Use maximin reasoning for high-stakes, irreversible decisions under ambiguity

Choose the option whose worst plausible outcome is most survivable — when you can’t compute expected value, optimize the floor.

Why it works

Under genuine Knightian uncertainty, expected-value maximization is incoherent — you can’t compute expected value without a probability distribution. Gilboa and Schmeidler’s maxmin framework recommends choosing the option whose worst plausible outcome is best. This is conservative but appropriate when: (1) stakes are high, (2) outcomes are irreversible, and (3) the probability distribution is genuinely unknown. For low-stakes decisions, small bets are preferable because information value is high relative to cost.

How to do it

1. List all plausible scenarios, including tail cases.
2. For each option, identify the worst realistic outcome.
3. Choose the option where the worst outcome is most survivable or reversible.
4. Reserve this for high-stakes, irreversible decisions; for low-stakes decisions, run small bets instead.

Evidence

Maximin is a well-formalized decision rule in game theory and decision theory under ambiguity (Gilboa & Schmeidler, 1989). Its real-world effectiveness is untested in RCTs but it is logically defensible under genuine uncertainty and widely used in policy analysis. (mechanistic)

Maximin is extremely conservative and can be worse than expected-value reasoning when probabilities are actually estimable — use only when the probability distribution is genuinely unknown.

Sources

• Gilboa, I., & Schmeidler, D. (1989). Maxmin expected utility with non-unique prior. Journal of Mathematical Economics, 18(2), 141–153.

Common mistake

Applying maximin to low-stakes decisions where a small experiment would produce real data — maximin is a last resort for irreversible high-stakes choices, not an everyday decision rule.