15 KiB
Cognitive Bias Catalog
Quick Reference Table
| Bias | Category | Impact | Detection | Remediation |
|---|---|---|---|---|
| Confirmation | Confirmation | Seek supporting evidence only | Search for disconfirming evidence? | Red team your forecast |
| Desirability | Confirmation | Want outcome → believe it's likely | Do I want this outcome? | Outsource to neutral party |
| Availability | Availability | Recent/vivid events dominate | What recent news influenced me? | Look up actual statistics |
| Recency | Availability | Overweight recent data | Considering full history? | Expand time window |
| Anchoring | Anchoring | First number sticks | Too close to initial number? | Generate estimate first |
| Affect | Affect | Feelings override data | How do I feel about this? | Acknowledge, then set aside |
| Loss Aversion | Affect | Overweight downside | Weighting losses more? | Evaluate symmetrically |
| Overconfidence | Overconfidence | Intervals too narrow | Track calibration | Widen intervals to 20-80% |
| Dunning-Kruger | Overconfidence | Novices overestimate | How experienced am I? | Seek expert feedback |
| Optimism | Overconfidence | "Won't happen to me" | What's the base rate? | Apply base rate to self |
| Pessimism | Overconfidence | Overweight negatives | Only considering downsides? | List positive scenarios |
| Attribution Error | Attribution | Blame person, not situation | What situational factors? | Consider constraints first |
| Self-Serving | Attribution | Success=skill, failure=luck | Consistent attribution? | Same standard for both |
| Framing | Framing | Presentation changes answer | How is this framed? | Rephrase multiple ways |
| Narrative Fallacy | Framing | Simple stories mislead | Story too clean? | Prefer stats over stories |
| Sunk Cost | Temporal | Can't abandon past investment | Only future costs/benefits? | Decide as if starting fresh |
| Hindsight | Temporal | "Knew it all along" | Written record of prediction? | Record forecasts beforehand |
| Planning Fallacy | Temporal | Underestimate time/cost | Reference class timeline? | Add 2-3x buffer |
| Outcome Bias | Temporal | Judge by result not process | Evaluating process or outcome? | Judge by info available then |
| Clustering Illusion | Pattern | See patterns in randomness | Statistically significant? | Test significance |
| Gambler's Fallacy | Pattern | Expect short-term balancing | Are events independent? | Use actual probability |
| Base Rate Neglect | Bayesian | Ignore prior probabilities | Did I start with base rate? | Always start with base rate |
| Conjunction Fallacy | Bayesian | Specific > general | Is A&B > A alone? | P(A&B) ≤ P(A) always |
| Halo Effect | Social | One trait colors everything | Generalizing from one trait? | Assess dimensions separately |
| Authority Bias | Social | Overweight expert opinions | Expert's track record? | Evaluate evidence not credentials |
| Peak-End | Memory | Remember peaks/endings only | Remembering whole sequence? | Review full historical record |
Confirmation Cluster
Confirmation Bias
Definition: Search for, interpret, and recall information that confirms pre-existing beliefs.
Affects forecasting: Only look for supporting evidence, discount contradictions, selective memory.
Detect: Did I search for disconfirming evidence? Can I steelman the opposite view?
Remediate: Red team your forecast, list disconfirming evidence first, ask "How could I be wrong?"
Desirability Bias
Definition: Believing outcomes you want are more likely than they are.
Affects forecasting: Bullish on own startup, wishful thinking masquerading as analysis.
Detect: Do I want this outcome? Am I emotionally invested?
Remediate: Outsource to neutral party, imagine opposite outcome, forecast before declaring preference.
Availability Cluster
Availability Heuristic
Definition: Judging probability by how easily examples come to mind.
Affects forecasting: Overestimate vivid risks (terrorism), underestimate mundane (heart disease), media coverage distorts frequency perception.
Detect: What recent news am I thinking of? Is this vivid/emotional/recent?
Remediate: Look up actual statistics, use reference class not memorable examples.
Recency Bias
Definition: Overweighting recent events relative to historical patterns.
Affects forecasting: Extrapolate recent trends linearly, forget cycles and mean reversion.
Detect: How much history am I considering? Is forecast just recent trend?
Remediate: Expand time window (decades not months), check for cyclicality.
Anchoring Cluster
Anchoring Bias
Definition: Over-relying on first piece of information encountered.
