Decision Making Under Uncertainty Frameworks

How it works:

• For each option, calculate sum of (probability × utility) across all outcomes.
• Choose the option with highest expected utility.

When it dominates:

• Probabilities are reasonably well-known.
• Decision is repeated or high-stakes single.
• You have clear utility scale (your subjective valuation of outcomes).
• Time available for analysis.

When it fails:

• Probabilities are unknown or highly disputed.
• Utility scale is itself uncertain.
• Outcomes include 'unimaginable' scenarios that can't be utility-assessed.

Practical use:

• Investment decisions, portfolio construction.
• Career decisions where alternatives can be probabilistically assessed.
• Insurance buying.
• Medical treatment choices (with NHS / NICE-published probabilities).

How it works:

• For each option, identify the worst-case outcome.
• For each option-outcome combination, calculate 'regret' = (best outcome you could have achieved) - (outcome you actually got).
• Find the maximum regret for each option (across all states of nature).
• Choose the option with the lowest maximum regret.

Why it matters:

• Some decisions are dominated by the question 'how badly will I feel if X happens'.
• Regret aversion is psychologically real + sometimes economically rational.
• Example: if you reject a job offer + it turns out to be your dream company, you'll regret it forever.

When it dominates:

• One-off decisions with high asymmetric downside.
• You have strong preferences against specific bad outcomes.
• You're psychologically vulnerable to regret + need to safeguard.

When it fails:

• Decisions where multiple bad outcomes are possible and you can't rank them.
• Repeated decisions where law of large numbers means individual regret matters less.

How it works:

• Define what 'acceptable' means (specific threshold criteria).
• Generate options.
• Take the first one that meets your threshold.
• Don't waste time optimising.

Why it matters:

• Herbert Simon (Nobel 1978) showed that satisficing often beats optimisation in real-world conditions where information-gathering has costs.
• Time + cognitive resources are scarce.
• 'Good enough' decisions made fast often beat 'optimal' decisions made slowly.

When it dominates:

• Time-constrained decisions.
• Decisions with diminishing returns to research.
• Routine + low-stakes choices.
• Most consumer decisions (which to buy at supermarket).

When it fails:

• High-stakes decisions where optimal vs acceptable creates significant value gap.
• Decisions where threshold criteria are hard to articulate.
• Long-term reversal cost is low (over time, you can switch and find better).

Real-world example:

• Choosing a restaurant: satisficing (find any with reasonable reviews) usually fine.
• Choosing a house to buy: optimisation usually better (large lifetime impact).

How it works:

• For each option, identify the best possible outcome.
• Choose the option with the highest 'best outcome'.
• Ignore downside (or accept it).

Why it matters:

• Some decisions are dominated by 'what's the upside if this works?'
• Small experimental bets often have asymmetric upside.
• Venture-capital reasoning: most fail, but ones that succeed return 100x+.

When it dominates:

• Small experimental bets where downside is limited.
• Option-acquisition decisions (low cost + high potential upside).
• Asymmetric-payoff opportunities (you can only lose 1x but win 100x).
• Personal experimentation (try a new skill, business idea).

When it fails:

• High-cost decisions where downside is meaningful.
• Decisions where you can't actually lose just '1x' (long-tail negative outcomes).

Real-world examples:

• Buying an option ticket: maximax thinking; small premium, big payoff if successful.
• Submitting one of N article pitches: each has small cost; one accepted is great upside.
• Going on a 30-min coffee meeting with someone new: small time cost, potentially big career upside.

Use EXPECTED UTILITY when:

• Probabilities reasonably well-known.
• Repeated or high-stakes single decision.
• You have utility scale.
• Time available for analysis.
• Example: portfolio construction, career change with researchable industries.

Use MINIMAX REGRET when:

• One-off decision with strong asymmetric downside.
• Specific bad outcomes you'd particularly hate.
• Probabilities are uncertain or disputed.
• Example: choosing whether to email an ex; selling a beloved car.

Use SATISFICING when:

• Time-constrained.
• Diminishing returns to research.
• Routine, low-stakes, reversible.
• Example: restaurant choice; appliance purchase; routine consumer decisions.

Use MAXIMAX when:

• Limited downside.
• Asymmetric upside.
• Experimental + option-acquiring.
• Example: trying a new business idea on a small scale; reaching out to a connection.

Real-world decisions often benefit from hybrid framework use:

• EU + minimax floor: maximise expected utility BUT ensure worst-case isn't disastrous (e.g. portfolio strategy).
• Satisfice + maximax: find acceptable option first, then push for upside within acceptable (e.g. job search).
• EU + satisficing constraint: maximise expected utility BUT only consider options that meet basic acceptability criteria (e.g. property buying).
• Maximax + minimax safety net: pursue upside but only if downside is limited (e.g. starting side business).

Why hybrids work:

• Real decisions have multiple dimensions.
• Single-criterion optimisation can ignore important constraints.
• Layering frameworks captures more nuance.

1. Using EU when probabilities are guesses: false precision; minimax regret often better.
2. Using minimax regret for low-stakes decisions: leads to risk-paralysis on trivial choices.
3. Using satisficing on high-stakes irreversible decisions: under-optimisation costs lifetime value.
4. Using maximax on decisions with real downside: ignores meaningful negative outcomes.
5. Sticking to one framework regardless of context: 'I always optimise everything' = decision paralysis.
6. Not articulating your framework: implicit framework choice misses better fits.