Smart Choices cheat sheet

A decision is a commitment of resources under uncertainty; the quality of a decision is judged by the process and information available at the time, not only by the outcome.

Bounded rationality: decision makers have limited attention, memory, and computation, so they satisfice (accept the first option meeting a threshold) rather than optimize.

Expected value = sum over outcomes of (probability × payoff). Choose the option with the highest expected value when repeated many times and risk attitude is neutral.

Expected utility = sum over outcomes of (probability × utility of payoff). Utility curves capture risk aversion (concave), risk neutrality (linear), and risk seeking (convex).

System 1 thinking is fast, automatic, and associative; System 2 is slow, effortful, and rule-based. Most errors occur when System 1 answers a question that requires System 2.

A decision frame is the mental representation of the choice: which options, consequences, and reference points are included. Changing the frame can reverse a preference.

Sunk costs are unrecoverable and irrelevant to the forward-looking comparison of options; only marginal costs and benefits of remaining alternatives matter.

Opportunity cost is the value of the best forgone alternative and must be included in every option's cost, including the option of doing nothing.

The decision cycle: define the problem, specify objectives, create alternatives, evaluate consequences, make trade-offs, choose, implement, and audit the result.

A problem statement should name the trigger, the gap between desired and actual state, and the decision owner; a statement that presupposes a solution is a solution masquerading as a problem.

Widen the frame before narrowing: ask whether the decision is a choice among given options, a design problem, or a problem of learning what the situation is.

Gain versus loss framing: outcomes described as losses loom larger than equivalent gains (loss aversion), so the same choice can flip when the reference point moves.

Reference points are set by the status quo, by expectations, or by comparison with others; identify the reference point explicitly before evaluating outcomes.

Attribute framing changes evaluations: a program described as 90% survival is rated more favorably than the identical program described as 10% mortality.

Narrow framing treats each decision in isolation; broad framing bundles decisions and usually produces more consistent risk preferences and better diversification.

The 'decision hierarchy' check: confirm the decision is not a sub-decision of a larger one whose answer would make this choice trivial.

Premature commitment to one option is the most common framing failure; generate at least three genuinely different options before evaluating any of them.

Objectives are the things the decision maker cares about; fundamental objectives answer 'why does this matter?' and means objectives are instrumental steps toward them.

The 'why?' ladder moves upward from means to fundamental objectives; the 'how?' ladder moves downward from fundamental objectives to concrete means.

A fundamental-objectives hierarchy should be mutually exclusive and collectively exhaustive at each level, and small enough to compare (typically 5-9 top-level objectives).

Alternatives must be under the decision maker's control and mutually exclusive in execution; a portfolio of actions can itself be one alternative.

The status quo is always an alternative and must be evaluated on the same criteria as the others; omitting it biases toward change.

Brainstorming rules: defer judgment, encourage wild ideas, build on others, and aim for quantity before quality; evaluation is a separate step.

A strategy table lists alternatives as rows and objectives as columns; a dominated alternative is one that is worse on at least one objective and no better on any other.

The 'even swap' method can eliminate an objective: adjust two alternatives so one objective is equalized, then judge whether the trade-off is acceptable.

Value of information = expected value with the information minus expected value without it, minus the cost of obtaining it; information is worth acquiring only if it could change the choice.

Information has value only when it is decision-relevant, i.e., when it can shift the preferred alternative or the confidence in it.

Base rates anchor probability estimates; combine base-rate information with case-specific evidence rather than relying on either alone.

Confirmatory bias: people seek evidence consistent with a favored hypothesis. Actively search for disconfirming evidence before concluding.

Anchoring: an initial number pulls subsequent estimates toward it, even when the anchor is arbitrary. Form independent estimates before seeing others'.

Availability bias: events that are easy to recall are judged more frequent. Use recorded frequencies rather than memory when data exist.

Sample size matters: small samples produce extreme results by chance. Regression to the mean predicts that extreme performance will be followed by less extreme performance without any causal change.

Expert judgment should be elicited with structured, independent, and documented estimates; aggregation across independent judges reduces random error.