Affects forecasting: First number becomes estimate, can't adjust sufficiently from anchor.
Detect: What was first number I heard? Am I too close to it?
Remediate: Generate own estimate first, use multiple independent sources.
Priming
Definition: Prior stimulus influences subsequent response.
Affects forecasting: Reading disaster primes pessimism, context shapes judgment unconsciously.
Detect: What did I just read/see/hear? Is mood affecting forecast?
Remediate: Clear mind before forecasting, wait between exposure and estimation.
Affect Cluster
Affect Heuristic
Definition: Letting feelings about something determine beliefs about it.
Affects forecasting: Like it → think it's safe, dislike it → think it's dangerous.
Detect: How do I feel about this? Would I forecast differently if neutral?
Remediate: Acknowledge emotion then set aside, focus on base rates and evidence.
Loss Aversion
Definition: Losses hurt more than equivalent gains feel good (2:1 ratio).
Affects forecasting: Overweight downside scenarios, status quo bias, asymmetric risk evaluation.
Detect: Am I weighting losses more? Would I accept bet if gains/losses swapped?
Remediate: Evaluate gains and losses symmetrically, use expected value calculation.
Overconfidence Cluster
Overconfidence Bias
Definition: Confidence exceeds actual accuracy.
Affects forecasting: 90% intervals capture truth 50% of time, narrow ranges, extreme probabilities.
Detect: Track calibration, are intervals too narrow? Can I be surprised?
Remediate: Widen confidence intervals, track calibration, use 20-80% as default.
Dunning-Kruger Effect
Definition: Unskilled overestimate competence; experts underestimate.
Affects forecasting: Novices predict with false precision, don't know what they don't know.
Detect: How experienced am I in this domain? Do experts agree?
Remediate: If novice widen intervals, seek expert feedback, learn domain deeply first.
Optimism Bias
Definition: Believing you're less likely than others to experience negatives.
Affects forecasting: "My startup is different" (90% fail), "This time is different" (rarely is).
Detect: What's base rate for people like me? Am I assuming I'm special?
Remediate: Use reference class for yourself, apply base rates, assume average then adjust slightly.
Pessimism Bias
Definition: Overweighting negative outcomes, underweighting positive.
Affects forecasting: Disaster predictions rarely materialize, underestimate human adaptability.
Detect: Only considering downsides? What positive scenarios missing?
Remediate: Explicitly list positive scenarios, consider adaptive responses.
Attribution Cluster
Fundamental Attribution Error
Definition: Overattribute behavior to personality, underattribute to situation.
Affects forecasting: "CEO is brilliant" ignores market conditions, predict based on person not circumstances.
Detect: What situational factors am I ignoring? How much is luck vs. skill?
Remediate: Consider situational constraints first, estimate luck vs. skill proportion.
Self-Serving Bias
Definition: Attribute success to skill, failure to bad luck.
Affects forecasting: Can't learn from mistakes (was luck!), overconfident after wins (was skill!).
Detect: Would I explain someone else's outcome this way? Do I attribute consistently?
Remediate: Apply same standard to wins and losses, assume 50% luck/50% skill, focus on process.
Framing Cluster
Framing Effect
Definition: Same information, different presentation, different decision.
Affects forecasting: "90% survival" vs "10% death" changes estimate, format matters.
Detect: How is question framed? Do I get same answer both ways?
Remediate: Rephrase multiple ways, convert to neutral format, use frequency (100 out of 1000).
Narrative Fallacy
Definition: Constructing simple stories to explain complex reality.
Affects forecasting: Post-hoc explanations feel compelling, smooth narratives overpower messy data.
Detect: Is story too clean? Can I fit multiple narratives to same data?
Remediate: Prefer statistics over stories, generate alternative narratives, use base rates.
Temporal Biases
Sunk Cost Fallacy
Definition: Continuing endeavor because of past investment, not future value.
Affects forecasting: "Invested $1M, can't stop now", hold losing positions too long.
Detect: If I started today, would I choose this? Considering only future costs/benefits?
Remediate: Consider only forward-looking value, treat sunk costs as irrelevant.
Hindsight Bias
Definition: After outcome known, "I knew it all along."
Affects forecasting: Can't recall prior uncertainty, overestimate predictability, can't learn from surprises.
Detect: What did I actually predict beforehand? Written record exists?
