A practical guide to thinking in probabilities instead of certainties — the single most important skill for making better decisions under uncertainty. Covers Bayesian reasoning, calibration, base rates, expected value, and practical exercises.
- Why Probabilistic Thinking?
- Core Concepts
- Practical Applications
- Common Probability Mistakes
- Exercises
- Resources
- License
The world is uncertain. Yet most people think in black and white: "It will work" or "It won't work." "The market will go up" or "The market will go down."
Probabilistic thinkers say: "There's a 70% chance it will work, a 20% chance of partial success, and a 10% chance of failure. Given those probabilities, the expected value is positive."
This matters because:
- Better calibration — You're less often surprised
- Better decisions — Expected value beats gut feeling
- Less overconfidence — You acknowledge what you don't know
- Better communication — "I'm 80% confident" is more useful than "I think so"
- Faster updating — New information adjusts your probabilities, not your identity
"The test of a first-rate intelligence is the ability to hold two opposed ideas in the mind at the same time and still retain the ability to function." — F. Scott Fitzgerald
Instead of: "This project will take 3 months" Say: "There's a 50% chance it takes 2-4 months, a 90% chance it takes 1-6 months"
Instead of: "The stock is worth $50" Say: "I estimate it's worth $40-$65, with my best guess at $50"
Template:
## Probability Range: [Estimate]
| Confidence | Range |
|:-:|:-:|
| 50% (likely range) | [low] to [high] |
| 80% (wider range) | [low] to [high] |
| 95% (almost certain) | [low] to [high] |
| Best single estimate | [point] |Why ranges matter:
- A single point estimate hides your uncertainty
- Ranges communicate how much you know (and don't)
- Wide ranges = low confidence = more investigation needed
- Narrow ranges = high confidence (but check for overconfidence)
Before analyzing the specific case, ask: "What usually happens in situations like this?"
The formula:
Start with the base rate
+ Adjust for specific evidence
= Better estimate
Examples:
| Question | Your Intuition | Base Rate | Adjusted Estimate |
|---|---|---|---|
| Will my startup succeed? | "Probably!" (80%) | ~10% survive 5 years | 15-25% (if you have strong evidence) |
| Will this hire work out? | "Great interview!" (90%) | ~50% of hires succeed long-term | 60-70% |
| Will this project be on time? | "Definitely" (95%) | ~70% of projects are late | 50-60% |
| Will this relationship last? | "Forever!" (99%) | ~40-50% of marriages end in divorce | 60-70% |
The key: Your specific evidence should SHIFT the base rate, not REPLACE it. A great interview doesn't make a hire 90% likely to succeed — it moves the base rate from 50% to maybe 65%.
Named after: Thomas Bayes (18th-century statistician)
Core idea: Update your beliefs proportionally to the strength of new evidence.
Simplified process:
1. Start with a PRIOR probability (your belief before new evidence)
2. Observe new EVIDENCE
3. Ask: "How likely is this evidence if my belief is TRUE?"
4. Ask: "How likely is this evidence if my belief is FALSE?"
5. Calculate the POSTERIOR probability (updated belief)
Intuitive example:
Prior: "There's a 20% chance it will rain today" (based on season)
New evidence: The sky is dark and cloudy.
- How likely are dark clouds if it WILL rain? Very likely (90%)
- How likely are dark clouds if it WON'T rain? Somewhat likely (30%)
Updated belief: ~43% chance of rain
More evidence: The weather app says 80% rain probability.
Updated belief: ~75% chance of rain
Practical rules:
- Strong evidence (very likely if true, very unlikely if false) → big update
- Weak evidence (similarly likely whether true or false) → small update
- Extraordinary claims require extraordinary evidence — a low prior needs very strong evidence to shift it significantly
Formula:
Expected Value = Σ (Probability × Value) for all outcomes
Decision rule: Choose the option with the highest expected value (adjusted for risk tolerance).
Template:
## Expected Value: [Decision]
### Option A:
| Outcome | Probability | Value | P × V |
|---------|:-:|:-:|:-:|
| Great outcome | __% | +$____ | $____ |
| OK outcome | __% | +$____ | $____ |
| Bad outcome | __% | -$____ | -$____ |
| **Expected Value** | 100% | | **$____** |
### Option B:
| Outcome | Probability | Value | P × V |
|---------|:-:|:-:|:-:|
| Great outcome | __% | +$____ | $____ |
| OK outcome | __% | +$____ | $____ |
| Bad outcome | __% | -$____ | -$____ |
| **Expected Value** | 100% | | **$____** |
### Choose: Option with higher EV = ____
### Risk check: Can I survive the worst case?
- Option A worst case: ____ → Survivable? Y/N
- Option B worst case: ____ → Survivable? Y/NImportant: Expected value maximization works for repeated decisions. For one-time, high-stakes decisions, also consider:
- Can you survive the worst case? (Ruin probability)
- How asymmetric are the outcomes? (Upside vs. downside)
- How reversible is the decision?
Definition: Your confidence levels match reality. When you say "90% sure," you're right ~90% of the time.
Most people are overconfident: When they say 90% confident, they're right only 50-70% of the time.