Probability is a degree of belief (subjective) or a long-run frequency (objective); both obey the same rules: 0 ≤ P ≤ 1 and probabilities of mutually exclusive, exhaustive events sum to 1.

Joint probability of independent events = product of their probabilities; probability of A or B = P(A) + P(B) − P(A and B).

Conditional probability: P(A|B) = P(A and B) / P(B). Bayes' rule updates a prior into a posterior: P(A|B) = P(B|A)·P(A) / P(B).

A decision tree orders choices (squares) and chance events (circles); solve by backward induction (rollback), choosing the maximum at decision nodes and computing expected value at chance nodes.

Risk attitude determines the certainty equivalent: the guaranteed amount a person would accept in place of a gamble. Risk-averse people set it below the expected value.

Sensitivity analysis identifies which input, if changed, would reverse the decision; the decision is robust if no plausible change flips it.

Risk mitigation options include avoiding, reducing, transferring (insurance, contracts), and accepting; each has a cost that must be weighed against the reduction in expected loss.

A probability distribution over outcomes, not a single point estimate, is required to represent uncertainty; state ranges and the reasoning behind them.

Trade-offs exist whenever no alternative is best on every objective; a decision is dominated only if another is at least as good on all objectives and strictly better on one.

Multi-attribute scoring: assign weights to objectives (weights sum to 1), score each alternative on each objective, and compute the weighted sum to rank alternatives.

Swing weighting sets weights by asking how much each objective's range (best to worst) matters relative to the others, not how important the objective is in the abstract.

The additive value model assumes preferential independence: the trade-off between two objectives does not depend on the levels of other objectives. Check this assumption.

Lexicographic rules rank by the single most important objective first; they are simple but can ignore large differences on secondary objectives.

Conjoint or even-swap reasoning reveals implicit trade-off rates: how many units of one objective a person will give up for one unit of another.

Value-focused thinking derives alternatives from values rather than evaluating pre-existing options, which usually produces better alternatives.

An indifference point marks two outcomes the decision maker values equally; it is the empirical basis for building a value function.

Groups outperform individuals on judgment tasks when members are independent and diverse; they underperform when members share information and pressure toward consensus.

Groupthink symptoms include illusion of unanimity, self-censorship, pressure on dissenters, and stereotyping of outsiders; prevention includes assigned devil's advocate and anonymous input.

Common-knowledge bias: groups discuss information everyone already shares more than unique information held by single members; explicitly solicit unshared facts.

Aggregation rules matter: majority rule, plurality, weighted voting, and Delphi each produce different winners from the same preferences; Condorcet cycles can make majority rule intransitive.

Stakeholder analysis maps each party's interest, influence, and position on the decision; high-influence, high-interest parties need early engagement.

Diversity of perspective improves decision quality when paired with psychological safety; without safety, dissent is suppressed.

Accountability and pre-commitment to criteria before discussion reduce bias and post-hoc rationalization in group settings.

The Delphi method collects anonymous, iterated expert estimates with controlled feedback, avoiding dominance and anchoring by high-status members.

A decision is not complete until resources, owners, deadlines, and success metrics are assigned; an unassigned decision is a suggestion.

Pre-mortem: before committing, imagine the decision has failed and generate reasons; this surfaces risks that optimism suppresses.

Implementation intentions (if situation X occurs, then I will do Y) raise follow-through by linking the action to a concrete cue.

Set explicit tripwires: predetermined thresholds or dates at which the decision will be reviewed, so that sunk-cost bias does not lock in a failing course.

Separate the decision review from the outcome review: evaluate whether the process was sound, then whether the result was favorable; good process can yield bad outcomes.

Decision logs record the options considered, the information used, the reasoning, and the expected outcome, enabling later calibration and learning.

Commitment devices (contracts, public pledges, escrow) raise the cost of abandoning an intended course and counter present bias.

Post-decision dissonance motivates people to upgrade the attractiveness of the chosen option and downgrade the rejected one; awareness of this effect supports honest evaluation.