AI summary by TwinMind
Bayes' Theorem, Clearly Explained!!!!
StatQuest with Josh Starmer
13m 59s
Conditional Probability Review
- StatLand population survey on candy/soda preferences
- 2 people: loved both candy and soda
- 4 people: only loved candy
- 5 people: only loved soda
- 3 people: loved neither
- Total population: 14 people
- Calculated probabilities by dividing counts by total (14)
- Determined marginal probabilities for soda lovers vs non-soda lovers
- Determined marginal probabilities for candy lovers vs non-candy lovers
Conditional Probability Calculations
- P(not love candy but love soda | love soda) = 5/7 = 0.71
- Numerator: 5 people who don't love candy but love soda
- Denominator: 7 people who love soda
- Alternative calculation: divided numerator and denominator by total population (14)
- Result unchanged (0.71)
- Numerator becomes unconditional probability
- Denominator becomes marginal probability of loving soda
Notation and Interpretation
- Conditional probability = probability of event scaled by existing knowledge
- Standard notation omits redundant information
- "Someone who does not love candy given they love soda" vs
- "Someone who does not love candy and loves soda given they love soda"
- Longer notation helps understand relationship between calculation components
Reversing the Conditional
- Changed given condition from "loves soda" to "does not love candy"
- P(not love candy but love soda | not love candy) = 5/8 = 0.63
- Same numerator (5), different denominator (8 vs 7)
- Different scaling leads to different probability results
Deriving Bayes Theorem
- StatsQuatch challenge: solve conditional probabilities without knowing joint probability
- Algebraic manipulation of conditional probability equations
- Multiply both sides by marginal probabilities
- Set equations equal (both equal joint probability)
- Solve for each conditional probability term
- Result: P(A|B) = P(B|A) × P(A) / P(B)
- Double derivation showing relationship works both ways
General Formula
- Let A = "does not love candy", B = "loves soda"
- Standard Bayes theorem formula derived from specific example
- Conditional probability given one condition can be derived from conditional probability given other condition
Practical Application Without Complete Data
- Real-world scenario: incomplete information
- P(not love candy | love soda) = 0.71 (known)
- P(love soda) ≈ 0.6 (estimated/guess)
- P(not love candy) = 0.57 (known)
- Applied Bayes theorem: P(love soda | not love candy) ≈ 0.75
- Result differs from complete data calculation (0.63) due to estimation error
Importance and Philosophy
- Bayes theorem essential when complete data unavailable
- Example: surveying entire population of India impractical
- Bayesian statistics philosophy: understanding implications of making educated guesses
- Bayes theorem foundation for broader Bayesian statistical approach
Standard Notation Clarification
- Conventional notation omits redundant information
- P(love soda | not love candy) instead of full description
- P(not love candy | love soda) instead of full description
- Both conditional probabilities refer to same event (yellow area/square)
- Only difference: given knowledge/condition
- Longer notation preferred for clarity about referring to same underlying event
