How to use it
- 1. No probabilitiesThese five criteria are for when you genuinely cannot put numbers on the states. That is uncertainty, not risk.
- 2. Watch them disagreeThe same table, five defensible criteria, and often five different answers. Personality is the input.
- 3. Read the regret matrixRegret = what you lost by not having been clairvoyant. Minimax regret hates being badly wrong.
- 4. Slide Hurwicz αα=0 is pure pessimism, α=1 pure optimism. Find the α where your choice flips.
Which warehouse do we build?
Payoff table ($k profit)
| Boom | Steady | Slump |
Hurwicz optimism α
The verdict
Payoffs. Green = that row’s best outcome, red = its worst. Maximin reads only the red column; maximax reads only the green.
Regret matrix: for each state, best-in-column − this cell. The number that matters is each row’s maximum - minimax regret picks the smallest of those.
Hurwicz score as α sweeps 0→1. Each line is an alternative; where lines cross, your decision flips. The vertical line is your α.
Try these experiments
- Find a table where all five criteria disagree. It is easier than you would like - and it means “the rational choice” is not well defined without a stance.
- Add a single alternative that is never best and never worst: watch minimax regret change its answer. Regret violates independence of irrelevant alternatives.
- Slide α slowly and find the flip point. If your decision changes at α=0.48 vs 0.52, the choice was never really about the data.
About this tool
The concept
Decisions under uncertainty (no probabilities) differ from decisions under risk (probabilities known). With no probabilities, expected value is unavailable and you must choose a criterion.
How it works
Maximin: max of row minima. Maximax: max of row maxima. Hurwicz: α·max + (1−α)·min. Laplace: assume equal probabilities and average. Savage: minimise the maximum regret.
Winning strategy
Pick the criterion before you see the table - otherwise you will pick whichever one endorses the answer you already wanted. If you do have probabilities, use the Decision Tree tool instead.