Calibration exercise:
| Statement | Your Confidence (%) | Actually True? |
|---|---|---|
| "The Earth is closer to the Sun than Mars" | __% | |
| "Brazil has more people than Russia" | __% | |
| "The Eiffel Tower is taller than 300 meters" | __% | |
| [Add your own predictions] | __% |
Track over time:
## Calibration Log
| Month | Predictions at 90%+ | Actually correct | Calibration |
|-------|:-:|:-:|:-:|
| Jan | 10 | 7 | 70% (overconfident) |
| Feb | 12 | 9 | 75% (still overconfident) |
| Mar | 8 | 7 | 87.5% (improving) |Tips for better calibration:
- Widen your confidence intervals (most people's are too narrow)
- Track predictions in writing
- Seek feedback on past predictions
- Practice with calibration games and exercises
Before any initiative, assign probabilities to outcomes:
## Pre-Mortem Probability: [Project/Decision]
| Outcome | Probability | If This Happens, We Will... |
|---------|:-:|---|
| Wild success | __% | |
| Moderate success | __% | |
| Break even | __% | |
| Partial failure | __% | |
| Complete failure | __% | |
| **Total** | **100%** | |
### Are we comfortable with this distribution?
### What could we do to shift probability toward success?
### What's our plan for the failure scenarios?## Probabilistic Business Case: [Initiative]
### Revenue Scenarios
| Scenario | Probability | Annual Revenue | EV |
|----------|:-:|:-:|:-:|
| Bull case | __% | $____ | $____ |
| Base case | __% | $____ | $____ |
| Bear case | __% | $____ | $____ |
| **Expected Revenue** | 100% | | **$____** |
### Cost: $____
### Expected Profit: $____ - $____ = $____
### Decision criteria:
- [ ] Expected profit is positive
- [ ] We can survive the bear case
- [ ] The bull case justifies the effortMargin of safety = buying below expected value to account for being wrong.
## Investment Analysis: [Asset]
### Scenario Valuation
| Scenario | Probability | Fair Value | Weighted |
|----------|:-:|:-:|:-:|
| Bull | __% | $____ | $____ |
| Base | __% | $____ | $____ |
| Bear | __% | $____ | $____ |
| **Expected Fair Value** | 100% | | **$____** |
### Current price: $____
### Expected return: (____ - ____) / ____ = ____%
### Margin of safety: ____%
### Kelly Criterion (position sizing):
Kelly % = (bp - q) / b
Where b = odds, p = win probability, q = loss probability
Suggested position size: ___% of portfolio
(Most practitioners use half-Kelly for safety)Even daily decisions benefit from probabilistic thinking:
| Decision | Deterministic Thinking | Probabilistic Thinking |
|---|---|---|
| Bring umbrella? | "Will it rain? Yes/No" | "30% chance of rain. Umbrella costs 0 effort. Bring it." |
| Take the highway? | "Highway is faster" | "80% chance highway saves 10 min. 20% chance accident adds 30 min. EV: +2 min." |
| Accept job offer? | "It's a good company" | "70% chance this is better than current job. 20% equivalent. 10% worse." |
| Mistake | What It Looks Like | Fix |
|---|---|---|
| Neglecting base rates | "My startup is different!" | Start with the base rate, then adjust |
| Conjunction fallacy | "She's a feminist bank teller" seems more likely than "she's a bank teller" | A + B can never be more likely than A alone |
| Gambler's fallacy | "It's been red 5 times, black is due" | Independent events have no memory |
| Ignoring sample size | Drawing conclusions from small samples | Larger samples = more reliable |
| Survivorship bias | "All successful CEOs did X" | What about the failures who also did X? |
| Overconfidence | 90% confidence intervals that are right 50% of the time | Track and calibrate |
| Anchoring | First number heard distorts estimate | Generate your own estimate first |
| Binary thinking | "It will work" vs. "It won't work" | Express as probability: "65% chance" |
| Availability bias | Vivid events seem more probable | Look up actual frequencies |
Make 10 predictions this week with explicit confidence levels (e.g., "80% confident the meeting will run over time"). Track results. Adjust.
Estimate something you don't know, breaking it into components:
- How many dentists in your city?
- How many flights take off worldwide each day?
- How much does your neighbor spend on groceries per year?
Pick a belief. State your current confidence (e.g., "70% confident AI will replace most white-collar jobs within 20 years"). Read one article that disagrees. What's your updated confidence?
For your next three decisions this week, quickly calculate expected value. Does it change what you choose?
For your current project, assign probabilities to five outcomes (wild success through complete failure). Do the numbers add to 100%? Are you comfortable with them?
Books:
- Superforecasting — Philip Tetlock
- Thinking, Fast and Slow — Daniel Kahneman
- The Signal and the Noise — Nate Silver
- How to Measure Anything — Douglas Hubbard
- Thinking in Bets — Annie Duke
- The Drunkard's Walk — Leonard Mlodinow
For decision principles that embrace uncertainty and probabilistic reasoning, explore KeepRule — a curated library of mental models from the world's best thinkers, organized for practical application.
Have a probabilistic thinking technique or exercise? PRs welcome.
MIT License — see LICENSE for details.