7 Debiasing and Better Decisions

Intuition and debiasing

Good decisions do not require eliminating intuition. Heuristics—quick rules of thumb—often save time and work well in familiar, predictable situations. They are more likely to mislead when evidence is sparse, outcomes are uncertain, or a situation feels unique but resembles many past cases.

Expertise is most dependable when the environment has learnable regularities and people receive useful feedback. Confidence or years of experience alone do not guarantee accuracy. Debiasing can reduce predictable errors, but it is not a guarantee of correctness: results vary by method and setting, so interventions should be matched to the decision.

A practical decision routine

For a consequential choice, use a structured routine to clarify the decision, test alternatives, and respond to uncertainty.

1. Define the decision. State what must be chosen, by when, and what would count as a good outcome. Separate the decision from the outcome: a reasonable choice made under uncertainty can still turn out badly.

2. Generate alternatives. Include “wait,” “gather more information,” or “do nothing” where relevant. Before settling on a favored explanation, identify at least one credible alternative and what evidence would distinguish it. This can counter confirmation bias.

3. Use the outside view. Find a reference class of genuinely similar past cases. Ask how often they succeeded, how long they took, and what they cost. Adjust that baseline for important differences in the present case. This helps counter the planning fallacy: focusing on a vivid story about a particular plan while overlooking the distribution of outcomes in comparable cases.

4. Make uncertainty explicit. List key assumptions, unknowns, plausible outcomes, and their consequences. Estimate probabilities when possible instead of using vague labels such as “likely.” Distinguish what is known from what is inferred.

5. Run a premortem. Imagine the decision has failed and write down plausible reasons. Ask which risks could be prevented, monitored, or planned for. Use this to surface risks, not as proof that failure is likely. Considering alternatives and imagining how a decision could go wrong are practical structured prompts.

6. Set a review point. Decide what new evidence would change the choice, when to check progress, and whether the decision can be reversed. For a reversible, low-cost choice, a small trial may be more useful than extended speculation. For an irreversible or high-impact choice, seek stronger evidence and independent review.

Calibrating confidence

Calibration concerns a series of judgments, not whether any single uncertain forecast comes true. Among $100$ forecasts made at $70\%$ confidence, for example, roughly $70$ should prove correct over time if the forecaster is calibrated.

Accuracy and calibration are distinct. A forecaster can correctly rank which events are more likely while systematically assigning probabilities that are too high or too low.

Track forecasts and outcomes

Keep a decision journal with the question, date, probability estimate, reasons, key assumptions, and the outcome that will resolve the forecast. Later, compare estimates with outcomes. Group forecasts into broad confidence ranges, such as $60\text{–}70\%$, and check how often they came true. Look for patterns: whether high-confidence judgments are especially error-prone, estimates are too extreme, or forecasts are worse in unfamiliar domains.

Feedback can support learning, but it does not guarantee improved calibration in every task. Use a meaningful set of resolved judgments rather than drawing conclusions from one or two outcomes.

Using the Brier score

A scoring rule can make forecast review more systematic. For a yes-or-no event, the Brier score is:

Here, $p$ is the forecast probability and $o$ is the outcome: $o=1$ if the event occurs and $o=0$ if it does not. Lower scores are better across a set of forecasts. For example, forecasting $80\%$ and then seeing the event occur gives:

Scoring is useful for tracking a series of forecasts, not for judging a single decision in isolation.

Improving group and organizational decisions

For repeated or group decisions, use a consistent checklist, record estimates before discussion, and ask participants to state what evidence would change their minds. Independent initial estimates can reduce the risk that early opinions anchor the whole group.

Review decisions after outcomes are known, including near misses and decisions that went well for the wrong reasons. When possible, change the decision environment as well: make relevant base rates easy to see, standardize how information is presented, and build reminders into routine workflows. Debiasing approaches include both individual strategies and changes to decision structure; no single technique works universally.

Putting the process together

Better judgment comes from matching intuition to the task, checking individual stories against base rates, considering credible alternatives, making uncertainty explicit, and learning from recorded forecasts. Use structured prompts and feedback as aids, not guarantees. Evaluate decisions by the quality of the reasoning available at the time, not by the outcome alone.