Remediate: Write forecasts before outcome, record confidence levels, review predictions regularly.
Planning Fallacy
Definition: Underestimate time, costs, risks; overestimate benefits.
Affects forecasting: Projects take 2-3x longer than planned, inside view ignores reference class.
Detect: How long did similar projects take? Using inside view only?
Remediate: Use reference class forecasting, add 2-3x buffer, consider outside view first.
Outcome Bias
Definition: Judging decision quality by result, not by information available at time.
Affects forecasting: Good outcome ≠ good decision, can't separate luck from skill.
Detect: What did I know when I decided? Evaluating process or outcome?
Remediate: Judge decisions by process not results, evaluate with info available then.
Pattern Recognition Biases
Clustering Illusion
Definition: Seeing patterns in random data.
Affects forecasting: "Winning streak" in random sequence, stock "trends" that are noise, "hot hand" fallacy.
Detect: Is this statistically significant? Could this be random chance?
Remediate: Test statistical significance, use appropriate sample size, consider null hypothesis.
Gambler's Fallacy
Definition: Believing random events "balance out" in short run.
Affects forecasting: "Due for a win" after losses, expecting mean reversion too quickly.
Detect: Are these events independent? Does past affect future probability?
Remediate: Recognize independent events, don't expect short-term balancing.
Bayesian Reasoning Failures
Base Rate Neglect
Definition: Ignoring prior probabilities, focusing only on new evidence.
Affects forecasting: "Test is 90% accurate" ignores base rate, vivid case study overrides statistics.
Detect: What's the base rate? Did I start with prior probability?
Remediate: Always start with base rate, update incrementally with evidence.
Conjunction Fallacy
Definition: Believing specific scenario is more probable than general one.
Affects forecasting: "Librarian who likes poetry" > "Librarian", detailed scenarios feel more likely.
Detect: Is A&B more likely than A alone? Confusing plausibility with probability?
Remediate: Remember P(A&B) ≤ P(A), strip away narrative details.
Social Biases
Halo Effect
Definition: One positive trait colors perception of everything else.
Affects forecasting: Successful CEO → good at everything, one win → forecaster must be skilled.
Detect: Am I generalizing from one trait? Are dimensions actually correlated?
Remediate: Assess dimensions separately, don't assume correlation, judge each forecast independently.
Authority Bias
Definition: Overweight opinions of authorities, underweight evidence.
Affects forecasting: "Expert said so" → must be true, defer to credentials over data.
Detect: What's expert's track record? Does evidence support claim?
Remediate: Evaluate expert track record, consider evidence not just credentials.
Memory Biases
Peak-End Rule
Definition: Judging experience by peak and end, ignoring duration and average.
Affects forecasting: Remember market peak, ignore average returns, distorted recall of sequences.
Detect: Am I remembering whole sequence? What was average not just peak/end?
Remediate: Review full historical record, calculate averages not memorable moments.
Rosy Retrospection
Definition: Remembering past as better than it was.
Affects forecasting: "Things were better in old days", underestimate historical problems.
Detect: What do contemporary records show? Am I romanticizing the past?
Remediate: Consult historical data not memory, read contemporary accounts.
Application to Forecasting
Pre-Forecast Checklist
- What's the base rate? (Base rate neglect)
- Am I anchored on a number? (Anchoring)
- Do I want this outcome? (Desirability bias)
- What recent events am I recalling? (Availability)
- Am I overconfident? (Overconfidence)
During Forecast
- Did I search for disconfirming evidence? (Confirmation)
- Am I using inside or outside view? (Planning fallacy)
- Is this pattern real or random? (Clustering illusion)
- Am I framing this question neutrally? (Framing)
- What would change my mind? (Motivated reasoning)
Post-Forecast Review
- Record what I predicted before (Hindsight bias)
- Judge decision by process, not outcome (Outcome bias)
- Attribute success/failure consistently (Self-serving bias)
- Update calibration tracking (Overconfidence)
- What did I learn? (Growth mindset)
Bias Remediation Framework
Five principles:
- Awareness: Know which biases affect you most
- Process: Use checklists and frameworks
- Calibration: Track accuracy over time
- Humility: Assume you're biased, not immune
- Updating: Learn from mistakes, adjust process
Key insight: You can't eliminate biases, but you can design systems that compensate for them.
Return to: Main Skill