produces it.
The thinking behind it
Theodore J. Gordon Theodore J. Gordon, a futurist who had worked with Olaf Helmer at RAND on early Delphi forecasting, developed cross-impact analysis with Helmer in 1966 to model how events change one another's likelihood. With H. Hayward he published the first journal report of the method in Futures in 1968. He was among the founders of the Institute for the Future, and later wrote the chapters on Delphi and cross-impact analysis in the Millennium Project's Futures Research Methodology, Version 3.0 (2009), edited with Jerome C. Glenn.
Olaf Helmer Olaf Helmer, a logician at the RAND Corporation from 1946 to 1968, developed the Delphi method of anonymous, iterated expert forecasting with Norman Dalkey and Nicholas Rescher, reported in An Experimental Application of the Delphi Method to the Use of Experts (Management Science, 1963). With Theodore J. Gordon he devised cross-impact analysis, first played as a card game called Future, and in 1968 he co-founded the Institute for the Future.
H. Hayward H. Hayward co-wrote with Theodore J. Gordon the first journal report on the method, Initial Experiments with the Cross Impact Matrix Method of Forecasting (Futures, 1968). Their paper sets a list of forecast events in a matrix and adjusts each event's probability for the occurrence of the others, the same adjustment the figure on this page makes on three linked events.
How to run it
- List the events. Five to ten future events that matter to the decision, each stated so it either happens or does not by a set date.
- Estimate each one alone. Get a probability for each before any discussion of how they connect.
- Ask the cross question for every pair. If this one happens, how much likelier or less likely does that one become?
- Fill the matrix. Record each pair's effect with a line of reasoning, and leave unlinked pairs at zero.
- Run many futures. Simulate the events in order a thousand times or more, each drawn on odds adjusted by those before it.
- Read the joint outcomes. Count how often each combination occurs, starting with the ones the decision depends on.
- Name the signposts. The events that move the important combinations most are the ones to watch; tie the decision to them.
Where it goes wrong
The first failure is invented precision. Pairwise effects are judgments, and a matrix of fifty of them can look far more exact than anyone's knowledge. Spot it when nobody can give the reason behind a number, and recover by keeping only the links someone can explain and rerunning with each one set higher and lower.
The second is double counting. When two events share a cause, linking each to the other and both to the cause inflates their joint odds. The third is reading the counts as a forecast of one future. The output says which combinations are likely, and its value lies in the signposts it points to more than in any single percentage.
When to use something else
- Scenario Planning. When the uncertainties are too deep for probabilities and the team needs a few coherent stories of the future.
- Monte Carlo Simulation. When the inputs are quantities with ranges, such as costs and durations, more than events that happen or not.
- Delphi Forecasts and Predictions. When the probabilities themselves need building from a panel of experts before any matrix is filled.
Where it sits in SOLVE eX
Go to the source
- Cross-impact matrix method, developed for the Kaiser Aluminum Future game, 1966
- Initial Experiments with the Cross Impact Matrix Method of Forecasting, Futures 1(2), 1968